2007-06-12 21:07:21 +08:00
|
|
|
/*
|
2008-09-30 03:18:18 +08:00
|
|
|
* Copyright (C) 2007,2008 Oracle. All rights reserved.
|
2007-06-12 21:07:21 +08:00
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public
|
|
|
|
* License v2 as published by the Free Software Foundation.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public
|
|
|
|
* License along with this program; if not, write to the
|
|
|
|
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
|
|
|
|
* Boston, MA 021110-1307, USA.
|
|
|
|
*/
|
|
|
|
|
2007-10-16 04:22:39 +08:00
|
|
|
#include <linux/sched.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
|
|
|
#include <linux/slab.h>
|
2012-05-16 23:18:50 +08:00
|
|
|
#include <linux/rbtree.h>
|
2016-03-30 22:01:12 +08:00
|
|
|
#include <linux/vmalloc.h>
|
2007-02-02 22:18:22 +08:00
|
|
|
#include "ctree.h"
|
|
|
|
#include "disk-io.h"
|
2007-03-24 03:56:19 +08:00
|
|
|
#include "transaction.h"
|
2007-10-16 04:14:19 +08:00
|
|
|
#include "print-tree.h"
|
2008-06-26 04:01:30 +08:00
|
|
|
#include "locking.h"
|
2007-02-23 21:38:36 +08:00
|
|
|
|
2007-03-17 04:20:31 +08:00
|
|
|
static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
|
|
|
|
*root, struct btrfs_path *path, int level);
|
2017-01-18 15:24:37 +08:00
|
|
|
static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
|
|
|
|
const struct btrfs_key *ins_key, struct btrfs_path *path,
|
|
|
|
int data_size, int extend);
|
2007-10-16 04:14:19 +08:00
|
|
|
static int push_node_left(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *dst,
|
2008-04-24 22:54:32 +08:00
|
|
|
struct extent_buffer *src, int empty);
|
2007-10-16 04:14:19 +08:00
|
|
|
static int balance_node_right(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *dst_buf,
|
|
|
|
struct extent_buffer *src_buf);
|
2013-04-16 13:18:22 +08:00
|
|
|
static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
|
|
|
|
int level, int slot);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
|
2012-05-26 17:43:17 +08:00
|
|
|
struct extent_buffer *eb);
|
2007-02-21 05:40:44 +08:00
|
|
|
|
2007-04-04 21:36:31 +08:00
|
|
|
struct btrfs_path *btrfs_alloc_path(void)
|
2007-04-02 22:50:19 +08:00
|
|
|
{
|
2016-09-13 03:35:52 +08:00
|
|
|
return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
|
2007-04-02 22:50:19 +08:00
|
|
|
}
|
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
/*
|
|
|
|
* set all locked nodes in the path to blocking locks. This should
|
|
|
|
* be done before scheduling
|
|
|
|
*/
|
|
|
|
noinline void btrfs_set_path_blocking(struct btrfs_path *p)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
|
2011-07-17 03:23:14 +08:00
|
|
|
if (!p->nodes[i] || !p->locks[i])
|
|
|
|
continue;
|
|
|
|
btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
|
|
|
|
if (p->locks[i] == BTRFS_READ_LOCK)
|
|
|
|
p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
|
|
|
|
else if (p->locks[i] == BTRFS_WRITE_LOCK)
|
|
|
|
p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* reset all the locked nodes in the patch to spinning locks.
|
2009-02-13 03:09:45 +08:00
|
|
|
*
|
|
|
|
* held is used to keep lockdep happy, when lockdep is enabled
|
|
|
|
* we set held to a blocking lock before we go around and
|
|
|
|
* retake all the spinlocks in the path. You can safely use NULL
|
|
|
|
* for held
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
*/
|
2009-02-13 03:09:45 +08:00
|
|
|
noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
|
2011-07-17 03:23:14 +08:00
|
|
|
struct extent_buffer *held, int held_rw)
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
{
|
|
|
|
int i;
|
2009-02-13 03:09:45 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
if (held) {
|
|
|
|
btrfs_set_lock_blocking_rw(held, held_rw);
|
|
|
|
if (held_rw == BTRFS_WRITE_LOCK)
|
|
|
|
held_rw = BTRFS_WRITE_LOCK_BLOCKING;
|
|
|
|
else if (held_rw == BTRFS_READ_LOCK)
|
|
|
|
held_rw = BTRFS_READ_LOCK_BLOCKING;
|
|
|
|
}
|
2009-02-13 03:09:45 +08:00
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
|
|
|
|
for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
|
2011-07-17 03:23:14 +08:00
|
|
|
if (p->nodes[i] && p->locks[i]) {
|
|
|
|
btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
|
|
|
|
if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
|
|
|
|
p->locks[i] = BTRFS_WRITE_LOCK;
|
|
|
|
else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
|
|
|
|
p->locks[i] = BTRFS_READ_LOCK;
|
|
|
|
}
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
2009-02-13 03:09:45 +08:00
|
|
|
|
|
|
|
if (held)
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_lock_blocking_rw(held, held_rw);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/* this also releases the path */
|
2007-04-04 21:36:31 +08:00
|
|
|
void btrfs_free_path(struct btrfs_path *p)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2010-12-26 05:22:30 +08:00
|
|
|
if (!p)
|
|
|
|
return;
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(p);
|
2007-04-04 21:36:31 +08:00
|
|
|
kmem_cache_free(btrfs_path_cachep, p);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* path release drops references on the extent buffers in the path
|
|
|
|
* and it drops any locks held by this path
|
|
|
|
*
|
|
|
|
* It is safe to call this on paths that no locks or extent buffers held.
|
|
|
|
*/
|
2011-04-21 07:20:15 +08:00
|
|
|
noinline void btrfs_release_path(struct btrfs_path *p)
|
2007-02-02 22:18:22 +08:00
|
|
|
{
|
|
|
|
int i;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2007-03-13 22:46:10 +08:00
|
|
|
for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
|
2008-06-26 04:01:31 +08:00
|
|
|
p->slots[i] = 0;
|
2007-02-02 22:18:22 +08:00
|
|
|
if (!p->nodes[i])
|
2008-06-26 04:01:30 +08:00
|
|
|
continue;
|
|
|
|
if (p->locks[i]) {
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
|
2008-06-26 04:01:30 +08:00
|
|
|
p->locks[i] = 0;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(p->nodes[i]);
|
2008-06-26 04:01:31 +08:00
|
|
|
p->nodes[i] = NULL;
|
2007-02-02 22:18:22 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* safely gets a reference on the root node of a tree. A lock
|
|
|
|
* is not taken, so a concurrent writer may put a different node
|
|
|
|
* at the root of the tree. See btrfs_lock_root_node for the
|
|
|
|
* looping required.
|
|
|
|
*
|
|
|
|
* The extent buffer returned by this has a reference taken, so
|
|
|
|
* it won't disappear. It may stop being the root of the tree
|
|
|
|
* at any time because there are no locks held.
|
|
|
|
*/
|
2008-06-26 04:01:30 +08:00
|
|
|
struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct extent_buffer *eb;
|
2011-03-24 02:54:42 +08:00
|
|
|
|
2012-03-10 05:01:49 +08:00
|
|
|
while (1) {
|
|
|
|
rcu_read_lock();
|
|
|
|
eb = rcu_dereference(root->node);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* RCU really hurts here, we could free up the root node because
|
2016-05-20 09:18:45 +08:00
|
|
|
* it was COWed but we may not get the new root node yet so do
|
2012-03-10 05:01:49 +08:00
|
|
|
* the inc_not_zero dance and if it doesn't work then
|
|
|
|
* synchronize_rcu and try again.
|
|
|
|
*/
|
|
|
|
if (atomic_inc_not_zero(&eb->refs)) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
synchronize_rcu();
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/* loop around taking references on and locking the root node of the
|
|
|
|
* tree until you end up with a lock on the root. A locked buffer
|
|
|
|
* is returned, with a reference held.
|
|
|
|
*/
|
2008-06-26 04:01:30 +08:00
|
|
|
struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2008-06-26 04:01:30 +08:00
|
|
|
eb = btrfs_root_node(root);
|
|
|
|
btrfs_tree_lock(eb);
|
2011-03-24 02:54:42 +08:00
|
|
|
if (eb == root->node)
|
2008-06-26 04:01:30 +08:00
|
|
|
break;
|
|
|
|
btrfs_tree_unlock(eb);
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
}
|
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
/* loop around taking references on and locking the root node of the
|
|
|
|
* tree until you end up with a lock on the root. A locked buffer
|
|
|
|
* is returned, with a reference held.
|
|
|
|
*/
|
2013-04-26 04:41:01 +08:00
|
|
|
static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
|
2011-07-17 03:23:14 +08:00
|
|
|
{
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
eb = btrfs_root_node(root);
|
|
|
|
btrfs_tree_read_lock(eb);
|
|
|
|
if (eb == root->node)
|
|
|
|
break;
|
|
|
|
btrfs_tree_read_unlock(eb);
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
}
|
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/* cowonly root (everything not a reference counted cow subvolume), just get
|
|
|
|
* put onto a simple dirty list. transaction.c walks this to make sure they
|
|
|
|
* get properly updated on disk.
|
|
|
|
*/
|
2008-03-25 03:01:56 +08:00
|
|
|
static void add_root_to_dirty_list(struct btrfs_root *root)
|
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
|
|
|
|
2014-12-17 00:54:43 +08:00
|
|
|
if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
|
|
|
|
!test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
|
|
|
|
return;
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
spin_lock(&fs_info->trans_lock);
|
2014-12-17 00:54:43 +08:00
|
|
|
if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
|
|
|
|
/* Want the extent tree to be the last on the list */
|
|
|
|
if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID)
|
|
|
|
list_move_tail(&root->dirty_list,
|
2016-06-23 06:54:23 +08:00
|
|
|
&fs_info->dirty_cowonly_roots);
|
2014-12-17 00:54:43 +08:00
|
|
|
else
|
|
|
|
list_move(&root->dirty_list,
|
2016-06-23 06:54:23 +08:00
|
|
|
&fs_info->dirty_cowonly_roots);
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
2016-06-23 06:54:23 +08:00
|
|
|
spin_unlock(&fs_info->trans_lock);
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* used by snapshot creation to make a copy of a root for a tree with
|
|
|
|
* a given objectid. The buffer with the new root node is returned in
|
|
|
|
* cow_ret, and this func returns zero on success or a negative error code.
|
|
|
|
*/
|
2007-12-18 09:14:01 +08:00
|
|
|
int btrfs_copy_root(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct extent_buffer *buf,
|
|
|
|
struct extent_buffer **cow_ret, u64 new_root_objectid)
|
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-12-18 09:14:01 +08:00
|
|
|
struct extent_buffer *cow;
|
|
|
|
int ret = 0;
|
|
|
|
int level;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-12-18 09:14:01 +08:00
|
|
|
|
2014-04-02 19:51:05 +08:00
|
|
|
WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
2016-06-23 06:54:23 +08:00
|
|
|
trans->transid != fs_info->running_transaction->transid);
|
2014-04-02 19:51:05 +08:00
|
|
|
WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
|
|
|
trans->transid != root->last_trans);
|
2007-12-18 09:14:01 +08:00
|
|
|
|
|
|
|
level = btrfs_header_level(buf);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (level == 0)
|
|
|
|
btrfs_item_key(buf, &disk_key, 0);
|
|
|
|
else
|
|
|
|
btrfs_node_key(buf, &disk_key, 0);
|
2008-09-24 01:14:14 +08:00
|
|
|
|
2014-06-15 07:54:12 +08:00
|
|
|
cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
|
|
|
|
&disk_key, level, buf->start, 0);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (IS_ERR(cow))
|
2007-12-18 09:14:01 +08:00
|
|
|
return PTR_ERR(cow);
|
|
|
|
|
2016-11-09 01:30:31 +08:00
|
|
|
copy_extent_buffer_full(cow, buf);
|
2007-12-18 09:14:01 +08:00
|
|
|
btrfs_set_header_bytenr(cow, cow->start);
|
|
|
|
btrfs_set_header_generation(cow, trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
|
|
|
|
btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
|
|
|
|
BTRFS_HEADER_FLAG_RELOC);
|
|
|
|
if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
|
|
|
|
btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
|
|
|
|
else
|
|
|
|
btrfs_set_header_owner(cow, new_root_objectid);
|
2007-12-18 09:14:01 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
write_extent_buffer_fsid(cow, fs_info->fsid);
|
2008-11-18 10:11:30 +08:00
|
|
|
|
2007-12-18 09:14:01 +08:00
|
|
|
WARN_ON(btrfs_header_generation(buf) > trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 1);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
else
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 0);
|
2007-12-19 05:25:45 +08:00
|
|
|
|
2007-12-18 09:14:01 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
btrfs_mark_buffer_dirty(cow);
|
|
|
|
*cow_ret = cow;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-05-16 23:18:50 +08:00
|
|
|
enum mod_log_op {
|
|
|
|
MOD_LOG_KEY_REPLACE,
|
|
|
|
MOD_LOG_KEY_ADD,
|
|
|
|
MOD_LOG_KEY_REMOVE,
|
|
|
|
MOD_LOG_KEY_REMOVE_WHILE_FREEING,
|
|
|
|
MOD_LOG_KEY_REMOVE_WHILE_MOVING,
|
|
|
|
MOD_LOG_MOVE_KEYS,
|
|
|
|
MOD_LOG_ROOT_REPLACE,
|
|
|
|
};
|
|
|
|
|
|
|
|
struct tree_mod_move {
|
|
|
|
int dst_slot;
|
|
|
|
int nr_items;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct tree_mod_root {
|
|
|
|
u64 logical;
|
|
|
|
u8 level;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct tree_mod_elem {
|
|
|
|
struct rb_node node;
|
2016-01-21 18:25:59 +08:00
|
|
|
u64 logical;
|
2012-06-21 17:08:04 +08:00
|
|
|
u64 seq;
|
2012-05-16 23:18:50 +08:00
|
|
|
enum mod_log_op op;
|
|
|
|
|
|
|
|
/* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
|
|
|
|
int slot;
|
|
|
|
|
|
|
|
/* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
|
|
|
|
u64 generation;
|
|
|
|
|
|
|
|
/* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
|
|
|
|
struct btrfs_disk_key key;
|
|
|
|
u64 blockptr;
|
|
|
|
|
|
|
|
/* this is used for op == MOD_LOG_MOVE_KEYS */
|
|
|
|
struct tree_mod_move move;
|
|
|
|
|
|
|
|
/* this is used for op == MOD_LOG_ROOT_REPLACE */
|
|
|
|
struct tree_mod_root old_root;
|
|
|
|
};
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
2012-06-21 17:08:04 +08:00
|
|
|
read_lock(&fs_info->tree_mod_log_lock);
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
|
|
|
|
{
|
|
|
|
read_unlock(&fs_info->tree_mod_log_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
|
|
|
|
{
|
|
|
|
write_lock(&fs_info->tree_mod_log_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
|
|
|
|
{
|
|
|
|
write_unlock(&fs_info->tree_mod_log_lock);
|
|
|
|
}
|
|
|
|
|
2013-04-25 00:57:33 +08:00
|
|
|
/*
|
2014-05-14 08:30:47 +08:00
|
|
|
* Pull a new tree mod seq number for our operation.
|
2013-04-25 00:57:33 +08:00
|
|
|
*/
|
2014-05-14 08:30:47 +08:00
|
|
|
static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
|
2013-04-25 00:57:33 +08:00
|
|
|
{
|
|
|
|
return atomic64_inc_return(&fs_info->tree_mod_seq);
|
|
|
|
}
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
/*
|
|
|
|
* This adds a new blocker to the tree mod log's blocker list if the @elem
|
|
|
|
* passed does not already have a sequence number set. So when a caller expects
|
|
|
|
* to record tree modifications, it should ensure to set elem->seq to zero
|
|
|
|
* before calling btrfs_get_tree_mod_seq.
|
|
|
|
* Returns a fresh, unused tree log modification sequence number, even if no new
|
|
|
|
* blocker was added.
|
|
|
|
*/
|
|
|
|
u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
|
|
|
|
struct seq_list *elem)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_write_lock(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
spin_lock(&fs_info->tree_mod_seq_lock);
|
2012-06-21 17:08:04 +08:00
|
|
|
if (!elem->seq) {
|
2014-05-14 08:30:47 +08:00
|
|
|
elem->seq = btrfs_inc_tree_mod_seq(fs_info);
|
2012-06-21 17:08:04 +08:00
|
|
|
list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
|
|
|
|
}
|
2012-05-16 23:18:50 +08:00
|
|
|
spin_unlock(&fs_info->tree_mod_seq_lock);
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
|
2014-05-14 08:30:47 +08:00
|
|
|
return elem->seq;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
|
|
|
|
struct seq_list *elem)
|
|
|
|
{
|
|
|
|
struct rb_root *tm_root;
|
|
|
|
struct rb_node *node;
|
|
|
|
struct rb_node *next;
|
|
|
|
struct seq_list *cur_elem;
|
|
|
|
struct tree_mod_elem *tm;
|
|
|
|
u64 min_seq = (u64)-1;
|
|
|
|
u64 seq_putting = elem->seq;
|
|
|
|
|
|
|
|
if (!seq_putting)
|
|
|
|
return;
|
|
|
|
|
|
|
|
spin_lock(&fs_info->tree_mod_seq_lock);
|
|
|
|
list_del(&elem->list);
|
2012-06-21 17:08:04 +08:00
|
|
|
elem->seq = 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
|
|
|
list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
|
2012-06-21 17:08:04 +08:00
|
|
|
if (cur_elem->seq < min_seq) {
|
2012-05-16 23:18:50 +08:00
|
|
|
if (seq_putting > cur_elem->seq) {
|
|
|
|
/*
|
|
|
|
* blocker with lower sequence number exists, we
|
|
|
|
* cannot remove anything from the log
|
|
|
|
*/
|
2012-06-21 17:08:04 +08:00
|
|
|
spin_unlock(&fs_info->tree_mod_seq_lock);
|
|
|
|
return;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
min_seq = cur_elem->seq;
|
|
|
|
}
|
|
|
|
}
|
2012-06-21 17:08:04 +08:00
|
|
|
spin_unlock(&fs_info->tree_mod_seq_lock);
|
|
|
|
|
2012-05-16 23:18:50 +08:00
|
|
|
/*
|
|
|
|
* anything that's lower than the lowest existing (read: blocked)
|
|
|
|
* sequence number can be removed from the tree.
|
|
|
|
*/
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_write_lock(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
tm_root = &fs_info->tree_mod_log;
|
|
|
|
for (node = rb_first(tm_root); node; node = next) {
|
|
|
|
next = rb_next(node);
|
2016-12-19 22:53:41 +08:00
|
|
|
tm = rb_entry(node, struct tree_mod_elem, node);
|
2012-06-21 17:08:04 +08:00
|
|
|
if (tm->seq > min_seq)
|
2012-05-16 23:18:50 +08:00
|
|
|
continue;
|
|
|
|
rb_erase(node, tm_root);
|
|
|
|
kfree(tm);
|
|
|
|
}
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_write_unlock(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* key order of the log:
|
2016-01-21 18:25:59 +08:00
|
|
|
* node/leaf start address -> sequence
|
2012-05-16 23:18:50 +08:00
|
|
|
*
|
2016-01-21 18:25:59 +08:00
|
|
|
* The 'start address' is the logical address of the *new* root node
|
|
|
|
* for root replace operations, or the logical address of the affected
|
|
|
|
* block for all other operations.
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
*
|
|
|
|
* Note: must be called with write lock (tree_mod_log_write_lock).
|
2012-05-16 23:18:50 +08:00
|
|
|
*/
|
|
|
|
static noinline int
|
|
|
|
__tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
|
|
|
|
{
|
|
|
|
struct rb_root *tm_root;
|
|
|
|
struct rb_node **new;
|
|
|
|
struct rb_node *parent = NULL;
|
|
|
|
struct tree_mod_elem *cur;
|
2013-07-02 04:18:19 +08:00
|
|
|
|
|
|
|
BUG_ON(!tm);
|
|
|
|
|
2014-05-14 08:30:47 +08:00
|
|
|
tm->seq = btrfs_inc_tree_mod_seq(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
|
|
|
|
tm_root = &fs_info->tree_mod_log;
|
|
|
|
new = &tm_root->rb_node;
|
|
|
|
while (*new) {
|
2016-12-19 22:53:41 +08:00
|
|
|
cur = rb_entry(*new, struct tree_mod_elem, node);
|
2012-05-16 23:18:50 +08:00
|
|
|
parent = *new;
|
2016-01-21 18:25:59 +08:00
|
|
|
if (cur->logical < tm->logical)
|
2012-05-16 23:18:50 +08:00
|
|
|
new = &((*new)->rb_left);
|
2016-01-21 18:25:59 +08:00
|
|
|
else if (cur->logical > tm->logical)
|
2012-05-16 23:18:50 +08:00
|
|
|
new = &((*new)->rb_right);
|
2012-06-21 17:08:04 +08:00
|
|
|
else if (cur->seq < tm->seq)
|
2012-05-16 23:18:50 +08:00
|
|
|
new = &((*new)->rb_left);
|
2012-06-21 17:08:04 +08:00
|
|
|
else if (cur->seq > tm->seq)
|
2012-05-16 23:18:50 +08:00
|
|
|
new = &((*new)->rb_right);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
else
|
|
|
|
return -EEXIST;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
rb_link_node(&tm->node, parent, new);
|
|
|
|
rb_insert_color(&tm->node, tm_root);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
/*
|
|
|
|
* Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
|
|
|
|
* returns zero with the tree_mod_log_lock acquired. The caller must hold
|
|
|
|
* this until all tree mod log insertions are recorded in the rb tree and then
|
|
|
|
* call tree_mod_log_write_unlock() to release.
|
|
|
|
*/
|
2012-05-31 20:59:09 +08:00
|
|
|
static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *eb) {
|
|
|
|
smp_mb();
|
|
|
|
if (list_empty(&(fs_info)->tree_mod_seq_list))
|
|
|
|
return 1;
|
2012-06-21 17:08:04 +08:00
|
|
|
if (eb && btrfs_header_level(eb) == 0)
|
|
|
|
return 1;
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
|
|
|
|
tree_mod_log_write_lock(fs_info);
|
|
|
|
if (list_empty(&(fs_info)->tree_mod_seq_list)) {
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2012-05-31 20:59:09 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
/* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
|
|
|
|
static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *eb)
|
|
|
|
{
|
|
|
|
smp_mb();
|
|
|
|
if (list_empty(&(fs_info)->tree_mod_seq_list))
|
|
|
|
return 0;
|
|
|
|
if (eb && btrfs_header_level(eb) == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct tree_mod_elem *
|
|
|
|
alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
|
|
|
|
enum mod_log_op op, gfp_t flags)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
2012-06-21 17:08:04 +08:00
|
|
|
struct tree_mod_elem *tm;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
2013-07-02 04:18:19 +08:00
|
|
|
tm = kzalloc(sizeof(*tm), flags);
|
|
|
|
if (!tm)
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return NULL;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
2016-01-21 18:25:59 +08:00
|
|
|
tm->logical = eb->start;
|
2012-05-16 23:18:50 +08:00
|
|
|
if (op != MOD_LOG_KEY_ADD) {
|
|
|
|
btrfs_node_key(eb, &tm->key, slot);
|
|
|
|
tm->blockptr = btrfs_node_blockptr(eb, slot);
|
|
|
|
}
|
|
|
|
tm->op = op;
|
|
|
|
tm->slot = slot;
|
|
|
|
tm->generation = btrfs_node_ptr_generation(eb, slot);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
RB_CLEAR_NODE(&tm->node);
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return tm;
|
2012-06-21 17:08:04 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static noinline int
|
2013-07-02 04:18:19 +08:00
|
|
|
tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *eb, int slot,
|
|
|
|
enum mod_log_op op, gfp_t flags)
|
2012-06-21 17:08:04 +08:00
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
struct tree_mod_elem *tm;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!tree_mod_need_log(fs_info, eb))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
tm = alloc_tree_mod_elem(eb, slot, op, flags);
|
|
|
|
if (!tm)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
if (tree_mod_dont_log(fs_info, eb)) {
|
|
|
|
kfree(tm);
|
2012-06-21 17:08:04 +08:00
|
|
|
return 0;
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
ret = __tree_mod_log_insert(fs_info, tm);
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
if (ret)
|
|
|
|
kfree(tm);
|
2012-06-21 17:08:04 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
2012-06-21 17:08:04 +08:00
|
|
|
}
|
|
|
|
|
2012-05-16 23:18:50 +08:00
|
|
|
static noinline int
|
|
|
|
tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *eb, int dst_slot, int src_slot,
|
|
|
|
int nr_items, gfp_t flags)
|
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
struct tree_mod_elem *tm = NULL;
|
|
|
|
struct tree_mod_elem **tm_list = NULL;
|
|
|
|
int ret = 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
int i;
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
int locked = 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tree_mod_need_log(fs_info, eb))
|
2012-05-31 21:02:32 +08:00
|
|
|
return 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
2015-02-21 01:00:26 +08:00
|
|
|
tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tm_list)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
tm = kzalloc(sizeof(*tm), flags);
|
|
|
|
if (!tm) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
|
2016-01-21 18:25:59 +08:00
|
|
|
tm->logical = eb->start;
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
tm->slot = src_slot;
|
|
|
|
tm->move.dst_slot = dst_slot;
|
|
|
|
tm->move.nr_items = nr_items;
|
|
|
|
tm->op = MOD_LOG_MOVE_KEYS;
|
|
|
|
|
|
|
|
for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
|
|
|
|
tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
|
|
|
|
MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
|
|
|
|
if (!tm_list[i]) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (tree_mod_dont_log(fs_info, eb))
|
|
|
|
goto free_tms;
|
|
|
|
locked = 1;
|
|
|
|
|
2012-10-23 21:02:12 +08:00
|
|
|
/*
|
|
|
|
* When we override something during the move, we log these removals.
|
|
|
|
* This can only happen when we move towards the beginning of the
|
|
|
|
* buffer, i.e. dst_slot < src_slot.
|
|
|
|
*/
|
2012-05-16 23:18:50 +08:00
|
|
|
for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
ret = __tree_mod_log_insert(fs_info, tm_list[i]);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
ret = __tree_mod_log_insert(fs_info, tm);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
kfree(tm_list);
|
2012-05-31 21:02:32 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return 0;
|
|
|
|
free_tms:
|
|
|
|
for (i = 0; i < nr_items; i++) {
|
|
|
|
if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
|
|
|
|
rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
|
|
|
|
kfree(tm_list[i]);
|
|
|
|
}
|
|
|
|
if (locked)
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
kfree(tm_list);
|
|
|
|
kfree(tm);
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
static inline int
|
|
|
|
__tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
|
|
|
|
struct tree_mod_elem **tm_list,
|
|
|
|
int nritems)
|
2012-06-21 17:08:04 +08:00
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
int i, j;
|
2012-06-21 17:08:04 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
for (i = nritems - 1; i >= 0; i--) {
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
ret = __tree_mod_log_insert(fs_info, tm_list[i]);
|
|
|
|
if (ret) {
|
|
|
|
for (j = nritems - 1; j > i; j--)
|
|
|
|
rb_erase(&tm_list[j]->node,
|
|
|
|
&fs_info->tree_mod_log);
|
|
|
|
return ret;
|
|
|
|
}
|
2012-06-21 17:08:04 +08:00
|
|
|
}
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
|
|
|
|
return 0;
|
2012-06-21 17:08:04 +08:00
|
|
|
}
|
|
|
|
|
2012-05-16 23:18:50 +08:00
|
|
|
static noinline int
|
|
|
|
tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *old_root,
|
2013-04-13 21:19:53 +08:00
|
|
|
struct extent_buffer *new_root, gfp_t flags,
|
|
|
|
int log_removal)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
struct tree_mod_elem *tm = NULL;
|
|
|
|
struct tree_mod_elem **tm_list = NULL;
|
|
|
|
int nritems = 0;
|
|
|
|
int ret = 0;
|
|
|
|
int i;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tree_mod_need_log(fs_info, NULL))
|
2012-06-21 17:08:04 +08:00
|
|
|
return 0;
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (log_removal && btrfs_header_level(old_root) > 0) {
|
|
|
|
nritems = btrfs_header_nritems(old_root);
|
2015-02-21 01:00:26 +08:00
|
|
|
tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
flags);
|
|
|
|
if (!tm_list) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
for (i = 0; i < nritems; i++) {
|
|
|
|
tm_list[i] = alloc_tree_mod_elem(old_root, i,
|
|
|
|
MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
|
|
|
|
if (!tm_list[i]) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2013-03-20 21:49:48 +08:00
|
|
|
|
2013-07-02 04:18:19 +08:00
|
|
|
tm = kzalloc(sizeof(*tm), flags);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tm) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
2012-05-16 23:18:50 +08:00
|
|
|
|
2016-01-21 18:25:59 +08:00
|
|
|
tm->logical = new_root->start;
|
2012-05-16 23:18:50 +08:00
|
|
|
tm->old_root.logical = old_root->start;
|
|
|
|
tm->old_root.level = btrfs_header_level(old_root);
|
|
|
|
tm->generation = btrfs_header_generation(old_root);
|
|
|
|
tm->op = MOD_LOG_ROOT_REPLACE;
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (tree_mod_dont_log(fs_info, NULL))
|
|
|
|
goto free_tms;
|
|
|
|
|
|
|
|
if (tm_list)
|
|
|
|
ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
|
|
|
|
if (!ret)
|
|
|
|
ret = __tree_mod_log_insert(fs_info, tm);
|
|
|
|
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
|
|
|
kfree(tm_list);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
free_tms:
|
|
|
|
if (tm_list) {
|
|
|
|
for (i = 0; i < nritems; i++)
|
|
|
|
kfree(tm_list[i]);
|
|
|
|
kfree(tm_list);
|
|
|
|
}
|
|
|
|
kfree(tm);
|
|
|
|
|
|
|
|
return ret;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct tree_mod_elem *
|
|
|
|
__tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
|
|
|
|
int smallest)
|
|
|
|
{
|
|
|
|
struct rb_root *tm_root;
|
|
|
|
struct rb_node *node;
|
|
|
|
struct tree_mod_elem *cur = NULL;
|
|
|
|
struct tree_mod_elem *found = NULL;
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_read_lock(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
tm_root = &fs_info->tree_mod_log;
|
|
|
|
node = tm_root->rb_node;
|
|
|
|
while (node) {
|
2016-12-19 22:53:41 +08:00
|
|
|
cur = rb_entry(node, struct tree_mod_elem, node);
|
2016-01-21 18:25:59 +08:00
|
|
|
if (cur->logical < start) {
|
2012-05-16 23:18:50 +08:00
|
|
|
node = node->rb_left;
|
2016-01-21 18:25:59 +08:00
|
|
|
} else if (cur->logical > start) {
|
2012-05-16 23:18:50 +08:00
|
|
|
node = node->rb_right;
|
2012-06-21 17:08:04 +08:00
|
|
|
} else if (cur->seq < min_seq) {
|
2012-05-16 23:18:50 +08:00
|
|
|
node = node->rb_left;
|
|
|
|
} else if (!smallest) {
|
|
|
|
/* we want the node with the highest seq */
|
|
|
|
if (found)
|
2012-06-21 17:08:04 +08:00
|
|
|
BUG_ON(found->seq > cur->seq);
|
2012-05-16 23:18:50 +08:00
|
|
|
found = cur;
|
|
|
|
node = node->rb_left;
|
2012-06-21 17:08:04 +08:00
|
|
|
} else if (cur->seq > min_seq) {
|
2012-05-16 23:18:50 +08:00
|
|
|
/* we want the node with the smallest seq */
|
|
|
|
if (found)
|
2012-06-21 17:08:04 +08:00
|
|
|
BUG_ON(found->seq < cur->seq);
|
2012-05-16 23:18:50 +08:00
|
|
|
found = cur;
|
|
|
|
node = node->rb_right;
|
|
|
|
} else {
|
|
|
|
found = cur;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_read_unlock(fs_info);
|
2012-05-16 23:18:50 +08:00
|
|
|
|
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* this returns the element from the log with the smallest time sequence
|
|
|
|
* value that's in the log (the oldest log item). any element with a time
|
|
|
|
* sequence lower than min_seq will be ignored.
|
|
|
|
*/
|
|
|
|
static struct tree_mod_elem *
|
|
|
|
tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
|
|
|
|
u64 min_seq)
|
|
|
|
{
|
|
|
|
return __tree_mod_log_search(fs_info, start, min_seq, 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* this returns the element from the log with the largest time sequence
|
|
|
|
* value that's in the log (the most recent log item). any element with
|
|
|
|
* a time sequence lower than min_seq will be ignored.
|
|
|
|
*/
|
|
|
|
static struct tree_mod_elem *
|
|
|
|
tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
|
|
|
|
{
|
|
|
|
return __tree_mod_log_search(fs_info, start, min_seq, 0);
|
|
|
|
}
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
static noinline int
|
2012-05-16 23:18:50 +08:00
|
|
|
tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
|
|
|
|
struct extent_buffer *src, unsigned long dst_offset,
|
2013-04-13 21:19:53 +08:00
|
|
|
unsigned long src_offset, int nr_items)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
int ret = 0;
|
|
|
|
struct tree_mod_elem **tm_list = NULL;
|
|
|
|
struct tree_mod_elem **tm_list_add, **tm_list_rem;
|
2012-05-16 23:18:50 +08:00
|
|
|
int i;
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
int locked = 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tree_mod_need_log(fs_info, NULL))
|
|
|
|
return 0;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
2013-07-02 04:18:19 +08:00
|
|
|
if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return 0;
|
|
|
|
|
2015-02-21 01:00:26 +08:00
|
|
|
tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
GFP_NOFS);
|
|
|
|
if (!tm_list)
|
|
|
|
return -ENOMEM;
|
2012-05-16 23:18:50 +08:00
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
tm_list_add = tm_list;
|
|
|
|
tm_list_rem = tm_list + nr_items;
|
2012-05-16 23:18:50 +08:00
|
|
|
for (i = 0; i < nr_items; i++) {
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
|
|
|
|
MOD_LOG_KEY_REMOVE, GFP_NOFS);
|
|
|
|
if (!tm_list_rem[i]) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
|
|
|
|
tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
|
|
|
|
MOD_LOG_KEY_ADD, GFP_NOFS);
|
|
|
|
if (!tm_list_add[i]) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (tree_mod_dont_log(fs_info, NULL))
|
|
|
|
goto free_tms;
|
|
|
|
locked = 1;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_items; i++) {
|
|
|
|
ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
|
|
|
ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
kfree(tm_list);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
free_tms:
|
|
|
|
for (i = 0; i < nr_items * 2; i++) {
|
|
|
|
if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
|
|
|
|
rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
|
|
|
|
kfree(tm_list[i]);
|
|
|
|
}
|
|
|
|
if (locked)
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
kfree(tm_list);
|
|
|
|
|
|
|
|
return ret;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void
|
|
|
|
tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
|
|
|
|
int dst_offset, int src_offset, int nr_items)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
|
|
|
|
nr_items, GFP_NOFS);
|
|
|
|
BUG_ON(ret < 0);
|
|
|
|
}
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
static noinline void
|
2012-05-16 23:18:50 +08:00
|
|
|
tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
|
2012-10-19 20:52:15 +08:00
|
|
|
struct extent_buffer *eb, int slot, int atomic)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-12-13 03:19:52 +08:00
|
|
|
ret = tree_mod_log_insert_key(fs_info, eb, slot,
|
2013-07-02 04:18:19 +08:00
|
|
|
MOD_LOG_KEY_REPLACE,
|
|
|
|
atomic ? GFP_ATOMIC : GFP_NOFS);
|
2012-05-16 23:18:50 +08:00
|
|
|
BUG_ON(ret < 0);
|
|
|
|
}
|
|
|
|
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
static noinline int
|
2012-06-21 17:08:04 +08:00
|
|
|
tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
struct tree_mod_elem **tm_list = NULL;
|
|
|
|
int nritems = 0;
|
|
|
|
int i;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (btrfs_header_level(eb) == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!tree_mod_need_log(fs_info, NULL))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nritems = btrfs_header_nritems(eb);
|
2015-02-21 01:00:26 +08:00
|
|
|
tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (!tm_list)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
for (i = 0; i < nritems; i++) {
|
|
|
|
tm_list[i] = alloc_tree_mod_elem(eb, i,
|
|
|
|
MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
|
|
|
|
if (!tm_list[i]) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto free_tms;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-05-31 20:59:09 +08:00
|
|
|
if (tree_mod_dont_log(fs_info, eb))
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
goto free_tms;
|
|
|
|
|
|
|
|
ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
|
|
|
|
tree_mod_log_write_unlock(fs_info);
|
|
|
|
if (ret)
|
|
|
|
goto free_tms;
|
|
|
|
kfree(tm_list);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
free_tms:
|
|
|
|
for (i = 0; i < nritems; i++)
|
|
|
|
kfree(tm_list[i]);
|
|
|
|
kfree(tm_list);
|
|
|
|
|
|
|
|
return ret;
|
2012-05-16 23:18:50 +08:00
|
|
|
}
|
|
|
|
|
2012-06-21 17:08:04 +08:00
|
|
|
static noinline void
|
2012-05-16 23:18:50 +08:00
|
|
|
tree_mod_log_set_root_pointer(struct btrfs_root *root,
|
2013-04-13 21:19:53 +08:00
|
|
|
struct extent_buffer *new_root_node,
|
|
|
|
int log_removal)
|
2012-05-16 23:18:50 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
ret = tree_mod_log_insert_root(root->fs_info, root->node,
|
2013-04-13 21:19:53 +08:00
|
|
|
new_root_node, GFP_NOFS, log_removal);
|
2012-05-16 23:18:50 +08:00
|
|
|
BUG_ON(ret < 0);
|
|
|
|
}
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
/*
|
|
|
|
* check if the tree block can be shared by multiple trees
|
|
|
|
*/
|
|
|
|
int btrfs_block_can_be_shared(struct btrfs_root *root,
|
|
|
|
struct extent_buffer *buf)
|
|
|
|
{
|
|
|
|
/*
|
2016-05-20 09:18:45 +08:00
|
|
|
* Tree blocks not in reference counted trees and tree roots
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
* are never shared. If a block was allocated after the last
|
|
|
|
* snapshot and the block was not allocated by tree relocation,
|
|
|
|
* we know the block is not shared.
|
|
|
|
*/
|
2014-04-02 19:51:05 +08:00
|
|
|
if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
buf != root->node && buf != root->commit_root &&
|
|
|
|
(btrfs_header_generation(buf) <=
|
|
|
|
btrfs_root_last_snapshot(&root->root_item) ||
|
|
|
|
btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
|
|
|
|
return 1;
|
|
|
|
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
|
2014-04-02 19:51:05 +08:00
|
|
|
if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
|
|
|
|
return 1;
|
|
|
|
#endif
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct extent_buffer *buf,
|
2010-05-16 22:46:25 +08:00
|
|
|
struct extent_buffer *cow,
|
|
|
|
int *last_ref)
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
u64 refs;
|
|
|
|
u64 owner;
|
|
|
|
u64 flags;
|
|
|
|
u64 new_flags = 0;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Backrefs update rules:
|
|
|
|
*
|
|
|
|
* Always use full backrefs for extent pointers in tree block
|
|
|
|
* allocated by tree relocation.
|
|
|
|
*
|
|
|
|
* If a shared tree block is no longer referenced by its owner
|
|
|
|
* tree (btrfs_header_owner(buf) == root->root_key.objectid),
|
|
|
|
* use full backrefs for extent pointers in tree block.
|
|
|
|
*
|
|
|
|
* If a tree block is been relocating
|
|
|
|
* (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
|
|
|
|
* use full backrefs for extent pointers in tree block.
|
|
|
|
* The reason for this is some operations (such as drop tree)
|
|
|
|
* are only allowed for blocks use full backrefs.
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (btrfs_block_can_be_shared(root, buf)) {
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
|
2013-03-08 03:22:04 +08:00
|
|
|
btrfs_header_level(buf), 1,
|
|
|
|
&refs, &flags);
|
2011-08-09 04:20:18 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2011-08-30 05:17:04 +08:00
|
|
|
if (refs == 0) {
|
|
|
|
ret = -EROFS;
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_handle_fs_error(fs_info, ret, NULL);
|
2011-08-30 05:17:04 +08:00
|
|
|
return ret;
|
|
|
|
}
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
} else {
|
|
|
|
refs = 1;
|
|
|
|
if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
|
|
|
|
btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
|
|
|
|
flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
|
|
|
|
else
|
|
|
|
flags = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
owner = btrfs_header_owner(buf);
|
|
|
|
BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
|
|
|
|
!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
|
|
|
|
|
|
|
|
if (refs > 1) {
|
|
|
|
if ((owner == root->root_key.objectid ||
|
|
|
|
root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
|
|
|
|
!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, buf, 1);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
|
|
|
if (root->root_key.objectid ==
|
|
|
|
BTRFS_TREE_RELOC_OBJECTID) {
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_dec_ref(trans, root, buf, 0);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 1);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
|
|
|
|
} else {
|
|
|
|
|
|
|
|
if (root->root_key.objectid ==
|
|
|
|
BTRFS_TREE_RELOC_OBJECTID)
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 1);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
else
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 0);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
if (new_flags != 0) {
|
2013-05-10 01:49:30 +08:00
|
|
|
int level = btrfs_header_level(buf);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = btrfs_set_disk_extent_flags(trans, fs_info,
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
buf->start,
|
|
|
|
buf->len,
|
2013-05-10 01:49:30 +08:00
|
|
|
new_flags, level, 0);
|
2011-08-09 04:20:18 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
|
|
|
|
if (root->root_key.objectid ==
|
|
|
|
BTRFS_TREE_RELOC_OBJECTID)
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 1);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
else
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_inc_ref(trans, root, cow, 0);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
2014-07-03 01:54:25 +08:00
|
|
|
ret = btrfs_dec_ref(trans, root, buf, 1);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* -ENOMEM */
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, buf);
|
2010-05-16 22:46:25 +08:00
|
|
|
*last_ref = 1;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
2009-01-06 10:25:51 +08:00
|
|
|
* does the dirty work in cow of a single block. The parent block (if
|
|
|
|
* supplied) is updated to point to the new cow copy. The new buffer is marked
|
|
|
|
* dirty and returned locked. If you modify the block it needs to be marked
|
|
|
|
* dirty again.
|
2008-09-30 03:18:18 +08:00
|
|
|
*
|
|
|
|
* search_start -- an allocation hint for the new block
|
|
|
|
*
|
2009-01-06 10:25:51 +08:00
|
|
|
* empty_size -- a hint that you plan on doing more cow. This is the size in
|
|
|
|
* bytes the allocator should try to find free next to the block it returns.
|
|
|
|
* This is just a hint and may be ignored by the allocator.
|
2008-09-30 03:18:18 +08:00
|
|
|
*/
|
2009-01-06 10:25:51 +08:00
|
|
|
static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_root *root,
|
|
|
|
struct extent_buffer *buf,
|
|
|
|
struct extent_buffer *parent, int parent_slot,
|
|
|
|
struct extent_buffer **cow_ret,
|
2009-03-13 22:24:59 +08:00
|
|
|
u64 search_start, u64 empty_size)
|
2007-03-03 05:08:05 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *cow;
|
2011-08-09 04:20:18 +08:00
|
|
|
int level, ret;
|
2010-05-16 22:46:25 +08:00
|
|
|
int last_ref = 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
int unlock_orig = 0;
|
2016-09-23 03:11:34 +08:00
|
|
|
u64 parent_start = 0;
|
2007-12-11 22:25:06 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (*cow_ret == buf)
|
|
|
|
unlock_orig = 1;
|
|
|
|
|
2009-03-09 23:45:38 +08:00
|
|
|
btrfs_assert_tree_locked(buf);
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2014-04-02 19:51:05 +08:00
|
|
|
WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
2016-06-23 06:54:23 +08:00
|
|
|
trans->transid != fs_info->running_transaction->transid);
|
2014-04-02 19:51:05 +08:00
|
|
|
WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
|
|
|
|
trans->transid != root->last_trans);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-12-11 22:25:06 +08:00
|
|
|
level = btrfs_header_level(buf);
|
2008-09-24 01:14:14 +08:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (level == 0)
|
|
|
|
btrfs_item_key(buf, &disk_key, 0);
|
|
|
|
else
|
|
|
|
btrfs_node_key(buf, &disk_key, 0);
|
|
|
|
|
2016-09-23 03:11:34 +08:00
|
|
|
if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
|
|
|
|
parent_start = parent->start;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
2014-06-15 07:54:12 +08:00
|
|
|
cow = btrfs_alloc_tree_block(trans, root, parent_start,
|
|
|
|
root->root_key.objectid, &disk_key, level,
|
|
|
|
search_start, empty_size);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (IS_ERR(cow))
|
|
|
|
return PTR_ERR(cow);
|
2007-08-08 04:15:09 +08:00
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
/* cow is set to blocking by btrfs_init_new_buffer */
|
|
|
|
|
2016-11-09 01:30:31 +08:00
|
|
|
copy_extent_buffer_full(cow, buf);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_header_bytenr(cow, cow->start);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_generation(cow, trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
|
|
|
|
btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
|
|
|
|
BTRFS_HEADER_FLAG_RELOC);
|
|
|
|
if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
|
|
|
|
btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
|
|
|
|
else
|
|
|
|
btrfs_set_header_owner(cow, root->root_key.objectid);
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
write_extent_buffer_fsid(cow, fs_info->fsid);
|
2008-11-18 10:11:30 +08:00
|
|
|
|
2011-08-09 04:20:18 +08:00
|
|
|
ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
|
2011-08-30 05:30:39 +08:00
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
2011-08-30 05:30:39 +08:00
|
|
|
return ret;
|
|
|
|
}
|
Btrfs: update space balancing code
This patch updates the space balancing code to utilize the new
backref format. Before, btrfs-vol -b would break any COW links
on data blocks or metadata. This was slow and caused the amount
of space used to explode if a large number of snapshots were present.
The new code can keeps the sharing of all data extents and
most of the tree blocks.
To maintain the sharing of data extents, the space balance code uses
a seperate inode hold data extent pointers, then updates the references
to point to the new location.
To maintain the sharing of tree blocks, the space balance code uses
reloc trees to relocate tree blocks in reference counted roots.
There is one reloc tree for each subvol, and all reloc trees share
same root key objectid. Reloc trees are snapshots of the latest
committed roots of subvols (root->commit_root).
To relocate a tree block referenced by a subvol, there are two steps.
COW the block through subvol's reloc tree, then update block pointer in
the subvol to point to the new block. Since all reloc trees share
same root key objectid, doing special handing for tree blocks
owned by them is easy. Once a tree block has been COWed in one
reloc tree, we can use the resulting new block directly when the
same block is required to COW again through other reloc trees.
In this way, relocated tree blocks are shared between reloc trees,
so they are also shared between subvols.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-26 22:09:34 +08:00
|
|
|
|
2014-04-02 19:51:05 +08:00
|
|
|
if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
|
2013-08-31 03:09:51 +08:00
|
|
|
ret = btrfs_reloc_cow_block(trans, root, buf, cow);
|
2015-08-06 21:56:58 +08:00
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
2013-08-31 03:09:51 +08:00
|
|
|
return ret;
|
2015-08-06 21:56:58 +08:00
|
|
|
}
|
2013-08-31 03:09:51 +08:00
|
|
|
}
|
2010-05-16 22:49:59 +08:00
|
|
|
|
2007-03-03 05:08:05 +08:00
|
|
|
if (buf == root->node) {
|
2008-06-26 04:01:30 +08:00
|
|
|
WARN_ON(parent && parent != buf);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
|
|
|
|
btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
|
|
|
|
parent_start = buf->start;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
extent_buffer_get(cow);
|
2013-04-13 21:19:53 +08:00
|
|
|
tree_mod_log_set_root_pointer(root, cow, 1);
|
2011-03-24 02:54:42 +08:00
|
|
|
rcu_assign_pointer(root->node, cow);
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2010-05-16 22:46:25 +08:00
|
|
|
btrfs_free_tree_block(trans, root, buf, parent_start,
|
2012-05-16 23:04:52 +08:00
|
|
|
last_ref);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(buf);
|
2008-03-25 03:01:56 +08:00
|
|
|
add_root_to_dirty_list(root);
|
2007-03-03 05:08:05 +08:00
|
|
|
} else {
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
WARN_ON(trans->transid != btrfs_header_generation(parent));
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_insert_key(fs_info, parent, parent_slot,
|
2013-07-02 04:18:19 +08:00
|
|
|
MOD_LOG_KEY_REPLACE, GFP_NOFS);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_blockptr(parent, parent_slot,
|
2007-10-16 04:15:53 +08:00
|
|
|
cow->start);
|
2007-12-11 22:25:06 +08:00
|
|
|
btrfs_set_node_ptr_generation(parent, parent_slot,
|
|
|
|
trans->transid);
|
2007-03-31 02:27:56 +08:00
|
|
|
btrfs_mark_buffer_dirty(parent);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (last_ref) {
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_free_eb(fs_info, buf);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
2010-05-16 22:46:25 +08:00
|
|
|
btrfs_free_tree_block(trans, root, buf, parent_start,
|
2012-05-16 23:04:52 +08:00
|
|
|
last_ref);
|
2007-03-03 05:08:05 +08:00
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
if (unlock_orig)
|
|
|
|
btrfs_tree_unlock(buf);
|
2012-03-10 05:01:49 +08:00
|
|
|
free_extent_buffer_stale(buf);
|
2007-06-29 03:57:36 +08:00
|
|
|
btrfs_mark_buffer_dirty(cow);
|
2007-04-02 22:50:19 +08:00
|
|
|
*cow_ret = cow;
|
2007-03-03 05:08:05 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-05-17 00:25:47 +08:00
|
|
|
/*
|
|
|
|
* returns the logical address of the oldest predecessor of the given root.
|
|
|
|
* entries older than time_seq are ignored.
|
|
|
|
*/
|
|
|
|
static struct tree_mod_elem *
|
|
|
|
__tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
|
2013-04-13 21:19:54 +08:00
|
|
|
struct extent_buffer *eb_root, u64 time_seq)
|
2012-05-17 00:25:47 +08:00
|
|
|
{
|
|
|
|
struct tree_mod_elem *tm;
|
|
|
|
struct tree_mod_elem *found = NULL;
|
2013-04-13 21:19:54 +08:00
|
|
|
u64 root_logical = eb_root->start;
|
2012-05-17 00:25:47 +08:00
|
|
|
int looped = 0;
|
|
|
|
|
|
|
|
if (!time_seq)
|
2013-08-15 00:12:25 +08:00
|
|
|
return NULL;
|
2012-05-17 00:25:47 +08:00
|
|
|
|
|
|
|
/*
|
2016-01-21 18:25:59 +08:00
|
|
|
* the very last operation that's logged for a root is the
|
|
|
|
* replacement operation (if it is replaced at all). this has
|
|
|
|
* the logical address of the *new* root, making it the very
|
|
|
|
* first operation that's logged for this root.
|
2012-05-17 00:25:47 +08:00
|
|
|
*/
|
|
|
|
while (1) {
|
|
|
|
tm = tree_mod_log_search_oldest(fs_info, root_logical,
|
|
|
|
time_seq);
|
|
|
|
if (!looped && !tm)
|
2013-08-15 00:12:25 +08:00
|
|
|
return NULL;
|
2012-05-17 00:25:47 +08:00
|
|
|
/*
|
2012-06-21 16:59:13 +08:00
|
|
|
* if there are no tree operation for the oldest root, we simply
|
|
|
|
* return it. this should only happen if that (old) root is at
|
|
|
|
* level 0.
|
2012-05-17 00:25:47 +08:00
|
|
|
*/
|
2012-06-21 16:59:13 +08:00
|
|
|
if (!tm)
|
|
|
|
break;
|
2012-05-17 00:25:47 +08:00
|
|
|
|
2012-06-21 16:59:13 +08:00
|
|
|
/*
|
|
|
|
* if there's an operation that's not a root replacement, we
|
|
|
|
* found the oldest version of our root. normally, we'll find a
|
|
|
|
* MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
|
|
|
|
*/
|
2012-05-17 00:25:47 +08:00
|
|
|
if (tm->op != MOD_LOG_ROOT_REPLACE)
|
|
|
|
break;
|
|
|
|
|
|
|
|
found = tm;
|
|
|
|
root_logical = tm->old_root.logical;
|
|
|
|
looped = 1;
|
|
|
|
}
|
|
|
|
|
2012-06-05 22:41:24 +08:00
|
|
|
/* if there's no old root to return, return what we found instead */
|
|
|
|
if (!found)
|
|
|
|
found = tm;
|
|
|
|
|
2012-05-17 00:25:47 +08:00
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* tm is a pointer to the first operation to rewind within eb. then, all
|
2016-05-20 09:18:45 +08:00
|
|
|
* previous operations will be rewound (until we reach something older than
|
2012-05-17 00:25:47 +08:00
|
|
|
* time_seq).
|
|
|
|
*/
|
|
|
|
static void
|
2013-06-30 11:15:19 +08:00
|
|
|
__tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
|
|
|
|
u64 time_seq, struct tree_mod_elem *first_tm)
|
2012-05-17 00:25:47 +08:00
|
|
|
{
|
|
|
|
u32 n;
|
|
|
|
struct rb_node *next;
|
|
|
|
struct tree_mod_elem *tm = first_tm;
|
|
|
|
unsigned long o_dst;
|
|
|
|
unsigned long o_src;
|
|
|
|
unsigned long p_size = sizeof(struct btrfs_key_ptr);
|
|
|
|
|
|
|
|
n = btrfs_header_nritems(eb);
|
2013-06-30 11:15:19 +08:00
|
|
|
tree_mod_log_read_lock(fs_info);
|
2012-06-21 17:08:04 +08:00
|
|
|
while (tm && tm->seq >= time_seq) {
|
2012-05-17 00:25:47 +08:00
|
|
|
/*
|
|
|
|
* all the operations are recorded with the operator used for
|
|
|
|
* the modification. as we're going backwards, we do the
|
|
|
|
* opposite of each operation here.
|
|
|
|
*/
|
|
|
|
switch (tm->op) {
|
|
|
|
case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
|
|
|
|
BUG_ON(tm->slot < n);
|
2013-01-31 08:54:56 +08:00
|
|
|
/* Fallthrough */
|
Btrfs: MOD_LOG_KEY_REMOVE_WHILE_MOVING never change node's nritems
Key MOD_LOG_KEY_REMOVE_WHILE_MOVING means that we're doing memmove inside
an extent buffer node, and the node's number of items remains unchanged
(unless we are inserting a single pointer, but we have MOD_LOG_KEY_ADD for that).
So we don't need to increase node's number of items during rewinding,
otherwise we may get an node larger than leafsize and cause general protection
errors later.
Here is the details,
- If we do memory move for inserting a single pointer, we need to
add node's nritems by one, and we honor MOD_LOG_KEY_ADD for adding.
- If we do memory move for deleting a single pointer, we need to
decrease node's nritems by one, and we honor MOD_LOG_KEY_REMOVE for
deleting.
- If we do memory move for balance left/right, we need to decrease
node's nritems, and we honor MOD_LOG_KEY_REMOVE for balaning.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2012-10-19 17:50:52 +08:00
|
|
|
case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
|
2012-12-19 04:43:18 +08:00
|
|
|
case MOD_LOG_KEY_REMOVE:
|
2012-05-17 00:25:47 +08:00
|
|
|
btrfs_set_node_key(eb, &tm->key, tm->slot);
|
|
|
|
btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
|
|
|
|
btrfs_set_node_ptr_generation(eb, tm->slot,
|
|
|
|
tm->generation);
|
2012-12-19 04:43:18 +08:00
|
|
|
n++;
|
2012-05-17 00:25:47 +08:00
|
|
|
break;
|
|
|
|
case MOD_LOG_KEY_REPLACE:
|
|
|
|
BUG_ON(tm->slot >= n);
|
|
|
|
btrfs_set_node_key(eb, &tm->key, tm->slot);
|
|
|
|
btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
|
|
|
|
btrfs_set_node_ptr_generation(eb, tm->slot,
|
|
|
|
tm->generation);
|
|
|
|
break;
|
|
|
|
case MOD_LOG_KEY_ADD:
|
2012-06-22 20:52:13 +08:00
|
|
|
/* if a move operation is needed it's in the log */
|
2012-05-17 00:25:47 +08:00
|
|
|
n--;
|
|
|
|
break;
|
|
|
|
case MOD_LOG_MOVE_KEYS:
|
2012-06-01 01:24:36 +08:00
|
|
|
o_dst = btrfs_node_key_ptr_offset(tm->slot);
|
|
|
|
o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
|
|
|
|
memmove_extent_buffer(eb, o_dst, o_src,
|
2012-05-17 00:25:47 +08:00
|
|
|
tm->move.nr_items * p_size);
|
|
|
|
break;
|
|
|
|
case MOD_LOG_ROOT_REPLACE:
|
|
|
|
/*
|
|
|
|
* this operation is special. for roots, this must be
|
|
|
|
* handled explicitly before rewinding.
|
|
|
|
* for non-roots, this operation may exist if the node
|
|
|
|
* was a root: root A -> child B; then A gets empty and
|
|
|
|
* B is promoted to the new root. in the mod log, we'll
|
|
|
|
* have a root-replace operation for B, a tree block
|
|
|
|
* that is no root. we simply ignore that operation.
|
|
|
|
*/
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
next = rb_next(&tm->node);
|
|
|
|
if (!next)
|
|
|
|
break;
|
2016-12-19 22:53:41 +08:00
|
|
|
tm = rb_entry(next, struct tree_mod_elem, node);
|
2016-01-21 18:25:59 +08:00
|
|
|
if (tm->logical != first_tm->logical)
|
2012-05-17 00:25:47 +08:00
|
|
|
break;
|
|
|
|
}
|
2013-06-30 11:15:19 +08:00
|
|
|
tree_mod_log_read_unlock(fs_info);
|
2012-05-17 00:25:47 +08:00
|
|
|
btrfs_set_header_nritems(eb, n);
|
|
|
|
}
|
|
|
|
|
2013-04-13 21:19:55 +08:00
|
|
|
/*
|
2016-05-20 09:18:45 +08:00
|
|
|
* Called with eb read locked. If the buffer cannot be rewound, the same buffer
|
2013-04-13 21:19:55 +08:00
|
|
|
* is returned. If rewind operations happen, a fresh buffer is returned. The
|
|
|
|
* returned buffer is always read-locked. If the returned buffer is not the
|
|
|
|
* input buffer, the lock on the input buffer is released and the input buffer
|
|
|
|
* is freed (its refcount is decremented).
|
|
|
|
*/
|
2012-05-17 00:25:47 +08:00
|
|
|
static struct extent_buffer *
|
2013-08-08 04:57:23 +08:00
|
|
|
tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
|
|
|
|
struct extent_buffer *eb, u64 time_seq)
|
2012-05-17 00:25:47 +08:00
|
|
|
{
|
|
|
|
struct extent_buffer *eb_rewin;
|
|
|
|
struct tree_mod_elem *tm;
|
|
|
|
|
|
|
|
if (!time_seq)
|
|
|
|
return eb;
|
|
|
|
|
|
|
|
if (btrfs_header_level(eb) == 0)
|
|
|
|
return eb;
|
|
|
|
|
|
|
|
tm = tree_mod_log_search(fs_info, eb->start, time_seq);
|
|
|
|
if (!tm)
|
|
|
|
return eb;
|
|
|
|
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
|
|
|
btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
|
|
|
|
|
2012-05-17 00:25:47 +08:00
|
|
|
if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
|
|
|
|
BUG_ON(tm->slot != 0);
|
2016-06-15 21:22:56 +08:00
|
|
|
eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
|
2013-08-08 02:54:37 +08:00
|
|
|
if (!eb_rewin) {
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_tree_read_unlock_blocking(eb);
|
2013-08-08 02:54:37 +08:00
|
|
|
free_extent_buffer(eb);
|
|
|
|
return NULL;
|
|
|
|
}
|
2012-05-17 00:25:47 +08:00
|
|
|
btrfs_set_header_bytenr(eb_rewin, eb->start);
|
|
|
|
btrfs_set_header_backref_rev(eb_rewin,
|
|
|
|
btrfs_header_backref_rev(eb));
|
|
|
|
btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
|
2012-06-01 01:24:36 +08:00
|
|
|
btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
|
2012-05-17 00:25:47 +08:00
|
|
|
} else {
|
|
|
|
eb_rewin = btrfs_clone_extent_buffer(eb);
|
2013-08-08 02:54:37 +08:00
|
|
|
if (!eb_rewin) {
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_tree_read_unlock_blocking(eb);
|
2013-08-08 02:54:37 +08:00
|
|
|
free_extent_buffer(eb);
|
|
|
|
return NULL;
|
|
|
|
}
|
2012-05-17 00:25:47 +08:00
|
|
|
}
|
|
|
|
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
|
|
|
|
btrfs_tree_read_unlock_blocking(eb);
|
2012-05-17 00:25:47 +08:00
|
|
|
free_extent_buffer(eb);
|
|
|
|
|
2013-04-13 21:19:55 +08:00
|
|
|
extent_buffer_get(eb_rewin);
|
|
|
|
btrfs_tree_read_lock(eb_rewin);
|
2013-06-30 11:15:19 +08:00
|
|
|
__tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
|
2012-10-19 15:22:03 +08:00
|
|
|
WARN_ON(btrfs_header_nritems(eb_rewin) >
|
2016-06-15 21:22:56 +08:00
|
|
|
BTRFS_NODEPTRS_PER_BLOCK(fs_info));
|
2012-05-17 00:25:47 +08:00
|
|
|
|
|
|
|
return eb_rewin;
|
|
|
|
}
|
|
|
|
|
2012-06-04 22:54:57 +08:00
|
|
|
/*
|
|
|
|
* get_old_root() rewinds the state of @root's root node to the given @time_seq
|
|
|
|
* value. If there are no changes, the current root->root_node is returned. If
|
|
|
|
* anything changed in between, there's a fresh buffer allocated on which the
|
|
|
|
* rewind operations are done. In any case, the returned buffer is read locked.
|
|
|
|
* Returns NULL on error (with no locks held).
|
|
|
|
*/
|
2012-05-17 00:25:47 +08:00
|
|
|
static inline struct extent_buffer *
|
|
|
|
get_old_root(struct btrfs_root *root, u64 time_seq)
|
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2012-05-17 00:25:47 +08:00
|
|
|
struct tree_mod_elem *tm;
|
2013-04-13 21:19:54 +08:00
|
|
|
struct extent_buffer *eb = NULL;
|
|
|
|
struct extent_buffer *eb_root;
|
2012-10-25 21:30:19 +08:00
|
|
|
struct extent_buffer *old;
|
2012-06-05 22:41:24 +08:00
|
|
|
struct tree_mod_root *old_root = NULL;
|
2012-06-16 08:02:02 +08:00
|
|
|
u64 old_generation = 0;
|
2012-06-05 22:41:24 +08:00
|
|
|
u64 logical;
|
2012-05-17 00:25:47 +08:00
|
|
|
|
2013-04-13 21:19:54 +08:00
|
|
|
eb_root = btrfs_read_lock_root_node(root);
|
2016-06-23 06:54:23 +08:00
|
|
|
tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq);
|
2012-05-17 00:25:47 +08:00
|
|
|
if (!tm)
|
2013-04-13 21:19:54 +08:00
|
|
|
return eb_root;
|
2012-05-17 00:25:47 +08:00
|
|
|
|
2012-06-05 22:41:24 +08:00
|
|
|
if (tm->op == MOD_LOG_ROOT_REPLACE) {
|
|
|
|
old_root = &tm->old_root;
|
|
|
|
old_generation = tm->generation;
|
|
|
|
logical = old_root->logical;
|
|
|
|
} else {
|
2013-04-13 21:19:54 +08:00
|
|
|
logical = eb_root->start;
|
2012-06-05 22:41:24 +08:00
|
|
|
}
|
2012-05-17 00:25:47 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
tm = tree_mod_log_search(fs_info, logical, time_seq);
|
2012-10-23 17:27:33 +08:00
|
|
|
if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
|
2013-04-13 21:19:54 +08:00
|
|
|
btrfs_tree_read_unlock(eb_root);
|
|
|
|
free_extent_buffer(eb_root);
|
2016-06-23 06:54:24 +08:00
|
|
|
old = read_tree_block(fs_info, logical, 0);
|
2015-05-25 17:30:15 +08:00
|
|
|
if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
|
|
|
|
if (!IS_ERR(old))
|
|
|
|
free_extent_buffer(old);
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_warn(fs_info,
|
|
|
|
"failed to read tree block %llu from get_old_root",
|
|
|
|
logical);
|
2012-10-23 17:27:33 +08:00
|
|
|
} else {
|
2012-10-25 21:30:19 +08:00
|
|
|
eb = btrfs_clone_extent_buffer(old);
|
|
|
|
free_extent_buffer(old);
|
2012-10-23 17:27:33 +08:00
|
|
|
}
|
|
|
|
} else if (old_root) {
|
2013-04-13 21:19:54 +08:00
|
|
|
btrfs_tree_read_unlock(eb_root);
|
|
|
|
free_extent_buffer(eb_root);
|
2016-06-23 06:54:23 +08:00
|
|
|
eb = alloc_dummy_extent_buffer(fs_info, logical);
|
2012-10-23 17:27:33 +08:00
|
|
|
} else {
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
|
2013-04-13 21:19:54 +08:00
|
|
|
eb = btrfs_clone_extent_buffer(eb_root);
|
2013-08-08 04:57:23 +08:00
|
|
|
btrfs_tree_read_unlock_blocking(eb_root);
|
2013-04-13 21:19:54 +08:00
|
|
|
free_extent_buffer(eb_root);
|
2012-10-23 17:27:33 +08:00
|
|
|
}
|
|
|
|
|
2012-06-04 22:54:57 +08:00
|
|
|
if (!eb)
|
|
|
|
return NULL;
|
2012-10-23 20:21:05 +08:00
|
|
|
extent_buffer_get(eb);
|
2012-06-04 22:54:57 +08:00
|
|
|
btrfs_tree_read_lock(eb);
|
2012-06-05 22:41:24 +08:00
|
|
|
if (old_root) {
|
2012-05-17 00:25:47 +08:00
|
|
|
btrfs_set_header_bytenr(eb, eb->start);
|
|
|
|
btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
|
2013-04-13 21:19:54 +08:00
|
|
|
btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
|
2012-06-05 22:41:24 +08:00
|
|
|
btrfs_set_header_level(eb, old_root->level);
|
|
|
|
btrfs_set_header_generation(eb, old_generation);
|
2012-05-17 00:25:47 +08:00
|
|
|
}
|
2012-06-21 16:59:13 +08:00
|
|
|
if (tm)
|
2016-06-23 06:54:23 +08:00
|
|
|
__tree_mod_log_rewind(fs_info, eb, time_seq, tm);
|
2012-06-21 16:59:13 +08:00
|
|
|
else
|
|
|
|
WARN_ON(btrfs_header_level(eb) != 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
|
2012-05-17 00:25:47 +08:00
|
|
|
|
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
2012-10-23 17:28:27 +08:00
|
|
|
int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
|
|
|
|
{
|
|
|
|
struct tree_mod_elem *tm;
|
|
|
|
int level;
|
2013-04-13 21:19:54 +08:00
|
|
|
struct extent_buffer *eb_root = btrfs_root_node(root);
|
2012-10-23 17:28:27 +08:00
|
|
|
|
2013-04-13 21:19:54 +08:00
|
|
|
tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
|
2012-10-23 17:28:27 +08:00
|
|
|
if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
|
|
|
|
level = tm->old_root.level;
|
|
|
|
} else {
|
2013-04-13 21:19:54 +08:00
|
|
|
level = btrfs_header_level(eb_root);
|
2012-10-23 17:28:27 +08:00
|
|
|
}
|
2013-04-13 21:19:54 +08:00
|
|
|
free_extent_buffer(eb_root);
|
2012-10-23 17:28:27 +08:00
|
|
|
|
|
|
|
return level;
|
|
|
|
}
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
static inline int should_cow_block(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct extent_buffer *buf)
|
|
|
|
{
|
2016-06-21 21:52:41 +08:00
|
|
|
if (btrfs_is_testing(root->fs_info))
|
2014-05-08 05:06:09 +08:00
|
|
|
return 0;
|
2014-09-30 05:53:21 +08:00
|
|
|
|
2011-11-15 09:48:06 +08:00
|
|
|
/* ensure we can see the force_cow */
|
|
|
|
smp_rmb();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We do not need to cow a block if
|
|
|
|
* 1) this block is not created or changed in this transaction;
|
|
|
|
* 2) this block does not belong to TREE_RELOC tree;
|
|
|
|
* 3) the root is not forced COW.
|
|
|
|
*
|
|
|
|
* What is forced COW:
|
2016-05-20 09:18:45 +08:00
|
|
|
* when we create snapshot during committing the transaction,
|
2011-11-15 09:48:06 +08:00
|
|
|
* after we've finished coping src root, we must COW the shared
|
|
|
|
* block to ensure the metadata consistency.
|
|
|
|
*/
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (btrfs_header_generation(buf) == trans->transid &&
|
|
|
|
!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
|
|
|
|
!(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
|
2011-11-15 09:48:06 +08:00
|
|
|
btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
|
2014-04-02 19:51:05 +08:00
|
|
|
!test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
return 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* cows a single block, see __btrfs_cow_block for the real work.
|
2016-05-20 09:18:45 +08:00
|
|
|
* This version of it has extra checks so that a block isn't COWed more than
|
2008-09-30 03:18:18 +08:00
|
|
|
* once per transaction, as long as it hasn't been written yet
|
|
|
|
*/
|
2009-01-06 10:25:51 +08:00
|
|
|
noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_root *root, struct extent_buffer *buf,
|
|
|
|
struct extent_buffer *parent, int parent_slot,
|
2009-03-13 22:24:59 +08:00
|
|
|
struct extent_buffer **cow_ret)
|
2007-08-08 04:15:09 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-08-08 04:15:09 +08:00
|
|
|
u64 search_start;
|
2007-10-16 04:14:48 +08:00
|
|
|
int ret;
|
2008-01-09 04:46:30 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
if (trans->transaction != fs_info->running_transaction)
|
2012-11-03 18:58:34 +08:00
|
|
|
WARN(1, KERN_CRIT "trans %llu running %llu\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
trans->transid,
|
2016-06-23 06:54:23 +08:00
|
|
|
fs_info->running_transaction->transid);
|
2012-11-03 18:58:34 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
if (trans->transid != fs_info->generation)
|
2012-11-03 18:58:34 +08:00
|
|
|
WARN(1, KERN_CRIT "trans %llu running %llu\n",
|
2016-06-23 06:54:23 +08:00
|
|
|
trans->transid, fs_info->generation);
|
2008-01-09 04:46:30 +08:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (!should_cow_block(trans, root, buf)) {
|
2016-06-08 12:36:38 +08:00
|
|
|
trans->dirty = true;
|
2007-08-08 04:15:09 +08:00
|
|
|
*cow_ret = buf;
|
|
|
|
return 0;
|
|
|
|
}
|
2009-02-04 22:24:25 +08:00
|
|
|
|
2015-12-15 00:42:10 +08:00
|
|
|
search_start = buf->start & ~((u64)SZ_1G - 1);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
|
|
|
if (parent)
|
|
|
|
btrfs_set_lock_blocking(parent);
|
|
|
|
btrfs_set_lock_blocking(buf);
|
|
|
|
|
2007-10-16 04:14:48 +08:00
|
|
|
ret = __btrfs_cow_block(trans, root, buf, parent,
|
2009-03-13 22:24:59 +08:00
|
|
|
parent_slot, cow_ret, search_start, 0);
|
Btrfs: add initial tracepoint support for btrfs
Tracepoints can provide insight into why btrfs hits bugs and be greatly
helpful for debugging, e.g
dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0
dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0
btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0)
btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0)
btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8
flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA
flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0)
flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0)
flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0)
btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0)
btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0)
Here is what I have added:
1) ordere_extent:
btrfs_ordered_extent_add
btrfs_ordered_extent_remove
btrfs_ordered_extent_start
btrfs_ordered_extent_put
These provide critical information to understand how ordered_extents are
updated.
2) extent_map:
btrfs_get_extent
extent_map is used in both read and write cases, and it is useful for tracking
how btrfs specific IO is running.
3) writepage:
__extent_writepage
btrfs_writepage_end_io_hook
Pages are cirtical resourses and produce a lot of corner cases during writeback,
so it is valuable to know how page is written to disk.
4) inode:
btrfs_inode_new
btrfs_inode_request
btrfs_inode_evict
These can show where and when a inode is created, when a inode is evicted.
5) sync:
btrfs_sync_file
btrfs_sync_fs
These show sync arguments.
6) transaction:
btrfs_transaction_commit
In transaction based filesystem, it will be useful to know the generation and
who does commit.
7) back reference and cow:
btrfs_delayed_tree_ref
btrfs_delayed_data_ref
btrfs_delayed_ref_head
btrfs_cow_block
Btrfs natively supports back references, these tracepoints are helpful on
understanding btrfs's COW mechanism.
8) chunk:
btrfs_chunk_alloc
btrfs_chunk_free
Chunk is a link between physical offset and logical offset, and stands for space
infomation in btrfs, and these are helpful on tracing space things.
9) reserved_extent:
btrfs_reserved_extent_alloc
btrfs_reserved_extent_free
These can show how btrfs uses its space.
Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 19:18:59 +08:00
|
|
|
|
|
|
|
trace_btrfs_cow_block(root, buf, *cow_ret);
|
|
|
|
|
2007-10-16 04:14:48 +08:00
|
|
|
return ret;
|
2007-08-08 04:15:09 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* helper function for defrag to decide if two blocks pointed to by a
|
|
|
|
* node are actually close by
|
|
|
|
*/
|
2007-10-16 04:17:34 +08:00
|
|
|
static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
|
2007-08-08 04:15:09 +08:00
|
|
|
{
|
2007-10-16 04:17:34 +08:00
|
|
|
if (blocknr < other && other - (blocknr + blocksize) < 32768)
|
2007-08-08 04:15:09 +08:00
|
|
|
return 1;
|
2007-10-16 04:17:34 +08:00
|
|
|
if (blocknr > other && blocknr - (other + blocksize) < 32768)
|
2007-08-08 04:15:09 +08:00
|
|
|
return 1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-11-06 23:26:24 +08:00
|
|
|
/*
|
|
|
|
* compare two keys in a memcmp fashion
|
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
static int comp_keys(const struct btrfs_disk_key *disk,
|
|
|
|
const struct btrfs_key *k2)
|
2007-11-06 23:26:24 +08:00
|
|
|
{
|
|
|
|
struct btrfs_key k1;
|
|
|
|
|
|
|
|
btrfs_disk_key_to_cpu(&k1, disk);
|
|
|
|
|
2009-07-24 23:06:52 +08:00
|
|
|
return btrfs_comp_cpu_keys(&k1, k2);
|
2007-11-06 23:26:24 +08:00
|
|
|
}
|
|
|
|
|
2008-11-13 03:19:50 +08:00
|
|
|
/*
|
|
|
|
* same as comp_keys only with two btrfs_key's
|
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
|
2008-11-13 03:19:50 +08:00
|
|
|
{
|
|
|
|
if (k1->objectid > k2->objectid)
|
|
|
|
return 1;
|
|
|
|
if (k1->objectid < k2->objectid)
|
|
|
|
return -1;
|
|
|
|
if (k1->type > k2->type)
|
|
|
|
return 1;
|
|
|
|
if (k1->type < k2->type)
|
|
|
|
return -1;
|
|
|
|
if (k1->offset > k2->offset)
|
|
|
|
return 1;
|
|
|
|
if (k1->offset < k2->offset)
|
|
|
|
return -1;
|
|
|
|
return 0;
|
|
|
|
}
|
2007-11-06 23:26:24 +08:00
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* this is used by the defrag code to go through all the
|
|
|
|
* leaves pointed to by a node and reallocate them so that
|
|
|
|
* disk order is close to key order
|
|
|
|
*/
|
2007-08-08 04:15:09 +08:00
|
|
|
int btrfs_realloc_node(struct btrfs_trans_handle *trans,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_root *root, struct extent_buffer *parent,
|
2013-02-01 02:21:12 +08:00
|
|
|
int start_slot, u64 *last_ret,
|
2007-10-16 04:22:39 +08:00
|
|
|
struct btrfs_key *progress)
|
2007-08-08 04:15:09 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:17:34 +08:00
|
|
|
struct extent_buffer *cur;
|
2007-08-08 04:15:09 +08:00
|
|
|
u64 blocknr;
|
2008-05-13 00:59:19 +08:00
|
|
|
u64 gen;
|
2007-08-11 02:06:19 +08:00
|
|
|
u64 search_start = *last_ret;
|
|
|
|
u64 last_block = 0;
|
2007-08-08 04:15:09 +08:00
|
|
|
u64 other;
|
|
|
|
u32 parent_nritems;
|
|
|
|
int end_slot;
|
|
|
|
int i;
|
|
|
|
int err = 0;
|
2007-08-11 02:42:37 +08:00
|
|
|
int parent_level;
|
2007-10-16 04:17:34 +08:00
|
|
|
int uptodate;
|
|
|
|
u32 blocksize;
|
2007-11-06 23:26:24 +08:00
|
|
|
int progress_passed = 0;
|
|
|
|
struct btrfs_disk_key disk_key;
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2007-10-26 03:43:18 +08:00
|
|
|
parent_level = btrfs_header_level(parent);
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
WARN_ON(trans->transaction != fs_info->running_transaction);
|
|
|
|
WARN_ON(trans->transid != fs_info->generation);
|
2007-09-11 07:58:16 +08:00
|
|
|
|
2007-10-16 04:17:34 +08:00
|
|
|
parent_nritems = btrfs_header_nritems(parent);
|
2016-06-23 06:54:23 +08:00
|
|
|
blocksize = fs_info->nodesize;
|
2015-02-24 03:48:52 +08:00
|
|
|
end_slot = parent_nritems - 1;
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2015-02-24 03:48:52 +08:00
|
|
|
if (parent_nritems <= 1)
|
2007-08-08 04:15:09 +08:00
|
|
|
return 0;
|
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(parent);
|
|
|
|
|
2015-02-24 03:48:52 +08:00
|
|
|
for (i = start_slot; i <= end_slot; i++) {
|
2007-08-08 04:15:09 +08:00
|
|
|
int close = 1;
|
2007-10-16 04:22:39 +08:00
|
|
|
|
2007-11-06 23:26:24 +08:00
|
|
|
btrfs_node_key(parent, &disk_key, i);
|
|
|
|
if (!progress_passed && comp_keys(&disk_key, progress) < 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
progress_passed = 1;
|
2007-10-16 04:17:34 +08:00
|
|
|
blocknr = btrfs_node_blockptr(parent, i);
|
2008-05-13 00:59:19 +08:00
|
|
|
gen = btrfs_node_ptr_generation(parent, i);
|
2007-08-11 02:06:19 +08:00
|
|
|
if (last_block == 0)
|
|
|
|
last_block = blocknr;
|
2007-10-26 03:43:18 +08:00
|
|
|
|
2007-08-08 04:15:09 +08:00
|
|
|
if (i > 0) {
|
2007-10-16 04:17:34 +08:00
|
|
|
other = btrfs_node_blockptr(parent, i - 1);
|
|
|
|
close = close_blocks(blocknr, other, blocksize);
|
2007-08-08 04:15:09 +08:00
|
|
|
}
|
2015-02-24 03:48:52 +08:00
|
|
|
if (!close && i < end_slot) {
|
2007-10-16 04:17:34 +08:00
|
|
|
other = btrfs_node_blockptr(parent, i + 1);
|
|
|
|
close = close_blocks(blocknr, other, blocksize);
|
2007-08-08 04:15:09 +08:00
|
|
|
}
|
2007-08-11 02:06:19 +08:00
|
|
|
if (close) {
|
|
|
|
last_block = blocknr;
|
2007-08-08 04:15:09 +08:00
|
|
|
continue;
|
2007-08-11 02:06:19 +08:00
|
|
|
}
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
cur = find_extent_buffer(fs_info, blocknr);
|
2007-10-16 04:17:34 +08:00
|
|
|
if (cur)
|
2012-05-06 19:23:47 +08:00
|
|
|
uptodate = btrfs_buffer_uptodate(cur, gen, 0);
|
2007-10-16 04:17:34 +08:00
|
|
|
else
|
|
|
|
uptodate = 0;
|
2007-10-26 03:43:18 +08:00
|
|
|
if (!cur || !uptodate) {
|
2007-10-16 04:17:34 +08:00
|
|
|
if (!cur) {
|
2016-06-23 06:54:24 +08:00
|
|
|
cur = read_tree_block(fs_info, blocknr, gen);
|
2015-05-25 17:30:15 +08:00
|
|
|
if (IS_ERR(cur)) {
|
|
|
|
return PTR_ERR(cur);
|
|
|
|
} else if (!extent_buffer_uptodate(cur)) {
|
2013-04-24 02:17:42 +08:00
|
|
|
free_extent_buffer(cur);
|
2011-03-24 14:33:21 +08:00
|
|
|
return -EIO;
|
2013-04-24 02:17:42 +08:00
|
|
|
}
|
2007-10-16 04:17:34 +08:00
|
|
|
} else if (!uptodate) {
|
2012-05-29 17:10:13 +08:00
|
|
|
err = btrfs_read_buffer(cur, gen);
|
|
|
|
if (err) {
|
|
|
|
free_extent_buffer(cur);
|
|
|
|
return err;
|
|
|
|
}
|
2007-08-11 02:42:37 +08:00
|
|
|
}
|
2007-08-08 04:15:09 +08:00
|
|
|
}
|
2007-08-11 02:06:19 +08:00
|
|
|
if (search_start == 0)
|
2007-10-16 04:17:34 +08:00
|
|
|
search_start = last_block;
|
2007-08-11 02:06:19 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
btrfs_tree_lock(cur);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(cur);
|
2007-10-16 04:17:34 +08:00
|
|
|
err = __btrfs_cow_block(trans, root, cur, parent, i,
|
2008-06-26 04:01:31 +08:00
|
|
|
&cur, search_start,
|
2007-10-16 04:17:34 +08:00
|
|
|
min(16 * blocksize,
|
2009-03-13 22:24:59 +08:00
|
|
|
(end_slot - i) * blocksize));
|
2007-08-29 21:11:44 +08:00
|
|
|
if (err) {
|
2008-06-26 04:01:31 +08:00
|
|
|
btrfs_tree_unlock(cur);
|
2007-10-16 04:17:34 +08:00
|
|
|
free_extent_buffer(cur);
|
2007-08-08 04:15:09 +08:00
|
|
|
break;
|
2007-08-29 21:11:44 +08:00
|
|
|
}
|
2008-06-26 04:01:31 +08:00
|
|
|
search_start = cur->start;
|
|
|
|
last_block = cur->start;
|
2007-08-11 02:42:37 +08:00
|
|
|
*last_ret = search_start;
|
2008-06-26 04:01:31 +08:00
|
|
|
btrfs_tree_unlock(cur);
|
|
|
|
free_extent_buffer(cur);
|
2007-08-08 04:15:09 +08:00
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
2007-10-16 04:14:19 +08:00
|
|
|
* search for key in the extent_buffer. The items start at offset p,
|
|
|
|
* and they are item_size apart. There are 'max' items in p.
|
|
|
|
*
|
2007-02-03 00:05:29 +08:00
|
|
|
* the slot in the array is returned via slot, and it points to
|
|
|
|
* the place where you would insert key if it is not found in
|
|
|
|
* the array.
|
|
|
|
*
|
|
|
|
* slot may point to max if the key is bigger than all of the keys
|
|
|
|
*/
|
2008-09-06 04:13:11 +08:00
|
|
|
static noinline int generic_bin_search(struct extent_buffer *eb,
|
2017-01-18 15:24:37 +08:00
|
|
|
unsigned long p, int item_size,
|
|
|
|
const struct btrfs_key *key,
|
2008-09-06 04:13:11 +08:00
|
|
|
int max, int *slot)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
|
|
|
int low = 0;
|
|
|
|
int high = max;
|
|
|
|
int mid;
|
|
|
|
int ret;
|
2007-10-16 04:14:27 +08:00
|
|
|
struct btrfs_disk_key *tmp = NULL;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key unaligned;
|
|
|
|
unsigned long offset;
|
|
|
|
char *kaddr = NULL;
|
|
|
|
unsigned long map_start = 0;
|
|
|
|
unsigned long map_len = 0;
|
2007-10-16 04:14:27 +08:00
|
|
|
int err;
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2016-06-24 07:32:45 +08:00
|
|
|
if (low > high) {
|
|
|
|
btrfs_err(eb->fs_info,
|
|
|
|
"%s: low (%d) > high (%d) eb %llu owner %llu level %d",
|
|
|
|
__func__, low, high, eb->start,
|
|
|
|
btrfs_header_owner(eb), btrfs_header_level(eb));
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (low < high) {
|
2007-01-27 04:51:26 +08:00
|
|
|
mid = (low + high) / 2;
|
2007-10-16 04:14:19 +08:00
|
|
|
offset = p + mid * item_size;
|
|
|
|
|
2011-07-20 00:04:14 +08:00
|
|
|
if (!kaddr || offset < map_start ||
|
2007-10-16 04:14:19 +08:00
|
|
|
(offset + sizeof(struct btrfs_disk_key)) >
|
|
|
|
map_start + map_len) {
|
2008-12-09 05:43:10 +08:00
|
|
|
|
|
|
|
err = map_private_extent_buffer(eb, offset,
|
2007-10-16 04:14:27 +08:00
|
|
|
sizeof(struct btrfs_disk_key),
|
2011-07-20 00:04:14 +08:00
|
|
|
&kaddr, &map_start, &map_len);
|
2007-10-16 04:14:27 +08:00
|
|
|
|
|
|
|
if (!err) {
|
|
|
|
tmp = (struct btrfs_disk_key *)(kaddr + offset -
|
|
|
|
map_start);
|
2016-06-18 10:16:21 +08:00
|
|
|
} else if (err == 1) {
|
2007-10-16 04:14:27 +08:00
|
|
|
read_extent_buffer(eb, &unaligned,
|
|
|
|
offset, sizeof(unaligned));
|
|
|
|
tmp = &unaligned;
|
2016-06-18 10:16:21 +08:00
|
|
|
} else {
|
|
|
|
return err;
|
2007-10-16 04:14:27 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
} else {
|
|
|
|
tmp = (struct btrfs_disk_key *)(kaddr + offset -
|
|
|
|
map_start);
|
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
ret = comp_keys(tmp, key);
|
|
|
|
|
|
|
|
if (ret < 0)
|
|
|
|
low = mid + 1;
|
|
|
|
else if (ret > 0)
|
|
|
|
high = mid;
|
|
|
|
else {
|
|
|
|
*slot = mid;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
*slot = low;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2007-02-25 02:39:08 +08:00
|
|
|
/*
|
|
|
|
* simple bin_search frontend that does the right thing for
|
|
|
|
* leaves vs nodes
|
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
static int bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
|
2007-10-16 04:14:19 +08:00
|
|
|
int level, int *slot)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2012-03-30 15:14:27 +08:00
|
|
|
if (level == 0)
|
2007-10-16 04:14:19 +08:00
|
|
|
return generic_bin_search(eb,
|
|
|
|
offsetof(struct btrfs_leaf, items),
|
2007-03-13 08:12:07 +08:00
|
|
|
sizeof(struct btrfs_item),
|
2007-10-16 04:14:19 +08:00
|
|
|
key, btrfs_header_nritems(eb),
|
2007-03-13 00:01:18 +08:00
|
|
|
slot);
|
2012-03-30 15:14:27 +08:00
|
|
|
else
|
2007-10-16 04:14:19 +08:00
|
|
|
return generic_bin_search(eb,
|
|
|
|
offsetof(struct btrfs_node, ptrs),
|
2007-03-15 02:14:43 +08:00
|
|
|
sizeof(struct btrfs_key_ptr),
|
2007-10-16 04:14:19 +08:00
|
|
|
key, btrfs_header_nritems(eb),
|
2007-03-13 00:01:18 +08:00
|
|
|
slot);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2017-01-18 15:24:37 +08:00
|
|
|
int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
int level, int *slot)
|
|
|
|
{
|
|
|
|
return bin_search(eb, key, level, slot);
|
|
|
|
}
|
|
|
|
|
2010-05-16 22:46:25 +08:00
|
|
|
static void root_add_used(struct btrfs_root *root, u32 size)
|
|
|
|
{
|
|
|
|
spin_lock(&root->accounting_lock);
|
|
|
|
btrfs_set_root_used(&root->root_item,
|
|
|
|
btrfs_root_used(&root->root_item) + size);
|
|
|
|
spin_unlock(&root->accounting_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void root_sub_used(struct btrfs_root *root, u32 size)
|
|
|
|
{
|
|
|
|
spin_lock(&root->accounting_lock);
|
|
|
|
btrfs_set_root_used(&root->root_item,
|
|
|
|
btrfs_root_used(&root->root_item) - size);
|
|
|
|
spin_unlock(&root->accounting_lock);
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/* given a node and slot number, this reads the blocks it points to. The
|
|
|
|
* extent buffer is returned with a reference taken (but unlocked).
|
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
static noinline struct extent_buffer *
|
|
|
|
read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
|
|
|
|
int slot)
|
2007-03-02 01:04:21 +08:00
|
|
|
{
|
2008-05-13 00:59:19 +08:00
|
|
|
int level = btrfs_header_level(parent);
|
2013-04-24 02:17:42 +08:00
|
|
|
struct extent_buffer *eb;
|
|
|
|
|
2016-07-06 03:10:14 +08:00
|
|
|
if (slot < 0 || slot >= btrfs_header_nritems(parent))
|
|
|
|
return ERR_PTR(-ENOENT);
|
2008-05-13 00:59:19 +08:00
|
|
|
|
|
|
|
BUG_ON(level == 0);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
|
2013-04-24 02:17:42 +08:00
|
|
|
btrfs_node_ptr_generation(parent, slot));
|
2016-07-06 03:10:14 +08:00
|
|
|
if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
eb = ERR_PTR(-EIO);
|
2013-04-24 02:17:42 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return eb;
|
2007-03-02 01:04:21 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* node level balancing, used to make sure nodes are in proper order for
|
|
|
|
* item deletion. We balance from the top down, so we have to make sure
|
|
|
|
* that a deletion won't leave an node completely empty later on.
|
|
|
|
*/
|
2008-09-06 04:13:11 +08:00
|
|
|
static noinline int balance_level(struct btrfs_trans_handle *trans,
|
2008-01-03 23:01:48 +08:00
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, int level)
|
2007-03-02 01:04:21 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *right = NULL;
|
|
|
|
struct extent_buffer *mid;
|
|
|
|
struct extent_buffer *left = NULL;
|
|
|
|
struct extent_buffer *parent = NULL;
|
2007-03-02 01:04:21 +08:00
|
|
|
int ret = 0;
|
|
|
|
int wret;
|
|
|
|
int pslot;
|
|
|
|
int orig_slot = path->slots[level];
|
2007-03-02 04:16:26 +08:00
|
|
|
u64 orig_ptr;
|
2007-03-02 01:04:21 +08:00
|
|
|
|
|
|
|
if (level == 0)
|
|
|
|
return 0;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
mid = path->nodes[level];
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
|
|
|
|
path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(btrfs_header_generation(mid) != trans->transid);
|
|
|
|
|
2007-03-13 21:28:32 +08:00
|
|
|
orig_ptr = btrfs_node_blockptr(mid, orig_slot);
|
2007-03-02 04:16:26 +08:00
|
|
|
|
2011-09-06 16:55:34 +08:00
|
|
|
if (level < BTRFS_MAX_LEVEL - 1) {
|
2007-10-16 04:14:19 +08:00
|
|
|
parent = path->nodes[level + 1];
|
2011-09-06 16:55:34 +08:00
|
|
|
pslot = path->slots[level + 1];
|
|
|
|
}
|
2007-03-02 01:04:21 +08:00
|
|
|
|
2007-03-18 02:29:23 +08:00
|
|
|
/*
|
|
|
|
* deal with the case where there is only one pointer in the root
|
|
|
|
* by promoting the node below to a root
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
if (!parent) {
|
|
|
|
struct extent_buffer *child;
|
2007-03-02 01:04:21 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_nritems(mid) != 1)
|
2007-03-02 01:04:21 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* promote the child to a root */
|
2016-06-23 06:54:24 +08:00
|
|
|
child = read_node_slot(fs_info, mid, 0);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(child)) {
|
|
|
|
ret = PTR_ERR(child);
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_handle_fs_error(fs_info, ret, NULL);
|
2011-09-02 02:27:57 +08:00
|
|
|
goto enospc;
|
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_lock(child);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(child);
|
2009-03-13 22:24:59 +08:00
|
|
|
ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
|
2010-05-16 22:46:25 +08:00
|
|
|
if (ret) {
|
|
|
|
btrfs_tree_unlock(child);
|
|
|
|
free_extent_buffer(child);
|
|
|
|
goto enospc;
|
|
|
|
}
|
2008-02-02 03:58:07 +08:00
|
|
|
|
2013-04-13 21:19:53 +08:00
|
|
|
tree_mod_log_set_root_pointer(root, child, 1);
|
2011-03-24 02:54:42 +08:00
|
|
|
rcu_assign_pointer(root->node, child);
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2008-03-25 03:01:56 +08:00
|
|
|
add_root_to_dirty_list(root);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(child);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
path->locks[level] = 0;
|
2007-03-02 01:04:21 +08:00
|
|
|
path->nodes[level] = NULL;
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, mid);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(mid);
|
2007-03-02 01:04:21 +08:00
|
|
|
/* once for the path */
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(mid);
|
2010-05-16 22:46:25 +08:00
|
|
|
|
|
|
|
root_sub_used(root, mid->len);
|
2012-05-16 23:04:52 +08:00
|
|
|
btrfs_free_tree_block(trans, root, mid, 0, 1);
|
2007-03-02 01:04:21 +08:00
|
|
|
/* once for the root ptr */
|
2012-03-10 05:01:49 +08:00
|
|
|
free_extent_buffer_stale(mid);
|
2010-05-16 22:46:25 +08:00
|
|
|
return 0;
|
2007-03-02 01:04:21 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_nritems(mid) >
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
|
2007-03-02 01:04:21 +08:00
|
|
|
return 0;
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
left = read_node_slot(fs_info, parent, pslot - 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(left))
|
|
|
|
left = NULL;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (left) {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_lock(left);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(left);
|
2007-10-16 04:14:19 +08:00
|
|
|
wret = btrfs_cow_block(trans, root, left,
|
2009-03-13 22:24:59 +08:00
|
|
|
parent, pslot - 1, &left);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (wret) {
|
|
|
|
ret = wret;
|
|
|
|
goto enospc;
|
|
|
|
}
|
2007-08-28 04:49:44 +08:00
|
|
|
}
|
2016-07-06 03:10:14 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
right = read_node_slot(fs_info, parent, pslot + 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(right))
|
|
|
|
right = NULL;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (right) {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_lock(right);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(right);
|
2007-10-16 04:14:19 +08:00
|
|
|
wret = btrfs_cow_block(trans, root, right,
|
2009-03-13 22:24:59 +08:00
|
|
|
parent, pslot + 1, &right);
|
2007-08-28 04:49:44 +08:00
|
|
|
if (wret) {
|
|
|
|
ret = wret;
|
|
|
|
goto enospc;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* first, try to make some room in the middle buffer */
|
2007-10-16 04:14:19 +08:00
|
|
|
if (left) {
|
|
|
|
orig_slot += btrfs_header_nritems(left);
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = push_node_left(trans, fs_info, left, mid, 1);
|
2007-03-02 04:16:26 +08:00
|
|
|
if (wret < 0)
|
|
|
|
ret = wret;
|
2007-03-02 01:04:21 +08:00
|
|
|
}
|
2007-03-02 04:16:26 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* then try to empty the right most buffer into the middle
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
if (right) {
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = push_node_left(trans, fs_info, mid, right, 1);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (wret < 0 && wret != -ENOSPC)
|
2007-03-02 04:16:26 +08:00
|
|
|
ret = wret;
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_nritems(right) == 0) {
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, right);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(right);
|
2013-04-16 13:18:22 +08:00
|
|
|
del_ptr(root, path, level + 1, pslot + 1);
|
2010-05-16 22:46:25 +08:00
|
|
|
root_sub_used(root, right->len);
|
2012-05-16 23:04:52 +08:00
|
|
|
btrfs_free_tree_block(trans, root, right, 0, 1);
|
2012-03-10 05:01:49 +08:00
|
|
|
free_extent_buffer_stale(right);
|
2010-05-16 22:46:25 +08:00
|
|
|
right = NULL;
|
2007-03-02 01:04:21 +08:00
|
|
|
} else {
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key right_key;
|
|
|
|
btrfs_node_key(right, &right_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_set_node_key(fs_info, parent,
|
2012-10-19 20:52:15 +08:00
|
|
|
pslot + 1, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(parent, &right_key, pslot + 1);
|
|
|
|
btrfs_mark_buffer_dirty(parent);
|
2007-03-02 01:04:21 +08:00
|
|
|
}
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_nritems(mid) == 1) {
|
2007-03-02 04:16:26 +08:00
|
|
|
/*
|
|
|
|
* we're not allowed to leave a node with one item in the
|
|
|
|
* tree during a delete. A deletion from lower in the tree
|
|
|
|
* could try to delete the only pointer in this node.
|
|
|
|
* So, pull some keys from the left.
|
|
|
|
* There has to be a left pointer at this point because
|
|
|
|
* otherwise we would have pulled some pointers from the
|
|
|
|
* right
|
|
|
|
*/
|
2011-09-02 02:27:57 +08:00
|
|
|
if (!left) {
|
|
|
|
ret = -EROFS;
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_handle_fs_error(fs_info, ret, NULL);
|
2011-09-02 02:27:57 +08:00
|
|
|
goto enospc;
|
|
|
|
}
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = balance_node_right(trans, fs_info, mid, left);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (wret < 0) {
|
2007-03-02 04:16:26 +08:00
|
|
|
ret = wret;
|
2007-06-23 02:16:25 +08:00
|
|
|
goto enospc;
|
|
|
|
}
|
2008-04-25 02:42:46 +08:00
|
|
|
if (wret == 1) {
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = push_node_left(trans, fs_info, left, mid, 1);
|
2008-04-25 02:42:46 +08:00
|
|
|
if (wret < 0)
|
|
|
|
ret = wret;
|
|
|
|
}
|
2007-03-02 04:16:26 +08:00
|
|
|
BUG_ON(wret == 1);
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_nritems(mid) == 0) {
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, mid);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(mid);
|
2013-04-16 13:18:22 +08:00
|
|
|
del_ptr(root, path, level + 1, pslot);
|
2010-05-16 22:46:25 +08:00
|
|
|
root_sub_used(root, mid->len);
|
2012-05-16 23:04:52 +08:00
|
|
|
btrfs_free_tree_block(trans, root, mid, 0, 1);
|
2012-03-10 05:01:49 +08:00
|
|
|
free_extent_buffer_stale(mid);
|
2010-05-16 22:46:25 +08:00
|
|
|
mid = NULL;
|
2007-03-02 04:16:26 +08:00
|
|
|
} else {
|
|
|
|
/* update the parent key to reflect our changes */
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key mid_key;
|
|
|
|
btrfs_node_key(mid, &mid_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_set_node_key(fs_info, parent, pslot, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(parent, &mid_key, pslot);
|
|
|
|
btrfs_mark_buffer_dirty(parent);
|
2007-03-02 04:16:26 +08:00
|
|
|
}
|
2007-03-02 01:04:21 +08:00
|
|
|
|
2007-03-02 04:16:26 +08:00
|
|
|
/* update the path */
|
2007-10-16 04:14:19 +08:00
|
|
|
if (left) {
|
|
|
|
if (btrfs_header_nritems(left) > orig_slot) {
|
|
|
|
extent_buffer_get(left);
|
2008-06-26 04:01:30 +08:00
|
|
|
/* left was locked after cow */
|
2007-10-16 04:14:19 +08:00
|
|
|
path->nodes[level] = left;
|
2007-03-02 01:04:21 +08:00
|
|
|
path->slots[level + 1] -= 1;
|
|
|
|
path->slots[level] = orig_slot;
|
2008-06-26 04:01:30 +08:00
|
|
|
if (mid) {
|
|
|
|
btrfs_tree_unlock(mid);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(mid);
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-03-02 01:04:21 +08:00
|
|
|
} else {
|
2007-10-16 04:14:19 +08:00
|
|
|
orig_slot -= btrfs_header_nritems(left);
|
2007-03-02 01:04:21 +08:00
|
|
|
path->slots[level] = orig_slot;
|
|
|
|
}
|
|
|
|
}
|
2007-03-02 04:16:26 +08:00
|
|
|
/* double check we haven't messed things up */
|
2007-03-23 00:13:20 +08:00
|
|
|
if (orig_ptr !=
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_node_blockptr(path->nodes[level], path->slots[level]))
|
2007-03-02 04:16:26 +08:00
|
|
|
BUG();
|
2007-06-23 02:16:25 +08:00
|
|
|
enospc:
|
2008-06-26 04:01:30 +08:00
|
|
|
if (right) {
|
|
|
|
btrfs_tree_unlock(right);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(right);
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
|
|
|
if (left) {
|
|
|
|
if (path->nodes[level] != left)
|
|
|
|
btrfs_tree_unlock(left);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(left);
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-03-02 01:04:21 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/* Node balancing for insertion. Here we only split or push nodes around
|
|
|
|
* when they are completely full. This is also done top down, so we
|
|
|
|
* have to be pessimistic.
|
|
|
|
*/
|
2009-01-06 10:25:51 +08:00
|
|
|
static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
|
2008-01-03 23:01:48 +08:00
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, int level)
|
2007-04-21 01:16:02 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *right = NULL;
|
|
|
|
struct extent_buffer *mid;
|
|
|
|
struct extent_buffer *left = NULL;
|
|
|
|
struct extent_buffer *parent = NULL;
|
2007-04-21 01:16:02 +08:00
|
|
|
int ret = 0;
|
|
|
|
int wret;
|
|
|
|
int pslot;
|
|
|
|
int orig_slot = path->slots[level];
|
|
|
|
|
|
|
|
if (level == 0)
|
|
|
|
return 1;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
mid = path->nodes[level];
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(btrfs_header_generation(mid) != trans->transid);
|
2007-04-21 01:16:02 +08:00
|
|
|
|
2011-09-06 16:55:34 +08:00
|
|
|
if (level < BTRFS_MAX_LEVEL - 1) {
|
2007-10-16 04:14:19 +08:00
|
|
|
parent = path->nodes[level + 1];
|
2011-09-06 16:55:34 +08:00
|
|
|
pslot = path->slots[level + 1];
|
|
|
|
}
|
2007-04-21 01:16:02 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (!parent)
|
2007-04-21 01:16:02 +08:00
|
|
|
return 1;
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
left = read_node_slot(fs_info, parent, pslot - 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(left))
|
|
|
|
left = NULL;
|
2007-04-21 01:16:02 +08:00
|
|
|
|
|
|
|
/* first, try to make some room in the middle buffer */
|
2007-10-16 04:14:19 +08:00
|
|
|
if (left) {
|
2007-04-21 01:16:02 +08:00
|
|
|
u32 left_nr;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
|
|
|
btrfs_tree_lock(left);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(left);
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
left_nr = btrfs_header_nritems(left);
|
2016-06-23 06:54:23 +08:00
|
|
|
if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
|
2007-04-21 01:48:57 +08:00
|
|
|
wret = 1;
|
|
|
|
} else {
|
2007-10-16 04:14:19 +08:00
|
|
|
ret = btrfs_cow_block(trans, root, left, parent,
|
2009-03-13 22:24:59 +08:00
|
|
|
pslot - 1, &left);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (ret)
|
|
|
|
wret = 1;
|
|
|
|
else {
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = push_node_left(trans, fs_info,
|
2008-04-24 22:54:32 +08:00
|
|
|
left, mid, 0);
|
2007-06-23 02:16:25 +08:00
|
|
|
}
|
2007-04-21 01:48:57 +08:00
|
|
|
}
|
2007-04-21 01:16:02 +08:00
|
|
|
if (wret < 0)
|
|
|
|
ret = wret;
|
|
|
|
if (wret == 0) {
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-04-21 01:16:02 +08:00
|
|
|
orig_slot += left_nr;
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_node_key(mid, &disk_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_set_node_key(fs_info, parent, pslot, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(parent, &disk_key, pslot);
|
|
|
|
btrfs_mark_buffer_dirty(parent);
|
|
|
|
if (btrfs_header_nritems(left) > orig_slot) {
|
|
|
|
path->nodes[level] = left;
|
2007-04-21 01:16:02 +08:00
|
|
|
path->slots[level + 1] -= 1;
|
|
|
|
path->slots[level] = orig_slot;
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(mid);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(mid);
|
2007-04-21 01:16:02 +08:00
|
|
|
} else {
|
|
|
|
orig_slot -=
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_header_nritems(left);
|
2007-04-21 01:16:02 +08:00
|
|
|
path->slots[level] = orig_slot;
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(left);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(left);
|
2007-04-21 01:16:02 +08:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(left);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(left);
|
2007-04-21 01:16:02 +08:00
|
|
|
}
|
2016-06-23 06:54:24 +08:00
|
|
|
right = read_node_slot(fs_info, parent, pslot + 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(right))
|
|
|
|
right = NULL;
|
2007-04-21 01:16:02 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* then try to empty the right most buffer into the middle
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
if (right) {
|
2007-04-21 01:48:57 +08:00
|
|
|
u32 right_nr;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_lock(right);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_lock_blocking(right);
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
right_nr = btrfs_header_nritems(right);
|
2016-06-23 06:54:23 +08:00
|
|
|
if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
|
2007-04-21 01:48:57 +08:00
|
|
|
wret = 1;
|
|
|
|
} else {
|
2007-10-16 04:14:19 +08:00
|
|
|
ret = btrfs_cow_block(trans, root, right,
|
|
|
|
parent, pslot + 1,
|
2009-03-13 22:24:59 +08:00
|
|
|
&right);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (ret)
|
|
|
|
wret = 1;
|
|
|
|
else {
|
2016-06-23 06:54:24 +08:00
|
|
|
wret = balance_node_right(trans, fs_info,
|
2007-10-16 04:14:19 +08:00
|
|
|
right, mid);
|
2007-06-23 02:16:25 +08:00
|
|
|
}
|
2007-04-21 01:48:57 +08:00
|
|
|
}
|
2007-04-21 01:16:02 +08:00
|
|
|
if (wret < 0)
|
|
|
|
ret = wret;
|
|
|
|
if (wret == 0) {
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
|
|
|
|
btrfs_node_key(right, &disk_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_set_node_key(fs_info, parent,
|
2012-10-19 20:52:15 +08:00
|
|
|
pslot + 1, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(parent, &disk_key, pslot + 1);
|
|
|
|
btrfs_mark_buffer_dirty(parent);
|
|
|
|
|
|
|
|
if (btrfs_header_nritems(mid) <= orig_slot) {
|
|
|
|
path->nodes[level] = right;
|
2007-04-21 01:16:02 +08:00
|
|
|
path->slots[level + 1] += 1;
|
|
|
|
path->slots[level] = orig_slot -
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_header_nritems(mid);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(mid);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(mid);
|
2007-04-21 01:16:02 +08:00
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(right);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(right);
|
2007-04-21 01:16:02 +08:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(right);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(right);
|
2007-04-21 01:16:02 +08:00
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2007-08-08 03:52:22 +08:00
|
|
|
/*
|
2008-09-30 03:18:18 +08:00
|
|
|
* readahead one full node of leaves, finding things that are close
|
|
|
|
* to the block in 'slot', and triggering ra on them.
|
2007-08-08 03:52:22 +08:00
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
static void reada_for_search(struct btrfs_fs_info *fs_info,
|
2009-04-03 22:14:18 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
int level, int slot, u64 objectid)
|
2007-08-08 03:52:22 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *node;
|
2007-12-22 05:24:26 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-08-08 03:52:22 +08:00
|
|
|
u32 nritems;
|
|
|
|
u64 search;
|
2009-01-22 22:23:10 +08:00
|
|
|
u64 target;
|
2007-10-16 04:17:34 +08:00
|
|
|
u64 nread = 0;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *eb;
|
2007-10-16 04:17:34 +08:00
|
|
|
u32 nr;
|
|
|
|
u32 blocksize;
|
|
|
|
u32 nscan = 0;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2007-10-16 04:22:39 +08:00
|
|
|
if (level != 1)
|
2007-08-08 04:15:09 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
if (!path->nodes[level])
|
2007-08-08 03:52:22 +08:00
|
|
|
return;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
node = path->nodes[level];
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2007-08-08 03:52:22 +08:00
|
|
|
search = btrfs_node_blockptr(node, slot);
|
2016-06-23 06:54:23 +08:00
|
|
|
blocksize = fs_info->nodesize;
|
|
|
|
eb = find_extent_buffer(fs_info, search);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (eb) {
|
|
|
|
free_extent_buffer(eb);
|
2007-08-08 03:52:22 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-01-22 22:23:10 +08:00
|
|
|
target = search;
|
2007-10-16 04:17:34 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(node);
|
2007-10-16 04:17:34 +08:00
|
|
|
nr = slot;
|
2011-06-09 02:36:54 +08:00
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2015-11-27 23:31:35 +08:00
|
|
|
if (path->reada == READA_BACK) {
|
2007-10-16 04:17:34 +08:00
|
|
|
if (nr == 0)
|
|
|
|
break;
|
|
|
|
nr--;
|
2015-11-27 23:31:35 +08:00
|
|
|
} else if (path->reada == READA_FORWARD) {
|
2007-10-16 04:17:34 +08:00
|
|
|
nr++;
|
|
|
|
if (nr >= nritems)
|
|
|
|
break;
|
2007-08-08 03:52:22 +08:00
|
|
|
}
|
2015-11-27 23:31:35 +08:00
|
|
|
if (path->reada == READA_BACK && objectid) {
|
2007-12-22 05:24:26 +08:00
|
|
|
btrfs_node_key(node, &disk_key, nr);
|
|
|
|
if (btrfs_disk_key_objectid(&disk_key) != objectid)
|
|
|
|
break;
|
|
|
|
}
|
2007-10-16 04:17:34 +08:00
|
|
|
search = btrfs_node_blockptr(node, nr);
|
2009-01-22 22:23:10 +08:00
|
|
|
if ((search <= target && target - search <= 65536) ||
|
|
|
|
(search > target && search - target <= 65536)) {
|
2016-06-23 06:54:24 +08:00
|
|
|
readahead_tree_block(fs_info, search);
|
2007-10-16 04:17:34 +08:00
|
|
|
nread += blocksize;
|
|
|
|
}
|
|
|
|
nscan++;
|
2009-01-22 22:23:10 +08:00
|
|
|
if ((nread > 65536 || nscan > 32))
|
2007-10-16 04:17:34 +08:00
|
|
|
break;
|
2007-08-08 03:52:22 +08:00
|
|
|
}
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
|
2013-06-18 02:23:02 +08:00
|
|
|
struct btrfs_path *path, int level)
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
{
|
|
|
|
int slot;
|
|
|
|
int nritems;
|
|
|
|
struct extent_buffer *parent;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
u64 gen;
|
|
|
|
u64 block1 = 0;
|
|
|
|
u64 block2 = 0;
|
|
|
|
|
2009-04-21 03:50:10 +08:00
|
|
|
parent = path->nodes[level + 1];
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
if (!parent)
|
2013-06-18 02:23:02 +08:00
|
|
|
return;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
|
|
|
nritems = btrfs_header_nritems(parent);
|
2009-04-21 03:50:10 +08:00
|
|
|
slot = path->slots[level + 1];
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
|
|
|
if (slot > 0) {
|
|
|
|
block1 = btrfs_node_blockptr(parent, slot - 1);
|
|
|
|
gen = btrfs_node_ptr_generation(parent, slot - 1);
|
2016-06-23 06:54:23 +08:00
|
|
|
eb = find_extent_buffer(fs_info, block1);
|
2012-05-06 19:23:47 +08:00
|
|
|
/*
|
|
|
|
* if we get -eagain from btrfs_buffer_uptodate, we
|
|
|
|
* don't want to return eagain here. That will loop
|
|
|
|
* forever
|
|
|
|
*/
|
|
|
|
if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
block1 = 0;
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
}
|
2009-04-21 03:50:10 +08:00
|
|
|
if (slot + 1 < nritems) {
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
block2 = btrfs_node_blockptr(parent, slot + 1);
|
|
|
|
gen = btrfs_node_ptr_generation(parent, slot + 1);
|
2016-06-23 06:54:23 +08:00
|
|
|
eb = find_extent_buffer(fs_info, block2);
|
2012-05-06 19:23:47 +08:00
|
|
|
if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
block2 = 0;
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
}
|
2009-04-21 03:50:10 +08:00
|
|
|
|
2013-06-18 02:23:02 +08:00
|
|
|
if (block1)
|
2016-06-23 06:54:24 +08:00
|
|
|
readahead_tree_block(fs_info, block1);
|
2013-06-18 02:23:02 +08:00
|
|
|
if (block2)
|
2016-06-23 06:54:24 +08:00
|
|
|
readahead_tree_block(fs_info, block2);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
2009-01-06 10:25:51 +08:00
|
|
|
* when we walk down the tree, it is usually safe to unlock the higher layers
|
|
|
|
* in the tree. The exceptions are when our path goes through slot 0, because
|
|
|
|
* operations on the tree might require changing key pointers higher up in the
|
|
|
|
* tree.
|
2008-09-30 03:18:18 +08:00
|
|
|
*
|
2009-01-06 10:25:51 +08:00
|
|
|
* callers might also have set path->keep_locks, which tells this code to keep
|
|
|
|
* the lock if the path points to the last slot in the block. This is part of
|
|
|
|
* walking through the tree, and selecting the next slot in the higher block.
|
2008-09-30 03:18:18 +08:00
|
|
|
*
|
2009-01-06 10:25:51 +08:00
|
|
|
* lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
|
|
|
|
* if lowest_unlock is 1, level 0 won't be unlocked
|
2008-09-30 03:18:18 +08:00
|
|
|
*/
|
2008-09-06 04:13:11 +08:00
|
|
|
static noinline void unlock_up(struct btrfs_path *path, int level,
|
2012-03-20 03:54:38 +08:00
|
|
|
int lowest_unlock, int min_write_lock_level,
|
|
|
|
int *write_lock_level)
|
2008-06-26 04:01:30 +08:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
int skip_level = level;
|
2008-06-26 04:01:30 +08:00
|
|
|
int no_skips = 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
struct extent_buffer *t;
|
|
|
|
|
|
|
|
for (i = level; i < BTRFS_MAX_LEVEL; i++) {
|
|
|
|
if (!path->nodes[i])
|
|
|
|
break;
|
|
|
|
if (!path->locks[i])
|
|
|
|
break;
|
2008-06-26 04:01:30 +08:00
|
|
|
if (!no_skips && path->slots[i] == 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
skip_level = i + 1;
|
|
|
|
continue;
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
if (!no_skips && path->keep_locks) {
|
2008-06-26 04:01:30 +08:00
|
|
|
u32 nritems;
|
|
|
|
t = path->nodes[i];
|
|
|
|
nritems = btrfs_header_nritems(t);
|
2008-06-26 04:01:30 +08:00
|
|
|
if (nritems < 1 || path->slots[i] >= nritems - 1) {
|
2008-06-26 04:01:30 +08:00
|
|
|
skip_level = i + 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
if (skip_level < i && i >= lowest_unlock)
|
|
|
|
no_skips = 1;
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
t = path->nodes[i];
|
|
|
|
if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_unlock_rw(t, path->locks[i]);
|
2008-06-26 04:01:30 +08:00
|
|
|
path->locks[i] = 0;
|
2012-03-20 03:54:38 +08:00
|
|
|
if (write_lock_level &&
|
|
|
|
i > min_write_lock_level &&
|
|
|
|
i <= *write_lock_level) {
|
|
|
|
*write_lock_level = i - 1;
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
/*
|
|
|
|
* This releases any locks held in the path starting at level and
|
|
|
|
* going all the way up to the root.
|
|
|
|
*
|
|
|
|
* btrfs_search_slot will keep the lock held on higher nodes in a few
|
|
|
|
* corner cases, such as COW of the block at slot zero in the node. This
|
|
|
|
* ignores those rules, and it should only be called when there are no
|
|
|
|
* more updates to be done higher up in the tree.
|
|
|
|
*/
|
|
|
|
noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2013-04-06 04:51:15 +08:00
|
|
|
if (path->keep_locks)
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
for (i = level; i < BTRFS_MAX_LEVEL; i++) {
|
|
|
|
if (!path->nodes[i])
|
2009-02-04 22:31:42 +08:00
|
|
|
continue;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
if (!path->locks[i])
|
2009-02-04 22:31:42 +08:00
|
|
|
continue;
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
path->locks[i] = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
/*
|
|
|
|
* helper function for btrfs_search_slot. The goal is to find a block
|
|
|
|
* in cache without setting the path to blocking. If we find the block
|
|
|
|
* we return zero and the path is unchanged.
|
|
|
|
*
|
|
|
|
* If we can't find the block, we set the path blocking and do some
|
|
|
|
* reada. -EAGAIN is returned and the search must be repeated.
|
|
|
|
*/
|
|
|
|
static int
|
2017-01-31 04:23:42 +08:00
|
|
|
read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
|
|
|
|
struct extent_buffer **eb_ret, int level, int slot,
|
2017-02-11 01:44:32 +08:00
|
|
|
const struct btrfs_key *key)
|
2009-04-03 22:14:18 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2009-04-03 22:14:18 +08:00
|
|
|
u64 blocknr;
|
|
|
|
u64 gen;
|
|
|
|
struct extent_buffer *b = *eb_ret;
|
|
|
|
struct extent_buffer *tmp;
|
2009-05-15 01:24:30 +08:00
|
|
|
int ret;
|
2009-04-03 22:14:18 +08:00
|
|
|
|
|
|
|
blocknr = btrfs_node_blockptr(b, slot);
|
|
|
|
gen = btrfs_node_ptr_generation(b, slot);
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
tmp = find_extent_buffer(fs_info, blocknr);
|
2010-10-24 23:01:27 +08:00
|
|
|
if (tmp) {
|
2012-05-06 19:23:47 +08:00
|
|
|
/* first we do an atomic uptodate check */
|
2013-06-18 01:44:48 +08:00
|
|
|
if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
|
|
|
|
*eb_ret = tmp;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* the pages were up to date, but we failed
|
|
|
|
* the generation number check. Do a full
|
|
|
|
* read for the generation number that is correct.
|
|
|
|
* We must do this without dropping locks so
|
|
|
|
* we can trust our generation number
|
|
|
|
*/
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
|
|
|
|
/* now we're allowed to do a blocking uptodate check */
|
|
|
|
ret = btrfs_read_buffer(tmp, gen);
|
|
|
|
if (!ret) {
|
|
|
|
*eb_ret = tmp;
|
|
|
|
return 0;
|
2010-10-24 23:01:27 +08:00
|
|
|
}
|
2013-06-18 01:44:48 +08:00
|
|
|
free_extent_buffer(tmp);
|
|
|
|
btrfs_release_path(p);
|
|
|
|
return -EIO;
|
2009-04-03 22:14:18 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* reduce lock contention at high levels
|
|
|
|
* of the btree by dropping locks before
|
2009-05-15 01:24:30 +08:00
|
|
|
* we read. Don't release the lock on the current
|
|
|
|
* level because we need to walk this node to figure
|
|
|
|
* out which blocks to read.
|
2009-04-03 22:14:18 +08:00
|
|
|
*/
|
2009-04-21 03:50:10 +08:00
|
|
|
btrfs_unlock_up_safe(p, level + 1);
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
|
2010-10-24 23:01:27 +08:00
|
|
|
free_extent_buffer(tmp);
|
2015-11-27 23:31:35 +08:00
|
|
|
if (p->reada != READA_NONE)
|
2016-06-23 06:54:24 +08:00
|
|
|
reada_for_search(fs_info, p, level, slot, key->objectid);
|
2009-04-03 22:14:18 +08:00
|
|
|
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(p);
|
2009-05-15 01:24:30 +08:00
|
|
|
|
|
|
|
ret = -EAGAIN;
|
2016-06-23 06:54:24 +08:00
|
|
|
tmp = read_tree_block(fs_info, blocknr, 0);
|
2015-05-25 17:30:15 +08:00
|
|
|
if (!IS_ERR(tmp)) {
|
2009-05-15 01:24:30 +08:00
|
|
|
/*
|
|
|
|
* If the read above didn't mark this buffer up to date,
|
|
|
|
* it will never end up being up to date. Set ret to EIO now
|
|
|
|
* and give up so that our caller doesn't loop forever
|
|
|
|
* on our EAGAINs.
|
|
|
|
*/
|
2012-05-06 19:23:47 +08:00
|
|
|
if (!btrfs_buffer_uptodate(tmp, 0, 0))
|
2009-05-15 01:24:30 +08:00
|
|
|
ret = -EIO;
|
2009-04-03 22:14:18 +08:00
|
|
|
free_extent_buffer(tmp);
|
2016-06-07 03:01:23 +08:00
|
|
|
} else {
|
|
|
|
ret = PTR_ERR(tmp);
|
2009-05-15 01:24:30 +08:00
|
|
|
}
|
|
|
|
return ret;
|
2009-04-03 22:14:18 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* helper function for btrfs_search_slot. This does all of the checks
|
|
|
|
* for node-level blocks and does any balancing required based on
|
|
|
|
* the ins_len.
|
|
|
|
*
|
|
|
|
* If no extra work was required, zero is returned. If we had to
|
|
|
|
* drop the path, -EAGAIN is returned and btrfs_search_slot must
|
|
|
|
* start over
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
setup_nodes_for_search(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root, struct btrfs_path *p,
|
2011-07-17 03:23:14 +08:00
|
|
|
struct extent_buffer *b, int level, int ins_len,
|
|
|
|
int *write_lock_level)
|
2009-04-03 22:14:18 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2009-04-03 22:14:18 +08:00
|
|
|
int ret;
|
2016-06-23 06:54:23 +08:00
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
|
2009-04-03 22:14:18 +08:00
|
|
|
int sret;
|
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
if (*write_lock_level < level + 1) {
|
|
|
|
*write_lock_level = level + 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
btrfs_set_path_blocking(p);
|
2016-06-23 06:54:24 +08:00
|
|
|
reada_for_balance(fs_info, p, level);
|
2009-04-03 22:14:18 +08:00
|
|
|
sret = split_node(trans, root, p, level);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_path_blocking(p, NULL, 0);
|
2009-04-03 22:14:18 +08:00
|
|
|
|
|
|
|
BUG_ON(sret > 0);
|
|
|
|
if (sret) {
|
|
|
|
ret = sret;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
b = p->nodes[level];
|
|
|
|
} else if (ins_len < 0 && btrfs_header_nritems(b) <
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
|
2009-04-03 22:14:18 +08:00
|
|
|
int sret;
|
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
if (*write_lock_level < level + 1) {
|
|
|
|
*write_lock_level = level + 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
btrfs_set_path_blocking(p);
|
2016-06-23 06:54:24 +08:00
|
|
|
reada_for_balance(fs_info, p, level);
|
2009-04-03 22:14:18 +08:00
|
|
|
sret = balance_level(trans, root, p, level);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_path_blocking(p, NULL, 0);
|
2009-04-03 22:14:18 +08:00
|
|
|
|
|
|
|
if (sret) {
|
|
|
|
ret = sret;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
b = p->nodes[level];
|
|
|
|
if (!b) {
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(p);
|
2009-04-03 22:14:18 +08:00
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
BUG_ON(btrfs_header_nritems(b) == 1);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
again:
|
|
|
|
ret = -EAGAIN;
|
|
|
|
done:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-08-30 22:46:43 +08:00
|
|
|
static void key_search_validate(struct extent_buffer *b,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *key,
|
2013-08-30 22:46:43 +08:00
|
|
|
int level)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_BTRFS_ASSERT
|
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, key);
|
|
|
|
|
|
|
|
if (level == 0)
|
|
|
|
ASSERT(!memcmp_extent_buffer(b, &disk_key,
|
|
|
|
offsetof(struct btrfs_leaf, items[0].key),
|
|
|
|
sizeof(disk_key)));
|
|
|
|
else
|
|
|
|
ASSERT(!memcmp_extent_buffer(b, &disk_key,
|
|
|
|
offsetof(struct btrfs_node, ptrs[0].key),
|
|
|
|
sizeof(disk_key)));
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2017-01-18 15:24:37 +08:00
|
|
|
static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
|
2013-08-30 22:46:43 +08:00
|
|
|
int level, int *prev_cmp, int *slot)
|
|
|
|
{
|
|
|
|
if (*prev_cmp != 0) {
|
|
|
|
*prev_cmp = bin_search(b, key, level, slot);
|
|
|
|
return *prev_cmp;
|
|
|
|
}
|
|
|
|
|
|
|
|
key_search_validate(b, key, level);
|
|
|
|
*slot = 0;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-01-03 01:45:16 +08:00
|
|
|
int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
|
2013-11-05 11:33:33 +08:00
|
|
|
u64 iobjectid, u64 ioff, u8 key_type,
|
|
|
|
struct btrfs_key *found_key)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
struct btrfs_key key;
|
|
|
|
struct extent_buffer *eb;
|
2015-01-03 01:45:16 +08:00
|
|
|
|
|
|
|
ASSERT(path);
|
2015-01-03 02:36:14 +08:00
|
|
|
ASSERT(found_key);
|
2013-11-05 11:33:33 +08:00
|
|
|
|
|
|
|
key.type = key_type;
|
|
|
|
key.objectid = iobjectid;
|
|
|
|
key.offset = ioff;
|
|
|
|
|
|
|
|
ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
|
2015-01-03 02:36:14 +08:00
|
|
|
if (ret < 0)
|
2013-11-05 11:33:33 +08:00
|
|
|
return ret;
|
|
|
|
|
|
|
|
eb = path->nodes[0];
|
|
|
|
if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
|
|
|
|
ret = btrfs_next_leaf(fs_root, path);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
eb = path->nodes[0];
|
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
|
|
|
|
if (found_key->type != key.type ||
|
|
|
|
found_key->objectid != key.objectid)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* look for key in the tree. path is filled in with nodes along the way
|
|
|
|
* if key is found, we return zero and you can find the item in the leaf
|
|
|
|
* level of the path (level 0)
|
|
|
|
*
|
|
|
|
* If the key isn't found, the path points to the slot where it should
|
2007-03-01 05:35:06 +08:00
|
|
|
* be inserted, and 1 is returned. If there are other errors during the
|
|
|
|
* search a negative error number is returned.
|
2007-02-25 02:39:08 +08:00
|
|
|
*
|
|
|
|
* if ins_len > 0, nodes and leaves will be split as we walk down the
|
|
|
|
* tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
|
|
|
|
* possible)
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
|
|
|
|
const struct btrfs_key *key, struct btrfs_path *p,
|
|
|
|
int ins_len, int cow)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *b;
|
2007-01-27 04:51:26 +08:00
|
|
|
int slot;
|
|
|
|
int ret;
|
2009-07-22 21:59:00 +08:00
|
|
|
int err;
|
2007-01-27 04:51:26 +08:00
|
|
|
int level;
|
2008-06-26 04:01:30 +08:00
|
|
|
int lowest_unlock = 1;
|
2011-07-17 03:23:14 +08:00
|
|
|
int root_lock;
|
|
|
|
/* everything at write_lock_level or lower must be write locked */
|
|
|
|
int write_lock_level = 0;
|
2007-08-08 03:52:19 +08:00
|
|
|
u8 lowest_level = 0;
|
2012-03-20 03:54:38 +08:00
|
|
|
int min_write_lock_level;
|
2013-08-30 22:46:43 +08:00
|
|
|
int prev_cmp;
|
2007-08-08 03:52:19 +08:00
|
|
|
|
2007-08-08 04:15:09 +08:00
|
|
|
lowest_level = p->lowest_level;
|
2008-10-02 07:05:46 +08:00
|
|
|
WARN_ON(lowest_level && ins_len > 0);
|
2007-03-30 20:47:31 +08:00
|
|
|
WARN_ON(p->nodes[0] != NULL);
|
2013-12-23 19:53:02 +08:00
|
|
|
BUG_ON(!cow && ins_len);
|
Btrfs: nuke fs wide allocation mutex V2
This patch removes the giant fs_info->alloc_mutex and replaces it with a bunch
of little locks.
There is now a pinned_mutex, which is used when messing with the pinned_extents
extent io tree, and the extent_ins_mutex which is used with the pending_del and
extent_ins extent io trees.
The locking for the extent tree stuff was inspired by a patch that Yan Zheng
wrote to fix a race condition, I cleaned it up some and changed the locking
around a little bit, but the idea remains the same. Basically instead of
holding the extent_ins_mutex throughout the processing of an extent on the
extent_ins or pending_del trees, we just hold it while we're searching and when
we clear the bits on those trees, and lock the extent for the duration of the
operations on the extent.
Also to keep from getting hung up waiting to lock an extent, I've added a
try_lock_extent so if we cannot lock the extent, move on to the next one in the
tree and we'll come back to that one. I have tested this heavily and it does
not appear to break anything. This has to be applied on top of my
find_free_extent redo patch.
I tested this patch on top of Yan's space reblancing code and it worked fine.
The only thing that has changed since the last version is I pulled out all my
debugging stuff, apparently I forgot to run guilt refresh before I sent the
last patch out. Thank you,
Signed-off-by: Josef Bacik <jbacik@redhat.com>
2008-10-30 02:49:05 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
if (ins_len < 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
lowest_unlock = 2;
|
2008-08-02 03:11:20 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
/* when we are removing items, we might have to go up to level
|
|
|
|
* two as we update tree pointers Make sure we keep write
|
|
|
|
* for those levels as well
|
|
|
|
*/
|
|
|
|
write_lock_level = 2;
|
|
|
|
} else if (ins_len > 0) {
|
|
|
|
/*
|
|
|
|
* for inserting items, make sure we have a write lock on
|
|
|
|
* level 1 so we can update keys
|
|
|
|
*/
|
|
|
|
write_lock_level = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!cow)
|
|
|
|
write_lock_level = -1;
|
|
|
|
|
2013-04-06 04:51:15 +08:00
|
|
|
if (cow && (p->keep_locks || p->lowest_level))
|
2011-07-17 03:23:14 +08:00
|
|
|
write_lock_level = BTRFS_MAX_LEVEL;
|
|
|
|
|
2012-03-20 03:54:38 +08:00
|
|
|
min_write_lock_level = write_lock_level;
|
|
|
|
|
2007-03-02 01:04:21 +08:00
|
|
|
again:
|
2013-08-30 22:46:43 +08:00
|
|
|
prev_cmp = -1;
|
2011-07-17 03:23:14 +08:00
|
|
|
/*
|
|
|
|
* we try very hard to do read locks on the root
|
|
|
|
*/
|
|
|
|
root_lock = BTRFS_READ_LOCK;
|
|
|
|
level = 0;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (p->search_commit_root) {
|
2011-07-17 03:23:14 +08:00
|
|
|
/*
|
|
|
|
* the commit roots are read only
|
|
|
|
* so we always do read locks
|
|
|
|
*/
|
2014-03-29 05:16:01 +08:00
|
|
|
if (p->need_commit_sem)
|
2016-06-23 06:54:23 +08:00
|
|
|
down_read(&fs_info->commit_root_sem);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
b = root->commit_root;
|
|
|
|
extent_buffer_get(b);
|
2011-07-17 03:23:14 +08:00
|
|
|
level = btrfs_header_level(b);
|
2014-03-29 05:16:01 +08:00
|
|
|
if (p->need_commit_sem)
|
2016-06-23 06:54:23 +08:00
|
|
|
up_read(&fs_info->commit_root_sem);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (!p->skip_locking)
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_read_lock(b);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
} else {
|
2011-07-17 03:23:14 +08:00
|
|
|
if (p->skip_locking) {
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
b = btrfs_root_node(root);
|
2011-07-17 03:23:14 +08:00
|
|
|
level = btrfs_header_level(b);
|
|
|
|
} else {
|
|
|
|
/* we don't know the level of the root node
|
|
|
|
* until we actually have it read locked
|
|
|
|
*/
|
|
|
|
b = btrfs_read_lock_root_node(root);
|
|
|
|
level = btrfs_header_level(b);
|
|
|
|
if (level <= write_lock_level) {
|
|
|
|
/* whoops, must trade for write lock */
|
|
|
|
btrfs_tree_read_unlock(b);
|
|
|
|
free_extent_buffer(b);
|
|
|
|
b = btrfs_lock_root_node(root);
|
|
|
|
root_lock = BTRFS_WRITE_LOCK;
|
|
|
|
|
|
|
|
/* the level might have changed, check again */
|
|
|
|
level = btrfs_header_level(b);
|
|
|
|
}
|
|
|
|
}
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
2011-07-17 03:23:14 +08:00
|
|
|
p->nodes[level] = b;
|
|
|
|
if (!p->skip_locking)
|
|
|
|
p->locks[level] = root_lock;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2007-02-02 22:18:22 +08:00
|
|
|
while (b) {
|
2007-10-16 04:14:19 +08:00
|
|
|
level = btrfs_header_level(b);
|
2008-08-02 03:11:20 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* setup the path here so we can release it under lock
|
|
|
|
* contention with the cow code
|
|
|
|
*/
|
2007-03-03 05:08:05 +08:00
|
|
|
if (cow) {
|
2009-04-03 22:14:18 +08:00
|
|
|
/*
|
|
|
|
* if we don't really need to cow this block
|
|
|
|
* then we don't want to set the path blocking,
|
|
|
|
* so we test it here
|
|
|
|
*/
|
2016-06-08 12:36:38 +08:00
|
|
|
if (!should_cow_block(trans, root, b)) {
|
|
|
|
trans->dirty = true;
|
2008-08-02 03:11:20 +08:00
|
|
|
goto cow_done;
|
2016-06-08 12:36:38 +08:00
|
|
|
}
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
/*
|
|
|
|
* must have write locks on this node and the
|
|
|
|
* parent
|
|
|
|
*/
|
2012-11-08 02:44:13 +08:00
|
|
|
if (level > write_lock_level ||
|
|
|
|
(level + 1 > write_lock_level &&
|
|
|
|
level + 1 < BTRFS_MAX_LEVEL &&
|
|
|
|
p->nodes[level + 1])) {
|
2011-07-17 03:23:14 +08:00
|
|
|
write_lock_level = level + 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
2014-07-29 02:37:17 +08:00
|
|
|
btrfs_set_path_blocking(p);
|
2009-07-22 21:59:00 +08:00
|
|
|
err = btrfs_cow_block(trans, root, b,
|
|
|
|
p->nodes[level + 1],
|
|
|
|
p->slots[level + 1], &b);
|
|
|
|
if (err) {
|
|
|
|
ret = err;
|
2008-08-02 03:11:20 +08:00
|
|
|
goto done;
|
2007-06-23 02:16:25 +08:00
|
|
|
}
|
2007-03-03 05:08:05 +08:00
|
|
|
}
|
2008-08-02 03:11:20 +08:00
|
|
|
cow_done:
|
2007-02-02 22:18:22 +08:00
|
|
|
p->nodes[level] = b;
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_path_blocking(p, NULL, 0);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* we have a lock on b and as long as we aren't changing
|
|
|
|
* the tree, there is no way to for the items in b to change.
|
|
|
|
* It is safe to drop the lock on our parent before we
|
|
|
|
* go through the expensive btree search on b.
|
|
|
|
*
|
2013-12-23 19:53:02 +08:00
|
|
|
* If we're inserting or deleting (ins_len != 0), then we might
|
|
|
|
* be changing slot zero, which may require changing the parent.
|
|
|
|
* So, we can't drop the lock until after we know which slot
|
|
|
|
* we're operating on.
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
*/
|
2013-12-23 19:53:02 +08:00
|
|
|
if (!ins_len && !p->keep_locks) {
|
|
|
|
int u = level + 1;
|
|
|
|
|
|
|
|
if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
|
|
|
|
btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
|
|
|
|
p->locks[u] = 0;
|
|
|
|
}
|
|
|
|
}
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2013-08-30 22:46:43 +08:00
|
|
|
ret = key_search(b, key, level, &prev_cmp, &slot);
|
2016-06-18 10:16:21 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto done;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (level != 0) {
|
2009-07-22 21:59:00 +08:00
|
|
|
int dec = 0;
|
|
|
|
if (ret && slot > 0) {
|
|
|
|
dec = 1;
|
2007-01-27 04:51:26 +08:00
|
|
|
slot -= 1;
|
2009-07-22 21:59:00 +08:00
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
p->slots[level] = slot;
|
2009-07-22 21:59:00 +08:00
|
|
|
err = setup_nodes_for_search(trans, root, p, b, level,
|
2011-07-17 03:23:14 +08:00
|
|
|
ins_len, &write_lock_level);
|
2009-07-22 21:59:00 +08:00
|
|
|
if (err == -EAGAIN)
|
2009-04-03 22:14:18 +08:00
|
|
|
goto again;
|
2009-07-22 21:59:00 +08:00
|
|
|
if (err) {
|
|
|
|
ret = err;
|
2009-04-03 22:14:18 +08:00
|
|
|
goto done;
|
2009-07-22 21:59:00 +08:00
|
|
|
}
|
2009-04-03 22:14:18 +08:00
|
|
|
b = p->nodes[level];
|
|
|
|
slot = p->slots[level];
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
/*
|
|
|
|
* slot 0 is special, if we change the key
|
|
|
|
* we have to update the parent pointer
|
|
|
|
* which means we must have a write lock
|
|
|
|
* on the parent
|
|
|
|
*/
|
2013-12-23 19:53:02 +08:00
|
|
|
if (slot == 0 && ins_len &&
|
2011-07-17 03:23:14 +08:00
|
|
|
write_lock_level < level + 1) {
|
|
|
|
write_lock_level = level + 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
2012-03-20 03:54:38 +08:00
|
|
|
unlock_up(p, level, lowest_unlock,
|
|
|
|
min_write_lock_level, &write_lock_level);
|
2008-06-26 04:14:04 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (level == lowest_level) {
|
2009-07-22 21:59:00 +08:00
|
|
|
if (dec)
|
|
|
|
p->slots[level]++;
|
2008-09-26 22:05:38 +08:00
|
|
|
goto done;
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2008-05-13 00:59:19 +08:00
|
|
|
|
2017-01-31 04:23:42 +08:00
|
|
|
err = read_block_for_search(root, p, &b, level,
|
2017-02-11 01:44:32 +08:00
|
|
|
slot, key);
|
2009-07-22 21:59:00 +08:00
|
|
|
if (err == -EAGAIN)
|
2009-04-03 22:14:18 +08:00
|
|
|
goto again;
|
2009-07-22 21:59:00 +08:00
|
|
|
if (err) {
|
|
|
|
ret = err;
|
2009-05-15 01:24:30 +08:00
|
|
|
goto done;
|
2009-07-22 21:59:00 +08:00
|
|
|
}
|
2009-05-15 01:24:30 +08:00
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
if (!p->skip_locking) {
|
2011-07-17 03:23:14 +08:00
|
|
|
level = btrfs_header_level(b);
|
|
|
|
if (level <= write_lock_level) {
|
|
|
|
err = btrfs_try_tree_write_lock(b);
|
|
|
|
if (!err) {
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
btrfs_tree_lock(b);
|
|
|
|
btrfs_clear_path_blocking(p, b,
|
|
|
|
BTRFS_WRITE_LOCK);
|
|
|
|
}
|
|
|
|
p->locks[level] = BTRFS_WRITE_LOCK;
|
|
|
|
} else {
|
2014-11-20 02:25:09 +08:00
|
|
|
err = btrfs_tree_read_lock_atomic(b);
|
2011-07-17 03:23:14 +08:00
|
|
|
if (!err) {
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
btrfs_tree_read_lock(b);
|
|
|
|
btrfs_clear_path_blocking(p, b,
|
|
|
|
BTRFS_READ_LOCK);
|
|
|
|
}
|
|
|
|
p->locks[level] = BTRFS_READ_LOCK;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
2011-07-17 03:23:14 +08:00
|
|
|
p->nodes[level] = b;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
} else {
|
|
|
|
p->slots[level] = slot;
|
2008-12-17 23:21:48 +08:00
|
|
|
if (ins_len > 0 &&
|
2016-06-23 06:54:24 +08:00
|
|
|
btrfs_leaf_free_space(fs_info, b) < ins_len) {
|
2011-07-17 03:23:14 +08:00
|
|
|
if (write_lock_level < 1) {
|
|
|
|
write_lock_level = 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_path_blocking(p);
|
2009-07-22 21:59:00 +08:00
|
|
|
err = split_leaf(trans, root, key,
|
|
|
|
p, ins_len, ret == 0);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_path_blocking(p, NULL, 0);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2009-07-22 21:59:00 +08:00
|
|
|
BUG_ON(err > 0);
|
|
|
|
if (err) {
|
|
|
|
ret = err;
|
2008-08-02 03:11:20 +08:00
|
|
|
goto done;
|
|
|
|
}
|
2007-02-23 00:39:13 +08:00
|
|
|
}
|
2008-12-10 22:10:46 +08:00
|
|
|
if (!p->search_for_split)
|
2012-03-20 03:54:38 +08:00
|
|
|
unlock_up(p, level, lowest_unlock,
|
|
|
|
min_write_lock_level, &write_lock_level);
|
2008-08-02 03:11:20 +08:00
|
|
|
goto done;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
}
|
2008-08-02 03:11:20 +08:00
|
|
|
ret = 1;
|
|
|
|
done:
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
/*
|
|
|
|
* we don't really know what they plan on doing with the path
|
|
|
|
* from here on, so for now just mark it as blocking
|
|
|
|
*/
|
2009-03-13 23:00:37 +08:00
|
|
|
if (!p->leave_spinning)
|
|
|
|
btrfs_set_path_blocking(p);
|
2014-11-09 16:38:39 +08:00
|
|
|
if (ret < 0 && !p->skip_release_on_error)
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(p);
|
2008-08-02 03:11:20 +08:00
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2012-05-17 00:25:47 +08:00
|
|
|
/*
|
|
|
|
* Like btrfs_search_slot, this looks for a key in the given tree. It uses the
|
|
|
|
* current state of the tree together with the operations recorded in the tree
|
|
|
|
* modification log to search for the key in a previous version of this tree, as
|
|
|
|
* denoted by the time_seq parameter.
|
|
|
|
*
|
|
|
|
* Naturally, there is no support for insert, delete or cow operations.
|
|
|
|
*
|
|
|
|
* The resulting path and return value will be set up as if we called
|
|
|
|
* btrfs_search_slot at that point in time with ins_len and cow both set to 0.
|
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
|
2012-05-17 00:25:47 +08:00
|
|
|
struct btrfs_path *p, u64 time_seq)
|
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2012-05-17 00:25:47 +08:00
|
|
|
struct extent_buffer *b;
|
|
|
|
int slot;
|
|
|
|
int ret;
|
|
|
|
int err;
|
|
|
|
int level;
|
|
|
|
int lowest_unlock = 1;
|
|
|
|
u8 lowest_level = 0;
|
2013-09-25 02:09:34 +08:00
|
|
|
int prev_cmp = -1;
|
2012-05-17 00:25:47 +08:00
|
|
|
|
|
|
|
lowest_level = p->lowest_level;
|
|
|
|
WARN_ON(p->nodes[0] != NULL);
|
|
|
|
|
|
|
|
if (p->search_commit_root) {
|
|
|
|
BUG_ON(time_seq);
|
|
|
|
return btrfs_search_slot(NULL, root, key, p, 0, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
again:
|
|
|
|
b = get_old_root(root, time_seq);
|
|
|
|
level = btrfs_header_level(b);
|
|
|
|
p->locks[level] = BTRFS_READ_LOCK;
|
|
|
|
|
|
|
|
while (b) {
|
|
|
|
level = btrfs_header_level(b);
|
|
|
|
p->nodes[level] = b;
|
|
|
|
btrfs_clear_path_blocking(p, NULL, 0);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* we have a lock on b and as long as we aren't changing
|
|
|
|
* the tree, there is no way to for the items in b to change.
|
|
|
|
* It is safe to drop the lock on our parent before we
|
|
|
|
* go through the expensive btree search on b.
|
|
|
|
*/
|
|
|
|
btrfs_unlock_up_safe(p, level + 1);
|
|
|
|
|
2013-09-25 02:09:34 +08:00
|
|
|
/*
|
2016-05-20 09:18:45 +08:00
|
|
|
* Since we can unwind ebs we want to do a real search every
|
2013-09-25 02:09:34 +08:00
|
|
|
* time.
|
|
|
|
*/
|
|
|
|
prev_cmp = -1;
|
2013-08-30 22:46:43 +08:00
|
|
|
ret = key_search(b, key, level, &prev_cmp, &slot);
|
2012-05-17 00:25:47 +08:00
|
|
|
|
|
|
|
if (level != 0) {
|
|
|
|
int dec = 0;
|
|
|
|
if (ret && slot > 0) {
|
|
|
|
dec = 1;
|
|
|
|
slot -= 1;
|
|
|
|
}
|
|
|
|
p->slots[level] = slot;
|
|
|
|
unlock_up(p, level, lowest_unlock, 0, NULL);
|
|
|
|
|
|
|
|
if (level == lowest_level) {
|
|
|
|
if (dec)
|
|
|
|
p->slots[level]++;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
2017-01-31 04:23:42 +08:00
|
|
|
err = read_block_for_search(root, p, &b, level,
|
2017-02-11 01:44:32 +08:00
|
|
|
slot, key);
|
2012-05-17 00:25:47 +08:00
|
|
|
if (err == -EAGAIN)
|
|
|
|
goto again;
|
|
|
|
if (err) {
|
|
|
|
ret = err;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
level = btrfs_header_level(b);
|
2014-11-20 02:25:09 +08:00
|
|
|
err = btrfs_tree_read_lock_atomic(b);
|
2012-05-17 00:25:47 +08:00
|
|
|
if (!err) {
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
btrfs_tree_read_lock(b);
|
|
|
|
btrfs_clear_path_blocking(p, b,
|
|
|
|
BTRFS_READ_LOCK);
|
|
|
|
}
|
2016-06-23 06:54:23 +08:00
|
|
|
b = tree_mod_log_rewind(fs_info, p, b, time_seq);
|
2013-08-08 02:54:37 +08:00
|
|
|
if (!b) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto done;
|
|
|
|
}
|
2012-05-17 00:25:47 +08:00
|
|
|
p->locks[level] = BTRFS_READ_LOCK;
|
|
|
|
p->nodes[level] = b;
|
|
|
|
} else {
|
|
|
|
p->slots[level] = slot;
|
|
|
|
unlock_up(p, level, lowest_unlock, 0, NULL);
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ret = 1;
|
|
|
|
done:
|
|
|
|
if (!p->leave_spinning)
|
|
|
|
btrfs_set_path_blocking(p);
|
|
|
|
if (ret < 0)
|
|
|
|
btrfs_release_path(p);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2011-09-13 17:18:10 +08:00
|
|
|
/*
|
|
|
|
* helper to use instead of search slot if no exact match is needed but
|
|
|
|
* instead the next or previous item should be returned.
|
|
|
|
* When find_higher is true, the next higher item is returned, the next lower
|
|
|
|
* otherwise.
|
|
|
|
* When return_any and find_higher are both true, and no higher item is found,
|
|
|
|
* return the next lower instead.
|
|
|
|
* When return_any is true and find_higher is false, and no lower item is found,
|
|
|
|
* return the next higher instead.
|
|
|
|
* It returns 0 if any item is found, 1 if none is found (tree empty), and
|
|
|
|
* < 0 on error
|
|
|
|
*/
|
|
|
|
int btrfs_search_slot_for_read(struct btrfs_root *root,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *key,
|
|
|
|
struct btrfs_path *p, int find_higher,
|
|
|
|
int return_any)
|
2011-09-13 17:18:10 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
|
|
|
|
again:
|
|
|
|
ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
|
|
|
|
if (ret <= 0)
|
|
|
|
return ret;
|
|
|
|
/*
|
|
|
|
* a return value of 1 means the path is at the position where the
|
|
|
|
* item should be inserted. Normally this is the next bigger item,
|
|
|
|
* but in case the previous item is the last in a leaf, path points
|
|
|
|
* to the first free slot in the previous leaf, i.e. at an invalid
|
|
|
|
* item.
|
|
|
|
*/
|
|
|
|
leaf = p->nodes[0];
|
|
|
|
|
|
|
|
if (find_higher) {
|
|
|
|
if (p->slots[0] >= btrfs_header_nritems(leaf)) {
|
|
|
|
ret = btrfs_next_leaf(root, p);
|
|
|
|
if (ret <= 0)
|
|
|
|
return ret;
|
|
|
|
if (!return_any)
|
|
|
|
return 1;
|
|
|
|
/*
|
|
|
|
* no higher item found, return the next
|
|
|
|
* lower instead
|
|
|
|
*/
|
|
|
|
return_any = 0;
|
|
|
|
find_higher = 0;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
} else {
|
2011-09-13 17:18:10 +08:00
|
|
|
if (p->slots[0] == 0) {
|
|
|
|
ret = btrfs_prev_leaf(root, p);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
if (!ret) {
|
2014-01-12 05:28:54 +08:00
|
|
|
leaf = p->nodes[0];
|
|
|
|
if (p->slots[0] == btrfs_header_nritems(leaf))
|
|
|
|
p->slots[0]--;
|
2011-09-13 17:18:10 +08:00
|
|
|
return 0;
|
2011-09-13 17:18:10 +08:00
|
|
|
}
|
2011-09-13 17:18:10 +08:00
|
|
|
if (!return_any)
|
|
|
|
return 1;
|
|
|
|
/*
|
|
|
|
* no lower item found, return the next
|
|
|
|
* higher instead
|
|
|
|
*/
|
|
|
|
return_any = 0;
|
|
|
|
find_higher = 1;
|
|
|
|
btrfs_release_path(p);
|
|
|
|
goto again;
|
|
|
|
} else {
|
2011-09-13 17:18:10 +08:00
|
|
|
--p->slots[0];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* adjust the pointers going up the tree, starting at level
|
|
|
|
* making sure the right key of each node is points to 'key'.
|
|
|
|
* This is used after shifting pointers to the left, so it stops
|
|
|
|
* fixing up pointers when a given leaf/node is not in slot 0 of the
|
|
|
|
* higher levels
|
2007-03-01 05:35:06 +08:00
|
|
|
*
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2014-11-12 12:43:09 +08:00
|
|
|
static void fixup_low_keys(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_path *path,
|
2012-03-01 21:56:26 +08:00
|
|
|
struct btrfs_disk_key *key, int level)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
|
|
|
int i;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *t;
|
|
|
|
|
2007-03-13 22:46:10 +08:00
|
|
|
for (i = level; i < BTRFS_MAX_LEVEL; i++) {
|
2007-01-27 04:51:26 +08:00
|
|
|
int tslot = path->slots[i];
|
2007-02-02 22:18:22 +08:00
|
|
|
if (!path->nodes[i])
|
2007-01-27 04:51:26 +08:00
|
|
|
break;
|
2007-10-16 04:14:19 +08:00
|
|
|
t = path->nodes[i];
|
2014-11-12 12:43:09 +08:00
|
|
|
tree_mod_log_set_node_key(fs_info, t, tslot, 1);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(t, key, tslot);
|
2007-03-31 02:27:56 +08:00
|
|
|
btrfs_mark_buffer_dirty(path->nodes[i]);
|
2007-01-27 04:51:26 +08:00
|
|
|
if (tslot != 0)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-09-24 01:14:14 +08:00
|
|
|
/*
|
|
|
|
* update item key.
|
|
|
|
*
|
|
|
|
* This function isn't completely safe. It's the caller's responsibility
|
|
|
|
* that the new key won't break the order
|
|
|
|
*/
|
2014-11-12 12:43:09 +08:00
|
|
|
void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *new_key)
|
2008-09-24 01:14:14 +08:00
|
|
|
{
|
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
int slot;
|
|
|
|
|
|
|
|
eb = path->nodes[0];
|
|
|
|
slot = path->slots[0];
|
|
|
|
if (slot > 0) {
|
|
|
|
btrfs_item_key(eb, &disk_key, slot - 1);
|
2012-03-01 21:56:26 +08:00
|
|
|
BUG_ON(comp_keys(&disk_key, new_key) >= 0);
|
2008-09-24 01:14:14 +08:00
|
|
|
}
|
|
|
|
if (slot < btrfs_header_nritems(eb) - 1) {
|
|
|
|
btrfs_item_key(eb, &disk_key, slot + 1);
|
2012-03-01 21:56:26 +08:00
|
|
|
BUG_ON(comp_keys(&disk_key, new_key) <= 0);
|
2008-09-24 01:14:14 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, new_key);
|
|
|
|
btrfs_set_item_key(eb, &disk_key, slot);
|
|
|
|
btrfs_mark_buffer_dirty(eb);
|
|
|
|
if (slot == 0)
|
2014-11-12 12:43:09 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2008-09-24 01:14:14 +08:00
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* try to push data from one node into the next node left in the
|
2007-03-02 04:16:26 +08:00
|
|
|
* tree.
|
2007-03-01 05:35:06 +08:00
|
|
|
*
|
|
|
|
* returns 0 if some ptrs were pushed left, < 0 if there was some horrible
|
|
|
|
* error, and > 0 if there was no room in the left hand block.
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2008-01-03 23:01:48 +08:00
|
|
|
static int push_node_left(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
|
|
|
struct extent_buffer *dst,
|
2008-04-24 22:54:32 +08:00
|
|
|
struct extent_buffer *src, int empty)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
|
|
|
int push_items = 0;
|
2007-03-02 01:04:21 +08:00
|
|
|
int src_nritems;
|
|
|
|
int dst_nritems;
|
2007-03-01 05:35:06 +08:00
|
|
|
int ret = 0;
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
src_nritems = btrfs_header_nritems(src);
|
|
|
|
dst_nritems = btrfs_header_nritems(dst);
|
2016-06-23 06:54:23 +08:00
|
|
|
push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(btrfs_header_generation(src) != trans->transid);
|
|
|
|
WARN_ON(btrfs_header_generation(dst) != trans->transid);
|
2007-06-23 02:16:25 +08:00
|
|
|
|
2008-04-25 02:42:46 +08:00
|
|
|
if (!empty && src_nritems <= 8)
|
2008-04-24 22:54:32 +08:00
|
|
|
return 1;
|
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
if (push_items <= 0)
|
2007-01-27 04:51:26 +08:00
|
|
|
return 1;
|
|
|
|
|
2008-04-25 02:42:46 +08:00
|
|
|
if (empty) {
|
2008-04-24 22:54:32 +08:00
|
|
|
push_items = min(src_nritems, push_items);
|
2008-04-25 02:42:46 +08:00
|
|
|
if (push_items < src_nritems) {
|
|
|
|
/* leave at least 8 pointers in the node if
|
|
|
|
* we aren't going to empty it
|
|
|
|
*/
|
|
|
|
if (src_nritems - push_items < 8) {
|
|
|
|
if (push_items <= 8)
|
|
|
|
return 1;
|
|
|
|
push_items -= 8;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
push_items = min(src_nritems - 8, push_items);
|
2007-03-02 04:16:26 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
push_items);
|
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(dst, src,
|
|
|
|
btrfs_node_key_ptr_offset(dst_nritems),
|
|
|
|
btrfs_node_key_ptr_offset(0),
|
2009-01-06 10:25:51 +08:00
|
|
|
push_items * sizeof(struct btrfs_key_ptr));
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-03-02 01:04:21 +08:00
|
|
|
if (push_items < src_nritems) {
|
2012-10-19 15:22:03 +08:00
|
|
|
/*
|
|
|
|
* don't call tree_mod_log_eb_move here, key removal was already
|
|
|
|
* fully logged by tree_mod_log_eb_copy above.
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
|
|
|
|
btrfs_node_key_ptr_offset(push_items),
|
|
|
|
(src_nritems - push_items) *
|
|
|
|
sizeof(struct btrfs_key_ptr));
|
|
|
|
}
|
|
|
|
btrfs_set_header_nritems(src, src_nritems - push_items);
|
|
|
|
btrfs_set_header_nritems(dst, dst_nritems + push_items);
|
|
|
|
btrfs_mark_buffer_dirty(src);
|
|
|
|
btrfs_mark_buffer_dirty(dst);
|
2008-09-24 01:14:14 +08:00
|
|
|
|
2007-03-02 04:16:26 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* try to push data from one node into the next node right in the
|
|
|
|
* tree.
|
|
|
|
*
|
|
|
|
* returns 0 if some ptrs were pushed, < 0 if there was some horrible
|
|
|
|
* error, and > 0 if there was no room in the right hand block.
|
|
|
|
*
|
|
|
|
* this will only push up to 1/2 the contents of the left node over
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
static int balance_node_right(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *dst,
|
|
|
|
struct extent_buffer *src)
|
2007-03-02 04:16:26 +08:00
|
|
|
{
|
|
|
|
int push_items = 0;
|
|
|
|
int max_push;
|
|
|
|
int src_nritems;
|
|
|
|
int dst_nritems;
|
|
|
|
int ret = 0;
|
|
|
|
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(btrfs_header_generation(src) != trans->transid);
|
|
|
|
WARN_ON(btrfs_header_generation(dst) != trans->transid);
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
src_nritems = btrfs_header_nritems(src);
|
|
|
|
dst_nritems = btrfs_header_nritems(dst);
|
2016-06-23 06:54:23 +08:00
|
|
|
push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
|
2009-01-06 10:25:51 +08:00
|
|
|
if (push_items <= 0)
|
2007-03-02 04:16:26 +08:00
|
|
|
return 1;
|
2008-04-25 02:42:46 +08:00
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
if (src_nritems < 4)
|
2008-04-25 02:42:46 +08:00
|
|
|
return 1;
|
2007-03-02 04:16:26 +08:00
|
|
|
|
|
|
|
max_push = src_nritems / 2 + 1;
|
|
|
|
/* don't try to empty the node */
|
2009-01-06 10:25:51 +08:00
|
|
|
if (max_push >= src_nritems)
|
2007-03-02 04:16:26 +08:00
|
|
|
return 1;
|
2007-08-29 21:11:44 +08:00
|
|
|
|
2007-03-02 04:16:26 +08:00
|
|
|
if (max_push < push_items)
|
|
|
|
push_items = max_push;
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems);
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
|
|
|
|
btrfs_node_key_ptr_offset(0),
|
|
|
|
(dst_nritems) *
|
|
|
|
sizeof(struct btrfs_key_ptr));
|
2007-03-31 02:27:56 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
src_nritems - push_items, push_items);
|
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(dst, src,
|
|
|
|
btrfs_node_key_ptr_offset(0),
|
|
|
|
btrfs_node_key_ptr_offset(src_nritems - push_items),
|
2009-01-06 10:25:51 +08:00
|
|
|
push_items * sizeof(struct btrfs_key_ptr));
|
2007-03-02 04:16:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(src, src_nritems - push_items);
|
|
|
|
btrfs_set_header_nritems(dst, dst_nritems + push_items);
|
2007-03-02 04:16:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(src);
|
|
|
|
btrfs_mark_buffer_dirty(dst);
|
2008-09-24 01:14:14 +08:00
|
|
|
|
2007-03-01 05:35:06 +08:00
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2007-02-25 02:39:08 +08:00
|
|
|
/*
|
|
|
|
* helper function to insert a new root level in the tree.
|
|
|
|
* A new node is allocated, and a single item is inserted to
|
|
|
|
* point to the existing root
|
2007-03-01 05:35:06 +08:00
|
|
|
*
|
|
|
|
* returns zero on success or < 0 on failure.
|
2007-02-25 02:39:08 +08:00
|
|
|
*/
|
2009-01-06 10:25:51 +08:00
|
|
|
static noinline int insert_new_root(struct btrfs_trans_handle *trans,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_root *root,
|
2013-05-22 20:06:51 +08:00
|
|
|
struct btrfs_path *path, int level)
|
2007-02-23 00:39:13 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-12-11 22:25:06 +08:00
|
|
|
u64 lower_gen;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *lower;
|
|
|
|
struct extent_buffer *c;
|
2008-06-26 04:01:30 +08:00
|
|
|
struct extent_buffer *old;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key lower_key;
|
2007-02-23 00:39:13 +08:00
|
|
|
|
|
|
|
BUG_ON(path->nodes[level]);
|
|
|
|
BUG_ON(path->nodes[level-1] != root->node);
|
|
|
|
|
2007-12-11 22:25:06 +08:00
|
|
|
lower = path->nodes[level-1];
|
|
|
|
if (level == 1)
|
|
|
|
btrfs_item_key(lower, &lower_key, 0);
|
|
|
|
else
|
|
|
|
btrfs_node_key(lower, &lower_key, 0);
|
|
|
|
|
2014-06-15 07:54:12 +08:00
|
|
|
c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
|
|
|
|
&lower_key, level, root->node->start, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (IS_ERR(c))
|
|
|
|
return PTR_ERR(c);
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
root_add_used(root, fs_info->nodesize);
|
2010-05-16 22:46:25 +08:00
|
|
|
|
2016-11-09 01:09:03 +08:00
|
|
|
memzero_extent_buffer(c, 0, sizeof(struct btrfs_header));
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(c, 1);
|
|
|
|
btrfs_set_header_level(c, level);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_header_bytenr(c, c->start);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_generation(c, trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_owner(c, root->root_key.objectid);
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
write_extent_buffer_fsid(c, fs_info->fsid);
|
|
|
|
write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid);
|
2008-04-16 03:41:47 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(c, &lower_key, 0);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_node_blockptr(c, 0, lower->start);
|
2007-12-11 22:25:06 +08:00
|
|
|
lower_gen = btrfs_header_generation(lower);
|
2008-09-24 01:14:14 +08:00
|
|
|
WARN_ON(lower_gen != trans->transid);
|
2007-12-11 22:25:06 +08:00
|
|
|
|
|
|
|
btrfs_set_node_ptr_generation(c, 0, lower_gen);
|
2007-03-23 22:01:08 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(c);
|
2007-03-23 22:01:08 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
old = root->node;
|
2013-05-22 20:06:51 +08:00
|
|
|
tree_mod_log_set_root_pointer(root, c, 0);
|
2011-03-24 02:54:42 +08:00
|
|
|
rcu_assign_pointer(root->node, c);
|
2008-06-26 04:01:30 +08:00
|
|
|
|
|
|
|
/* the super has an extra ref to root->node */
|
|
|
|
free_extent_buffer(old);
|
|
|
|
|
2008-03-25 03:01:56 +08:00
|
|
|
add_root_to_dirty_list(root);
|
2007-10-16 04:14:19 +08:00
|
|
|
extent_buffer_get(c);
|
|
|
|
path->nodes[level] = c;
|
2015-01-15 14:52:03 +08:00
|
|
|
path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
|
2007-02-23 00:39:13 +08:00
|
|
|
path->slots[level] = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* worker function to insert a single pointer in a node.
|
|
|
|
* the node should have enough room for the pointer already
|
2007-02-25 02:39:08 +08:00
|
|
|
*
|
2007-02-03 00:05:29 +08:00
|
|
|
* slot and level indicate where you want the key to go, and
|
|
|
|
* blocknr is the block the key points to.
|
|
|
|
*/
|
2012-03-01 21:56:26 +08:00
|
|
|
static void insert_ptr(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info, struct btrfs_path *path,
|
2012-03-01 21:56:26 +08:00
|
|
|
struct btrfs_disk_key *key, u64 bytenr,
|
2012-06-21 17:01:06 +08:00
|
|
|
int slot, int level)
|
2007-02-03 00:05:29 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *lower;
|
2007-02-03 00:05:29 +08:00
|
|
|
int nritems;
|
2012-05-26 17:45:21 +08:00
|
|
|
int ret;
|
2007-02-23 00:39:13 +08:00
|
|
|
|
|
|
|
BUG_ON(!path->nodes[level]);
|
2010-05-16 22:46:25 +08:00
|
|
|
btrfs_assert_tree_locked(path->nodes[level]);
|
2007-10-16 04:14:19 +08:00
|
|
|
lower = path->nodes[level];
|
|
|
|
nritems = btrfs_header_nritems(lower);
|
2009-04-03 05:05:11 +08:00
|
|
|
BUG_ON(slot > nritems);
|
2016-06-23 06:54:23 +08:00
|
|
|
BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
|
2007-02-03 00:05:29 +08:00
|
|
|
if (slot != nritems) {
|
2012-06-21 17:01:06 +08:00
|
|
|
if (level)
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_eb_move(fs_info, lower, slot + 1,
|
2012-05-26 17:45:21 +08:00
|
|
|
slot, nritems - slot);
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(lower,
|
|
|
|
btrfs_node_key_ptr_offset(slot + 1),
|
|
|
|
btrfs_node_key_ptr_offset(slot),
|
2007-03-31 02:27:56 +08:00
|
|
|
(nritems - slot) * sizeof(struct btrfs_key_ptr));
|
2007-02-03 00:05:29 +08:00
|
|
|
}
|
2012-06-21 17:01:06 +08:00
|
|
|
if (level) {
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_insert_key(fs_info, lower, slot,
|
2013-07-02 04:18:19 +08:00
|
|
|
MOD_LOG_KEY_ADD, GFP_NOFS);
|
2012-05-26 17:45:21 +08:00
|
|
|
BUG_ON(ret < 0);
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_node_key(lower, key, slot);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_node_blockptr(lower, slot, bytenr);
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(trans->transid == 0);
|
|
|
|
btrfs_set_node_ptr_generation(lower, slot, trans->transid);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(lower, nritems + 1);
|
|
|
|
btrfs_mark_buffer_dirty(lower);
|
2007-02-03 00:05:29 +08:00
|
|
|
}
|
|
|
|
|
2007-02-25 02:39:08 +08:00
|
|
|
/*
|
|
|
|
* split the node at the specified level in path in two.
|
|
|
|
* The path is corrected to point to the appropriate node after the split
|
|
|
|
*
|
|
|
|
* Before splitting this tries to make some room in the node by pushing
|
|
|
|
* left and right, if either one works, it returns right away.
|
2007-03-01 05:35:06 +08:00
|
|
|
*
|
|
|
|
* returns 0 on success and < 0 on failure
|
2007-02-25 02:39:08 +08:00
|
|
|
*/
|
2008-09-06 04:13:11 +08:00
|
|
|
static noinline int split_node(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, int level)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *c;
|
|
|
|
struct extent_buffer *split;
|
|
|
|
struct btrfs_disk_key disk_key;
|
2007-01-27 04:51:26 +08:00
|
|
|
int mid;
|
2007-02-23 00:39:13 +08:00
|
|
|
int ret;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 c_nritems;
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
c = path->nodes[level];
|
2007-12-11 22:25:06 +08:00
|
|
|
WARN_ON(btrfs_header_generation(c) != trans->transid);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (c == root->node) {
|
2013-03-20 21:49:48 +08:00
|
|
|
/*
|
2013-04-13 21:19:53 +08:00
|
|
|
* trying to split the root, lets make a new one
|
|
|
|
*
|
2013-05-22 20:06:51 +08:00
|
|
|
* tree mod log: We don't log_removal old root in
|
2013-04-13 21:19:53 +08:00
|
|
|
* insert_new_root, because that root buffer will be kept as a
|
|
|
|
* normal node. We are going to log removal of half of the
|
|
|
|
* elements below with tree_mod_log_eb_copy. We're holding a
|
|
|
|
* tree lock on the buffer, which is why we cannot race with
|
|
|
|
* other tree_mod_log users.
|
2013-03-20 21:49:48 +08:00
|
|
|
*/
|
2013-05-22 20:06:51 +08:00
|
|
|
ret = insert_new_root(trans, root, path, level + 1);
|
2007-02-23 00:39:13 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2009-05-14 07:12:15 +08:00
|
|
|
} else {
|
2007-04-21 01:16:02 +08:00
|
|
|
ret = push_nodes_for_insert(trans, root, path, level);
|
2007-10-16 04:14:19 +08:00
|
|
|
c = path->nodes[level];
|
|
|
|
if (!ret && btrfs_header_nritems(c) <
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
|
2007-04-21 01:16:02 +08:00
|
|
|
return 0;
|
2007-06-23 02:16:25 +08:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2007-04-21 01:16:02 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
c_nritems = btrfs_header_nritems(c);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
mid = (c_nritems + 1) / 2;
|
|
|
|
btrfs_node_key(c, &disk_key, mid);
|
2007-12-11 22:25:06 +08:00
|
|
|
|
2014-06-15 07:54:12 +08:00
|
|
|
split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
|
|
|
|
&disk_key, level, c->start, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (IS_ERR(split))
|
|
|
|
return PTR_ERR(split);
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
root_add_used(root, fs_info->nodesize);
|
2010-05-16 22:46:25 +08:00
|
|
|
|
2016-11-09 01:09:03 +08:00
|
|
|
memzero_extent_buffer(split, 0, sizeof(struct btrfs_header));
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_level(split, btrfs_header_level(c));
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_header_bytenr(split, split->start);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_generation(split, trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_owner(split, root->root_key.objectid);
|
2016-06-23 06:54:23 +08:00
|
|
|
write_extent_buffer_fsid(split, fs_info->fsid);
|
|
|
|
write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid);
|
2007-06-23 02:16:25 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
if (ret) {
|
2016-06-11 06:19:25 +08:00
|
|
|
btrfs_abort_transaction(trans, ret);
|
Btrfs: fix tree mod logging
While running the test btrfs/004 from xfstests in a loop, it failed
about 1 time out of 20 runs in my desktop. The failure happened in
the backref walking part of the test, and the test's error message was
like this:
btrfs/004 93s ... [failed, exit status 1] - output mismatch (see /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad)
--- tests/btrfs/004.out 2013-11-26 18:25:29.263333714 +0000
+++ /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad 2013-12-10 15:25:10.327518516 +0000
@@ -1,3 +1,8 @@
QA output created by 004
*** test backref walking
-*** done
+unexpected output from
+ /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
+expected inum: 405, expected address: 454656, file: /home/fdmanana/btrfs-tests/scratch_1/snap1/p0/d6/d3d/d156/fce, got:
+
...
(Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests_2/results//btrfs/004.out.bad' to see the entire diff)
Ran: btrfs/004
Failures: btrfs/004
Failed 1 of 1 tests
But immediately after the test finished, the btrfs inspect-internal command
returned the expected output:
$ btrfs inspect-internal logical-resolve -P 141512704 /home/fdmanana/btrfs-tests/scratch_1
inode 405 offset 454656 root 258
inode 405 offset 454656 root 5
It turned out this was because the btrfs_search_old_slot() calls performed
during backref walking (backref.c:__resolve_indirect_ref) were not finding
anything. The reason for this turned out to be that the tree mod logging
code was not logging some node multi-step operations atomically, therefore
btrfs_search_old_slot() callers iterated often over an incomplete tree that
wasn't fully consistent with any tree state from the past. Besides missing
items, this often (but not always) resulted in -EIO errors during old slot
searches, reported in dmesg like this:
[ 4299.933936] ------------[ cut here ]------------
[ 4299.933949] WARNING: CPU: 0 PID: 23190 at fs/btrfs/ctree.c:1343 btrfs_search_old_slot+0x57b/0xab0 [btrfs]()
[ 4299.933950] Modules linked in: btrfs raid6_pq xor pci_stub vboxpci(O) vboxnetadp(O) vboxnetflt(O) vboxdrv(O) bnep rfcomm bluetooth parport_pc ppdev binfmt_misc joydev snd_hda_codec_h
[ 4299.933977] CPU: 0 PID: 23190 Comm: btrfs Tainted: G W O 3.12.0-fdm-btrfs-next-16+ #70
[ 4299.933978] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z77 Pro4, BIOS P1.50 09/04/2012
[ 4299.933979] 000000000000053f ffff8806f3fd98f8 ffffffff8176d284 0000000000000007
[ 4299.933982] 0000000000000000 ffff8806f3fd9938 ffffffff8104a81c ffff880659c64b70
[ 4299.933984] ffff880659c643d0 ffff8806599233d8 ffff880701e2e938 0000160000000000
[ 4299.933987] Call Trace:
[ 4299.933991] [<ffffffff8176d284>] dump_stack+0x55/0x76
[ 4299.933994] [<ffffffff8104a81c>] warn_slowpath_common+0x8c/0xc0
[ 4299.933997] [<ffffffff8104a86a>] warn_slowpath_null+0x1a/0x20
[ 4299.934003] [<ffffffffa065d3bb>] btrfs_search_old_slot+0x57b/0xab0 [btrfs]
[ 4299.934005] [<ffffffff81775f3b>] ? _raw_read_unlock+0x2b/0x50
[ 4299.934010] [<ffffffffa0655001>] ? __tree_mod_log_search+0x81/0xc0 [btrfs]
[ 4299.934019] [<ffffffffa06dd9b0>] __resolve_indirect_refs+0x130/0x5f0 [btrfs]
[ 4299.934027] [<ffffffffa06a21f1>] ? free_extent_buffer+0x61/0xc0 [btrfs]
[ 4299.934034] [<ffffffffa06de39c>] find_parent_nodes+0x1fc/0xe40 [btrfs]
[ 4299.934042] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934048] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934056] [<ffffffffa06df980>] iterate_extent_inodes+0xe0/0x250 [btrfs]
[ 4299.934058] [<ffffffff817762db>] ? _raw_spin_unlock+0x2b/0x50
[ 4299.934065] [<ffffffffa06dfb82>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
[ 4299.934071] [<ffffffffa06b13e0>] ? defrag_lookup_extent+0xe0/0xe0 [btrfs]
[ 4299.934078] [<ffffffffa06b7015>] btrfs_ioctl+0xf65/0x1f60 [btrfs]
[ 4299.934080] [<ffffffff811658b8>] ? handle_mm_fault+0x278/0xb00
[ 4299.934083] [<ffffffff81075563>] ? up_read+0x23/0x40
[ 4299.934085] [<ffffffff8177a41c>] ? __do_page_fault+0x20c/0x5a0
[ 4299.934088] [<ffffffff811b2946>] do_vfs_ioctl+0x96/0x570
[ 4299.934090] [<ffffffff81776e23>] ? error_sti+0x5/0x6
[ 4299.934093] [<ffffffff810b71e8>] ? trace_hardirqs_off_caller+0x28/0xd0
[ 4299.934096] [<ffffffff81776a09>] ? retint_swapgs+0xe/0x13
[ 4299.934098] [<ffffffff811b2eb1>] SyS_ioctl+0x91/0xb0
[ 4299.934100] [<ffffffff813eecde>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934102] [<ffffffff8177ef12>] system_call_fastpath+0x16/0x1b
[ 4299.934104] ---[ end trace 48f0cfc902491414 ]---
[ 4299.934378] btrfs bad fsid on block 0
These tree mod log operations that must be performed atomically, tree_mod_log_free_eb,
tree_mod_log_eb_copy, tree_mod_log_insert_root and tree_mod_log_insert_move, used to
be performed atomically before the following commit:
c8cc6341653721b54760480b0d0d9b5f09b46741
(Btrfs: stop using GFP_ATOMIC for the tree mod log allocations)
That change removed the atomicity of such operations. This patch restores the
atomicity while still not doing the GFP_ATOMIC allocations of tree_mod_elem
structures, so it has to do the allocations using GFP_NOFS before acquiring
the mod log lock.
This issue has been experienced by several users recently, such as for example:
http://www.spinics.net/lists/linux-btrfs/msg28574.html
After running the btrfs/004 test for 679 consecutive iterations with this
patch applied, I didn't ran into the issue anymore.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
2013-12-20 23:17:46 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(split, c,
|
|
|
|
btrfs_node_key_ptr_offset(0),
|
|
|
|
btrfs_node_key_ptr_offset(mid),
|
|
|
|
(c_nritems - mid) * sizeof(struct btrfs_key_ptr));
|
|
|
|
btrfs_set_header_nritems(split, c_nritems - mid);
|
|
|
|
btrfs_set_header_nritems(c, mid);
|
2007-03-01 05:35:06 +08:00
|
|
|
ret = 0;
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(c);
|
|
|
|
btrfs_mark_buffer_dirty(split);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
insert_ptr(trans, fs_info, path, &disk_key, split->start,
|
2012-06-21 17:01:06 +08:00
|
|
|
path->slots[level + 1] + 1, level + 1);
|
2007-03-01 05:35:06 +08:00
|
|
|
|
2007-02-24 19:24:44 +08:00
|
|
|
if (path->slots[level] >= mid) {
|
2007-02-23 00:39:13 +08:00
|
|
|
path->slots[level] -= mid;
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(c);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(c);
|
|
|
|
path->nodes[level] = split;
|
2007-02-23 00:39:13 +08:00
|
|
|
path->slots[level + 1] += 1;
|
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(split);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(split);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2007-03-01 05:35:06 +08:00
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* how many bytes are required to store the items in a leaf. start
|
|
|
|
* and nr indicate which items in the leaf to check. This totals up the
|
|
|
|
* space used both by the item structs and the item data
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
static int leaf_space_used(struct extent_buffer *l, int start, int nr)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2012-10-16 01:43:18 +08:00
|
|
|
struct btrfs_item *start_item;
|
|
|
|
struct btrfs_item *end_item;
|
|
|
|
struct btrfs_map_token token;
|
2007-01-27 04:51:26 +08:00
|
|
|
int data_len;
|
2007-10-16 04:14:19 +08:00
|
|
|
int nritems = btrfs_header_nritems(l);
|
2007-04-05 02:08:15 +08:00
|
|
|
int end = min(nritems, start + nr) - 1;
|
2007-01-27 04:51:26 +08:00
|
|
|
|
|
|
|
if (!nr)
|
|
|
|
return 0;
|
2012-10-16 01:43:18 +08:00
|
|
|
btrfs_init_map_token(&token);
|
2013-09-16 22:58:09 +08:00
|
|
|
start_item = btrfs_item_nr(start);
|
|
|
|
end_item = btrfs_item_nr(end);
|
2012-10-16 01:43:18 +08:00
|
|
|
data_len = btrfs_token_item_offset(l, start_item, &token) +
|
|
|
|
btrfs_token_item_size(l, start_item, &token);
|
|
|
|
data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
|
2007-03-13 08:12:07 +08:00
|
|
|
data_len += sizeof(struct btrfs_item) * nr;
|
2007-04-05 02:08:15 +08:00
|
|
|
WARN_ON(data_len < 0);
|
2007-01-27 04:51:26 +08:00
|
|
|
return data_len;
|
|
|
|
}
|
|
|
|
|
2007-04-05 02:08:15 +08:00
|
|
|
/*
|
|
|
|
* The space between the end of the leaf items and
|
|
|
|
* the start of the leaf data. IOW, how much room
|
|
|
|
* the leaf has left for both items and data
|
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
|
2008-09-06 04:13:11 +08:00
|
|
|
struct extent_buffer *leaf)
|
2007-04-05 02:08:15 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
int nritems = btrfs_header_nritems(leaf);
|
|
|
|
int ret;
|
2016-06-23 06:54:23 +08:00
|
|
|
|
|
|
|
ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (ret < 0) {
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_crit(fs_info,
|
|
|
|
"leaf free space ret %d, leaf data size %lu, used %d nritems %d",
|
|
|
|
ret,
|
|
|
|
(unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
|
|
|
|
leaf_space_used(leaf, 0, nritems), nritems);
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
|
|
|
return ret;
|
2007-04-05 02:08:15 +08:00
|
|
|
}
|
|
|
|
|
2010-07-07 22:51:48 +08:00
|
|
|
/*
|
|
|
|
* min slot controls the lowest index we're willing to push to the
|
|
|
|
* right. We'll push up to and including min_slot, but no lower
|
|
|
|
*/
|
2009-03-13 22:04:31 +08:00
|
|
|
static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2009-03-13 22:04:31 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
int data_size, int empty,
|
|
|
|
struct extent_buffer *right,
|
2010-07-07 22:51:48 +08:00
|
|
|
int free_space, u32 left_nritems,
|
|
|
|
u32 min_slot)
|
2007-02-25 01:47:20 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *left = path->nodes[0];
|
2009-03-13 22:04:31 +08:00
|
|
|
struct extent_buffer *upper = path->nodes[1];
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-02-25 01:47:20 +08:00
|
|
|
int slot;
|
2007-11-08 02:31:03 +08:00
|
|
|
u32 i;
|
2007-02-25 01:47:20 +08:00
|
|
|
int push_space = 0;
|
|
|
|
int push_items = 0;
|
2007-03-13 08:12:07 +08:00
|
|
|
struct btrfs_item *item;
|
2007-11-08 02:31:03 +08:00
|
|
|
u32 nr;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 right_nritems;
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 data_end;
|
2007-10-16 04:15:53 +08:00
|
|
|
u32 this_item_size;
|
2007-02-25 01:47:20 +08:00
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
btrfs_init_map_token(&token);
|
|
|
|
|
2007-11-08 02:31:03 +08:00
|
|
|
if (empty)
|
|
|
|
nr = 0;
|
|
|
|
else
|
2010-07-07 22:51:48 +08:00
|
|
|
nr = max_t(u32, 1, min_slot);
|
2007-11-08 02:31:03 +08:00
|
|
|
|
2008-09-24 01:14:14 +08:00
|
|
|
if (path->slots[0] >= left_nritems)
|
2008-12-17 23:21:48 +08:00
|
|
|
push_space += data_size;
|
2008-09-24 01:14:14 +08:00
|
|
|
|
2009-03-13 22:04:31 +08:00
|
|
|
slot = path->slots[1];
|
2007-11-08 02:31:03 +08:00
|
|
|
i = left_nritems - 1;
|
|
|
|
while (i >= nr) {
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2008-09-24 01:14:14 +08:00
|
|
|
if (!empty && push_items > 0) {
|
|
|
|
if (path->slots[0] > i)
|
|
|
|
break;
|
|
|
|
if (path->slots[0] == i) {
|
2016-06-23 06:54:24 +08:00
|
|
|
int space = btrfs_leaf_free_space(fs_info, left);
|
2008-09-24 01:14:14 +08:00
|
|
|
if (space + push_space * 2 > free_space)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
if (path->slots[0] == i)
|
2008-12-17 23:21:48 +08:00
|
|
|
push_space += data_size;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
|
|
|
this_item_size = btrfs_item_size(left, item);
|
|
|
|
if (this_item_size + sizeof(*item) + push_space > free_space)
|
2007-02-25 01:47:20 +08:00
|
|
|
break;
|
2008-09-24 01:14:14 +08:00
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
push_items++;
|
2007-10-16 04:15:53 +08:00
|
|
|
push_space += this_item_size + sizeof(*item);
|
2007-11-08 02:31:03 +08:00
|
|
|
if (i == 0)
|
|
|
|
break;
|
|
|
|
i--;
|
2007-10-16 04:15:53 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (push_items == 0)
|
|
|
|
goto out_unlock;
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2012-11-04 04:30:18 +08:00
|
|
|
WARN_ON(!empty && push_items == left_nritems);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
/* push left to right */
|
2007-10-16 04:14:19 +08:00
|
|
|
right_nritems = btrfs_header_nritems(right);
|
2007-11-08 02:31:03 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
push_space = btrfs_item_end_nr(left, left_nritems - push_items);
|
2016-06-23 06:54:24 +08:00
|
|
|
push_space -= leaf_data_end(fs_info, left);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
/* make room in the right data area */
|
2016-06-23 06:54:24 +08:00
|
|
|
data_end = leaf_data_end(fs_info, right);
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(right,
|
|
|
|
btrfs_leaf_data(right) + data_end - push_space,
|
|
|
|
btrfs_leaf_data(right) + data_end,
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
/* copy from the left data area */
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(right, left, btrfs_leaf_data(right) +
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
|
2016-06-23 06:54:24 +08:00
|
|
|
btrfs_leaf_data(left) + leaf_data_end(fs_info, left),
|
2007-03-31 02:27:56 +08:00
|
|
|
push_space);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
|
|
|
|
btrfs_item_nr_offset(0),
|
|
|
|
right_nritems * sizeof(struct btrfs_item));
|
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
/* copy the items from left to right */
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
|
|
|
|
btrfs_item_nr_offset(left_nritems - push_items),
|
|
|
|
push_items * sizeof(struct btrfs_item));
|
2007-02-25 01:47:20 +08:00
|
|
|
|
|
|
|
/* update the item pointers */
|
2007-03-13 00:01:18 +08:00
|
|
|
right_nritems += push_items;
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(right, right_nritems);
|
2016-06-23 06:54:23 +08:00
|
|
|
push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
|
2007-03-13 00:01:18 +08:00
|
|
|
for (i = 0; i < right_nritems; i++) {
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2012-03-03 20:40:03 +08:00
|
|
|
push_space -= btrfs_token_item_size(right, item, &token);
|
|
|
|
btrfs_set_token_item_offset(right, item, push_space, &token);
|
2007-10-16 04:15:53 +08:00
|
|
|
}
|
|
|
|
|
2007-03-13 00:01:18 +08:00
|
|
|
left_nritems -= push_items;
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(left, left_nritems);
|
2007-02-25 01:47:20 +08:00
|
|
|
|
2007-11-08 02:31:03 +08:00
|
|
|
if (left_nritems)
|
|
|
|
btrfs_mark_buffer_dirty(left);
|
2010-05-16 22:46:25 +08:00
|
|
|
else
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, left);
|
2010-05-16 22:46:25 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(right);
|
2007-04-19 04:15:28 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_key(right, &disk_key, 0);
|
|
|
|
btrfs_set_node_key(upper, &disk_key, slot + 1);
|
2007-03-31 02:27:56 +08:00
|
|
|
btrfs_mark_buffer_dirty(upper);
|
2007-03-03 05:08:05 +08:00
|
|
|
|
2007-02-25 01:47:20 +08:00
|
|
|
/* then fixup the leaf pointer in the path */
|
2007-03-13 00:01:18 +08:00
|
|
|
if (path->slots[0] >= left_nritems) {
|
|
|
|
path->slots[0] -= left_nritems;
|
2008-06-26 04:01:30 +08:00
|
|
|
if (btrfs_header_nritems(path->nodes[0]) == 0)
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, path->nodes[0]);
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(path->nodes[0]);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(path->nodes[0]);
|
|
|
|
path->nodes[0] = right;
|
2007-02-25 01:47:20 +08:00
|
|
|
path->slots[1] += 1;
|
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(right);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(right);
|
2007-02-25 01:47:20 +08:00
|
|
|
}
|
|
|
|
return 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
|
|
|
out_unlock:
|
|
|
|
btrfs_tree_unlock(right);
|
|
|
|
free_extent_buffer(right);
|
|
|
|
return 1;
|
2007-02-25 01:47:20 +08:00
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2009-03-13 22:04:31 +08:00
|
|
|
/*
|
|
|
|
* push some data in the path leaf to the right, trying to free up at
|
|
|
|
* least data_size bytes. returns zero if the push worked, nonzero otherwise
|
|
|
|
*
|
|
|
|
* returns 1 if the push failed because the other node didn't have enough
|
|
|
|
* room, 0 if everything worked out and < 0 if there were major errors.
|
2010-07-07 22:51:48 +08:00
|
|
|
*
|
|
|
|
* this will push starting from min_slot to the end of the leaf. It won't
|
|
|
|
* push any slot lower than min_slot
|
2009-03-13 22:04:31 +08:00
|
|
|
*/
|
|
|
|
static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
|
2010-07-07 22:51:48 +08:00
|
|
|
*root, struct btrfs_path *path,
|
|
|
|
int min_data_size, int data_size,
|
|
|
|
int empty, u32 min_slot)
|
2009-03-13 22:04:31 +08:00
|
|
|
{
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2009-03-13 22:04:31 +08:00
|
|
|
struct extent_buffer *left = path->nodes[0];
|
|
|
|
struct extent_buffer *right;
|
|
|
|
struct extent_buffer *upper;
|
|
|
|
int slot;
|
|
|
|
int free_space;
|
|
|
|
u32 left_nritems;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!path->nodes[1])
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
slot = path->slots[1];
|
|
|
|
upper = path->nodes[1];
|
|
|
|
if (slot >= btrfs_header_nritems(upper) - 1)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
btrfs_assert_tree_locked(path->nodes[1]);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
right = read_node_slot(fs_info, upper, slot + 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
/*
|
|
|
|
* slot + 1 is not valid or we fail to read the right node,
|
|
|
|
* no big deal, just return.
|
|
|
|
*/
|
|
|
|
if (IS_ERR(right))
|
2011-01-05 10:32:22 +08:00
|
|
|
return 1;
|
|
|
|
|
2009-03-13 22:04:31 +08:00
|
|
|
btrfs_tree_lock(right);
|
|
|
|
btrfs_set_lock_blocking(right);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
free_space = btrfs_leaf_free_space(fs_info, right);
|
2009-03-13 22:04:31 +08:00
|
|
|
if (free_space < data_size)
|
|
|
|
goto out_unlock;
|
|
|
|
|
|
|
|
/* cow and double check */
|
|
|
|
ret = btrfs_cow_block(trans, root, right, upper,
|
|
|
|
slot + 1, &right);
|
|
|
|
if (ret)
|
|
|
|
goto out_unlock;
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
free_space = btrfs_leaf_free_space(fs_info, right);
|
2009-03-13 22:04:31 +08:00
|
|
|
if (free_space < data_size)
|
|
|
|
goto out_unlock;
|
|
|
|
|
|
|
|
left_nritems = btrfs_header_nritems(left);
|
|
|
|
if (left_nritems == 0)
|
|
|
|
goto out_unlock;
|
|
|
|
|
2013-12-05 06:17:39 +08:00
|
|
|
if (path->slots[0] == left_nritems && !empty) {
|
|
|
|
/* Key greater than all keys in the leaf, right neighbor has
|
|
|
|
* enough room for it and we're not emptying our leaf to delete
|
|
|
|
* it, therefore use right neighbor to insert the new item and
|
|
|
|
* no need to touch/dirty our left leaft. */
|
|
|
|
btrfs_tree_unlock(left);
|
|
|
|
free_extent_buffer(left);
|
|
|
|
path->nodes[0] = right;
|
|
|
|
path->slots[0] = 0;
|
|
|
|
path->slots[1]++;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
return __push_leaf_right(trans, fs_info, path, min_data_size, empty,
|
2010-07-07 22:51:48 +08:00
|
|
|
right, free_space, left_nritems, min_slot);
|
2009-03-13 22:04:31 +08:00
|
|
|
out_unlock:
|
|
|
|
btrfs_tree_unlock(right);
|
|
|
|
free_extent_buffer(right);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* push some data in the path leaf to the left, trying to free up at
|
|
|
|
* least data_size bytes. returns zero if the push worked, nonzero otherwise
|
2010-07-07 22:51:48 +08:00
|
|
|
*
|
|
|
|
* max_slot can put a limit on how far into the leaf we'll push items. The
|
|
|
|
* item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
|
|
|
|
* items
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2009-03-13 22:04:31 +08:00
|
|
|
static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2009-03-13 22:04:31 +08:00
|
|
|
struct btrfs_path *path, int data_size,
|
|
|
|
int empty, struct extent_buffer *left,
|
2010-07-07 22:51:48 +08:00
|
|
|
int free_space, u32 right_nritems,
|
|
|
|
u32 max_slot)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
struct extent_buffer *right = path->nodes[0];
|
2007-01-27 04:51:26 +08:00
|
|
|
int i;
|
|
|
|
int push_space = 0;
|
|
|
|
int push_items = 0;
|
2007-03-13 08:12:07 +08:00
|
|
|
struct btrfs_item *item;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 old_left_nritems;
|
2007-11-08 02:31:03 +08:00
|
|
|
u32 nr;
|
2007-03-01 05:35:06 +08:00
|
|
|
int ret = 0;
|
2007-10-16 04:15:53 +08:00
|
|
|
u32 this_item_size;
|
|
|
|
u32 old_left_item_size;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
|
|
|
btrfs_init_map_token(&token);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2007-11-08 02:31:03 +08:00
|
|
|
if (empty)
|
2010-07-07 22:51:48 +08:00
|
|
|
nr = min(right_nritems, max_slot);
|
2007-11-08 02:31:03 +08:00
|
|
|
else
|
2010-07-07 22:51:48 +08:00
|
|
|
nr = min(right_nritems - 1, max_slot);
|
2007-11-08 02:31:03 +08:00
|
|
|
|
|
|
|
for (i = 0; i < nr; i++) {
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2008-09-24 01:14:14 +08:00
|
|
|
if (!empty && push_items > 0) {
|
|
|
|
if (path->slots[0] < i)
|
|
|
|
break;
|
|
|
|
if (path->slots[0] == i) {
|
2016-06-23 06:54:24 +08:00
|
|
|
int space = btrfs_leaf_free_space(fs_info, right);
|
2008-09-24 01:14:14 +08:00
|
|
|
if (space + push_space * 2 > free_space)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-27 04:51:26 +08:00
|
|
|
if (path->slots[0] == i)
|
2008-12-17 23:21:48 +08:00
|
|
|
push_space += data_size;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
|
|
|
this_item_size = btrfs_item_size(right, item);
|
|
|
|
if (this_item_size + sizeof(*item) + push_space > free_space)
|
2007-01-27 04:51:26 +08:00
|
|
|
break;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2007-01-27 04:51:26 +08:00
|
|
|
push_items++;
|
2007-10-16 04:15:53 +08:00
|
|
|
push_space += this_item_size + sizeof(*item);
|
|
|
|
}
|
|
|
|
|
2007-01-27 04:51:26 +08:00
|
|
|
if (push_items == 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
ret = 1;
|
|
|
|
goto out;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2013-10-31 13:00:08 +08:00
|
|
|
WARN_ON(!empty && push_items == btrfs_header_nritems(right));
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-01-27 04:51:26 +08:00
|
|
|
/* push data from right to left */
|
2007-10-16 04:14:19 +08:00
|
|
|
copy_extent_buffer(left, right,
|
|
|
|
btrfs_item_nr_offset(btrfs_header_nritems(left)),
|
|
|
|
btrfs_item_nr_offset(0),
|
|
|
|
push_items * sizeof(struct btrfs_item));
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
|
2009-01-06 10:25:51 +08:00
|
|
|
btrfs_item_offset_nr(right, push_items - 1);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
copy_extent_buffer(left, right, btrfs_leaf_data(left) +
|
2016-06-23 06:54:24 +08:00
|
|
|
leaf_data_end(fs_info, left) - push_space,
|
2007-03-31 02:27:56 +08:00
|
|
|
btrfs_leaf_data(right) +
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_offset_nr(right, push_items - 1),
|
2007-03-31 02:27:56 +08:00
|
|
|
push_space);
|
2007-10-16 04:14:19 +08:00
|
|
|
old_left_nritems = btrfs_header_nritems(left);
|
2008-12-17 23:21:48 +08:00
|
|
|
BUG_ON(old_left_nritems <= 0);
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
|
2007-03-13 08:12:07 +08:00
|
|
|
for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 ioff;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(left, item, &token);
|
|
|
|
btrfs_set_token_item_offset(left, item,
|
2016-06-23 06:54:23 +08:00
|
|
|
ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
|
2012-03-03 20:40:03 +08:00
|
|
|
&token);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(left, old_left_nritems + push_items);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
|
|
|
/* fixup right node */
|
2012-11-03 18:58:34 +08:00
|
|
|
if (push_items > right_nritems)
|
|
|
|
WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
|
2009-01-06 10:25:51 +08:00
|
|
|
right_nritems);
|
2007-11-08 02:31:03 +08:00
|
|
|
|
|
|
|
if (push_items < right_nritems) {
|
|
|
|
push_space = btrfs_item_offset_nr(right, push_items - 1) -
|
2016-06-23 06:54:24 +08:00
|
|
|
leaf_data_end(fs_info, right);
|
2007-11-08 02:31:03 +08:00
|
|
|
memmove_extent_buffer(right, btrfs_leaf_data(right) +
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
|
2007-11-08 02:31:03 +08:00
|
|
|
btrfs_leaf_data(right) +
|
2016-06-23 06:54:24 +08:00
|
|
|
leaf_data_end(fs_info, right), push_space);
|
2007-11-08 02:31:03 +08:00
|
|
|
|
|
|
|
memmove_extent_buffer(right, btrfs_item_nr_offset(0),
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_nr_offset(push_items),
|
|
|
|
(btrfs_header_nritems(right) - push_items) *
|
|
|
|
sizeof(struct btrfs_item));
|
2007-11-08 02:31:03 +08:00
|
|
|
}
|
2007-11-26 23:58:13 +08:00
|
|
|
right_nritems -= push_items;
|
|
|
|
btrfs_set_header_nritems(right, right_nritems);
|
2016-06-23 06:54:23 +08:00
|
|
|
push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
|
2007-10-16 04:14:19 +08:00
|
|
|
for (i = 0; i < right_nritems; i++) {
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
push_space = push_space - btrfs_token_item_size(right,
|
|
|
|
item, &token);
|
|
|
|
btrfs_set_token_item_offset(right, item, push_space, &token);
|
2007-10-16 04:15:53 +08:00
|
|
|
}
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(left);
|
2007-11-08 02:31:03 +08:00
|
|
|
if (right_nritems)
|
|
|
|
btrfs_mark_buffer_dirty(right);
|
2010-05-16 22:46:25 +08:00
|
|
|
else
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, right);
|
2007-05-11 23:33:21 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_key(right, &disk_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
|
|
|
/* then fixup the leaf pointer in the path */
|
|
|
|
if (path->slots[0] < push_items) {
|
|
|
|
path->slots[0] += old_left_nritems;
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(path->nodes[0]);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(path->nodes[0]);
|
|
|
|
path->nodes[0] = left;
|
2007-01-27 04:51:26 +08:00
|
|
|
path->slots[1] -= 1;
|
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_tree_unlock(left);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(left);
|
2007-01-27 04:51:26 +08:00
|
|
|
path->slots[0] -= push_items;
|
|
|
|
}
|
2007-02-02 22:18:22 +08:00
|
|
|
BUG_ON(path->slots[0] < 0);
|
2007-03-01 05:35:06 +08:00
|
|
|
return ret;
|
2008-06-26 04:01:30 +08:00
|
|
|
out:
|
|
|
|
btrfs_tree_unlock(left);
|
|
|
|
free_extent_buffer(left);
|
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2009-03-13 22:04:31 +08:00
|
|
|
/*
|
|
|
|
* push some data in the path leaf to the left, trying to free up at
|
|
|
|
* least data_size bytes. returns zero if the push worked, nonzero otherwise
|
2010-07-07 22:51:48 +08:00
|
|
|
*
|
|
|
|
* max_slot can put a limit on how far into the leaf we'll push items. The
|
|
|
|
* item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
|
|
|
|
* items
|
2009-03-13 22:04:31 +08:00
|
|
|
*/
|
|
|
|
static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
|
2010-07-07 22:51:48 +08:00
|
|
|
*root, struct btrfs_path *path, int min_data_size,
|
|
|
|
int data_size, int empty, u32 max_slot)
|
2009-03-13 22:04:31 +08:00
|
|
|
{
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2009-03-13 22:04:31 +08:00
|
|
|
struct extent_buffer *right = path->nodes[0];
|
|
|
|
struct extent_buffer *left;
|
|
|
|
int slot;
|
|
|
|
int free_space;
|
|
|
|
u32 right_nritems;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
slot = path->slots[1];
|
|
|
|
if (slot == 0)
|
|
|
|
return 1;
|
|
|
|
if (!path->nodes[1])
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
right_nritems = btrfs_header_nritems(right);
|
|
|
|
if (right_nritems == 0)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
btrfs_assert_tree_locked(path->nodes[1]);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
left = read_node_slot(fs_info, path->nodes[1], slot - 1);
|
2016-07-06 03:10:14 +08:00
|
|
|
/*
|
|
|
|
* slot - 1 is not valid or we fail to read the left node,
|
|
|
|
* no big deal, just return.
|
|
|
|
*/
|
|
|
|
if (IS_ERR(left))
|
2011-01-05 10:32:22 +08:00
|
|
|
return 1;
|
|
|
|
|
2009-03-13 22:04:31 +08:00
|
|
|
btrfs_tree_lock(left);
|
|
|
|
btrfs_set_lock_blocking(left);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
free_space = btrfs_leaf_free_space(fs_info, left);
|
2009-03-13 22:04:31 +08:00
|
|
|
if (free_space < data_size) {
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* cow and double check */
|
|
|
|
ret = btrfs_cow_block(trans, root, left,
|
|
|
|
path->nodes[1], slot - 1, &left);
|
|
|
|
if (ret) {
|
|
|
|
/* we hit -ENOSPC, but it isn't fatal here */
|
2012-03-12 23:03:00 +08:00
|
|
|
if (ret == -ENOSPC)
|
|
|
|
ret = 1;
|
2009-03-13 22:04:31 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
free_space = btrfs_leaf_free_space(fs_info, left);
|
2009-03-13 22:04:31 +08:00
|
|
|
if (free_space < data_size) {
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
return __push_leaf_left(trans, fs_info, path, min_data_size,
|
2010-07-07 22:51:48 +08:00
|
|
|
empty, left, free_space, right_nritems,
|
|
|
|
max_slot);
|
2009-03-13 22:04:31 +08:00
|
|
|
out:
|
|
|
|
btrfs_tree_unlock(left);
|
|
|
|
free_extent_buffer(left);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* split the path's leaf in two, making sure there is at least data_size
|
|
|
|
* available for the resulting leaf level of the path.
|
|
|
|
*/
|
2012-03-01 21:56:26 +08:00
|
|
|
static noinline void copy_for_split(struct btrfs_trans_handle *trans,
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2012-03-01 21:56:26 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
struct extent_buffer *l,
|
|
|
|
struct extent_buffer *right,
|
|
|
|
int slot, int mid, int nritems)
|
2009-03-13 22:04:31 +08:00
|
|
|
{
|
|
|
|
int data_copy_size;
|
|
|
|
int rt_data_off;
|
|
|
|
int i;
|
|
|
|
struct btrfs_disk_key disk_key;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
|
|
|
btrfs_init_map_token(&token);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
|
|
|
nritems = nritems - mid;
|
|
|
|
btrfs_set_header_nritems(right, nritems);
|
2016-06-23 06:54:24 +08:00
|
|
|
data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
|
|
|
copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
|
|
|
|
btrfs_item_nr_offset(mid),
|
|
|
|
nritems * sizeof(struct btrfs_item));
|
|
|
|
|
|
|
|
copy_extent_buffer(right, l,
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(fs_info) -
|
2009-03-13 22:04:31 +08:00
|
|
|
data_copy_size, btrfs_leaf_data(l) +
|
2016-06-23 06:54:24 +08:00
|
|
|
leaf_data_end(fs_info, l), data_copy_size);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
|
|
|
for (i = 0; i < nritems; i++) {
|
2013-09-16 22:58:09 +08:00
|
|
|
struct btrfs_item *item = btrfs_item_nr(i);
|
2009-03-13 22:04:31 +08:00
|
|
|
u32 ioff;
|
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(right, item, &token);
|
|
|
|
btrfs_set_token_item_offset(right, item,
|
|
|
|
ioff + rt_data_off, &token);
|
2009-03-13 22:04:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_set_header_nritems(l, mid);
|
|
|
|
btrfs_item_key(right, &disk_key, 0);
|
2016-06-23 06:54:24 +08:00
|
|
|
insert_ptr(trans, fs_info, path, &disk_key, right->start,
|
2012-06-21 17:01:06 +08:00
|
|
|
path->slots[1] + 1, 1);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
|
|
|
btrfs_mark_buffer_dirty(right);
|
|
|
|
btrfs_mark_buffer_dirty(l);
|
|
|
|
BUG_ON(path->slots[0] != slot);
|
|
|
|
|
|
|
|
if (mid <= slot) {
|
|
|
|
btrfs_tree_unlock(path->nodes[0]);
|
|
|
|
free_extent_buffer(path->nodes[0]);
|
|
|
|
path->nodes[0] = right;
|
|
|
|
path->slots[0] -= mid;
|
|
|
|
path->slots[1] += 1;
|
|
|
|
} else {
|
|
|
|
btrfs_tree_unlock(right);
|
|
|
|
free_extent_buffer(right);
|
|
|
|
}
|
|
|
|
|
|
|
|
BUG_ON(path->slots[0] < 0);
|
|
|
|
}
|
|
|
|
|
2010-07-07 22:51:48 +08:00
|
|
|
/*
|
|
|
|
* double splits happen when we need to insert a big item in the middle
|
|
|
|
* of a leaf. A double split can leave us with 3 mostly empty leaves:
|
|
|
|
* leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
|
|
|
|
* A B C
|
|
|
|
*
|
|
|
|
* We avoid this by trying to push the items on either side of our target
|
|
|
|
* into the adjacent leaves. If all goes well we can avoid the double split
|
|
|
|
* completely.
|
|
|
|
*/
|
|
|
|
static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
|
|
|
int data_size)
|
|
|
|
{
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2010-07-07 22:51:48 +08:00
|
|
|
int ret;
|
|
|
|
int progress = 0;
|
|
|
|
int slot;
|
|
|
|
u32 nritems;
|
2013-11-25 11:20:46 +08:00
|
|
|
int space_needed = data_size;
|
2010-07-07 22:51:48 +08:00
|
|
|
|
|
|
|
slot = path->slots[0];
|
2013-11-25 11:20:46 +08:00
|
|
|
if (slot < btrfs_header_nritems(path->nodes[0]))
|
2016-06-23 06:54:24 +08:00
|
|
|
space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
|
2010-07-07 22:51:48 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* try to push all the items after our slot into the
|
|
|
|
* right leaf
|
|
|
|
*/
|
2013-11-25 11:20:46 +08:00
|
|
|
ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
|
2010-07-07 22:51:48 +08:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (ret == 0)
|
|
|
|
progress++;
|
|
|
|
|
|
|
|
nritems = btrfs_header_nritems(path->nodes[0]);
|
|
|
|
/*
|
|
|
|
* our goal is to get our slot at the start or end of a leaf. If
|
|
|
|
* we've done so we're done
|
|
|
|
*/
|
|
|
|
if (path->slots[0] == 0 || path->slots[0] == nritems)
|
|
|
|
return 0;
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
|
2010-07-07 22:51:48 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* try to push all the items before our slot into the next leaf */
|
|
|
|
slot = path->slots[0];
|
2013-11-25 11:20:46 +08:00
|
|
|
ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
|
2010-07-07 22:51:48 +08:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (ret == 0)
|
|
|
|
progress++;
|
|
|
|
|
|
|
|
if (progress)
|
|
|
|
return 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* split the path's leaf in two, making sure there is at least data_size
|
|
|
|
* available for the resulting leaf level of the path.
|
2007-03-01 05:35:06 +08:00
|
|
|
*
|
|
|
|
* returns 0 if all went well and < 0 on failure.
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2008-09-06 04:13:11 +08:00
|
|
|
static noinline int split_leaf(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *ins_key,
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_path *path, int data_size,
|
|
|
|
int extend)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *l;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 nritems;
|
2007-02-02 22:18:22 +08:00
|
|
|
int mid;
|
|
|
|
int slot;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *right;
|
2014-11-12 12:43:09 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-04-05 02:08:15 +08:00
|
|
|
int ret = 0;
|
2007-03-01 05:35:06 +08:00
|
|
|
int wret;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
int split;
|
2007-10-26 03:42:57 +08:00
|
|
|
int num_doubles = 0;
|
2010-07-07 22:51:48 +08:00
|
|
|
int tried_avoid_double = 0;
|
2007-03-01 05:35:06 +08:00
|
|
|
|
2009-09-24 21:17:31 +08:00
|
|
|
l = path->nodes[0];
|
|
|
|
slot = path->slots[0];
|
|
|
|
if (extend && data_size + btrfs_item_size_nr(l, slot) +
|
2016-06-23 06:54:23 +08:00
|
|
|
sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
|
2009-09-24 21:17:31 +08:00
|
|
|
return -EOVERFLOW;
|
|
|
|
|
2007-03-18 02:29:23 +08:00
|
|
|
/* first try to make some room by pushing left and right */
|
2013-05-22 20:07:06 +08:00
|
|
|
if (data_size && path->nodes[1]) {
|
2013-11-25 11:20:46 +08:00
|
|
|
int space_needed = data_size;
|
|
|
|
|
|
|
|
if (slot < btrfs_header_nritems(l))
|
2016-06-23 06:54:24 +08:00
|
|
|
space_needed -= btrfs_leaf_free_space(fs_info, l);
|
2013-11-25 11:20:46 +08:00
|
|
|
|
|
|
|
wret = push_leaf_right(trans, root, path, space_needed,
|
|
|
|
space_needed, 0, 0);
|
2009-01-06 10:25:51 +08:00
|
|
|
if (wret < 0)
|
2007-03-13 23:17:52 +08:00
|
|
|
return wret;
|
2007-10-19 21:23:27 +08:00
|
|
|
if (wret) {
|
2013-11-25 11:20:46 +08:00
|
|
|
wret = push_leaf_left(trans, root, path, space_needed,
|
|
|
|
space_needed, 0, (u32)-1);
|
2007-10-19 21:23:27 +08:00
|
|
|
if (wret < 0)
|
|
|
|
return wret;
|
|
|
|
}
|
|
|
|
l = path->nodes[0];
|
2007-03-01 05:35:06 +08:00
|
|
|
|
2007-10-19 21:23:27 +08:00
|
|
|
/* did the pushes work? */
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, l) >= data_size)
|
2007-10-19 21:23:27 +08:00
|
|
|
return 0;
|
2007-10-16 04:18:25 +08:00
|
|
|
}
|
2007-03-01 05:35:06 +08:00
|
|
|
|
2007-02-23 00:39:13 +08:00
|
|
|
if (!path->nodes[1]) {
|
2013-05-22 20:06:51 +08:00
|
|
|
ret = insert_new_root(trans, root, path, 1);
|
2007-02-23 00:39:13 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
2007-10-26 03:42:57 +08:00
|
|
|
again:
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
split = 1;
|
2007-10-26 03:42:57 +08:00
|
|
|
l = path->nodes[0];
|
2007-02-02 22:18:22 +08:00
|
|
|
slot = path->slots[0];
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(l);
|
2009-01-06 10:25:51 +08:00
|
|
|
mid = (nritems + 1) / 2;
|
2007-06-23 02:16:25 +08:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (mid <= slot) {
|
|
|
|
if (nritems == 1 ||
|
|
|
|
leaf_space_used(l, mid, nritems - mid) + data_size >
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_LEAF_DATA_SIZE(fs_info)) {
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (slot >= nritems) {
|
|
|
|
split = 0;
|
|
|
|
} else {
|
|
|
|
mid = slot;
|
|
|
|
if (mid != nritems &&
|
|
|
|
leaf_space_used(l, mid, nritems - mid) +
|
2016-06-23 06:54:23 +08:00
|
|
|
data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
|
2010-07-07 22:51:48 +08:00
|
|
|
if (data_size && !tried_avoid_double)
|
|
|
|
goto push_for_double;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
split = 2;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (leaf_space_used(l, 0, mid) + data_size >
|
2016-06-23 06:54:23 +08:00
|
|
|
BTRFS_LEAF_DATA_SIZE(fs_info)) {
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (!extend && data_size && slot == 0) {
|
|
|
|
split = 0;
|
|
|
|
} else if ((extend || !data_size) && slot == 0) {
|
|
|
|
mid = 1;
|
|
|
|
} else {
|
|
|
|
mid = slot;
|
|
|
|
if (mid != nritems &&
|
|
|
|
leaf_space_used(l, mid, nritems - mid) +
|
2016-06-23 06:54:23 +08:00
|
|
|
data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
|
2010-07-07 22:51:48 +08:00
|
|
|
if (data_size && !tried_avoid_double)
|
|
|
|
goto push_for_double;
|
2013-10-31 13:03:04 +08:00
|
|
|
split = 2;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (split == 0)
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, ins_key);
|
|
|
|
else
|
|
|
|
btrfs_item_key(l, &disk_key, mid);
|
|
|
|
|
2014-06-15 07:54:12 +08:00
|
|
|
right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
|
|
|
|
&disk_key, 0, l->start, 0);
|
2010-05-16 22:46:25 +08:00
|
|
|
if (IS_ERR(right))
|
2007-10-16 04:14:19 +08:00
|
|
|
return PTR_ERR(right);
|
2010-05-16 22:46:25 +08:00
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
root_add_used(root, fs_info->nodesize);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2016-11-09 01:09:03 +08:00
|
|
|
memzero_extent_buffer(right, 0, sizeof(struct btrfs_header));
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_header_bytenr(right, right->start);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_generation(right, trans->transid);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_owner(right, root->root_key.objectid);
|
|
|
|
btrfs_set_header_level(right, 0);
|
2016-11-10 00:44:25 +08:00
|
|
|
write_extent_buffer_fsid(right, fs_info->fsid);
|
|
|
|
write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid);
|
2009-03-13 22:04:31 +08:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (split == 0) {
|
|
|
|
if (mid <= slot) {
|
|
|
|
btrfs_set_header_nritems(right, 0);
|
2016-06-23 06:54:24 +08:00
|
|
|
insert_ptr(trans, fs_info, path, &disk_key,
|
|
|
|
right->start, path->slots[1] + 1, 1);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_tree_unlock(path->nodes[0]);
|
|
|
|
free_extent_buffer(path->nodes[0]);
|
|
|
|
path->nodes[0] = right;
|
|
|
|
path->slots[0] = 0;
|
|
|
|
path->slots[1] += 1;
|
|
|
|
} else {
|
|
|
|
btrfs_set_header_nritems(right, 0);
|
2016-06-23 06:54:24 +08:00
|
|
|
insert_ptr(trans, fs_info, path, &disk_key,
|
|
|
|
right->start, path->slots[1], 1);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_tree_unlock(path->nodes[0]);
|
|
|
|
free_extent_buffer(path->nodes[0]);
|
|
|
|
path->nodes[0] = right;
|
|
|
|
path->slots[0] = 0;
|
2012-03-01 21:56:26 +08:00
|
|
|
if (path->slots[1] == 0)
|
2014-11-12 12:43:09 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2007-04-05 02:08:15 +08:00
|
|
|
}
|
2016-09-08 05:48:28 +08:00
|
|
|
/*
|
|
|
|
* We create a new leaf 'right' for the required ins_len and
|
|
|
|
* we'll do btrfs_mark_buffer_dirty() on this leaf after copying
|
|
|
|
* the content of ins_len to 'right'.
|
|
|
|
*/
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
return ret;
|
2007-04-05 02:08:15 +08:00
|
|
|
}
|
2007-02-03 00:05:29 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
|
2008-09-24 01:14:14 +08:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
if (split == 2) {
|
2007-10-26 03:42:57 +08:00
|
|
|
BUG_ON(num_doubles != 0);
|
|
|
|
num_doubles++;
|
|
|
|
goto again;
|
2007-04-19 04:15:28 +08:00
|
|
|
}
|
2009-03-13 22:04:31 +08:00
|
|
|
|
2012-03-01 21:56:26 +08:00
|
|
|
return 0;
|
2010-07-07 22:51:48 +08:00
|
|
|
|
|
|
|
push_for_double:
|
|
|
|
push_for_double_split(trans, root, path, data_size);
|
|
|
|
tried_avoid_double = 1;
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
|
2010-07-07 22:51:48 +08:00
|
|
|
return 0;
|
|
|
|
goto again;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, int ins_len)
|
2008-12-10 22:10:46 +08:00
|
|
|
{
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2009-11-12 17:33:58 +08:00
|
|
|
struct btrfs_key key;
|
2008-12-10 22:10:46 +08:00
|
|
|
struct extent_buffer *leaf;
|
2009-11-12 17:33:58 +08:00
|
|
|
struct btrfs_file_extent_item *fi;
|
|
|
|
u64 extent_len = 0;
|
|
|
|
u32 item_size;
|
|
|
|
int ret;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
leaf = path->nodes[0];
|
2009-11-12 17:33:58 +08:00
|
|
|
btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
|
|
|
|
|
|
|
|
BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
|
|
|
|
key.type != BTRFS_EXTENT_CSUM_KEY);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
|
2009-11-12 17:33:58 +08:00
|
|
|
return 0;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
item_size = btrfs_item_size_nr(leaf, path->slots[0]);
|
2009-11-12 17:33:58 +08:00
|
|
|
if (key.type == BTRFS_EXTENT_DATA_KEY) {
|
|
|
|
fi = btrfs_item_ptr(leaf, path->slots[0],
|
|
|
|
struct btrfs_file_extent_item);
|
|
|
|
extent_len = btrfs_file_extent_num_bytes(leaf, fi);
|
|
|
|
}
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
path->keep_locks = 1;
|
2009-11-12 17:33:58 +08:00
|
|
|
path->search_for_split = 1;
|
|
|
|
ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
|
2008-12-10 22:10:46 +08:00
|
|
|
path->search_for_split = 0;
|
2015-01-20 20:40:53 +08:00
|
|
|
if (ret > 0)
|
|
|
|
ret = -EAGAIN;
|
2009-11-12 17:33:58 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto err;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
ret = -EAGAIN;
|
|
|
|
leaf = path->nodes[0];
|
2015-01-20 20:40:53 +08:00
|
|
|
/* if our item isn't there, return now */
|
|
|
|
if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
|
2009-11-12 17:33:58 +08:00
|
|
|
goto err;
|
|
|
|
|
2010-04-02 21:20:18 +08:00
|
|
|
/* the leaf has changed, it now has room. return now */
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
|
2010-04-02 21:20:18 +08:00
|
|
|
goto err;
|
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
if (key.type == BTRFS_EXTENT_DATA_KEY) {
|
|
|
|
fi = btrfs_item_ptr(leaf, path->slots[0],
|
|
|
|
struct btrfs_file_extent_item);
|
|
|
|
if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
|
|
|
|
goto err;
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
|
|
|
|
2009-03-13 23:00:37 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2009-11-12 17:33:58 +08:00
|
|
|
ret = split_leaf(trans, root, &key, path, ins_len, 1);
|
2010-05-16 22:46:25 +08:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
path->keep_locks = 0;
|
2009-03-13 23:00:37 +08:00
|
|
|
btrfs_unlock_up_safe(path, 1);
|
2009-11-12 17:33:58 +08:00
|
|
|
return 0;
|
|
|
|
err:
|
|
|
|
path->keep_locks = 0;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-02-11 01:49:53 +08:00
|
|
|
static noinline int split_item(struct btrfs_fs_info *fs_info,
|
2009-11-12 17:33:58 +08:00
|
|
|
struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *new_key,
|
2009-11-12 17:33:58 +08:00
|
|
|
unsigned long split_offset)
|
|
|
|
{
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
struct btrfs_item *item;
|
|
|
|
struct btrfs_item *new_item;
|
|
|
|
int slot;
|
|
|
|
char *buf;
|
|
|
|
u32 nritems;
|
|
|
|
u32 item_size;
|
|
|
|
u32 orig_offset;
|
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
|
2009-03-13 23:00:37 +08:00
|
|
|
leaf = path->nodes[0];
|
2016-06-23 06:54:24 +08:00
|
|
|
BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
|
2009-03-13 23:00:37 +08:00
|
|
|
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
|
|
|
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(path->slots[0]);
|
2008-12-10 22:10:46 +08:00
|
|
|
orig_offset = btrfs_item_offset(leaf, item);
|
|
|
|
item_size = btrfs_item_size(leaf, item);
|
|
|
|
|
|
|
|
buf = kmalloc(item_size, GFP_NOFS);
|
2009-11-12 17:33:58 +08:00
|
|
|
if (!buf)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2008-12-10 22:10:46 +08:00
|
|
|
read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
|
|
|
|
path->slots[0]), item_size);
|
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
slot = path->slots[0] + 1;
|
2008-12-10 22:10:46 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
|
|
|
if (slot != nritems) {
|
|
|
|
/* shift the items */
|
|
|
|
memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
|
2009-11-12 17:33:58 +08:00
|
|
|
btrfs_item_nr_offset(slot),
|
|
|
|
(nritems - slot) * sizeof(struct btrfs_item));
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, new_key);
|
|
|
|
btrfs_set_item_key(leaf, &disk_key, slot);
|
|
|
|
|
2013-09-16 22:58:09 +08:00
|
|
|
new_item = btrfs_item_nr(slot);
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
btrfs_set_item_offset(leaf, new_item, orig_offset);
|
|
|
|
btrfs_set_item_size(leaf, new_item, item_size - split_offset);
|
|
|
|
|
|
|
|
btrfs_set_item_offset(leaf, item,
|
|
|
|
orig_offset + item_size - split_offset);
|
|
|
|
btrfs_set_item_size(leaf, item, split_offset);
|
|
|
|
|
|
|
|
btrfs_set_header_nritems(leaf, nritems + 1);
|
|
|
|
|
|
|
|
/* write the data for the start of the original item */
|
|
|
|
write_extent_buffer(leaf, buf,
|
|
|
|
btrfs_item_ptr_offset(leaf, path->slots[0]),
|
|
|
|
split_offset);
|
|
|
|
|
|
|
|
/* write the data for the new item */
|
|
|
|
write_extent_buffer(leaf, buf + split_offset,
|
|
|
|
btrfs_item_ptr_offset(leaf, slot),
|
|
|
|
item_size - split_offset);
|
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
|
2008-12-10 22:10:46 +08:00
|
|
|
kfree(buf);
|
2009-11-12 17:33:58 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This function splits a single item into two items,
|
|
|
|
* giving 'new_key' to the new item and splitting the
|
|
|
|
* old one at split_offset (from the start of the item).
|
|
|
|
*
|
|
|
|
* The path may be released by this operation. After
|
|
|
|
* the split, the path is pointing to the old item. The
|
|
|
|
* new item is going to be in the same node as the old one.
|
|
|
|
*
|
|
|
|
* Note, the item being split must be smaller enough to live alone on
|
|
|
|
* a tree block with room for one extra struct btrfs_item
|
|
|
|
*
|
|
|
|
* This allows us to split the item in place, keeping a lock on the
|
|
|
|
* leaf the entire time.
|
|
|
|
*/
|
|
|
|
int btrfs_split_item(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *new_key,
|
2009-11-12 17:33:58 +08:00
|
|
|
unsigned long split_offset)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
ret = setup_leaf_for_split(trans, root, path,
|
|
|
|
sizeof(struct btrfs_item));
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2017-02-11 01:49:53 +08:00
|
|
|
ret = split_item(root->fs_info, path, new_key, split_offset);
|
2008-12-10 22:10:46 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-11-12 17:33:58 +08:00
|
|
|
/*
|
|
|
|
* This function duplicate a item, giving 'new_key' to the new item.
|
|
|
|
* It guarantees both items live in the same tree leaf and the new item
|
|
|
|
* is contiguous with the original item.
|
|
|
|
*
|
|
|
|
* This allows us to split file extent in place, keeping a lock on the
|
|
|
|
* leaf the entire time.
|
|
|
|
*/
|
|
|
|
int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *new_key)
|
2009-11-12 17:33:58 +08:00
|
|
|
{
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
int ret;
|
|
|
|
u32 item_size;
|
|
|
|
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
item_size = btrfs_item_size_nr(leaf, path->slots[0]);
|
|
|
|
ret = setup_leaf_for_split(trans, root, path,
|
|
|
|
item_size + sizeof(struct btrfs_item));
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
path->slots[0]++;
|
2013-04-16 13:18:22 +08:00
|
|
|
setup_items_for_insert(root, path, new_key, &item_size,
|
2012-03-01 21:56:26 +08:00
|
|
|
item_size, item_size +
|
|
|
|
sizeof(struct btrfs_item), 1);
|
2009-11-12 17:33:58 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
memcpy_extent_buffer(leaf,
|
|
|
|
btrfs_item_ptr_offset(leaf, path->slots[0]),
|
|
|
|
btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
|
|
|
|
item_size);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* make the item pointed to by the path smaller. new_size indicates
|
|
|
|
* how small to make it, and from_end tells us if we just chop bytes
|
|
|
|
* off the end of the item or if we shift the item to chop bytes off
|
|
|
|
* the front.
|
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_path *path, u32 new_size, int from_end)
|
2007-04-18 01:26:50 +08:00
|
|
|
{
|
|
|
|
int slot;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
|
|
|
struct btrfs_item *item;
|
2007-04-18 01:26:50 +08:00
|
|
|
u32 nritems;
|
|
|
|
unsigned int data_end;
|
|
|
|
unsigned int old_data_start;
|
|
|
|
unsigned int old_size;
|
|
|
|
unsigned int size_diff;
|
|
|
|
int i;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
|
|
|
btrfs_init_map_token(&token);
|
2007-04-18 01:26:50 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
2007-11-01 23:28:41 +08:00
|
|
|
slot = path->slots[0];
|
|
|
|
|
|
|
|
old_size = btrfs_item_size_nr(leaf, slot);
|
|
|
|
if (old_size == new_size)
|
2012-03-01 21:56:26 +08:00
|
|
|
return;
|
2007-04-18 01:26:50 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
2016-06-23 06:54:24 +08:00
|
|
|
data_end = leaf_data_end(fs_info, leaf);
|
2007-04-18 01:26:50 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
old_data_start = btrfs_item_offset_nr(leaf, slot);
|
2007-11-01 23:28:41 +08:00
|
|
|
|
2007-04-18 01:26:50 +08:00
|
|
|
size_diff = old_size - new_size;
|
|
|
|
|
|
|
|
BUG_ON(slot < 0);
|
|
|
|
BUG_ON(slot >= nritems);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* item0..itemN ... dataN.offset..dataN.size .. data0.size
|
|
|
|
*/
|
|
|
|
/* first correct the data pointers */
|
|
|
|
for (i = slot; i < nritems; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 ioff;
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(leaf, item, &token);
|
|
|
|
btrfs_set_token_item_offset(leaf, item,
|
|
|
|
ioff + size_diff, &token);
|
2007-04-18 01:26:50 +08:00
|
|
|
}
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2007-04-18 01:26:50 +08:00
|
|
|
/* shift the data */
|
2007-11-01 23:28:41 +08:00
|
|
|
if (from_end) {
|
|
|
|
memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
|
|
|
|
data_end + size_diff, btrfs_leaf_data(leaf) +
|
|
|
|
data_end, old_data_start + new_size - data_end);
|
|
|
|
} else {
|
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
u64 offset;
|
|
|
|
|
|
|
|
btrfs_item_key(leaf, &disk_key, slot);
|
|
|
|
|
|
|
|
if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
|
|
|
|
unsigned long ptr;
|
|
|
|
struct btrfs_file_extent_item *fi;
|
|
|
|
|
|
|
|
fi = btrfs_item_ptr(leaf, slot,
|
|
|
|
struct btrfs_file_extent_item);
|
|
|
|
fi = (struct btrfs_file_extent_item *)(
|
|
|
|
(unsigned long)fi - size_diff);
|
|
|
|
|
|
|
|
if (btrfs_file_extent_type(leaf, fi) ==
|
|
|
|
BTRFS_FILE_EXTENT_INLINE) {
|
|
|
|
ptr = btrfs_item_ptr_offset(leaf, slot);
|
|
|
|
memmove_extent_buffer(leaf, ptr,
|
2009-01-06 10:25:51 +08:00
|
|
|
(unsigned long)fi,
|
2014-07-24 23:34:58 +08:00
|
|
|
BTRFS_FILE_EXTENT_INLINE_DATA_START);
|
2007-11-01 23:28:41 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
|
|
|
|
data_end + size_diff, btrfs_leaf_data(leaf) +
|
|
|
|
data_end, old_data_start - data_end);
|
|
|
|
|
|
|
|
offset = btrfs_disk_key_offset(&disk_key);
|
|
|
|
btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
|
|
|
|
btrfs_set_item_key(leaf, &disk_key, slot);
|
|
|
|
if (slot == 0)
|
2016-06-23 06:54:23 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2007-11-01 23:28:41 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(slot);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_item_size(leaf, item, new_size);
|
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-04-18 01:26:50 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
|
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2007-04-18 01:26:50 +08:00
|
|
|
BUG();
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
2007-04-18 01:26:50 +08:00
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
2013-05-07 18:23:30 +08:00
|
|
|
* make the item pointed to by the path bigger, data_size is the added size.
|
2008-09-30 03:18:18 +08:00
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
|
2012-03-01 21:56:26 +08:00
|
|
|
u32 data_size)
|
2007-04-16 21:22:45 +08:00
|
|
|
{
|
|
|
|
int slot;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
|
|
|
struct btrfs_item *item;
|
2007-04-16 21:22:45 +08:00
|
|
|
u32 nritems;
|
|
|
|
unsigned int data_end;
|
|
|
|
unsigned int old_data;
|
|
|
|
unsigned int old_size;
|
|
|
|
int i;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
|
|
|
btrfs_init_map_token(&token);
|
2007-04-16 21:22:45 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
2007-04-16 21:22:45 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
2016-06-23 06:54:24 +08:00
|
|
|
data_end = leaf_data_end(fs_info, leaf);
|
2007-04-16 21:22:45 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
|
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2007-04-16 21:22:45 +08:00
|
|
|
BUG();
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
2007-04-16 21:22:45 +08:00
|
|
|
slot = path->slots[0];
|
2007-10-16 04:14:19 +08:00
|
|
|
old_data = btrfs_item_end_nr(leaf, slot);
|
2007-04-16 21:22:45 +08:00
|
|
|
|
|
|
|
BUG_ON(slot < 0);
|
2007-10-16 04:18:25 +08:00
|
|
|
if (slot >= nritems) {
|
2016-06-23 06:54:24 +08:00
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_crit(fs_info, "slot %d too large, nritems %d",
|
|
|
|
slot, nritems);
|
2007-10-16 04:18:25 +08:00
|
|
|
BUG_ON(1);
|
|
|
|
}
|
2007-04-16 21:22:45 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* item0..itemN ... dataN.offset..dataN.size .. data0.size
|
|
|
|
*/
|
|
|
|
/* first correct the data pointers */
|
|
|
|
for (i = slot; i < nritems; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 ioff;
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(leaf, item, &token);
|
|
|
|
btrfs_set_token_item_offset(leaf, item,
|
|
|
|
ioff - data_size, &token);
|
2007-04-16 21:22:45 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-04-16 21:22:45 +08:00
|
|
|
/* shift the data */
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
|
2007-04-16 21:22:45 +08:00
|
|
|
data_end - data_size, btrfs_leaf_data(leaf) +
|
|
|
|
data_end, old_data - data_end);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-04-16 21:22:45 +08:00
|
|
|
data_end = old_data;
|
2007-10-16 04:14:19 +08:00
|
|
|
old_size = btrfs_item_size_nr(leaf, slot);
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(slot);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_item_size(leaf, item, old_size + data_size);
|
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-04-16 21:22:45 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
|
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2007-04-16 21:22:45 +08:00
|
|
|
BUG();
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
2007-04-16 21:22:45 +08:00
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
2009-03-13 22:04:31 +08:00
|
|
|
* this is a helper for btrfs_insert_empty_items, the main goal here is
|
|
|
|
* to save stack depth by doing the bulk of the work in a function
|
|
|
|
* that doesn't call btrfs_search_slot
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2013-04-16 13:18:22 +08:00
|
|
|
void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *cpu_key, u32 *data_size,
|
2012-03-01 21:56:26 +08:00
|
|
|
u32 total_data, u32 total_size, int nr)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_item *item;
|
2008-01-30 04:15:18 +08:00
|
|
|
int i;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 nritems;
|
2007-01-27 04:51:26 +08:00
|
|
|
unsigned int data_end;
|
2007-03-13 04:22:34 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
2009-03-13 22:04:31 +08:00
|
|
|
struct extent_buffer *leaf;
|
|
|
|
int slot;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
2014-07-29 02:34:35 +08:00
|
|
|
if (path->slots[0] == 0) {
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, cpu_key);
|
2016-06-23 06:54:23 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2014-07-29 02:34:35 +08:00
|
|
|
}
|
|
|
|
btrfs_unlock_up_safe(path, 1);
|
|
|
|
|
2012-03-03 20:40:03 +08:00
|
|
|
btrfs_init_map_token(&token);
|
2007-03-13 04:22:34 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
2009-03-13 22:04:31 +08:00
|
|
|
slot = path->slots[0];
|
2007-02-03 00:05:29 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
2016-06-23 06:54:24 +08:00
|
|
|
data_end = leaf_data_end(fs_info, leaf);
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
|
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_crit(fs_info, "not enough freespace need %u have %d",
|
2016-06-23 06:54:24 +08:00
|
|
|
total_size, btrfs_leaf_free_space(fs_info, leaf));
|
2007-01-27 04:51:26 +08:00
|
|
|
BUG();
|
2007-04-05 02:08:15 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-01-27 04:51:26 +08:00
|
|
|
if (slot != nritems) {
|
2007-10-16 04:14:19 +08:00
|
|
|
unsigned int old_data = btrfs_item_end_nr(leaf, slot);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (old_data < data_end) {
|
2016-06-23 06:54:24 +08:00
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
|
2016-09-20 22:05:00 +08:00
|
|
|
slot, old_data, data_end);
|
2007-10-16 04:14:19 +08:00
|
|
|
BUG_ON(1);
|
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
/*
|
|
|
|
* item0..itemN ... dataN.offset..dataN.size .. data0.size
|
|
|
|
*/
|
|
|
|
/* first correct the data pointers */
|
2007-03-13 08:12:07 +08:00
|
|
|
for (i = slot; i < nritems; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 ioff;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2016-09-20 22:05:01 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(leaf, item, &token);
|
|
|
|
btrfs_set_token_item_offset(leaf, item,
|
|
|
|
ioff - total_data, &token);
|
2007-03-13 08:12:07 +08:00
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
/* shift the items */
|
2008-01-30 04:15:18 +08:00
|
|
|
memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_nr_offset(slot),
|
2007-03-31 02:27:56 +08:00
|
|
|
(nritems - slot) * sizeof(struct btrfs_item));
|
2007-01-27 04:51:26 +08:00
|
|
|
|
|
|
|
/* shift the data */
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
|
2008-01-30 04:15:18 +08:00
|
|
|
data_end - total_data, btrfs_leaf_data(leaf) +
|
2007-03-31 02:27:56 +08:00
|
|
|
data_end, old_data - data_end);
|
2007-01-27 04:51:26 +08:00
|
|
|
data_end = old_data;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-03-16 00:56:47 +08:00
|
|
|
/* setup the item for the new data */
|
2008-01-30 04:15:18 +08:00
|
|
|
for (i = 0; i < nr; i++) {
|
|
|
|
btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
|
|
|
|
btrfs_set_item_key(leaf, &disk_key, slot + i);
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(slot + i);
|
2012-03-03 20:40:03 +08:00
|
|
|
btrfs_set_token_item_offset(leaf, item,
|
|
|
|
data_end - data_size[i], &token);
|
2008-01-30 04:15:18 +08:00
|
|
|
data_end -= data_size[i];
|
2012-03-03 20:40:03 +08:00
|
|
|
btrfs_set_token_item_size(leaf, item, data_size[i], &token);
|
2008-01-30 04:15:18 +08:00
|
|
|
}
|
2009-03-13 22:04:31 +08:00
|
|
|
|
2008-01-30 04:15:18 +08:00
|
|
|
btrfs_set_header_nritems(leaf, nritems + nr);
|
2009-03-13 23:00:37 +08:00
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-03-01 05:35:06 +08:00
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
|
|
|
|
btrfs_print_leaf(fs_info, leaf);
|
2007-01-27 04:51:26 +08:00
|
|
|
BUG();
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
2009-03-13 22:04:31 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a key and some data, insert items into the tree.
|
|
|
|
* This does all the path init required, making room in the tree if needed.
|
|
|
|
*/
|
|
|
|
int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
2017-01-18 15:24:37 +08:00
|
|
|
const struct btrfs_key *cpu_key, u32 *data_size,
|
2009-03-13 22:04:31 +08:00
|
|
|
int nr)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
int slot;
|
|
|
|
int i;
|
|
|
|
u32 total_size = 0;
|
|
|
|
u32 total_data = 0;
|
|
|
|
|
|
|
|
for (i = 0; i < nr; i++)
|
|
|
|
total_data += data_size[i];
|
|
|
|
|
|
|
|
total_size = total_data + (nr * sizeof(struct btrfs_item));
|
|
|
|
ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
|
|
|
|
if (ret == 0)
|
|
|
|
return -EEXIST;
|
|
|
|
if (ret < 0)
|
2012-03-01 21:56:26 +08:00
|
|
|
return ret;
|
2009-03-13 22:04:31 +08:00
|
|
|
|
|
|
|
slot = path->slots[0];
|
|
|
|
BUG_ON(slot < 0);
|
|
|
|
|
2013-04-16 13:18:22 +08:00
|
|
|
setup_items_for_insert(root, path, cpu_key, data_size,
|
2009-03-13 22:04:31 +08:00
|
|
|
total_data, total_size, nr);
|
2012-03-01 21:56:26 +08:00
|
|
|
return 0;
|
2007-03-16 00:56:47 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a key and some data, insert an item into the tree.
|
|
|
|
* This does all the path init required, making room in the tree if needed.
|
|
|
|
*/
|
2017-01-18 15:24:37 +08:00
|
|
|
int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
|
|
|
|
const struct btrfs_key *cpu_key, void *data,
|
|
|
|
u32 data_size)
|
2007-03-16 00:56:47 +08:00
|
|
|
{
|
|
|
|
int ret = 0;
|
2007-04-02 22:50:19 +08:00
|
|
|
struct btrfs_path *path;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
|
|
|
unsigned long ptr;
|
2007-03-16 00:56:47 +08:00
|
|
|
|
2007-04-02 22:50:19 +08:00
|
|
|
path = btrfs_alloc_path();
|
2011-03-23 16:14:16 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2007-04-02 22:50:19 +08:00
|
|
|
ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
|
2007-03-16 00:56:47 +08:00
|
|
|
if (!ret) {
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
|
|
|
|
write_extent_buffer(leaf, data, ptr, data_size);
|
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-03-16 00:56:47 +08:00
|
|
|
}
|
2007-04-02 22:50:19 +08:00
|
|
|
btrfs_free_path(path);
|
2007-03-01 05:35:06 +08:00
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
2007-02-24 19:24:44 +08:00
|
|
|
* delete the pointer from a given node.
|
2007-02-03 00:05:29 +08:00
|
|
|
*
|
2008-09-30 03:18:18 +08:00
|
|
|
* the tree should have been previously balanced so the deletion does not
|
|
|
|
* empty a node.
|
2007-02-03 00:05:29 +08:00
|
|
|
*/
|
2013-04-16 13:18:22 +08:00
|
|
|
static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
|
|
|
|
int level, int slot)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *parent = path->nodes[level];
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 nritems;
|
2012-05-26 17:45:21 +08:00
|
|
|
int ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(parent);
|
2009-01-06 10:25:51 +08:00
|
|
|
if (slot != nritems - 1) {
|
2012-10-19 17:50:54 +08:00
|
|
|
if (level)
|
2016-06-23 06:54:23 +08:00
|
|
|
tree_mod_log_eb_move(fs_info, parent, slot,
|
2012-05-26 17:45:21 +08:00
|
|
|
slot + 1, nritems - slot - 1);
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(parent,
|
|
|
|
btrfs_node_key_ptr_offset(slot),
|
|
|
|
btrfs_node_key_ptr_offset(slot + 1),
|
2007-03-31 02:27:56 +08:00
|
|
|
sizeof(struct btrfs_key_ptr) *
|
|
|
|
(nritems - slot - 1));
|
2012-12-19 08:35:32 +08:00
|
|
|
} else if (level) {
|
2016-06-23 06:54:23 +08:00
|
|
|
ret = tree_mod_log_insert_key(fs_info, parent, slot,
|
2013-07-02 04:18:19 +08:00
|
|
|
MOD_LOG_KEY_REMOVE, GFP_NOFS);
|
2012-12-19 08:35:32 +08:00
|
|
|
BUG_ON(ret < 0);
|
2007-03-02 01:04:21 +08:00
|
|
|
}
|
2012-05-26 17:45:21 +08:00
|
|
|
|
2007-03-13 00:01:18 +08:00
|
|
|
nritems--;
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_nritems(parent, nritems);
|
2007-03-13 00:01:18 +08:00
|
|
|
if (nritems == 0 && parent == root->node) {
|
2007-10-16 04:14:19 +08:00
|
|
|
BUG_ON(btrfs_header_level(root->node) != 1);
|
2007-03-02 01:04:21 +08:00
|
|
|
/* just turn the root into a leaf and break */
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_header_level(root->node, 0);
|
2007-03-02 01:04:21 +08:00
|
|
|
} else if (slot == 0) {
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
|
|
|
|
btrfs_node_key(parent, &disk_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, level + 1);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2007-03-31 02:27:56 +08:00
|
|
|
btrfs_mark_buffer_dirty(parent);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2008-10-02 07:05:46 +08:00
|
|
|
/*
|
|
|
|
* a helper function to delete the leaf pointed to by path->slots[1] and
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
* path->nodes[1].
|
2008-10-02 07:05:46 +08:00
|
|
|
*
|
|
|
|
* This deletes the pointer in path->nodes[1] and frees the leaf
|
|
|
|
* block extent. zero is returned if it all worked out, < 0 otherwise.
|
|
|
|
*
|
|
|
|
* The path must have already been setup for deleting the leaf, including
|
|
|
|
* all the proper balancing. path->nodes[1] must be locked.
|
|
|
|
*/
|
2012-03-01 21:56:26 +08:00
|
|
|
static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
|
|
|
struct extent_buffer *leaf)
|
2008-10-02 07:05:46 +08:00
|
|
|
{
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
WARN_ON(btrfs_header_generation(leaf) != trans->transid);
|
2013-04-16 13:18:22 +08:00
|
|
|
del_ptr(root, path, 1, path->slots[1]);
|
2008-10-02 07:05:46 +08:00
|
|
|
|
2009-02-04 22:31:28 +08:00
|
|
|
/*
|
|
|
|
* btrfs_free_extent is expensive, we want to make sure we
|
|
|
|
* aren't holding any locks when we call it
|
|
|
|
*/
|
|
|
|
btrfs_unlock_up_safe(path, 0);
|
|
|
|
|
2010-05-16 22:46:25 +08:00
|
|
|
root_sub_used(root, leaf->len);
|
|
|
|
|
2012-03-10 05:01:49 +08:00
|
|
|
extent_buffer_get(leaf);
|
2012-05-16 23:04:52 +08:00
|
|
|
btrfs_free_tree_block(trans, root, leaf, 0, 1);
|
2012-03-10 05:01:49 +08:00
|
|
|
free_extent_buffer_stale(leaf);
|
2008-10-02 07:05:46 +08:00
|
|
|
}
|
2007-02-03 00:05:29 +08:00
|
|
|
/*
|
|
|
|
* delete the item at the leaf level in path. If that empties
|
|
|
|
* the leaf, remove it from the tree
|
|
|
|
*/
|
2008-01-30 04:11:36 +08:00
|
|
|
int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, int slot, int nr)
|
2007-01-27 04:51:26 +08:00
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
|
|
|
struct btrfs_item *item;
|
2015-08-24 00:01:42 +08:00
|
|
|
u32 last_off;
|
|
|
|
u32 dsize = 0;
|
2007-03-01 05:35:06 +08:00
|
|
|
int ret = 0;
|
|
|
|
int wret;
|
2008-01-30 04:11:36 +08:00
|
|
|
int i;
|
2007-03-13 00:01:18 +08:00
|
|
|
u32 nritems;
|
2012-03-03 20:40:03 +08:00
|
|
|
struct btrfs_map_token token;
|
|
|
|
|
|
|
|
btrfs_init_map_token(&token);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
2008-01-30 04:11:36 +08:00
|
|
|
last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
|
|
|
|
|
|
|
|
for (i = 0; i < nr; i++)
|
|
|
|
dsize += btrfs_item_size_nr(leaf, slot + i);
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
2007-01-27 04:51:26 +08:00
|
|
|
|
2008-01-30 04:11:36 +08:00
|
|
|
if (slot + nr != nritems) {
|
2016-06-23 06:54:24 +08:00
|
|
|
int data_end = leaf_data_end(fs_info, leaf);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
|
2007-03-31 02:27:56 +08:00
|
|
|
data_end + dsize,
|
|
|
|
btrfs_leaf_data(leaf) + data_end,
|
2008-01-30 04:11:36 +08:00
|
|
|
last_off - data_end);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2008-01-30 04:11:36 +08:00
|
|
|
for (i = slot + nr; i < nritems; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
u32 ioff;
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2013-09-16 22:58:09 +08:00
|
|
|
item = btrfs_item_nr(i);
|
2012-03-03 20:40:03 +08:00
|
|
|
ioff = btrfs_token_item_offset(leaf, item, &token);
|
|
|
|
btrfs_set_token_item_offset(leaf, item,
|
|
|
|
ioff + dsize, &token);
|
2007-03-13 08:12:07 +08:00
|
|
|
}
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
|
2008-01-30 04:11:36 +08:00
|
|
|
btrfs_item_nr_offset(slot + nr),
|
2007-03-31 02:27:56 +08:00
|
|
|
sizeof(struct btrfs_item) *
|
2008-01-30 04:11:36 +08:00
|
|
|
(nritems - slot - nr));
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2008-01-30 04:11:36 +08:00
|
|
|
btrfs_set_header_nritems(leaf, nritems - nr);
|
|
|
|
nritems -= nr;
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/* delete the leaf if we've emptied it */
|
2007-03-13 00:01:18 +08:00
|
|
|
if (nritems == 0) {
|
2007-10-16 04:14:19 +08:00
|
|
|
if (leaf == root->node) {
|
|
|
|
btrfs_set_header_level(leaf, 0);
|
2007-02-23 21:38:36 +08:00
|
|
|
} else {
|
2010-05-16 22:46:25 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2017-02-11 01:47:57 +08:00
|
|
|
clean_tree_block(fs_info, leaf);
|
2012-03-01 21:56:26 +08:00
|
|
|
btrfs_del_leaf(trans, root, path, leaf);
|
2007-02-23 21:38:36 +08:00
|
|
|
}
|
2007-01-27 04:51:26 +08:00
|
|
|
} else {
|
2007-03-13 00:01:18 +08:00
|
|
|
int used = leaf_space_used(leaf, 0, nritems);
|
2007-03-01 05:35:06 +08:00
|
|
|
if (slot == 0) {
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_disk_key disk_key;
|
|
|
|
|
|
|
|
btrfs_item_key(leaf, &disk_key, 0);
|
2016-06-23 06:54:23 +08:00
|
|
|
fixup_low_keys(fs_info, path, &disk_key, 1);
|
2007-03-01 05:35:06 +08:00
|
|
|
}
|
|
|
|
|
2007-02-03 00:05:29 +08:00
|
|
|
/* delete the leaf if it is mostly empty */
|
2016-06-23 06:54:23 +08:00
|
|
|
if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
|
2007-01-27 04:51:26 +08:00
|
|
|
/* push_leaf_left fixes the path.
|
|
|
|
* make sure the path still points to our leaf
|
|
|
|
* for possible call to del_ptr below
|
|
|
|
*/
|
2007-01-27 05:38:42 +08:00
|
|
|
slot = path->slots[1];
|
2007-10-16 04:14:19 +08:00
|
|
|
extent_buffer_get(leaf);
|
|
|
|
|
2009-03-13 23:00:37 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2010-07-07 22:51:48 +08:00
|
|
|
wret = push_leaf_left(trans, root, path, 1, 1,
|
|
|
|
1, (u32)-1);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (wret < 0 && wret != -ENOSPC)
|
2007-03-01 05:35:06 +08:00
|
|
|
ret = wret;
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
if (path->nodes[0] == leaf &&
|
|
|
|
btrfs_header_nritems(leaf)) {
|
2010-07-07 22:51:48 +08:00
|
|
|
wret = push_leaf_right(trans, root, path, 1,
|
|
|
|
1, 1, 0);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (wret < 0 && wret != -ENOSPC)
|
2007-03-01 05:35:06 +08:00
|
|
|
ret = wret;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
|
|
|
if (btrfs_header_nritems(leaf) == 0) {
|
2008-10-02 07:05:46 +08:00
|
|
|
path->slots[1] = slot;
|
2012-03-01 21:56:26 +08:00
|
|
|
btrfs_del_leaf(trans, root, path, leaf);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(leaf);
|
2012-03-01 21:56:26 +08:00
|
|
|
ret = 0;
|
2007-02-24 19:24:44 +08:00
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
/* if we're still in the path, make sure
|
|
|
|
* we're dirty. Otherwise, one of the
|
|
|
|
* push_leaf functions must have already
|
|
|
|
* dirtied this buffer
|
|
|
|
*/
|
|
|
|
if (path->nodes[0] == leaf)
|
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(leaf);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
2007-03-23 22:01:08 +08:00
|
|
|
} else {
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
}
|
2007-03-01 05:35:06 +08:00
|
|
|
return ret;
|
2007-01-27 04:51:26 +08:00
|
|
|
}
|
|
|
|
|
2007-12-11 22:25:06 +08:00
|
|
|
/*
|
2008-06-26 04:01:30 +08:00
|
|
|
* search the tree again to find a leaf with lesser keys
|
2007-12-11 22:25:06 +08:00
|
|
|
* returns 0 if it found something or 1 if there are no lesser leaves.
|
|
|
|
* returns < 0 on io errors.
|
2008-09-30 03:18:18 +08:00
|
|
|
*
|
|
|
|
* This may release the path, and so you may lose any locks held at the
|
|
|
|
* time you call it.
|
2007-12-11 22:25:06 +08:00
|
|
|
*/
|
2013-10-23 00:18:51 +08:00
|
|
|
int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
|
2007-12-11 22:25:06 +08:00
|
|
|
{
|
2008-06-26 04:01:30 +08:00
|
|
|
struct btrfs_key key;
|
|
|
|
struct btrfs_disk_key found_key;
|
|
|
|
int ret;
|
2007-12-11 22:25:06 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
|
2007-12-11 22:25:06 +08:00
|
|
|
|
2013-10-15 07:12:27 +08:00
|
|
|
if (key.offset > 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
key.offset--;
|
2013-10-15 07:12:27 +08:00
|
|
|
} else if (key.type > 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
key.type--;
|
2013-10-15 07:12:27 +08:00
|
|
|
key.offset = (u64)-1;
|
|
|
|
} else if (key.objectid > 0) {
|
2008-06-26 04:01:30 +08:00
|
|
|
key.objectid--;
|
2013-10-15 07:12:27 +08:00
|
|
|
key.type = (u8)-1;
|
|
|
|
key.offset = (u64)-1;
|
|
|
|
} else {
|
2008-06-26 04:01:30 +08:00
|
|
|
return 1;
|
2013-10-15 07:12:27 +08:00
|
|
|
}
|
2007-12-11 22:25:06 +08:00
|
|
|
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2008-06-26 04:01:30 +08:00
|
|
|
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
btrfs_item_key(path->nodes[0], &found_key, 0);
|
|
|
|
ret = comp_keys(&found_key, &key);
|
2014-06-09 20:22:13 +08:00
|
|
|
/*
|
|
|
|
* We might have had an item with the previous key in the tree right
|
|
|
|
* before we released our path. And after we released our path, that
|
|
|
|
* item might have been pushed to the first slot (0) of the leaf we
|
|
|
|
* were holding due to a tree balance. Alternatively, an item with the
|
|
|
|
* previous key can exist as the only element of a leaf (big fat item).
|
|
|
|
* Therefore account for these 2 cases, so that our callers (like
|
|
|
|
* btrfs_previous_item) don't miss an existing item with a key matching
|
|
|
|
* the previous key we computed above.
|
|
|
|
*/
|
|
|
|
if (ret <= 0)
|
2008-06-26 04:01:30 +08:00
|
|
|
return 0;
|
|
|
|
return 1;
|
2007-12-11 22:25:06 +08:00
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
/*
|
|
|
|
* A helper function to walk down the tree starting at min_key, and looking
|
2013-02-01 02:21:12 +08:00
|
|
|
* for nodes or leaves that are have a minimum transaction id.
|
|
|
|
* This is used by the btree defrag code, and tree logging
|
2008-06-26 04:01:31 +08:00
|
|
|
*
|
|
|
|
* This does not cow, but it does stuff the starting key it finds back
|
|
|
|
* into min_key, so you can call btrfs_search_slot with cow=1 on the
|
|
|
|
* key and get a writable path.
|
|
|
|
*
|
|
|
|
* This does lock as it descends, and path->keep_locks should be set
|
|
|
|
* to 1 by the caller.
|
|
|
|
*
|
|
|
|
* This honors path->lowest_level to prevent descent past a given level
|
|
|
|
* of the tree.
|
|
|
|
*
|
2008-09-30 03:18:18 +08:00
|
|
|
* min_trans indicates the oldest transaction that you are interested
|
|
|
|
* in walking through. Any nodes or leaves older than min_trans are
|
|
|
|
* skipped over (without reading them).
|
|
|
|
*
|
2008-06-26 04:01:31 +08:00
|
|
|
* returns zero if something useful was found, < 0 on error and 1 if there
|
|
|
|
* was nothing in the tree that matched the search criteria.
|
|
|
|
*/
|
|
|
|
int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
|
2013-02-01 02:21:12 +08:00
|
|
|
struct btrfs_path *path,
|
2008-06-26 04:01:31 +08:00
|
|
|
u64 min_trans)
|
|
|
|
{
|
2016-06-23 06:54:24 +08:00
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
2008-06-26 04:01:31 +08:00
|
|
|
struct extent_buffer *cur;
|
|
|
|
struct btrfs_key found_key;
|
|
|
|
int slot;
|
2008-07-25 00:19:49 +08:00
|
|
|
int sret;
|
2008-06-26 04:01:31 +08:00
|
|
|
u32 nritems;
|
|
|
|
int level;
|
|
|
|
int ret = 1;
|
2014-08-05 02:37:21 +08:00
|
|
|
int keep_locks = path->keep_locks;
|
2008-06-26 04:01:31 +08:00
|
|
|
|
2014-08-05 02:37:21 +08:00
|
|
|
path->keep_locks = 1;
|
2008-06-26 04:01:31 +08:00
|
|
|
again:
|
2011-07-17 03:23:14 +08:00
|
|
|
cur = btrfs_read_lock_root_node(root);
|
2008-06-26 04:01:31 +08:00
|
|
|
level = btrfs_header_level(cur);
|
2008-09-06 04:13:11 +08:00
|
|
|
WARN_ON(path->nodes[level]);
|
2008-06-26 04:01:31 +08:00
|
|
|
path->nodes[level] = cur;
|
2011-07-17 03:23:14 +08:00
|
|
|
path->locks[level] = BTRFS_READ_LOCK;
|
2008-06-26 04:01:31 +08:00
|
|
|
|
|
|
|
if (btrfs_header_generation(cur) < min_trans) {
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2008-06-26 04:01:31 +08:00
|
|
|
nritems = btrfs_header_nritems(cur);
|
|
|
|
level = btrfs_header_level(cur);
|
2008-07-25 00:19:49 +08:00
|
|
|
sret = bin_search(cur, min_key, level, &slot);
|
2008-06-26 04:01:31 +08:00
|
|
|
|
2008-10-02 07:05:46 +08:00
|
|
|
/* at the lowest level, we're done, setup the path and exit */
|
|
|
|
if (level == path->lowest_level) {
|
2008-09-06 04:13:11 +08:00
|
|
|
if (slot >= nritems)
|
|
|
|
goto find_next_key;
|
2008-06-26 04:01:31 +08:00
|
|
|
ret = 0;
|
|
|
|
path->slots[level] = slot;
|
|
|
|
btrfs_item_key_to_cpu(cur, &found_key, slot);
|
|
|
|
goto out;
|
|
|
|
}
|
2008-07-25 00:19:49 +08:00
|
|
|
if (sret && slot > 0)
|
|
|
|
slot--;
|
2008-06-26 04:01:31 +08:00
|
|
|
/*
|
2013-02-01 02:21:12 +08:00
|
|
|
* check this node pointer against the min_trans parameters.
|
|
|
|
* If it is too old, old, skip to the next one.
|
2008-06-26 04:01:31 +08:00
|
|
|
*/
|
2009-01-06 10:25:51 +08:00
|
|
|
while (slot < nritems) {
|
2008-06-26 04:01:31 +08:00
|
|
|
u64 gen;
|
2008-09-06 04:13:11 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
gen = btrfs_node_ptr_generation(cur, slot);
|
|
|
|
if (gen < min_trans) {
|
|
|
|
slot++;
|
|
|
|
continue;
|
|
|
|
}
|
2013-02-01 02:21:12 +08:00
|
|
|
break;
|
2008-06-26 04:01:31 +08:00
|
|
|
}
|
2008-09-06 04:13:11 +08:00
|
|
|
find_next_key:
|
2008-06-26 04:01:31 +08:00
|
|
|
/*
|
|
|
|
* we didn't find a candidate key in this node, walk forward
|
|
|
|
* and find another one
|
|
|
|
*/
|
|
|
|
if (slot >= nritems) {
|
2008-09-06 04:13:11 +08:00
|
|
|
path->slots[level] = slot;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2008-09-06 04:13:11 +08:00
|
|
|
sret = btrfs_find_next_key(root, path, min_key, level,
|
2013-02-01 02:21:12 +08:00
|
|
|
min_trans);
|
2008-09-06 04:13:11 +08:00
|
|
|
if (sret == 0) {
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2008-06-26 04:01:31 +08:00
|
|
|
goto again;
|
|
|
|
} else {
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/* save our key for returning back */
|
|
|
|
btrfs_node_key_to_cpu(cur, &found_key, slot);
|
|
|
|
path->slots[level] = slot;
|
|
|
|
if (level == path->lowest_level) {
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2016-06-23 06:54:24 +08:00
|
|
|
cur = read_node_slot(fs_info, cur, slot);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(cur)) {
|
|
|
|
ret = PTR_ERR(cur);
|
|
|
|
goto out;
|
|
|
|
}
|
2008-06-26 04:01:31 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_read_lock(cur);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2011-07-17 03:23:14 +08:00
|
|
|
path->locks[level - 1] = BTRFS_READ_LOCK;
|
2008-06-26 04:01:31 +08:00
|
|
|
path->nodes[level - 1] = cur;
|
2012-03-20 03:54:38 +08:00
|
|
|
unlock_up(path, level, 1, 0, NULL);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_clear_path_blocking(path, NULL, 0);
|
2008-06-26 04:01:31 +08:00
|
|
|
}
|
|
|
|
out:
|
2014-08-05 02:37:21 +08:00
|
|
|
path->keep_locks = keep_locks;
|
|
|
|
if (ret == 0) {
|
|
|
|
btrfs_unlock_up_safe(path, path->lowest_level + 1);
|
|
|
|
btrfs_set_path_blocking(path);
|
2008-06-26 04:01:31 +08:00
|
|
|
memcpy(min_key, &found_key, sizeof(found_key));
|
2014-08-05 02:37:21 +08:00
|
|
|
}
|
2008-06-26 04:01:31 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
static int tree_move_down(struct btrfs_fs_info *fs_info,
|
2012-06-06 03:07:48 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
int *level, int root_level)
|
|
|
|
{
|
2016-07-06 03:10:14 +08:00
|
|
|
struct extent_buffer *eb;
|
|
|
|
|
2012-08-08 04:25:13 +08:00
|
|
|
BUG_ON(*level == 0);
|
2016-06-23 06:54:24 +08:00
|
|
|
eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (IS_ERR(eb))
|
|
|
|
return PTR_ERR(eb);
|
|
|
|
|
|
|
|
path->nodes[*level - 1] = eb;
|
2012-06-06 03:07:48 +08:00
|
|
|
path->slots[*level - 1] = 0;
|
|
|
|
(*level)--;
|
2016-07-06 03:10:14 +08:00
|
|
|
return 0;
|
2012-06-06 03:07:48 +08:00
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
static int tree_move_next_or_upnext(struct btrfs_fs_info *fs_info,
|
2012-06-06 03:07:48 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
int *level, int root_level)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
int nritems;
|
|
|
|
nritems = btrfs_header_nritems(path->nodes[*level]);
|
|
|
|
|
|
|
|
path->slots[*level]++;
|
|
|
|
|
2012-08-08 04:25:13 +08:00
|
|
|
while (path->slots[*level] >= nritems) {
|
2012-06-06 03:07:48 +08:00
|
|
|
if (*level == root_level)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
/* move upnext */
|
|
|
|
path->slots[*level] = 0;
|
|
|
|
free_extent_buffer(path->nodes[*level]);
|
|
|
|
path->nodes[*level] = NULL;
|
|
|
|
(*level)++;
|
|
|
|
path->slots[*level]++;
|
|
|
|
|
|
|
|
nritems = btrfs_header_nritems(path->nodes[*level]);
|
|
|
|
ret = 1;
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Returns 1 if it had to move up and next. 0 is returned if it moved only next
|
|
|
|
* or down.
|
|
|
|
*/
|
2016-06-23 06:54:24 +08:00
|
|
|
static int tree_advance(struct btrfs_fs_info *fs_info,
|
2012-06-06 03:07:48 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
int *level, int root_level,
|
|
|
|
int allow_down,
|
|
|
|
struct btrfs_key *key)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (*level == 0 || !allow_down) {
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = tree_move_next_or_upnext(fs_info, path, level,
|
|
|
|
root_level);
|
2012-06-06 03:07:48 +08:00
|
|
|
} else {
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = tree_move_down(fs_info, path, level, root_level);
|
2012-06-06 03:07:48 +08:00
|
|
|
}
|
|
|
|
if (ret >= 0) {
|
|
|
|
if (*level == 0)
|
|
|
|
btrfs_item_key_to_cpu(path->nodes[*level], key,
|
|
|
|
path->slots[*level]);
|
|
|
|
else
|
|
|
|
btrfs_node_key_to_cpu(path->nodes[*level], key,
|
|
|
|
path->slots[*level]);
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:24 +08:00
|
|
|
static int tree_compare_item(struct btrfs_path *left_path,
|
2012-06-06 03:07:48 +08:00
|
|
|
struct btrfs_path *right_path,
|
|
|
|
char *tmp_buf)
|
|
|
|
{
|
|
|
|
int cmp;
|
|
|
|
int len1, len2;
|
|
|
|
unsigned long off1, off2;
|
|
|
|
|
|
|
|
len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
|
|
|
|
len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
|
|
|
|
if (len1 != len2)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
|
|
|
|
off2 = btrfs_item_ptr_offset(right_path->nodes[0],
|
|
|
|
right_path->slots[0]);
|
|
|
|
|
|
|
|
read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
|
|
|
|
|
|
|
|
cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
|
|
|
|
if (cmp)
|
|
|
|
return 1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define ADVANCE 1
|
|
|
|
#define ADVANCE_ONLY_NEXT -1
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This function compares two trees and calls the provided callback for
|
|
|
|
* every changed/new/deleted item it finds.
|
|
|
|
* If shared tree blocks are encountered, whole subtrees are skipped, making
|
|
|
|
* the compare pretty fast on snapshotted subvolumes.
|
|
|
|
*
|
|
|
|
* This currently works on commit roots only. As commit roots are read only,
|
|
|
|
* we don't do any locking. The commit roots are protected with transactions.
|
|
|
|
* Transactions are ended and rejoined when a commit is tried in between.
|
|
|
|
*
|
|
|
|
* This function checks for modifications done to the trees while comparing.
|
|
|
|
* If it detects a change, it aborts immediately.
|
|
|
|
*/
|
|
|
|
int btrfs_compare_trees(struct btrfs_root *left_root,
|
|
|
|
struct btrfs_root *right_root,
|
|
|
|
btrfs_changed_cb_t changed_cb, void *ctx)
|
|
|
|
{
|
2016-06-23 06:54:23 +08:00
|
|
|
struct btrfs_fs_info *fs_info = left_root->fs_info;
|
2012-06-06 03:07:48 +08:00
|
|
|
int ret;
|
|
|
|
int cmp;
|
|
|
|
struct btrfs_path *left_path = NULL;
|
|
|
|
struct btrfs_path *right_path = NULL;
|
|
|
|
struct btrfs_key left_key;
|
|
|
|
struct btrfs_key right_key;
|
|
|
|
char *tmp_buf = NULL;
|
|
|
|
int left_root_level;
|
|
|
|
int right_root_level;
|
|
|
|
int left_level;
|
|
|
|
int right_level;
|
|
|
|
int left_end_reached;
|
|
|
|
int right_end_reached;
|
|
|
|
int advance_left;
|
|
|
|
int advance_right;
|
|
|
|
u64 left_blockptr;
|
|
|
|
u64 right_blockptr;
|
2014-02-21 05:15:25 +08:00
|
|
|
u64 left_gen;
|
|
|
|
u64 right_gen;
|
2012-06-06 03:07:48 +08:00
|
|
|
|
|
|
|
left_path = btrfs_alloc_path();
|
|
|
|
if (!left_path) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
right_path = btrfs_alloc_path();
|
|
|
|
if (!right_path) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
tmp_buf = kmalloc(fs_info->nodesize, GFP_KERNEL | __GFP_NOWARN);
|
2012-06-06 03:07:48 +08:00
|
|
|
if (!tmp_buf) {
|
2016-06-23 06:54:23 +08:00
|
|
|
tmp_buf = vmalloc(fs_info->nodesize);
|
2016-03-30 22:01:12 +08:00
|
|
|
if (!tmp_buf) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
2012-06-06 03:07:48 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
left_path->search_commit_root = 1;
|
|
|
|
left_path->skip_locking = 1;
|
|
|
|
right_path->search_commit_root = 1;
|
|
|
|
right_path->skip_locking = 1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Strategy: Go to the first items of both trees. Then do
|
|
|
|
*
|
|
|
|
* If both trees are at level 0
|
|
|
|
* Compare keys of current items
|
|
|
|
* If left < right treat left item as new, advance left tree
|
|
|
|
* and repeat
|
|
|
|
* If left > right treat right item as deleted, advance right tree
|
|
|
|
* and repeat
|
|
|
|
* If left == right do deep compare of items, treat as changed if
|
|
|
|
* needed, advance both trees and repeat
|
|
|
|
* If both trees are at the same level but not at level 0
|
|
|
|
* Compare keys of current nodes/leafs
|
|
|
|
* If left < right advance left tree and repeat
|
|
|
|
* If left > right advance right tree and repeat
|
|
|
|
* If left == right compare blockptrs of the next nodes/leafs
|
|
|
|
* If they match advance both trees but stay at the same level
|
|
|
|
* and repeat
|
|
|
|
* If they don't match advance both trees while allowing to go
|
|
|
|
* deeper and repeat
|
|
|
|
* If tree levels are different
|
|
|
|
* Advance the tree that needs it and repeat
|
|
|
|
*
|
|
|
|
* Advancing a tree means:
|
|
|
|
* If we are at level 0, try to go to the next slot. If that's not
|
|
|
|
* possible, go one level up and repeat. Stop when we found a level
|
|
|
|
* where we could go to the next slot. We may at this point be on a
|
|
|
|
* node or a leaf.
|
|
|
|
*
|
|
|
|
* If we are not at level 0 and not on shared tree blocks, go one
|
|
|
|
* level deeper.
|
|
|
|
*
|
|
|
|
* If we are not at level 0 and on shared tree blocks, go one slot to
|
|
|
|
* the right if possible or go up and right.
|
|
|
|
*/
|
|
|
|
|
2016-06-23 06:54:23 +08:00
|
|
|
down_read(&fs_info->commit_root_sem);
|
2012-06-06 03:07:48 +08:00
|
|
|
left_level = btrfs_header_level(left_root->commit_root);
|
|
|
|
left_root_level = left_level;
|
|
|
|
left_path->nodes[left_level] = left_root->commit_root;
|
|
|
|
extent_buffer_get(left_path->nodes[left_level]);
|
|
|
|
|
|
|
|
right_level = btrfs_header_level(right_root->commit_root);
|
|
|
|
right_root_level = right_level;
|
|
|
|
right_path->nodes[right_level] = right_root->commit_root;
|
|
|
|
extent_buffer_get(right_path->nodes[right_level]);
|
2016-06-23 06:54:23 +08:00
|
|
|
up_read(&fs_info->commit_root_sem);
|
2012-06-06 03:07:48 +08:00
|
|
|
|
|
|
|
if (left_level == 0)
|
|
|
|
btrfs_item_key_to_cpu(left_path->nodes[left_level],
|
|
|
|
&left_key, left_path->slots[left_level]);
|
|
|
|
else
|
|
|
|
btrfs_node_key_to_cpu(left_path->nodes[left_level],
|
|
|
|
&left_key, left_path->slots[left_level]);
|
|
|
|
if (right_level == 0)
|
|
|
|
btrfs_item_key_to_cpu(right_path->nodes[right_level],
|
|
|
|
&right_key, right_path->slots[right_level]);
|
|
|
|
else
|
|
|
|
btrfs_node_key_to_cpu(right_path->nodes[right_level],
|
|
|
|
&right_key, right_path->slots[right_level]);
|
|
|
|
|
|
|
|
left_end_reached = right_end_reached = 0;
|
|
|
|
advance_left = advance_right = 0;
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
if (advance_left && !left_end_reached) {
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = tree_advance(fs_info, left_path, &left_level,
|
2012-06-06 03:07:48 +08:00
|
|
|
left_root_level,
|
|
|
|
advance_left != ADVANCE_ONLY_NEXT,
|
|
|
|
&left_key);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (ret == -1)
|
2012-06-06 03:07:48 +08:00
|
|
|
left_end_reached = ADVANCE;
|
2016-07-06 03:10:14 +08:00
|
|
|
else if (ret < 0)
|
|
|
|
goto out;
|
2012-06-06 03:07:48 +08:00
|
|
|
advance_left = 0;
|
|
|
|
}
|
|
|
|
if (advance_right && !right_end_reached) {
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = tree_advance(fs_info, right_path, &right_level,
|
2012-06-06 03:07:48 +08:00
|
|
|
right_root_level,
|
|
|
|
advance_right != ADVANCE_ONLY_NEXT,
|
|
|
|
&right_key);
|
2016-07-06 03:10:14 +08:00
|
|
|
if (ret == -1)
|
2012-06-06 03:07:48 +08:00
|
|
|
right_end_reached = ADVANCE;
|
2016-07-06 03:10:14 +08:00
|
|
|
else if (ret < 0)
|
|
|
|
goto out;
|
2012-06-06 03:07:48 +08:00
|
|
|
advance_right = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (left_end_reached && right_end_reached) {
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
} else if (left_end_reached) {
|
|
|
|
if (right_level == 0) {
|
|
|
|
ret = changed_cb(left_root, right_root,
|
|
|
|
left_path, right_path,
|
|
|
|
&right_key,
|
|
|
|
BTRFS_COMPARE_TREE_DELETED,
|
|
|
|
ctx);
|
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
continue;
|
|
|
|
} else if (right_end_reached) {
|
|
|
|
if (left_level == 0) {
|
|
|
|
ret = changed_cb(left_root, right_root,
|
|
|
|
left_path, right_path,
|
|
|
|
&left_key,
|
|
|
|
BTRFS_COMPARE_TREE_NEW,
|
|
|
|
ctx);
|
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
advance_left = ADVANCE;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (left_level == 0 && right_level == 0) {
|
|
|
|
cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
|
|
|
|
if (cmp < 0) {
|
|
|
|
ret = changed_cb(left_root, right_root,
|
|
|
|
left_path, right_path,
|
|
|
|
&left_key,
|
|
|
|
BTRFS_COMPARE_TREE_NEW,
|
|
|
|
ctx);
|
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
advance_left = ADVANCE;
|
|
|
|
} else if (cmp > 0) {
|
|
|
|
ret = changed_cb(left_root, right_root,
|
|
|
|
left_path, right_path,
|
|
|
|
&right_key,
|
|
|
|
BTRFS_COMPARE_TREE_DELETED,
|
|
|
|
ctx);
|
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
} else {
|
2014-09-26 01:35:02 +08:00
|
|
|
enum btrfs_compare_tree_result result;
|
2013-08-17 04:52:55 +08:00
|
|
|
|
2012-08-08 04:25:13 +08:00
|
|
|
WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
|
2016-06-23 06:54:24 +08:00
|
|
|
ret = tree_compare_item(left_path, right_path,
|
|
|
|
tmp_buf);
|
2013-08-17 04:52:55 +08:00
|
|
|
if (ret)
|
2014-09-26 01:35:02 +08:00
|
|
|
result = BTRFS_COMPARE_TREE_CHANGED;
|
2013-08-17 04:52:55 +08:00
|
|
|
else
|
2014-09-26 01:35:02 +08:00
|
|
|
result = BTRFS_COMPARE_TREE_SAME;
|
2013-08-17 04:52:55 +08:00
|
|
|
ret = changed_cb(left_root, right_root,
|
|
|
|
left_path, right_path,
|
2014-09-26 01:35:02 +08:00
|
|
|
&left_key, result, ctx);
|
2013-08-17 04:52:55 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
2012-06-06 03:07:48 +08:00
|
|
|
advance_left = ADVANCE;
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
}
|
|
|
|
} else if (left_level == right_level) {
|
|
|
|
cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
|
|
|
|
if (cmp < 0) {
|
|
|
|
advance_left = ADVANCE;
|
|
|
|
} else if (cmp > 0) {
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
} else {
|
|
|
|
left_blockptr = btrfs_node_blockptr(
|
|
|
|
left_path->nodes[left_level],
|
|
|
|
left_path->slots[left_level]);
|
|
|
|
right_blockptr = btrfs_node_blockptr(
|
|
|
|
right_path->nodes[right_level],
|
|
|
|
right_path->slots[right_level]);
|
2014-02-21 05:15:25 +08:00
|
|
|
left_gen = btrfs_node_ptr_generation(
|
|
|
|
left_path->nodes[left_level],
|
|
|
|
left_path->slots[left_level]);
|
|
|
|
right_gen = btrfs_node_ptr_generation(
|
|
|
|
right_path->nodes[right_level],
|
|
|
|
right_path->slots[right_level]);
|
|
|
|
if (left_blockptr == right_blockptr &&
|
|
|
|
left_gen == right_gen) {
|
2012-06-06 03:07:48 +08:00
|
|
|
/*
|
|
|
|
* As we're on a shared block, don't
|
|
|
|
* allow to go deeper.
|
|
|
|
*/
|
|
|
|
advance_left = ADVANCE_ONLY_NEXT;
|
|
|
|
advance_right = ADVANCE_ONLY_NEXT;
|
|
|
|
} else {
|
|
|
|
advance_left = ADVANCE;
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (left_level < right_level) {
|
|
|
|
advance_right = ADVANCE;
|
|
|
|
} else {
|
|
|
|
advance_left = ADVANCE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
btrfs_free_path(left_path);
|
|
|
|
btrfs_free_path(right_path);
|
2016-03-30 22:01:12 +08:00
|
|
|
kvfree(tmp_buf);
|
2012-06-06 03:07:48 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
/*
|
|
|
|
* this is similar to btrfs_next_leaf, but does not try to preserve
|
|
|
|
* and fixup the path. It looks for and returns the next key in the
|
2013-02-01 02:21:12 +08:00
|
|
|
* tree based on the current path and the min_trans parameters.
|
2008-06-26 04:01:31 +08:00
|
|
|
*
|
|
|
|
* 0 is returned if another key is found, < 0 if there are any errors
|
|
|
|
* and 1 is returned if there are no higher keys in the tree
|
|
|
|
*
|
|
|
|
* path->keep_locks should be set to 1 on the search made before
|
|
|
|
* calling this function.
|
|
|
|
*/
|
2008-06-26 04:01:31 +08:00
|
|
|
int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
|
2013-02-01 02:21:12 +08:00
|
|
|
struct btrfs_key *key, int level, u64 min_trans)
|
2008-06-26 04:01:31 +08:00
|
|
|
{
|
|
|
|
int slot;
|
|
|
|
struct extent_buffer *c;
|
|
|
|
|
2008-12-09 05:43:10 +08:00
|
|
|
WARN_ON(!path->keep_locks);
|
2009-01-06 10:25:51 +08:00
|
|
|
while (level < BTRFS_MAX_LEVEL) {
|
2008-06-26 04:01:31 +08:00
|
|
|
if (!path->nodes[level])
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
slot = path->slots[level] + 1;
|
|
|
|
c = path->nodes[level];
|
2008-06-26 04:01:31 +08:00
|
|
|
next:
|
2008-06-26 04:01:31 +08:00
|
|
|
if (slot >= btrfs_header_nritems(c)) {
|
2009-07-22 21:59:00 +08:00
|
|
|
int ret;
|
|
|
|
int orig_lowest;
|
|
|
|
struct btrfs_key cur_key;
|
|
|
|
if (level + 1 >= BTRFS_MAX_LEVEL ||
|
|
|
|
!path->nodes[level + 1])
|
2008-06-26 04:01:31 +08:00
|
|
|
return 1;
|
2009-07-22 21:59:00 +08:00
|
|
|
|
|
|
|
if (path->locks[level + 1]) {
|
|
|
|
level++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
slot = btrfs_header_nritems(c) - 1;
|
|
|
|
if (level == 0)
|
|
|
|
btrfs_item_key_to_cpu(c, &cur_key, slot);
|
|
|
|
else
|
|
|
|
btrfs_node_key_to_cpu(c, &cur_key, slot);
|
|
|
|
|
|
|
|
orig_lowest = path->lowest_level;
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2009-07-22 21:59:00 +08:00
|
|
|
path->lowest_level = level;
|
|
|
|
ret = btrfs_search_slot(NULL, root, &cur_key, path,
|
|
|
|
0, 0);
|
|
|
|
path->lowest_level = orig_lowest;
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
c = path->nodes[level];
|
|
|
|
slot = path->slots[level];
|
|
|
|
if (ret == 0)
|
|
|
|
slot++;
|
|
|
|
goto next;
|
2008-06-26 04:01:31 +08:00
|
|
|
}
|
2009-07-22 21:59:00 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
if (level == 0)
|
|
|
|
btrfs_item_key_to_cpu(c, key, slot);
|
2008-06-26 04:01:31 +08:00
|
|
|
else {
|
|
|
|
u64 gen = btrfs_node_ptr_generation(c, slot);
|
|
|
|
|
|
|
|
if (gen < min_trans) {
|
|
|
|
slot++;
|
|
|
|
goto next;
|
|
|
|
}
|
2008-06-26 04:01:31 +08:00
|
|
|
btrfs_node_key_to_cpu(c, key, slot);
|
2008-06-26 04:01:31 +08:00
|
|
|
}
|
2008-06-26 04:01:31 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2007-02-25 02:39:08 +08:00
|
|
|
/*
|
2008-06-26 04:01:30 +08:00
|
|
|
* search the tree again to find a leaf with greater keys
|
2007-03-01 05:46:22 +08:00
|
|
|
* returns 0 if it found something or 1 if there are no greater leaves.
|
|
|
|
* returns < 0 on io errors.
|
2007-02-25 02:39:08 +08:00
|
|
|
*/
|
2007-03-13 22:46:10 +08:00
|
|
|
int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
|
2012-06-11 14:29:29 +08:00
|
|
|
{
|
|
|
|
return btrfs_next_old_leaf(root, path, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
|
|
|
|
u64 time_seq)
|
2007-02-21 05:40:44 +08:00
|
|
|
{
|
|
|
|
int slot;
|
2009-04-03 22:14:18 +08:00
|
|
|
int level;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *c;
|
2009-04-03 22:14:18 +08:00
|
|
|
struct extent_buffer *next;
|
2008-06-26 04:01:30 +08:00
|
|
|
struct btrfs_key key;
|
|
|
|
u32 nritems;
|
|
|
|
int ret;
|
2009-04-03 22:14:18 +08:00
|
|
|
int old_spinning = path->leave_spinning;
|
2011-07-17 03:23:14 +08:00
|
|
|
int next_rw_lock = 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
|
|
|
|
nritems = btrfs_header_nritems(path->nodes[0]);
|
2009-01-06 10:25:51 +08:00
|
|
|
if (nritems == 0)
|
2008-06-26 04:01:30 +08:00
|
|
|
return 1;
|
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
|
|
|
|
again:
|
|
|
|
level = 1;
|
|
|
|
next = NULL;
|
2011-07-17 03:23:14 +08:00
|
|
|
next_rw_lock = 0;
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2009-04-03 22:14:18 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
path->keep_locks = 1;
|
2011-07-27 04:01:59 +08:00
|
|
|
path->leave_spinning = 1;
|
2009-04-03 22:14:18 +08:00
|
|
|
|
2012-06-11 14:29:29 +08:00
|
|
|
if (time_seq)
|
|
|
|
ret = btrfs_search_old_slot(root, &key, path, time_seq);
|
|
|
|
else
|
|
|
|
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
|
2008-06-26 04:01:30 +08:00
|
|
|
path->keep_locks = 0;
|
|
|
|
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
nritems = btrfs_header_nritems(path->nodes[0]);
|
2008-06-26 04:01:30 +08:00
|
|
|
/*
|
|
|
|
* by releasing the path above we dropped all our locks. A balance
|
|
|
|
* could have added more items next to the key that used to be
|
|
|
|
* at the very end of the block. So, check again here and
|
|
|
|
* advance the path if there are now more items available.
|
|
|
|
*/
|
2008-06-26 04:01:30 +08:00
|
|
|
if (nritems > 0 && path->slots[0] < nritems - 1) {
|
2009-07-22 21:59:00 +08:00
|
|
|
if (ret == 0)
|
|
|
|
path->slots[0]++;
|
2009-04-03 22:14:18 +08:00
|
|
|
ret = 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
goto done;
|
|
|
|
}
|
Btrfs: fix leaf corruption after __btrfs_drop_extents
Several reports about leaf corruption has been floating on the list, one of them
points to __btrfs_drop_extents(), and we find that the leaf becomes corrupted
after __btrfs_drop_extents(), it's really a rare case but it does exist.
The problem turns out to be btrfs_next_leaf() called in __btrfs_drop_extents().
So in btrfs_next_leaf(), we release the current path to re-search the last key of
the leaf for locating next leaf, and we've taken it into account that there might
be balance operations between leafs during this 'unlock and re-lock' dance, so
we check the path again and advance it if there are now more items available.
But things are a bit different if that last key happens to be removed and balance
gets a bigger key as the last one, and btrfs_search_slot will return it with
ret > 0, IOW, nothing change in this leaf except the new last key, then we think
we're okay because there is no more item balanced in, fine, we thinks we can
go to the next leaf.
However, we should return that bigger key, otherwise we deserve leaf corruption,
for example, in endio, skipping that key means that __btrfs_drop_extents() thinks
it has dropped all extent matched the required range and finish_ordered_io can
safely insert a new extent, but it actually doesn't and ends up a leaf
corruption.
One may be asking that why our locking on extent io tree doesn't work as
expected, ie. it should avoid this kind of race situation. But in
__btrfs_drop_extents(), we don't always find extents which are included within
our locking range, IOW, extents can start before our searching start, in this
case locking on extent io tree doesn't protect us from the race.
This takes the special case into account.
Reviewed-by: Filipe Manana <fdmanana@gmail.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
2014-06-09 11:04:49 +08:00
|
|
|
/*
|
|
|
|
* So the above check misses one case:
|
|
|
|
* - after releasing the path above, someone has removed the item that
|
|
|
|
* used to be at the very end of the block, and balance between leafs
|
|
|
|
* gets another one with bigger key.offset to replace it.
|
|
|
|
*
|
|
|
|
* This one should be returned as well, or we can get leaf corruption
|
|
|
|
* later(esp. in __btrfs_drop_extents()).
|
|
|
|
*
|
|
|
|
* And a bit more explanation about this check,
|
|
|
|
* with ret > 0, the key isn't found, the path points to the slot
|
|
|
|
* where it should be inserted, so the path->slots[0] item must be the
|
|
|
|
* bigger one.
|
|
|
|
*/
|
|
|
|
if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
|
|
|
|
ret = 0;
|
|
|
|
goto done;
|
|
|
|
}
|
2007-02-21 05:40:44 +08:00
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (level < BTRFS_MAX_LEVEL) {
|
2009-04-03 22:14:18 +08:00
|
|
|
if (!path->nodes[level]) {
|
|
|
|
ret = 1;
|
|
|
|
goto done;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2007-02-21 05:40:44 +08:00
|
|
|
slot = path->slots[level] + 1;
|
|
|
|
c = path->nodes[level];
|
2007-10-16 04:14:19 +08:00
|
|
|
if (slot >= btrfs_header_nritems(c)) {
|
2007-02-21 05:40:44 +08:00
|
|
|
level++;
|
2009-04-03 22:14:18 +08:00
|
|
|
if (level == BTRFS_MAX_LEVEL) {
|
|
|
|
ret = 1;
|
|
|
|
goto done;
|
|
|
|
}
|
2007-02-21 05:40:44 +08:00
|
|
|
continue;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (next) {
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_unlock_rw(next, next_rw_lock);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(next);
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2009-04-03 22:14:18 +08:00
|
|
|
next = c;
|
2011-07-17 03:23:14 +08:00
|
|
|
next_rw_lock = path->locks[level];
|
2017-01-31 04:23:42 +08:00
|
|
|
ret = read_block_for_search(root, path, &next, level,
|
2017-02-11 01:44:32 +08:00
|
|
|
slot, &key);
|
2009-04-03 22:14:18 +08:00
|
|
|
if (ret == -EAGAIN)
|
|
|
|
goto again;
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2009-05-15 01:24:30 +08:00
|
|
|
if (ret < 0) {
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2009-05-15 01:24:30 +08:00
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (!path->skip_locking) {
|
2011-07-17 03:23:14 +08:00
|
|
|
ret = btrfs_try_tree_read_lock(next);
|
2012-06-22 20:51:15 +08:00
|
|
|
if (!ret && time_seq) {
|
|
|
|
/*
|
|
|
|
* If we don't get the lock, we may be racing
|
|
|
|
* with push_leaf_left, holding that lock while
|
|
|
|
* itself waiting for the leaf we've currently
|
|
|
|
* locked. To solve this situation, we give up
|
|
|
|
* on our lock and cycle.
|
|
|
|
*/
|
2012-07-04 21:42:48 +08:00
|
|
|
free_extent_buffer(next);
|
2012-06-22 20:51:15 +08:00
|
|
|
btrfs_release_path(path);
|
|
|
|
cond_resched();
|
|
|
|
goto again;
|
|
|
|
}
|
2009-04-03 22:14:18 +08:00
|
|
|
if (!ret) {
|
|
|
|
btrfs_set_path_blocking(path);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_read_lock(next);
|
2011-07-27 04:01:59 +08:00
|
|
|
btrfs_clear_path_blocking(path, next,
|
2011-07-17 03:23:14 +08:00
|
|
|
BTRFS_READ_LOCK);
|
2009-04-03 22:14:18 +08:00
|
|
|
}
|
2011-07-27 04:01:59 +08:00
|
|
|
next_rw_lock = BTRFS_READ_LOCK;
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-02-21 05:40:44 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
path->slots[level] = slot;
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2007-02-21 05:40:44 +08:00
|
|
|
level--;
|
|
|
|
c = path->nodes[level];
|
2008-06-26 04:01:30 +08:00
|
|
|
if (path->locks[level])
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_unlock_rw(c, path->locks[level]);
|
2009-04-03 22:14:18 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(c);
|
2007-02-21 05:40:44 +08:00
|
|
|
path->nodes[level] = next;
|
|
|
|
path->slots[level] = 0;
|
2008-06-26 04:01:31 +08:00
|
|
|
if (!path->skip_locking)
|
2011-07-17 03:23:14 +08:00
|
|
|
path->locks[level] = next_rw_lock;
|
2007-02-21 05:40:44 +08:00
|
|
|
if (!level)
|
|
|
|
break;
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
|
2017-01-31 04:23:42 +08:00
|
|
|
ret = read_block_for_search(root, path, &next, level,
|
2017-02-11 01:44:32 +08:00
|
|
|
0, &key);
|
2009-04-03 22:14:18 +08:00
|
|
|
if (ret == -EAGAIN)
|
|
|
|
goto again;
|
|
|
|
|
2009-05-15 01:24:30 +08:00
|
|
|
if (ret < 0) {
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2009-05-15 01:24:30 +08:00
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
if (!path->skip_locking) {
|
2011-07-17 03:23:14 +08:00
|
|
|
ret = btrfs_try_tree_read_lock(next);
|
2009-04-03 22:14:18 +08:00
|
|
|
if (!ret) {
|
|
|
|
btrfs_set_path_blocking(path);
|
2011-07-17 03:23:14 +08:00
|
|
|
btrfs_tree_read_lock(next);
|
2011-07-27 04:01:59 +08:00
|
|
|
btrfs_clear_path_blocking(path, next,
|
2011-07-17 03:23:14 +08:00
|
|
|
BTRFS_READ_LOCK);
|
|
|
|
}
|
2011-07-27 04:01:59 +08:00
|
|
|
next_rw_lock = BTRFS_READ_LOCK;
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-02-21 05:40:44 +08:00
|
|
|
}
|
2009-04-03 22:14:18 +08:00
|
|
|
ret = 0;
|
2008-06-26 04:01:30 +08:00
|
|
|
done:
|
2012-03-20 03:54:38 +08:00
|
|
|
unlock_up(path, 0, 1, 0, NULL);
|
2009-04-03 22:14:18 +08:00
|
|
|
path->leave_spinning = old_spinning;
|
|
|
|
if (!old_spinning)
|
|
|
|
btrfs_set_path_blocking(path);
|
|
|
|
|
|
|
|
return ret;
|
2007-02-21 05:40:44 +08:00
|
|
|
}
|
2008-03-25 03:01:56 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
/*
|
|
|
|
* this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
|
|
|
|
* searching until it gets past min_objectid or finds an item of 'type'
|
|
|
|
*
|
|
|
|
* returns 0 if something is found, 1 if nothing was found and < 0 on error
|
|
|
|
*/
|
2008-03-25 03:01:56 +08:00
|
|
|
int btrfs_previous_item(struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, u64 min_objectid,
|
|
|
|
int type)
|
|
|
|
{
|
|
|
|
struct btrfs_key found_key;
|
|
|
|
struct extent_buffer *leaf;
|
2008-09-06 04:13:11 +08:00
|
|
|
u32 nritems;
|
2008-03-25 03:01:56 +08:00
|
|
|
int ret;
|
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2008-03-25 03:01:56 +08:00
|
|
|
if (path->slots[0] == 0) {
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 22:25:08 +08:00
|
|
|
btrfs_set_path_blocking(path);
|
2008-03-25 03:01:56 +08:00
|
|
|
ret = btrfs_prev_leaf(root, path);
|
|
|
|
if (ret != 0)
|
|
|
|
return ret;
|
|
|
|
} else {
|
|
|
|
path->slots[0]--;
|
|
|
|
}
|
|
|
|
leaf = path->nodes[0];
|
2008-09-06 04:13:11 +08:00
|
|
|
nritems = btrfs_header_nritems(leaf);
|
|
|
|
if (nritems == 0)
|
|
|
|
return 1;
|
|
|
|
if (path->slots[0] == nritems)
|
|
|
|
path->slots[0]--;
|
|
|
|
|
2008-03-25 03:01:56 +08:00
|
|
|
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
|
2008-09-06 04:13:11 +08:00
|
|
|
if (found_key.objectid < min_objectid)
|
|
|
|
break;
|
2009-07-24 23:06:53 +08:00
|
|
|
if (found_key.type == type)
|
|
|
|
return 0;
|
2008-09-06 04:13:11 +08:00
|
|
|
if (found_key.objectid == min_objectid &&
|
|
|
|
found_key.type < type)
|
|
|
|
break;
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|
2014-01-12 21:38:33 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* search in extent tree to find a previous Metadata/Data extent item with
|
|
|
|
* min objecitd.
|
|
|
|
*
|
|
|
|
* returns 0 if something is found, 1 if nothing was found and < 0 on error
|
|
|
|
*/
|
|
|
|
int btrfs_previous_extent_item(struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, u64 min_objectid)
|
|
|
|
{
|
|
|
|
struct btrfs_key found_key;
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
u32 nritems;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
if (path->slots[0] == 0) {
|
|
|
|
btrfs_set_path_blocking(path);
|
|
|
|
ret = btrfs_prev_leaf(root, path);
|
|
|
|
if (ret != 0)
|
|
|
|
return ret;
|
|
|
|
} else {
|
|
|
|
path->slots[0]--;
|
|
|
|
}
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
nritems = btrfs_header_nritems(leaf);
|
|
|
|
if (nritems == 0)
|
|
|
|
return 1;
|
|
|
|
if (path->slots[0] == nritems)
|
|
|
|
path->slots[0]--;
|
|
|
|
|
|
|
|
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
|
|
|
|
if (found_key.objectid < min_objectid)
|
|
|
|
break;
|
|
|
|
if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
|
|
|
|
found_key.type == BTRFS_METADATA_ITEM_KEY)
|
|
|
|
return 0;
|
|
|
|
if (found_key.objectid == min_objectid &&
|
|
|
|
found_key.type < BTRFS_EXTENT_ITEM_KEY)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|