self-tests code assumes 4k as the sectorsize and nodesize. This commit
fix hardcoded 4K. Enables the self-tests code to be executed on non-4k
page sized systems (e.g. ppc64).
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Feifei Xu <xufeifei@linux.vnet.ibm.com>
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
On ppc64, bytes_per_bitmap will be (65536*8*65536). Hence append UL to
fix integer overflow.
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Feifei Xu <xufeifei@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
On a ppc64 machine using 64K as the block size, assume that the RB
tree at btrfs_free_space_ctl->free_space_offset contains following
two entries:
1. A bitmap entry having an offset value of 0 and having the bits
corresponding to the address range [128M+512K, 128M+768K] set.
2. An extent entry corresponding to the address range
[128M-256K, 128M-128K]
In such a scenario, test_check_exists() invoked for checking the
existence of address range [128M+768K, 256M] can lead to an
infinite loop as explained below:
- Checking for the extent entry fails.
- Checking for a bitmap entry results in the free space info in
range [128M+512K, 128M+768K] beng returned.
- rb_prev(info) returns NULL because the bitmap entry starting from
offset 0 comes first in the RB tree.
- current_node = bitmap node.
- while (current_node)
tmp = rb_next(bitmap_node);/*tmp is extent based free space entry*/
Since extent based free space entry's last address is smaller
than the address being searched for (i.e. 128M+768K) we
incorrectly again obtain the extent node as the "next right node"
of the RB tree and thus end up looping infinitely.
This patch fixes the issue by checking the "tmp" variable which point
to the most recently searched free space node.
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Feifei Xu <xufeifei@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time
ago with promise that one day it will be possible to implement page
cache with bigger chunks than PAGE_SIZE.
This promise never materialized. And unlikely will.
We have many places where PAGE_CACHE_SIZE assumed to be equal to
PAGE_SIZE. And it's constant source of confusion on whether
PAGE_CACHE_* or PAGE_* constant should be used in a particular case,
especially on the border between fs and mm.
Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much
breakage to be doable.
Let's stop pretending that pages in page cache are special. They are
not.
The changes are pretty straight-forward:
- <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN};
- page_cache_get() -> get_page();
- page_cache_release() -> put_page();
This patch contains automated changes generated with coccinelle using
script below. For some reason, coccinelle doesn't patch header files.
I've called spatch for them manually.
The only adjustment after coccinelle is revert of changes to
PAGE_CAHCE_ALIGN definition: we are going to drop it later.
There are few places in the code where coccinelle didn't reach. I'll
fix them manually in a separate patch. Comments and documentation also
will be addressed with the separate patch.
virtual patch
@@
expression E;
@@
- E << (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
expression E;
@@
- E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
@@
- PAGE_CACHE_SHIFT
+ PAGE_SHIFT
@@
@@
- PAGE_CACHE_SIZE
+ PAGE_SIZE
@@
@@
- PAGE_CACHE_MASK
+ PAGE_MASK
@@
expression E;
@@
- PAGE_CACHE_ALIGN(E)
+ PAGE_ALIGN(E)
@@
expression E;
@@
- page_cache_get(E)
+ get_page(E)
@@
expression E;
@@
- page_cache_release(E)
+ put_page(E)
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use list_for_each_entry_safe() instead of list_for_each_safe() to
simplify the code.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We use many constants to represent size and offset value. And to make
code readable we use '256 * 1024 * 1024' instead of '268435456' to
represent '256MB'. However we can make far more readable with 'SZ_256MB'
which is defined in the 'linux/sizes.h'.
So this patch replaces 'xxx * 1024 * 1024' kind of expression with
single 'SZ_xxxMB' if 'xxx' is a power of 2 then 'xxx * SZ_1M' if 'xxx' is
not a power of 2. And I haven't touched to '4096' & '8192' because it's
more intuitive than 'SZ_4KB' & 'SZ_8KB'.
Signed-off-by: Byongho Lee <bhlee.kernel@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Pull btrfs fixes from Chris Mason:
"A couple of small fixes"
* 'for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
Btrfs: check prepare_uptodate_page() error code earlier
Btrfs: check for empty bitmap list in setup_cluster_bitmaps
btrfs: fix misleading warning when space cache failed to load
Btrfs: fix transaction handle leak in balance
Btrfs: fix unprotected list move from unused_bgs to deleted_bgs list
Dave Jones found a warning from kasan in setup_cluster_bitmaps()
==================================================================
BUG: KASAN: stack-out-of-bounds in setup_cluster_bitmap+0xc4/0x5a0 at
addr ffff88039bef6828
Read of size 8 by task nfsd/1009
page:ffffea000e6fbd80 count:0 mapcount:0 mapping: (null)
index:0x0
flags: 0x8000000000000000()
page dumped because: kasan: bad access detected
CPU: 1 PID: 1009 Comm: nfsd Tainted: G W
4.4.0-rc3-backup-debug+ #1
ffff880065647b50 000000006bb712c2 ffff88039bef6640 ffffffffa680a43e
0000004559c00000 ffff88039bef66c8 ffffffffa62638d1 ffffffffa61121c0
ffff8803a5769de8 0000000000000296 ffff8803a5769df0 0000000000046280
Call Trace:
[<ffffffffa680a43e>] dump_stack+0x4b/0x6d
[<ffffffffa62638d1>] kasan_report_error+0x501/0x520
[<ffffffffa61121c0>] ? debug_show_all_locks+0x1e0/0x1e0
[<ffffffffa6263948>] kasan_report+0x58/0x60
[<ffffffffa6814b00>] ? rb_last+0x10/0x40
[<ffffffffa66f8af4>] ? setup_cluster_bitmap+0xc4/0x5a0
[<ffffffffa6262ead>] __asan_load8+0x5d/0x70
[<ffffffffa66f8af4>] setup_cluster_bitmap+0xc4/0x5a0
[<ffffffffa66f675a>] ? setup_cluster_no_bitmap+0x6a/0x400
[<ffffffffa66fcd16>] btrfs_find_space_cluster+0x4b6/0x640
[<ffffffffa66fc860>] ? btrfs_alloc_from_cluster+0x4e0/0x4e0
[<ffffffffa66fc36e>] ? btrfs_return_cluster_to_free_space+0x9e/0xb0
[<ffffffffa702dc37>] ? _raw_spin_unlock+0x27/0x40
[<ffffffffa666a1a1>] find_free_extent+0xba1/0x1520
Andrey noticed this was because we were doing list_first_entry on a list
that might be empty. Rework the tests a bit so we don't do that.
Signed-off-by: Chris Mason <clm@fb.com>
Reprorted-by: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Reported-by: Dave Jones <dsj@fb.com>
When an inconsistent space cache is detected during loading we log a
warning that users frequently mistake as instruction to invalidate the
cache manually, even though this is not required. Fix the message to
indicate that the cache will be rebuilt automatically.
Signed-off-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com>
Acked-by: Filipe Manana <fdmanana@suse.com>
Merge second patch-bomb from Andrew Morton:
- most of the rest of MM
- procfs
- lib/ updates
- printk updates
- bitops infrastructure tweaks
- checkpatch updates
- nilfs2 update
- signals
- various other misc bits: coredump, seqfile, kexec, pidns, zlib, ipc,
dma-debug, dma-mapping, ...
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (102 commits)
ipc,msg: drop dst nil validation in copy_msg
include/linux/zutil.h: fix usage example of zlib_adler32()
panic: release stale console lock to always get the logbuf printed out
dma-debug: check nents in dma_sync_sg*
dma-mapping: tidy up dma_parms default handling
pidns: fix set/getpriority and ioprio_set/get in PRIO_USER mode
kexec: use file name as the output message prefix
fs, seqfile: always allow oom killer
seq_file: reuse string_escape_str()
fs/seq_file: use seq_* helpers in seq_hex_dump()
coredump: change zap_threads() and zap_process() to use for_each_thread()
coredump: ensure all coredumping tasks have SIGNAL_GROUP_COREDUMP
signal: remove jffs2_garbage_collect_thread()->allow_signal(SIGCONT)
signal: introduce kernel_signal_stop() to fix jffs2_garbage_collect_thread()
signal: turn dequeue_signal_lock() into kernel_dequeue_signal()
signals: kill block_all_signals() and unblock_all_signals()
nilfs2: fix gcc uninitialized-variable warnings in powerpc build
nilfs2: fix gcc unused-but-set-variable warnings
MAINTAINERS: nilfs2: add header file for tracing
nilfs2: add tracepoints for analyzing reading and writing metadata files
...
There are many places which use mapping_gfp_mask to restrict a more
generic gfp mask which would be used for allocations which are not
directly related to the page cache but they are performed in the same
context.
Let's introduce a helper function which makes the restriction explicit and
easier to track. This patch doesn't introduce any functional changes.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When we make ctl->unit allocations from a bitmap there is no point in searching
for the next 0 in the bitmap. If we've found a bit we're done and can just exit
the loop. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
We can waste a lot of time searching through bitmaps when we are heavily
fragmented trying to find large contiguous areas that don't exist in the bitmap.
So keep track of the max extent size when we do a full search of a bitmap so
that next time around we can just skip the expensive searching if our max size
is less than what we are looking for. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
If we are extremely fragmented then we won't be able to create a free_cluster.
So if this happens set last_ptr->fragmented so that all future allcations will
give up trying to create a cluster. When we unpin extents we will unset
->fragmented if we free up a sufficient amount of space in a block group.
Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
In tracking down these weird bitmap problems it was helpful to artificially
create an extremely fragmented file system. These mount options let us either
fragment data or metadata or both. With these options I could reproduce all
sorts of weird latencies and hangs that occur under extreme fragmentation and
get them fixed. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
Just fix a typo in the code comment.
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When we clear the dirty bits in btrfs_delete_unused_bgs for extents
in the empty block group, it results in btrfs_finish_extent_commit being
unable to discard the freed extents.
The block group removal patch added an alternate path to forget extents
other than btrfs_finish_extent_commit. As a result, any extents that
would be freed when the block group is removed aren't discarded. In my
test run, with a large copy of mixed sized files followed by removal, it
left nearly 2/3 of extents undiscarded.
To clean up the block groups, we add the removed block group onto a list
that will be discarded after transaction commit.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Tested-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
If the call to btrfs_truncate_inode_items() failed and we don't have a block
group, we were unlocking the cache_write_mutex without having locked it (we
do it only if we have a block group).
Fixes: 1bbc621ef2 ("Btrfs: allow block group cache writeout
outside critical section in commit")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
If the writeback of an inode cache failed we were unnecessarilly
attempting to release again the delalloc metadata that we previously
reserved. However attempting to do this a second time triggers an
assertion at drop_outstanding_extent() because we have no more
outstanding extents for our inode cache's inode. If we were able
to start writeback of the cache the reserved metadata space is
released at btrfs_finished_ordered_io(), even if an error happens
during writeback.
So make sure we don't repeat the metadata space release if writeback
started for our inode cache.
This issue was trivial to reproduce by running the fstest btrfs/088
with "-o inode_cache", which triggered the assertion leading to a
BUG() call and requiring a reboot in order to run the remaining
fstests. Trace produced by btrfs/088:
[255289.385904] BTRFS: assertion failed: BTRFS_I(inode)->outstanding_extents >= num_extents, file: fs/btrfs/extent-tree.c, line: 5276
[255289.388094] ------------[ cut here ]------------
[255289.389184] kernel BUG at fs/btrfs/ctree.h:4057!
[255289.390125] invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
(...)
[255289.392068] Call Trace:
[255289.392068] [<ffffffffa035e774>] drop_outstanding_extent+0x3d/0x6d [btrfs]
[255289.392068] [<ffffffffa0364988>] btrfs_delalloc_release_metadata+0x54/0xe3 [btrfs]
[255289.392068] [<ffffffffa03b4174>] btrfs_write_out_ino_cache+0x95/0xad [btrfs]
[255289.392068] [<ffffffffa036f5c4>] btrfs_save_ino_cache+0x275/0x2dc [btrfs]
[255289.392068] [<ffffffffa03e2d83>] commit_fs_roots.isra.12+0xaa/0x137 [btrfs]
[255289.392068] [<ffffffff8107d33d>] ? trace_hardirqs_on+0xd/0xf
[255289.392068] [<ffffffffa037841f>] ? btrfs_commit_transaction+0x4b1/0x9c9 [btrfs]
[255289.392068] [<ffffffff814351a4>] ? _raw_spin_unlock+0x32/0x46
[255289.392068] [<ffffffffa037842e>] btrfs_commit_transaction+0x4c0/0x9c9 [btrfs]
(...)
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
We were passing a flags value that differed from the intention in commit
2b10826800 ("Btrfs: don't use highmem for free space cache pages").
This caused problems in a ARM machine, leaving btrfs unusable there.
Reported-by: Merlijn Wajer <merlijn@wizzup.org>
Tested-by: Merlijn Wajer <merlijn@wizzup.org>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
If io_ctl_prepare_pages fails, the pages in io_ctl.pages are not valid.
When we try to access them later, things will blow up in various ways.
Also fix the comment about the return value, which is an errno on error,
not -1, and update the cases where it was not.
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Signed-off-by: Chris Mason <clm@fb.com>
__btrfs_write_out_cache is holding the ctl->tree_lock while it prepares
a list of bitmaps to record in the free space cache. It was dropping
the lock while it worked on other components, which made a window for
free_bitmap() to free the bitmap struct without removing it from the
list.
This changes things to hold the lock the whole time, and also makes sure
we hold the lock during enospc cleanup.
Reported-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
The code to fix stalls during free spache cache IO wasn't using
the correct root when waiting on the IO for inode caches. This
is only a problem when the inode cache is enabled with
mount -o inode_cache
This fixes the inode cache writeout to preserve any error values and
makes sure not to override the root when inode cache writeout is done.
Reported-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
We loop through all of the dirty block groups during commit and write
the free space cache. In order to make sure the cache is currect, we do
this while no other writers are allowed in the commit.
If a large number of block groups are dirty, this can introduce long
stalls during the final stages of the commit, which can block new procs
trying to change the filesystem.
This commit changes the block group cache writeout to take appropriate
locks and allow it to run earlier in the commit. We'll still have to
redo some of the block groups, but it means we can get most of the work
out of the way without blocking the entire FS.
Signed-off-by: Chris Mason <clm@fb.com>
In order to create the free space cache concurrently with FS modifications,
we need to take a few block group locks.
The cache code also does kmap, which would schedule with the locks held.
Instead of going through kmap_atomic, lets just use lowmem for the cache
pages.
Signed-off-by: Chris Mason <clm@fb.com>
Block group cache writeout is currently waiting on the pages for each
block group cache before moving on to writing the next one. This commit
switches things around to send down all the caches and then wait on them
in batches.
The end result is much faster, since we're keeping the disk pipeline
full.
Signed-off-by: Chris Mason <clm@fb.com>
We'll need to put the io_ctl into the block_group cache struct, so
name it struct btrfs_io_ctl and move it into ctree.h
Signed-off-by: Chris Mason <clm@fb.com>
When we are deleting large files with large extents, we are building up
a huge set of delayed refs for processing. Truncate isn't checking
often enough to see if we need to back off and process those, or let
a commit proceed.
The end result is long stalls after the rm, and very long commit times.
During the commits, other processes back up waiting to start new
transactions and we get into trouble.
Signed-off-by: Chris Mason <clm@fb.com>
The divisor is derived from nodesize or PAGE_SIZE, fits into 32bit type.
Get rid of a few more do_div instances.
Signed-off-by: David Sterba <dsterba@suse.cz>
Convert kmalloc(nr * size, ..) to kmalloc_array that does additional
overflow checks, the zeroing variant is kcalloc.
Signed-off-by: David Sterba <dsterba@suse.cz>
Switch to div_u64 if the divisor is a numeric constant or sum of
sizeof()s. We can remove a few instances of do_div that has the hidden
semtantics of changing the 1st argument.
Small power-of-two divisors are converted to bitshifts, large values are
kept intact for clarity.
Signed-off-by: David Sterba <dsterba@suse.cz>
Clean the opencoded variant, cond_resched_lock also checks the lock for
contention so it might help in some cases that were not covered by
simple need_resched().
Signed-off-by: David Sterba <dsterba@suse.cz>
o removed an unecessary INIT_LIST_HEAD after LIST_HEAD
o merge a declare & INIT_LIST_HEAD pair into one LIST_HEAD
Signed-off-by: Gui Hecheng <guihc.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
Currently any time we try to update the block groups on disk we will walk _all_
block groups and check for the ->dirty flag to see if it is set. This function
can get called several times during a commit. So if you have several terabytes
of data you will be a very sad panda as we will loop through _all_ of the block
groups several times, which makes the commit take a while which slows down the
rest of the file system operations.
This patch introduces a dirty list for the block groups that we get added to
when we dirty the block group for the first time. Then we simply update any
block groups that have been dirtied since the last time we called
btrfs_write_dirty_block_groups. This allows us to clean up how we write the
free space cache out so it is much cleaner. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
It doesn't do anything special, it just calls btrfs_discard_extent(),
so just remove it.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
The call to remove_extent_mapping() actually deletes the extent map
from the list it's included in - fs_info->pinned_chunks - and that
list is protected by the chunk mutex. Therefore make that call
while holding the chunk mutex and remove the redundant list delete
call because it's a noop.
This fixes an overlook of the patch titled
"Btrfs: fix race between fs trimming and block group remove/allocation"
following the same obvervation from the patch titled
"Btrfs: fix unprotected deletion from pending_chunks list".
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
There was a free space entry structure memeory leak if a block
group is remove while a free space entry is being trimmed, which
the following diagram explains:
CPU 1 CPU 2
btrfs_trim_block_group()
trim_no_bitmap()
remove free space entry from
block group cache's rbtree
do_trimming()
btrfs_remove_block_group()
btrfs_remove_free_space_cache()
add back free space entry to
block group's cache rbtree
btrfs_put_block_group()
(...)
btrfs_put_block_group()
kfree(bg->free_space_ctl)
kfree(bg)
The free space entry added after doing the discard of its respective
range ends up never being freed.
Detected after doing an "rmmod btrfs" after running the stress test
recently submitted for fstests:
[ 8234.642212] kmem_cache_destroy btrfs_free_space: Slab cache still has objects
[ 8234.642657] CPU: 1 PID: 32276 Comm: rmmod Tainted: G W L 3.17.0-rc5-btrfs-next-2+ #1
[ 8234.642660] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
[ 8234.642664] 0000000000000000 ffff8801af1b3eb8 ffffffff8140c7b6 ffff8801dbedd0c0
[ 8234.642670] ffff8801af1b3ed0 ffffffff811149ce 0000000000000000 ffff8801af1b3ee0
[ 8234.642676] ffffffffa042dbe7 ffff8801af1b3ef0 ffffffffa0487422 ffff8801af1b3f78
[ 8234.642682] Call Trace:
[ 8234.642692] [<ffffffff8140c7b6>] dump_stack+0x4d/0x66
[ 8234.642699] [<ffffffff811149ce>] kmem_cache_destroy+0x4d/0x92
[ 8234.642731] [<ffffffffa042dbe7>] btrfs_destroy_cachep+0x63/0x76 [btrfs]
[ 8234.642757] [<ffffffffa0487422>] exit_btrfs_fs+0x9/0xbe7 [btrfs]
[ 8234.642762] [<ffffffff810a76a5>] SyS_delete_module+0x155/0x1c6
[ 8234.642768] [<ffffffff8122a7eb>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 8234.642773] [<ffffffff814122d2>] system_call_fastpath+0x16/0x1b
This applies on top (depends on) of my previous patch titled:
"Btrfs: fix race between fs trimming and block group remove/allocation"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Trimming is completely transactionless, and the way it operates consists
of hiding free space entries from a block group, perform the trim/discard
and then make the free space entries visible again.
Therefore while a free space entry is being trimmed, we can have free space
cache writing running in parallel (as part of a transaction commit) which
will miss the free space entry. This means that an unmount (or crash/reboot)
after that transaction commit and mount again before another transaction
starts/commits after the discard finishes, we will have some free space
that won't be used again unless the free space cache is rebuilt. After the
unmount, fsck (btrfsck, btrfs check) reports the issue like the following
example:
*** fsck.btrfs output ***
checking extents
checking free space cache
There is no free space entry for 521764864-521781248
There is no free space entry for 521764864-1103101952
cache appears valid but isnt 29360128
Checking filesystem on /dev/sdc
UUID: b4789e27-4774-4626-98e9-ae8dfbfb0fb5
found 1235681286 bytes used err is -22
(...)
Another issue caused by this race is a crash while writing bitmap entries
to the cache, because while the cache writeout task accesses the bitmaps,
the trim task can be concurrently modifying the bitmap or worse might
be freeing the bitmap. The later case results in the following crash:
[55650.804460] general protection fault: 0000 [#1] SMP DEBUG_PAGEALLOC
[55650.804835] Modules linked in: btrfs dm_flakey dm_mod crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd fscache sunrpc loop parport_pc parport i2c_piix4 psmouse evdev pcspkr microcode processor i2ccore serio_raw thermal_sys button ext4 crc16 jbd2 mbcache sg sd_mod crc_t10dif sr_mod cdrom crct10dif_generic crct10dif_common ata_generic virtio_scsi floppy ata_piix libata virtio_pci virtio_ring virtio scsi_mod e1000 [last unloaded: btrfs]
[55650.806169] CPU: 1 PID: 31002 Comm: btrfs-transacti Tainted: G W 3.17.0-rc5-btrfs-next-1+ #1
[55650.806493] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
[55650.806867] task: ffff8800b12f6410 ti: ffff880071538000 task.ti: ffff880071538000
[55650.807166] RIP: 0010:[<ffffffffa037cf45>] [<ffffffffa037cf45>] write_bitmap_entries+0x65/0xbb [btrfs]
[55650.807514] RSP: 0018:ffff88007153bc30 EFLAGS: 00010246
[55650.807687] RAX: 000000005d1ec000 RBX: ffff8800a665df08 RCX: 0000000000000400
[55650.807885] RDX: ffff88005d1ec000 RSI: 6b6b6b6b6b6b6b6b RDI: ffff88005d1ec000
[55650.808017] RBP: ffff88007153bc58 R08: 00000000ddd51536 R09: 00000000000001e0
[55650.808017] R10: 0000000000000000 R11: 0000000000000037 R12: 6b6b6b6b6b6b6b6b
[55650.808017] R13: ffff88007153bca8 R14: 6b6b6b6b6b6b6b6b R15: ffff88007153bc98
[55650.808017] FS: 0000000000000000(0000) GS:ffff88023ec80000(0000) knlGS:0000000000000000
[55650.808017] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[55650.808017] CR2: 0000000002273b88 CR3: 00000000b18f6000 CR4: 00000000000006e0
[55650.808017] Stack:
[55650.808017] ffff88020e834e00 ffff880172d68db0 0000000000000000 ffff88019257c800
[55650.808017] ffff8801d42ea720 ffff88007153bd10 ffffffffa037d2fa ffff880224e99180
[55650.808017] ffff8801469a6188 ffff880224e99140 ffff880172d68c50 00000003000000b7
[55650.808017] Call Trace:
[55650.808017] [<ffffffffa037d2fa>] __btrfs_write_out_cache+0x1ea/0x37f [btrfs]
[55650.808017] [<ffffffffa037d959>] btrfs_write_out_cache+0xa1/0xd8 [btrfs]
[55650.808017] [<ffffffffa033936b>] btrfs_write_dirty_block_groups+0x4b5/0x505 [btrfs]
[55650.808017] [<ffffffffa03aa98e>] commit_cowonly_roots+0x15e/0x1f7 [btrfs]
[55650.808017] [<ffffffff813eb9c7>] ? _raw_spin_lock+0xe/0x10
[55650.808017] [<ffffffffa0346e46>] btrfs_commit_transaction+0x411/0x882 [btrfs]
[55650.808017] [<ffffffffa03432a4>] transaction_kthread+0xf2/0x1a4 [btrfs]
[55650.808017] [<ffffffffa03431b2>] ? btrfs_cleanup_transaction+0x3d8/0x3d8 [btrfs]
[55650.808017] [<ffffffff8105966b>] kthread+0xb7/0xbf
[55650.808017] [<ffffffff810595b4>] ? __kthread_parkme+0x67/0x67
[55650.808017] [<ffffffff813ebeac>] ret_from_fork+0x7c/0xb0
[55650.808017] [<ffffffff810595b4>] ? __kthread_parkme+0x67/0x67
[55650.808017] Code: 4c 89 ef 8d 70 ff e8 d4 fc ff ff 41 8b 45 34 41 39 45 30 7d 5c 31 f6 4c 89 ef e8 80 f6 ff ff 49 8b 7d 00 4c 89 f6 b9 00 04 00 00 <f3> a5 4c 89 ef 41 8b 45 30 8d 70 ff e8 a3 fc ff ff 41 8b 45 34
[55650.808017] RIP [<ffffffffa037cf45>] write_bitmap_entries+0x65/0xbb [btrfs]
[55650.808017] RSP <ffff88007153bc30>
[55650.815725] ---[ end trace 1c032e96b149ff86 ]---
Fix this by serializing both tasks in such a way that cache writeout
doesn't wait for the trim/discard of free space entries to finish and
doesn't miss any free space entry.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Our fs trim operation, which is completely transactionless (doesn't start
or joins an existing transaction) consists of visiting all block groups
and then for each one to iterate its free space entries and perform a
discard operation against the space range represented by the free space
entries. However before performing a discard, the corresponding free space
entry is removed from the free space rbtree, and when the discard completes
it is added back to the free space rbtree.
If a block group remove operation happens while the discard is ongoing (or
before it starts and after a free space entry is hidden), we end up not
waiting for the discard to complete, remove the extent map that maps
logical address to physical addresses and the corresponding chunk metadata
from the the chunk and device trees. After that and before the discard
completes, the current running transaction can finish and a new one start,
allowing for new block groups that map to the same physical addresses to
be allocated and written to.
So fix this by keeping the extent map in memory until the discard completes
so that the same physical addresses aren't reused before it completes.
If the physical locations that are under a discard operation end up being
used for a new metadata block group for example, and dirty metadata extents
are written before the discard finishes (the VM might call writepages() of
our btree inode's i_mapping for example, or an fsync log commit happens) we
end up overwriting metadata with zeroes, which leads to errors from fsck
like the following:
checking extents
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
read block failed check_tree_block
owner ref check failed [833912832 16384]
Errors found in extent allocation tree or chunk allocation
checking free space cache
checking fs roots
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
Check tree block failed, want=833912832, have=0
read block failed check_tree_block
root 5 root dir 256 error
root 5 inode 260 errors 2001, no inode item, link count wrong
unresolved ref dir 256 index 0 namelen 8 name foobar_3 filetype 1 errors 6, no dir index, no inode ref
root 5 inode 262 errors 2001, no inode item, link count wrong
unresolved ref dir 256 index 0 namelen 8 name foobar_5 filetype 1 errors 6, no dir index, no inode ref
root 5 inode 263 errors 2001, no inode item, link count wrong
(...)
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
While under random IO, a block group's free space cache eventually reaches
a state where it has a mix of extent entries and bitmap entries representing
free space regions.
As later free space regions are returned to the cache, some of them are merged
with existing extent entries if they are contiguous with them. But others are
not merged, because despite the existence of adjacent free space regions in
the cache, the merging doesn't happen because the existing free space regions
are represented in bitmap extents. Even when new free space regions are merged
with existing extent entries (enlarging the free space range they represent),
we create chances of having after an enlarged region that is contiguous with
some other region represented in a bitmap entry.
Both clustered and non-clustered space allocation work by iterating over our
extent and bitmap entries and skipping any that represents a region smaller
then the allocation request (and giving preference to extent entries before
bitmap entries). By having a contiguous free space region that is represented
by 2 (or more) entries (mix of extent and bitmap entries), we end up not
satisfying an allocation request with a size larger than the size of any of
the entries but no larger than the sum of their sizes. Making the caller assume
we're under a ENOSPC condition or force it to allocate multiple smaller space
regions (as we do for file data writes), which adds extra overhead and more
chances of causing fragmentation due to the smaller regions being all spread
apart from each other (more likely when under concurrency).
For example, if we have the following in the cache:
* extent entry representing free space range: [128Mb - 256Kb, 128Mb[
* bitmap entry covering the range [128Mb, 256Mb[, but only with the bits
representing the range [128Mb, 128Mb + 768Kb[ set - that is, only that
space in this 128Mb area is marked as free
An allocation request for 1Mb, starting at offset not greater than 128Mb - 256Kb,
would fail before, despite the existence of such contiguous free space area in the
cache. The caller could only allocate up to 768Kb of space at once and later another
256Kb (or vice-versa). In between each smaller allocation request, another task
working on a different file/inode might come in and take that space, preventing the
former task of getting a contiguous 1Mb region of free space.
Therefore this change implements the ability to move free space from bitmap
entries into existing and new free space regions represented with extent
entries. This is done when a space region is added to the cache.
A test was added to the sanity tests that explains in detail the issue too.
Some performance test results with compilebench on a 4 cores machine, with
32Gb of ram and using an HDD follow.
Test: compilebench -D /mnt -i 30 -r 1000 --makej
Before this change:
intial create total runs 30 avg 69.02 MB/s (user 0.28s sys 0.57s)
compile total runs 30 avg 314.96 MB/s (user 0.12s sys 0.25s)
read compiled tree total runs 3 avg 27.14 MB/s (user 1.52s sys 0.90s)
delete compiled tree total runs 30 avg 3.14 seconds (user 0.15s sys 0.66s)
After this change:
intial create total runs 30 avg 68.37 MB/s (user 0.29s sys 0.55s)
compile total runs 30 avg 382.83 MB/s (user 0.12s sys 0.24s)
read compiled tree total runs 3 avg 27.82 MB/s (user 1.45s sys 0.97s)
delete compiled tree total runs 30 avg 3.18 seconds (user 0.17s sys 0.65s)
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
