commit cdce59a154 upstream.
Current code does not fully takes care of krealloc() error case, which
could lead to silent memory corruption or a kernel bug. This patch
fixes that.
Also it cleans up some duplicated error handling logic from various
functions in fast_commit.c file.
Reported-by: luo penghao <luo.penghao@zte.com.cn>
Suggested-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/62e8b6a1cce9359682051deb736a3c0953c9d1e9.1642416995.git.riteshh@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 897026aaa7 upstream.
While running "./check -I 200 generic/475" it sometimes gives below
kernel BUG(). Ideally we should not call ext4_write_inline_data() if
ext4_create_inline_data() has failed.
<log snip>
[73131.453234] kernel BUG at fs/ext4/inline.c:223!
<code snip>
212 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
213 void *buffer, loff_t pos, unsigned int len)
214 {
<...>
223 BUG_ON(!EXT4_I(inode)->i_inline_off);
224 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
This patch handles the error and prints out a emergency msg saying potential
data loss for the given inode (since we couldn't restore the original
inline_data due to some previous error).
[ 9571.070313] EXT4-fs (dm-0): error restoring inline_data for inode -- potential data loss! (inode 1703982, error -30)
Reported-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/9f4cd7dfd54fa58ff27270881823d94ddf78dd07.1642416995.git.riteshh@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 31a074a0c6 upstream.
For now in ext4_mb_new_blocks_simple, if we found a block which
should be excluded then will switch to next group, this may
probably cause 'group' run out of range.
Change to check next block in the same group when get a block should
be excluded. Also change the search range to EXT4_CLUSTERS_PER_GROUP
and add error checking.
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20220110035141.1980-3-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 599ea31d13 upstream.
During fast commit replay procedure, we clear inode blocks bitmap in
ext4_ext_clear_bb(), this may cause ext4_mb_new_blocks_simple() allocate
blocks still in use.
Make ext4_fc_record_regions() also record physical disk regions used by
inodes during replay procedure. Then ext4_mb_new_blocks_simple() can
excludes these blocks in use.
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Link: https://lore.kernel.org/r/20220110035141.1980-2-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 6eeaf88fd5 upstream.
We probably want to remove the indirect block to extents migration
feature after a deprecation window, but until then, let's fix a
potential data loss problem caused by the fact that we put the
tmp_inode on the orphan list. In the unlikely case where we crash and
do a journal recovery, the data blocks belonging to the inode being
migrated are also represented in the tmp_inode on the orphan list ---
and so its data blocks will get marked unallocated, and available for
reuse.
Instead, stop putting the tmp_inode on the oprhan list. So in the
case where we crash while migrating the inode, we'll leak an inode,
which is not a disaster. It will be easily fixed the next time we run
fsck, and it's better than potentially having blocks getting claimed
by two different files, and losing data as a result.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5c48a7df91 upstream.
Our syzkaller report an use-after-free issue that accessing the freed
buffer_head on the writeback page in __ext4_journalled_writepage(). The
problem is that if there was a truncate racing with the data=journalled
writeback procedure, the writeback length could become zero and
bget_one() refuse to get buffer_head's refcount, then the truncate
procedure release buffer once we drop page lock, finally, the last
ext4_walk_page_buffers() trigger the use-after-free problem.
sync truncate
ext4_sync_file()
file_write_and_wait_range()
ext4_setattr(0)
inode->i_size = 0
ext4_writepage()
len = 0
__ext4_journalled_writepage()
page_bufs = page_buffers(page)
ext4_walk_page_buffers(bget_one) <- does not get refcount
do_invalidatepage()
free_buffer_head()
ext4_walk_page_buffers(page_bufs) <- trigger use-after-free
After commit bdf96838ae ("ext4: fix race between truncate and
__ext4_journalled_writepage()"), we have already handled the racing
case, so the bget_one() and bput_one() are not needed. So this patch
simply remove these hunk, and recheck the i_size to make it safe.
Fixes: bdf96838ae ("ext4: fix race between truncate and __ext4_journalled_writepage()")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20211225090937.712867-1-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 9725958bb7 upstream.
If use FALLOC_FL_KEEP_SIZE to alloc unwritten range at bottom, the
inode->i_size will not include the unwritten range. When call
ftruncate with fast commit enabled, it will miss to track the
unwritten range.
Change to trace the full range during ftruncate.
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20211223032337.5198-3-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0b5b5a62b9 upstream.
For now ,we use ext4_punch_hole() during fast commit replay delete range
procedure. But it will be affected by inode->i_size, which may not
correct during fast commit replay procedure. The following test will
failed.
-create & write foo (len 1000K)
-falloc FALLOC_FL_ZERO_RANGE foo (range 400K - 600K)
-create & fsync bar
-falloc FALLOC_FL_PUNCH_HOLE foo (range 300K-500K)
-fsync foo
-crash before a full commit
After the fast_commit reply procedure, the range 400K-500K will not be
removed. Because in this case, when calling ext4_punch_hole() the
inode->i_size is 0, and it just retruns with doing nothing.
Change to use ext4_ext_remove_space() instead of ext4_punch_hole()
to remove blocks of inode directly.
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20211223032337.5198-2-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e81c9302a6 upstream.
When migrating to extents, the temporary inode will have it's own checksum
seed. This means that, when swapping the inodes data, the inode checksums
will be incorrect.
This can be fixed by recalculating the extents checksums again. Or simply
by copying the seed into the temporary inode.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=213357
Reported-by: Jeroen van Wolffelaar <jeroen@wolffelaar.nl>
Signed-off-by: Luís Henriques <lhenriques@suse.de>
Link: https://lore.kernel.org/r/20211214175058.19511-1-lhenriques@suse.de
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5e4d0eba1c upstream.
when call falloc with FALLOC_FL_ZERO_RANGE, to set an range to unwritten,
which has been already initialized. If the range is align to blocksize,
fast commit will not track range for this change.
Also track range for unwritten range in ext4_map_blocks().
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20211221022839.374606-1-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c27c29c6af upstream.
It is not guaranteed that __ext4_get_inode_loc will definitely set
err_blk pointer when it returns EIO. To avoid using uninitialized
variables, let's first set err_blk to 0.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20211201163421.2631661-1-harshads@google.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8c80fb312d upstream.
We found on older kernel (3.10) that in the scenario of insufficient
disk space, system may trigger an ABBA deadlock problem, it seems that
this problem still exists in latest kernel, try to fix it here. The
main process triggered by this problem is that task A occupies the PA
and waits for the jbd2 transaction finish, the jbd2 transaction waits
for the completion of task B's IO (plug_list), but task B waits for
the release of PA by task A to finish discard, which indirectly forms
an ABBA deadlock. The related calltrace is as follows:
Task A
vfs_write
ext4_mb_new_blocks()
ext4_mb_mark_diskspace_used() JBD2
jbd2_journal_get_write_access() -> jbd2_journal_commit_transaction()
->schedule() filemap_fdatawait()
| |
| Task B |
| do_unlinkat() |
| ext4_evict_inode() |
| jbd2_journal_begin_ordered_truncate() |
| filemap_fdatawrite_range() |
| ext4_mb_new_blocks() |
-ext4_mb_discard_group_preallocations() <-----
Here, try to cancel ext4_mb_discard_group_preallocations() internal
retry due to PA busy, and do a limited number of retries inside
ext4_mb_discard_preallocations(), which can circumvent the above
problems, but also has some advantages:
1. Since the PA is in a busy state, if other groups have free PAs,
keeping the current PA may help to reduce fragmentation.
2. Continue to traverse forward instead of waiting for the current
group PA to be released. In most scenarios, the PA discard time
can be reduced.
However, in the case of smaller free space, if only a few groups have
space, then due to multiple traversals of the group, it may increase
CPU overhead. But in contrast, I feel that the overall benefit is
better than the cost.
Signed-off-by: Chunguang Xu <brookxu@tencent.com>
Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/1637630277-23496-1-git-send-email-brookxu.cn@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 15fc69bbbb upstream.
When we hit an error when enabling quotas and setting inode flags, we do
not properly shutdown quota subsystem despite returning error from
Q_QUOTAON quotactl. This can lead to some odd situations like kernel
using quota file while it is still writeable for userspace. Make sure we
properly cleanup the quota subsystem in case of error.
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20211007155336.12493-2-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4013d47a53 upstream.
When we succeed in enabling some quota type but fail to enable another
one with quota feature, we correctly disable all enabled quota types.
However we forget to reset i_data_sem lockdep class. When the inode gets
freed and reused, it will inherit this lockdep class (i_data_sem is
initialized only when a slab is created) and thus eventually lockdep
barfs about possible deadlocks.
Reported-and-tested-by: syzbot+3b6f9218b1301ddda3e2@syzkaller.appspotmail.com
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20211007155336.12493-3-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 173b6e383d ]
A user reported FITRIM ioctl failing for him on ext4 on some devices
without apparent reason. After some debugging we've found out that
these devices (being LVM volumes) report rather large discard
granularity of 42MB and the filesystem had 1k blocksize and thus group
size of 8MB. Because ext4 FITRIM implementation puts discard
granularity into minlen, ext4_trim_fs() declared the trim request as
invalid. However just silently doing nothing seems to be a more
appropriate reaction to such combination of parameters since user did
not specify anything wrong.
CC: Lukas Czerner <lczerner@redhat.com>
Fixes: 5c2ed62fd4 ("ext4: Adjust minlen with discard_granularity in the FITRIM ioctl")
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20211112152202.26614-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 9c6e071913 upstream.
Now that we can check out overlapping extents in leaf block and
out-of-order index extents in index block. But the .ee_block in the
first extent of one leaf block should equal to the .ei_block in it's
parent index extent entry. This patch add a check to verify such
inconsistent between the index and leaf block.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://lore.kernel.org/r/20210908120850.4012324-3-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8dd27feced upstream.
After commit 5946d08937 ("ext4: check for overlapping extents in
ext4_valid_extent_entries()"), we can check out the overlapping extent
entry in leaf extent blocks. But the out-of-order extent entry in index
extent blocks could also trigger bad things if the filesystem is
inconsistent. So this patch add a check to figure out the out-of-order
index extents and return error.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210908120850.4012324-2-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0f2f87d51a upstream.
In the most error path of current extents updating operations are not
roll back partial updates properly when some bad things happens(.e.g in
ext4_ext_insert_extent()). So we may get an inconsistent extents tree
if journal has been aborted due to IO error, which may probability lead
to BUGON later when we accessing these extent entries in errors=continue
mode. This patch drop extent buffer's verify flag before updatng the
contents in ext4_ext_get_access(), and reset it after updating in
__ext4_ext_dirty(). After this patch we could force to check the extent
buffer if extents tree updating was break off, make sure the extents are
consistent.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210908120850.4012324-4-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 1811bc401a upstream.
After we drop i_data sem, we need to reload the ext4_ext_path
structure since the extent tree can change once i_data_sem is
released.
This addresses the BUG:
[52117.465187] ------------[ cut here ]------------
[52117.465686] kernel BUG at fs/ext4/extents.c:1756!
...
[52117.478306] Call Trace:
[52117.478565] ext4_ext_shift_extents+0x3ee/0x710
[52117.479020] ext4_fallocate+0x139c/0x1b40
[52117.479405] ? __do_sys_newfstat+0x6b/0x80
[52117.479805] vfs_fallocate+0x151/0x4b0
[52117.480177] ksys_fallocate+0x4a/0xa0
[52117.480533] __x64_sys_fallocate+0x22/0x30
[52117.480930] do_syscall_64+0x35/0x80
[52117.481277] entry_SYSCALL_64_after_hwframe+0x44/0xae
[52117.481769] RIP: 0033:0x7fa062f855ca
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20210903062748.4118886-4-yangerkun@huawei.com
Signed-off-by: yangerkun <yangerkun@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4268496e48 upstream.
Like ext4_ext_rm_leaf, we can ensure that there are enough credits
before every call that will consume credits. As part of this fix we
fold the functionality of ext4_access_path() into
ext4_ext_shift_path_extents(). This change is needed as a preparation
for the next bugfix patch.
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20210903062748.4118886-3-yangerkun@huawei.com
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 39fec6889d upstream.
Ext4 file system has default lazy inode table initialization setup once
it is mounted. However, it has issue on computing the next schedule time
that makes the timeout same amount in jiffies but different real time in
secs if with various HZ values. Therefore, fix by measuring the current
time in a more granular unit nanoseconds and make the next schedule time
independent of the HZ value.
Fixes: bfff68738f ("ext4: add support for lazy inode table initialization")
Signed-off-by: Shaoying Xu <shaoyi@amazon.com>
Cc: stable@vger.kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210902164412.9994-2-shaoyi@amazon.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3eda41df05 upstream.
This reverts commit 948ca5f30e.
Two crash reports from users running variations on 5.15-rc4 kernels
suggest that it is premature to enforce the state assertion in the
original commit. Both crashes were triggered by BUG calls in that
code, indicating that under some rare circumstance the buffer head
state did not match a delayed allocated block at the time the
block was written out. No reproducer is available. Resolving this
problem will require more time than remains in the current release
cycle, so reverting the original patch for the time being is necessary
to avoid any instability it may cause.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Link: https://lore.kernel.org/r/20211012171901.5352-1-enwlinux@gmail.com
Fixes: 948ca5f30e ("ext4: enforce buffer head state assertion in ext4_da_map_blocks")
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
allocation. Also fix error handling code paths in ext4_dx_readdir()
and ext4_fill_super(). Finally, avoid a grabbing a journal head in
the delayed allocation write in the common cases where we are
overwriting an pre-existing block or appending to an inode.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEK2m5VNv+CHkogTfJ8vlZVpUNgaMFAmFZ2SsACgkQ8vlZVpUN
gaN6DAgAkIeisL1EfQT0VwshEs8y7N6IoX8dydLSRLpNf5oWiJOv2CTY9Qpi6X/C
qNfuLsbJ2NXChvhIAM2hD82hvX21rYc6iqPxgho02VF4eYIP7NzLjwTFKnKbHPB5
TiF498nJTnkcmSrJUEXmSAEdLoCwa5THH9+9HVHXZrkLXPULBtOOJ85mDAcIzVhV
Zqb7yfbpWl0gnF0S0YjNATPtbhcC9EiC4MOVYVesRlgT9B3+k5q4fmVU0euTU9OH
F2H6TNG+Mg/19gTnDP5acB9+eXHvYEqMpe+CaDifR9iFE9PTG/Edhxr6z9roXhHr
kBvEVHSFH+YTEJXghnpS9YDd9Lwc9w==
=WKzd
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 fixes from Ted Ts'o:
"Fix a number of ext4 bugs in fast_commit, inline data, and delayed
allocation.
Also fix error handling code paths in ext4_dx_readdir() and
ext4_fill_super().
Finally, avoid a grabbing a journal head in the delayed allocation
write in the common cases where we are overwriting a pre-existing
block or appending to an inode"
* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
ext4: recheck buffer uptodate bit under buffer lock
ext4: fix potential infinite loop in ext4_dx_readdir()
ext4: flush s_error_work before journal destroy in ext4_fill_super
ext4: fix loff_t overflow in ext4_max_bitmap_size()
ext4: fix reserved space counter leakage
ext4: limit the number of blocks in one ADD_RANGE TLV
ext4: enforce buffer head state assertion in ext4_da_map_blocks
ext4: remove extent cache entries when truncating inline data
ext4: drop unnecessary journal handle in delalloc write
ext4: factor out write end code of inline file
ext4: correct the error path of ext4_write_inline_data_end()
ext4: check and update i_disksize properly
ext4: add error checking to ext4_ext_replay_set_iblocks()
Commit 8e33fadf94 ("ext4: remove an unnecessary if statement in
__ext4_get_inode_loc()") forget to recheck buffer's uptodate bit again
under buffer lock, which may overwrite the buffer if someone else have
already brought it uptodate and changed it.
Fixes: 8e33fadf94 ("ext4: remove an unnecessary if statement in __ext4_get_inode_loc()")
Cc: stable@kernel.org
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210910080316.70421-1-yi.zhang@huawei.com
When ext4_htree_fill_tree() fails, ext4_dx_readdir() can run into an
infinite loop since if info->last_pos != ctx->pos this will reset the
directory scan and reread the failing entry. For example:
1. a dx_dir which has 3 block, block 0 as dx_root block, block 1/2 as
leaf block which own the ext4_dir_entry_2
2. block 1 read ok and call_filldir which will fill the dirent and update
the ctx->pos
3. block 2 read fail, but we has already fill some dirent, so we will
return back to userspace will a positive return val(see ksys_getdents64)
4. the second ext4_dx_readdir will reset the world since info->last_pos
!= ctx->pos, and will also init the curr_hash which pos to block 1
5. So we will read block1 too, and once block2 still read fail, we can
only fill one dirent because the hash of the entry in block1(besides
the last one) won't greater than curr_hash
6. this time, we forget update last_pos too since the read for block2
will fail, and since we has got the one entry, ksys_getdents64 can
return success
7. Latter we will trapped in a loop with step 4~6
Cc: stable@kernel.org
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210914111415.3921954-1-yangerkun@huawei.com
The error path in ext4_fill_super forget to flush s_error_work before
journal destroy, and it may trigger the follow bug since
flush_stashed_error_work can run concurrently with journal destroy
without any protection for sbi->s_journal.
[32031.740193] EXT4-fs (loop66): get root inode failed
[32031.740484] EXT4-fs (loop66): mount failed
[32031.759805] ------------[ cut here ]------------
[32031.759807] kernel BUG at fs/jbd2/transaction.c:373!
[32031.760075] invalid opcode: 0000 [#1] SMP PTI
[32031.760336] CPU: 5 PID: 1029268 Comm: kworker/5:1 Kdump: loaded
4.18.0
[32031.765112] Call Trace:
[32031.765375] ? __switch_to_asm+0x35/0x70
[32031.765635] ? __switch_to_asm+0x41/0x70
[32031.765893] ? __switch_to_asm+0x35/0x70
[32031.766148] ? __switch_to_asm+0x41/0x70
[32031.766405] ? _cond_resched+0x15/0x40
[32031.766665] jbd2__journal_start+0xf1/0x1f0 [jbd2]
[32031.766934] jbd2_journal_start+0x19/0x20 [jbd2]
[32031.767218] flush_stashed_error_work+0x30/0x90 [ext4]
[32031.767487] process_one_work+0x195/0x390
[32031.767747] worker_thread+0x30/0x390
[32031.768007] ? process_one_work+0x390/0x390
[32031.768265] kthread+0x10d/0x130
[32031.768521] ? kthread_flush_work_fn+0x10/0x10
[32031.768778] ret_from_fork+0x35/0x40
static int start_this_handle(...)
BUG_ON(journal->j_flags & JBD2_UNMOUNT); <---- Trigger this
Besides, after we enable fast commit, ext4_fc_replay can add work to
s_error_work but return success, so the latter journal destroy in
ext4_load_journal can trigger this problem too.
Fix this problem with two steps:
1. Call ext4_commit_super directly in ext4_handle_error for the case
that called from ext4_fc_replay
2. Since it's hard to pair the init and flush for s_error_work, we'd
better add a extras flush_work before journal destroy in
ext4_fill_super
Besides, this patch will call ext4_commit_super in ext4_handle_error for
any nojournal case too. But it seems safe since the reason we call
schedule_work was that we should save error info to sb through journal
if available. Conversely, for the nojournal case, it seems useless delay
commit superblock to s_error_work.
Fixes: c92dc85684 ("ext4: defer saving error info from atomic context")
Fixes: 2d01ddc866 ("ext4: save error info to sb through journal if available")
Cc: stable@kernel.org
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210924093917.1953239-1-yangerkun@huawei.com
We should use unsigned long long rather than loff_t to avoid
overflow in ext4_max_bitmap_size() for comparison before returning.
w/o this patch sbi->s_bitmap_maxbytes was becoming a negative
value due to overflow of upper_limit (with has_huge_files as true)
Below is a quick test to trigger it on a 64KB pagesize system.
sudo mkfs.ext4 -b 65536 -O ^has_extents,^64bit /dev/loop2
sudo mount /dev/loop2 /mnt
sudo echo "hello" > /mnt/hello -> This will error out with
"echo: write error: File too large"
Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/594f409e2c543e90fd836b78188dfa5c575065ba.1622867594.git.riteshh@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When ext4_insert_delayed block receives and recovers from an error from
ext4_es_insert_delayed_block(), e.g., ENOMEM, it does not release the
space it has reserved for that block insertion as it should. One effect
of this bug is that s_dirtyclusters_counter is not decremented and
remains incorrectly elevated until the file system has been unmounted.
This can result in premature ENOSPC returns and apparent loss of free
space.
Another effect of this bug is that
/sys/fs/ext4/<dev>/delayed_allocation_blocks can remain non-zero even
after syncfs has been executed on the filesystem.
Besides, add check for s_dirtyclusters_counter when inode is going to be
evicted and freed. s_dirtyclusters_counter can still keep non-zero until
inode is written back in .evict_inode(), and thus the check is delayed
to .destroy_inode().
Fixes: 51865fda28 ("ext4: let ext4 maintain extent status tree")
Cc: stable@kernel.org
Suggested-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com>
Reviewed-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210823061358.84473-1-jefflexu@linux.alibaba.com
Now EXT4_FC_TAG_ADD_RANGE uses ext4_extent to track the
newly-added blocks, but the limit on the max value of
ee_len field is ignored, and it can lead to BUG_ON as
shown below when running command "fallocate -l 128M file"
on a fast_commit-enabled fs:
kernel BUG at fs/ext4/ext4_extents.h:199!
invalid opcode: 0000 [#1] SMP PTI
CPU: 3 PID: 624 Comm: fallocate Not tainted 5.14.0-rc6+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
RIP: 0010:ext4_fc_write_inode_data+0x1f3/0x200
Call Trace:
? ext4_fc_write_inode+0xf2/0x150
ext4_fc_commit+0x93b/0xa00
? ext4_fallocate+0x1ad/0x10d0
ext4_sync_file+0x157/0x340
? ext4_sync_file+0x157/0x340
vfs_fsync_range+0x49/0x80
do_fsync+0x3d/0x70
__x64_sys_fsync+0x14/0x20
do_syscall_64+0x3b/0xc0
entry_SYSCALL_64_after_hwframe+0x44/0xae
Simply fixing it by limiting the number of blocks
in one EXT4_FC_TAG_ADD_RANGE TLV.
Fixes: aa75f4d3da ("ext4: main fast-commit commit path")
Cc: stable@kernel.org
Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210820044505.474318-1-houtao1@huawei.com
- Fix a race condition in the teardown path of raw mode pmem namespaces.
- Cleanup the code that filesystems use to detect filesystem-dax
capabilities of their underlying block device.
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQSbo+XnGs+rwLz9XGXfioYZHlFsZwUCYTlBMgAKCRDfioYZHlFs
ZwQLAQCPhwpuOP+Byn7NksotnfmyLNyniK0mX7Me7PoLiyq0oAEAmqBwlr9YP7E3
NPzWiBzqPCvDIv1YG4C3Vam7ue1osgM=
=33O+
-----END PGP SIGNATURE-----
Merge tag 'libnvdimm-for-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm
Pull libnvdimm updates from Dan Williams:
- Fix a race condition in the teardown path of raw mode pmem
namespaces.
- Cleanup the code that filesystems use to detect filesystem-dax
capabilities of their underlying block device.
* tag 'libnvdimm-for-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
dax: remove bdev_dax_supported
xfs: factor out a xfs_buftarg_is_dax helper
dax: stub out dax_supported for !CONFIG_FS_DAX
dax: remove __generic_fsdax_supported
dax: move the dax_read_lock() locking into dax_supported
dax: mark dax_get_by_host static
dm: use fs_dax_get_by_bdev instead of dax_get_by_host
dax: stop using bdevname
fsdax: improve the FS_DAX Kconfig description and help text
libnvdimm/pmem: Fix crash triggered when I/O in-flight during unbind
Remove the code that re-initializes a buffer head with an invalid block
number and BH_New and BH_Delay bits when a matching delayed and
unwritten block has been found in the extent status cache. Replace it
with assertions that verify the buffer head already has this state
correctly set. The current code masked an inline data truncation bug
that left stale entries in the extent status cache. With this change,
generic/130 can be used to reproduce and detect that bug.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210819144927.25163-3-enwlinux@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Conditionally remove all cached extents belonging to an inode
when truncating its inline data. It's only necessary to attempt to
remove cached extents when a conversion from inline to extent storage
has been initiated (!EXT4_STATE_MAY_INLINE_DATA). This avoids
unnecessary es lock overhead in the more common inline case.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210819144927.25163-2-enwlinux@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Fix a bug in how we update i_disksize, and the error path in
inline_data_end. Finally, drop an unnecessary creation of a journal
handle which was only needed for inline data, which can give us a
large performance gain in delayed allocation writes.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After we factor out the inline data write procedure from
ext4_da_write_end(), we don't need to start journal handle for the cases
of both buffer overwrite and append-write. If we need to update
i_disksize, mark_inode_dirty() do start handle and update inode buffer.
So we could just remove all the journal handle codes in the delalloc
write procedure.
After this patch, we could get a lot of performance improvement. Below
is the Unixbench comparison data test on my machine with 'Intel Xeon
Gold 5120' CPU and nvme SSD backend.
Test cmd:
./Run -c 56 -i 3 fstime fsbuffer fsdisk
Before this patch:
System Benchmarks Partial Index BASELINE RESULT INDEX
File Copy 1024 bufsize 2000 maxblocks 3960.0 422965.0 1068.1
File Copy 256 bufsize 500 maxblocks 1655.0 105077.0 634.9
File Copy 4096 bufsize 8000 maxblocks 5800.0 1429092.0 2464.0
======
System Benchmarks Index Score (Partial Only) 1186.6
After this patch:
System Benchmarks Partial Index BASELINE RESULT INDEX
File Copy 1024 bufsize 2000 maxblocks 3960.0 732716.0 1850.3
File Copy 256 bufsize 500 maxblocks 1655.0 184940.0 1117.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 2427152.0 4184.7
======
System Benchmarks Index Score (Partial Only) 2053.0
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210716122024.1105856-5-yi.zhang@huawei.com
Now that the inline_data file write end procedure are falled into the
common write end functions, it is not clear. Factor them out and do
some cleanup. This patch also drop ext4_da_write_inline_data_end()
and switch to use ext4_write_inline_data_end() instead because we also
need to do the same error processing if we failed to write data into
inline entry.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210716122024.1105856-4-yi.zhang@huawei.com
Current error path of ext4_write_inline_data_end() is not correct.
Firstly, it should pass out the error value if ext4_get_inode_loc()
return fail, or else it could trigger infinite loop if we inject error
here. And then it's better to add inode to orphan list if it return fail
in ext4_journal_stop(), otherwise we could not restore inline xattr
entry after power failure. Finally, we need to reset the 'ret' value if
ext4_write_inline_data_end() return success in ext4_write_end() and
ext4_journalled_write_end(), otherwise we could not get the error return
value of ext4_journal_stop().
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210716122024.1105856-3-yi.zhang@huawei.com
After commit 3da40c7b08 ("ext4: only call ext4_truncate when size <=
isize"), i_disksize could always be updated to i_size in ext4_setattr(),
and we could sure that i_disksize <= i_size since holding inode lock and
if i_disksize < i_size there are delalloc writes pending in the range
upto i_size. If the end of the current write is <= i_size, there's no
need to touch i_disksize since writeback will push i_disksize upto
i_size eventually. So we can switch to check i_size instead of
i_disksize in ext4_da_write_end() when write to the end of the file.
we also could remove ext4_mark_inode_dirty() together because we defer
inode dirtying to generic_write_end() or ext4_da_write_inline_data_end().
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://lore.kernel.org/r/20210716122024.1105856-2-yi.zhang@huawei.com
orphan_file feature, which eliminates bottlenecks when doing a large
number of parallel truncates and file deletions, and move the discard
operation out of the jbd2 commit thread when using the discard mount
option, to better support devices with slow discard operations.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEK2m5VNv+CHkogTfJ8vlZVpUNgaMFAmEw5gEACgkQ8vlZVpUN
gaMatgf9GKc2H3JUDGJVXrOE1EZWXzyDI+Tv6zt0iTr05zi1ahjGccAbmJXiAwxU
Zy5TGr53CpPcGUG+sO4NpVzcH8q7cQeG0pVx9OnzJUdVfmv+htoNE0aAqUY5L3vg
AxV4KPGgxPofRQa3QRE2LDFHIkNs7c0ncprdaAtxNztd09iFo7bIayt614mARK++
HIO7VOGrH5Wya8SSoqYHmlO0g5viy3ypP6CpysIQw0JifSlHYkmYBUJ0/hwPV/Xl
WfzmwQ9p43C9EXVmIN4++l674TDzkSn9ebITXOgkq4C8KjnFgyhKQIj5QVj81MvH
dac5jxsuLTXTLYnRpAQ/duV4jRd+Fw==
=+NN7
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 updates from Ted Ts'o:
"In addition to some ext4 bug fixes and cleanups, this cycle we add the
orphan_file feature, which eliminates bottlenecks when doing a large
number of parallel truncates and file deletions, and move the discard
operation out of the jbd2 commit thread when using the discard mount
option, to better support devices with slow discard operations"
* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (23 commits)
ext4: make the updating inode data procedure atomic
ext4: remove an unnecessary if statement in __ext4_get_inode_loc()
ext4: move inode eio simulation behind io completeion
ext4: Improve scalability of ext4 orphan file handling
ext4: Orphan file documentation
ext4: Speedup ext4 orphan inode handling
ext4: Move orphan inode handling into a separate file
ext4: Support for checksumming from journal triggers
ext4: fix race writing to an inline_data file while its xattrs are changing
jbd2: add sparse annotations for add_transaction_credits()
ext4: fix sparse warnings
ext4: Make sure quota files are not grabbed accidentally
ext4: fix e2fsprogs checksum failure for mounted filesystem
ext4: if zeroout fails fall back to splitting the extent node
ext4: reduce arguments of ext4_fc_add_dentry_tlv
ext4: flush background discard kwork when retry allocation
ext4: get discard out of jbd2 commit kthread contex
ext4: remove the repeated comment of ext4_trim_all_free
ext4: add new helper interface ext4_try_to_trim_range()
ext4: remove the 'group' parameter of ext4_trim_extent
...
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQSQHSd0lITzzeNWNm3h3BK/laaZPAUCYTDKKAAKCRDh3BK/laaZ
PG9PAQCUF0fdBlCKudwSEt5PV5xemycL9OCAlYCd7d4XbBIe9wEA6sVJL9J+OwV2
aF0NomiXtJccE+S9+byjVCyqSzQJGQQ=
=6L2Y
-----END PGP SIGNATURE-----
Merge tag 'ovl-update-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs
Pull overlayfs update from Miklos Szeredi:
- Copy up immutable/append/sync/noatime attributes (Amir Goldstein)
- Improve performance by enabling RCU lookup.
- Misc fixes and improvements
The reason this touches so many files is that the ->get_acl() method now
gets a "bool rcu" argument. The ->get_acl() API was updated based on
comments from Al and Linus:
Link: https://lore.kernel.org/linux-fsdevel/CAJfpeguQxpd6Wgc0Jd3ks77zcsAv_bn0q17L3VNnnmPKu11t8A@mail.gmail.com/
* tag 'ovl-update-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs:
ovl: enable RCU'd ->get_acl()
vfs: add rcu argument to ->get_acl() callback
ovl: fix BUG_ON() in may_delete() when called from ovl_cleanup()
ovl: use kvalloc in xattr copy-up
ovl: update ctime when changing fileattr
ovl: skip checking lower file's i_writecount on truncate
ovl: relax lookup error on mismatch origin ftype
ovl: do not set overlay.opaque for new directories
ovl: add ovl_allow_offline_changes() helper
ovl: disable decoding null uuid with redirect_dir
ovl: consistent behavior for immutable/append-only inodes
ovl: copy up sync/noatime fileattr flags
ovl: pass ovl_fs to ovl_check_setxattr()
fs: add generic helper for filling statx attribute flags
If the call to ext4_map_blocks() fails due to an corrupted file
system, ext4_ext_replay_set_iblocks() can get stuck in an infinite
loop. This could be reproduced by running generic/526 with a file
system that has inline_data and fast_commit enabled. The system will
repeatedly log to the console:
EXT4-fs warning (device dm-3): ext4_block_to_path:105: block 1074800922 > max in inode 131076
and the stack that it gets stuck in is:
ext4_block_to_path+0xe3/0x130
ext4_ind_map_blocks+0x93/0x690
ext4_map_blocks+0x100/0x660
skip_hole+0x47/0x70
ext4_ext_replay_set_iblocks+0x223/0x440
ext4_fc_replay_inode+0x29e/0x3b0
ext4_fc_replay+0x278/0x550
do_one_pass+0x646/0xc10
jbd2_journal_recover+0x14a/0x270
jbd2_journal_load+0xc4/0x150
ext4_load_journal+0x1f3/0x490
ext4_fill_super+0x22d4/0x2c00
With this patch, generic/526 still fails, but system is no longer
locking up in a tight loop. It's likely the root casue is that
fast_commit replay is corrupting file systems with inline_data, and we
probably need to add better error handling in the fast commit replay
code path beyond what is done here, which essentially just breaks the
infinite loop without reporting the to the higher levels of the code.
Fixes: 8016E29F4362 ("ext4: fast commit recovery path")
Cc: stable@kernel.org
Cc: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Some small fixes and cleanups for fs/crypto/:
- Fix ->getattr() for ext4, f2fs, and ubifs to report the correct
st_size for encrypted symlinks.
- Use base64url instead of a custom Base64 variant.
- Document struct fscrypt_operations.
-----BEGIN PGP SIGNATURE-----
iIoEABYIADIWIQSacvsUNc7UX4ntmEPzXCl4vpKOKwUCYS0HzhQcZWJpZ2dlcnNA
Z29vZ2xlLmNvbQAKCRDzXCl4vpKOK+XZAQDfvDE9gK4Ii2uE4Jb5XYv4M/BnVhoR
WIhNEoHROIGv+AEAtyfmeCMdpPobkWHFfAE1iBysl3iS56fibQhi2wqyuQI=
=s6Wi
-----END PGP SIGNATURE-----
Merge tag 'fscrypt-for-linus' of git://git.kernel.org/pub/scm/fs/fscrypt/fscrypt
Pull fscrypt updates from Eric Biggers:
"Some small fixes and cleanups for fs/crypto/:
- Fix ->getattr() for ext4, f2fs, and ubifs to report the correct
st_size for encrypted symlinks
- Use base64url instead of a custom Base64 variant
- Document struct fscrypt_operations"
* tag 'fscrypt-for-linus' of git://git.kernel.org/pub/scm/fs/fscrypt/fscrypt:
fscrypt: document struct fscrypt_operations
fscrypt: align Base64 encoding with RFC 4648 base64url
fscrypt: remove mention of symlink st_size quirk from documentation
ubifs: report correct st_size for encrypted symlinks
f2fs: report correct st_size for encrypted symlinks
ext4: report correct st_size for encrypted symlinks
fscrypt: add fscrypt_symlink_getattr() for computing st_size
Now that ext4_do_update_inode() return error before filling the whole
inode data if we fail to set inode blocks in ext4_inode_blocks_set().
This error should never happen in theory since sb->s_maxbytes should not
have allowed this, we have already init sb->s_maxbytes according to this
feature in ext4_fill_super(). So even through that could only happen due
to the filesystem corruption, we'd better to return after we finish
updating the inode because it may left an uninitialized buffer and we
could read this buffer later in "errors=continue" mode.
This patch make the updating inode data procedure atomic, call
EXT4_ERROR_INODE() after we dropping i_raw_lock after something bad
happened, make sure that the inode is integrated, and also drop a BUG_ON
and do some small cleanups.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210826130412.3921207-4-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The "if (!buffer_uptodate(bh))" hunk covered almost the whole code after
getting buffer in __ext4_get_inode_loc() which seems unnecessary, remove
it and switch to check ext4_buffer_uptodate(), it simplify code and make
it more readable.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210826130412.3921207-3-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
No EIO simulation is required if the buffer is uptodate, so move the
simulation behind read bio completeion just like inode/block bitmap
simulation does.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210826130412.3921207-2-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Even though the length of the critical section when adding / removing
orphaned inodes was significantly reduced by using orphan file, the
contention of lock protecting orphan file still appears high in profiles
for truncate / unlink intensive workloads with high number of threads.
This patch makes handling of orphan file completely lockless. Also to
reduce conflicts between CPUs different CPUs start searching for empty
slot in orphan file in different blocks.
Performance comparison of locked orphan file handling, lockless orphan
file handling, and completely disabled orphan inode handling
from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on
SAS SSD disk, average of 5 runs:
stress-orphan (microbenchmark truncating files byte-by-byte from N
processes in parallel)
Threads Time Time Time
Orphan locked Orphan lockless No orphan
1 0.945600 0.939400 0.891200
2 1.331800 1.246600 1.174400
4 1.995000 1.780600 1.713200
8 6.424200 4.900000 4.106000
16 14.937600 8.516400 8.138000
32 33.038200 24.565600 24.002200
64 60.823600 39.844600 38.440200
128 122.941400 70.950400 69.315000
So we can see that with lockless orphan file handling, addition /
deletion of orphaned inodes got almost completely out of picture even
for a microbenchmark stressing it.
For reaim creat_clo workload on ramdisk there are also noticeable gains
(average of 5 runs):
Clients Vanilla (ops/s) Patched (ops/s)
creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%*
creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%)
creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%*
creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%*
creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%*
creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%*
creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%*
creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%*
creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%*
creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%*
creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%*
creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%*
creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%*
creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%*
creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%*
creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%*
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Ext4 orphan inode handling is a bottleneck for workloads which heavily
truncate / unlink small files since it contends on the global
s_orphan_mutex lock (and generally it's difficult to improve scalability
of the ondisk linked list of orphaned inodes).
This patch implements new way of handling orphan inodes. Instead of
linking orphaned inode into a linked list, we store it's inode number in
a new special file which we call "orphan file". Only if there's no more
space in the orphan file (too many inodes are currently orphaned) we
fall back to using old style linked list. Currently we protect
operations in the orphan file with a spinlock for simplicity but even in
this setting we can substantially reduce the length of the critical
section and thus speedup some workloads. In the next patch we improve
this by making orphan handling lockless.
Note that the change is backwards compatible when the filesystem is
clean - the existence of the orphan file is a compat feature, we set
another ro-compat feature indicating orphan file needs scanning for
orphaned inodes when mounting filesystem read-write. This ro-compat
feature gets cleared on unmount / remount read-only.
Some performance data from 80 CPU Xeon Server with 512 GB of RAM,
filesystem located on SSD, average of 5 runs:
stress-orphan (microbenchmark truncating files byte-by-byte from N
processes in parallel)
Threads Time Time
Vanilla Patched
1 1.057200 0.945600
2 1.680400 1.331800
4 2.547000 1.995000
8 7.049400 6.424200
16 14.827800 14.937600
32 40.948200 33.038200
64 87.787400 60.823600
128 206.504000 122.941400
So we can see significant wins all over the board.
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
