Copying the file system while it is mounted as read-only results in
a mount failure:
[~]# mkfs.ext4 -F /dev/sdc
[~]# mount /dev/sdc -o ro /mnt/test
[~]# dd if=/dev/sdc of=/dev/sda bs=1M
[~]# mount /dev/sda /mnt/test1
[ 1094.849826] JBD2: journal checksum error
[ 1094.850927] EXT4-fs (sda): Could not load journal inode
mount: mount /dev/sda on /mnt/test1 failed: Bad message
The process described above is just an abstracted way I came up with to
reproduce the issue. In the actual scenario, the file system was mounted
read-only and then copied while it was still mounted. It was found that
the mount operation failed. The user intended to verify the data or use
it as a backup, and this action was performed during a version upgrade.
Above issue may happen as follows:
ext4_fill_super
set_journal_csum_feature_set(sb)
if (ext4_has_metadata_csum(sb))
incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
if (test_opt(sb, JOURNAL_CHECKSUM)
jbd2_journal_set_features(sbi->s_journal, compat, 0, incompat);
lock_buffer(journal->j_sb_buffer);
sb->s_feature_incompat |= cpu_to_be32(incompat);
//The data in the journal sb was modified, but the checksum was not
updated, so the data remaining in memory has a mismatch between the
data and the checksum.
unlock_buffer(journal->j_sb_buffer);
In this case, the journal sb copied over is in a state where the checksum
and data are inconsistent, so mounting fails.
To solve the above issue, update the checksum in memory after modifying
the journal sb.
Fixes: 4fd5ea43bc ("jbd2: checksum journal superblock")
Signed-off-by: Ye Bin <yebin10@huawei.com>
Reviewed-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Message-ID: <20251103010123.3753631-1-yebin@huaweicloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
When jbd2_journal_abort() is called, the provided error code is stored
in the journal superblock. Some existing calls hard-code -EIO even when
the actual failure is not I/O related.
This patch updates those calls to pass more accurate error codes,
allowing the superblock to record the true cause of failure. This helps
improve diagnostics and debugging clarity when analyzing journal aborts.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Message-ID: <20251031210501.7337-1-wen.gang.wang@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
There's issue when file system corrupted:
------------[ cut here ]------------
kernel BUG at fs/jbd2/transaction.c:1289!
Oops: invalid opcode: 0000 [#1] SMP KASAN PTI
CPU: 5 UID: 0 PID: 2031 Comm: mkdir Not tainted 6.18.0-rc1-next
RIP: 0010:jbd2_journal_get_create_access+0x3b6/0x4d0
RSP: 0018:ffff888117aafa30 EFLAGS: 00010202
RAX: 0000000000000000 RBX: ffff88811a86b000 RCX: ffffffff89a63534
RDX: 1ffff110200ec602 RSI: 0000000000000004 RDI: ffff888100763010
RBP: ffff888100763000 R08: 0000000000000001 R09: ffff888100763028
R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000000
R13: ffff88812c432000 R14: ffff88812c608000 R15: ffff888120bfc000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f91d6970c99 CR3: 00000001159c4000 CR4: 00000000000006f0
Call Trace:
<TASK>
__ext4_journal_get_create_access+0x42/0x170
ext4_getblk+0x319/0x6f0
ext4_bread+0x11/0x100
ext4_append+0x1e6/0x4a0
ext4_init_new_dir+0x145/0x1d0
ext4_mkdir+0x326/0x920
vfs_mkdir+0x45c/0x740
do_mkdirat+0x234/0x2f0
__x64_sys_mkdir+0xd6/0x120
do_syscall_64+0x5f/0xfa0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
The above issue occurs with us in errors=continue mode when accompanied by
storage failures. There have been many inconsistencies in the file system
data.
In the case of file system data inconsistency, for example, if the block
bitmap of a referenced block is not set, it can lead to the situation where
a block being committed is allocated and used again. As a result, the
following condition will not be satisfied then trigger BUG_ON. Of course,
it is entirely possible to construct a problematic image that can trigger
this BUG_ON through specific operations. In fact, I have constructed such
an image and easily reproduced this issue.
Therefore, J_ASSERT() holds true only under ideal conditions, but it may
not necessarily be satisfied in exceptional scenarios. Using J_ASSERT()
directly in abnormal situations would cause the system to crash, which is
clearly not what we want. So here we directly trigger a JBD abort instead
of immediately invoking BUG_ON.
Fixes: 470decc613 ("[PATCH] jbd2: initial copy of files from jbd")
Signed-off-by: Ye Bin <yebin10@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Message-ID: <20251025072657.307851-1-yebin@huaweicloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
jbd2 journal handling code doesn't want jbd2_might_wait_for_commit()
to be placed between start_this_handle() and stop_this_handle(). So it
marks the region with rwsem_acquire_read() and rwsem_release().
However, the annotation is too strong for that purpose. We don't have
to use more than try lock annotation for that.
rwsem_acquire_read() implies:
1. might be a waiter on contention of the lock.
2. enter to the critical section of the lock.
All we need in here is to act 2, not 1. So trylock version of
annotation is sufficient for that purpose. Now that dept partially
relies on lockdep annotaions, dept interpets rwsem_acquire_read() as a
potential wait and might report a deadlock by the wait.
Replace it with trylock version of annotation.
Signed-off-by: Byungchul Park <byungchul@sk.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: stable@kernel.org
Message-ID: <20251024073940.1063-1-byungchul@sk.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When releasing file system metadata blocks in jbd2_journal_forget(), if
this buffer has not yet been checkpointed, it may have already been
written back, currently be in the process of being written back, or has
not yet written back. jbd2_journal_forget() calls
jbd2_journal_try_remove_checkpoint() to check the buffer's status and
add it to the current transaction if it has not been written back. This
buffer can only be reallocated after the transaction is committed.
jbd2_journal_try_remove_checkpoint() attempts to lock the buffer and
check its dirty status while holding the buffer lock. If the buffer has
already been written back, everything proceeds normally. However, there
are two issues. First, the function returns immediately if the buffer is
locked by the write-back process. It does not wait for the write-back to
complete. Consequently, until the current transaction is committed and
the block is reallocated, there is no guarantee that the I/O will
complete. This means that ongoing I/O could write stale metadata to the
newly allocated block, potentially corrupting data. Second, the function
unlocks the buffer as soon as it detects that the buffer is still dirty.
If a concurrent write-back occurs immediately after this unlocking and
before clear_buffer_dirty() is called in jbd2_journal_forget(), data
corruption can theoretically still occur.
Although these two issues are unlikely to occur in practice since the
undergoing metadata writeback I/O does not take this long to complete,
it's better to explicitly ensure that all ongoing I/O operations are
completed.
Fixes: 597599268e ("jbd2: discard dirty data when forgetting an un-journalled buffer")
Cc: stable@kernel.org
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Message-ID: <20250916093337.3161016-2-yi.zhang@huaweicloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In commit 6a3afb6ac6 ("jbd2: increase the journal IO's priority"),
the priority of IOs initiated by jbd2 has been raised, exempting them
from WBT throttling.
Checkpoint is also a crucial operation of jbd2. While no serious issues
have been observed so far, it should still be reasonable to exempt
checkpoint from WBT throttling.
Signed-off-by: Julian Sun <sunjunchao@bytedance.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Both jbd2_log_do_checkpoint() and jbd2_journal_shrink_checkpoint_list()
periodically release j_list_lock after processing a batch of buffers to
avoid long hold times on the j_list_lock. However, since both functions
contend for j_list_lock, the combined time spent waiting and processing
can be significant.
jbd2_journal_shrink_checkpoint_list() explicitly calls cond_resched() when
need_resched() is true to avoid softlockups during prolonged operations.
But jbd2_log_do_checkpoint() only exits its loop when need_resched() is
true, relying on potentially sleeping functions like __flush_batch() or
wait_on_buffer() to trigger rescheduling. If those functions do not sleep,
the kernel may hit a softlockup.
watchdog: BUG: soft lockup - CPU#3 stuck for 156s! [kworker/u129:2:373]
CPU: 3 PID: 373 Comm: kworker/u129:2 Kdump: loaded Not tainted 6.6.0+ #10
Hardware name: Huawei TaiShan 2280 /BC11SPCD, BIOS 1.27 06/13/2017
Workqueue: writeback wb_workfn (flush-7:2)
pstate: 20000005 (nzCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : native_queued_spin_lock_slowpath+0x358/0x418
lr : jbd2_log_do_checkpoint+0x31c/0x438 [jbd2]
Call trace:
native_queued_spin_lock_slowpath+0x358/0x418
jbd2_log_do_checkpoint+0x31c/0x438 [jbd2]
__jbd2_log_wait_for_space+0xfc/0x2f8 [jbd2]
add_transaction_credits+0x3bc/0x418 [jbd2]
start_this_handle+0xf8/0x560 [jbd2]
jbd2__journal_start+0x118/0x228 [jbd2]
__ext4_journal_start_sb+0x110/0x188 [ext4]
ext4_do_writepages+0x3dc/0x740 [ext4]
ext4_writepages+0xa4/0x190 [ext4]
do_writepages+0x94/0x228
__writeback_single_inode+0x48/0x318
writeback_sb_inodes+0x204/0x590
__writeback_inodes_wb+0x54/0xf8
wb_writeback+0x2cc/0x3d8
wb_do_writeback+0x2e0/0x2f8
wb_workfn+0x80/0x2a8
process_one_work+0x178/0x3e8
worker_thread+0x234/0x3b8
kthread+0xf0/0x108
ret_from_fork+0x10/0x20
So explicitly call cond_resched() in jbd2_log_do_checkpoint() to avoid
softlockup.
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Link: https://patch.msgid.link/20250812063752.912130-1-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
jbd2_journal_blocks_per_page() returns the number of blocks in a single
page. Rename it to jbd2_journal_blocks_per_folio() and make it returns
the number of blocks in the largest folio, preparing for the calculation
of journal credits blocks when allocating blocks within a large folio in
the writeback path.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since handle->h_transaction may be a NULL pointer, so we should change it
to call is_handle_aborted(handle) first before dereferencing it.
And the following data-race was reported in my fuzzer:
==================================================================
BUG: KCSAN: data-race in jbd2_journal_dirty_metadata / jbd2_journal_dirty_metadata
write to 0xffff888011024104 of 4 bytes by task 10881 on cpu 1:
jbd2_journal_dirty_metadata+0x2a5/0x770 fs/jbd2/transaction.c:1556
__ext4_handle_dirty_metadata+0xe7/0x4b0 fs/ext4/ext4_jbd2.c:358
ext4_do_update_inode fs/ext4/inode.c:5220 [inline]
ext4_mark_iloc_dirty+0x32c/0xd50 fs/ext4/inode.c:5869
__ext4_mark_inode_dirty+0xe1/0x450 fs/ext4/inode.c:6074
ext4_dirty_inode+0x98/0xc0 fs/ext4/inode.c:6103
....
read to 0xffff888011024104 of 4 bytes by task 10880 on cpu 0:
jbd2_journal_dirty_metadata+0xf2/0x770 fs/jbd2/transaction.c:1512
__ext4_handle_dirty_metadata+0xe7/0x4b0 fs/ext4/ext4_jbd2.c:358
ext4_do_update_inode fs/ext4/inode.c:5220 [inline]
ext4_mark_iloc_dirty+0x32c/0xd50 fs/ext4/inode.c:5869
__ext4_mark_inode_dirty+0xe1/0x450 fs/ext4/inode.c:6074
ext4_dirty_inode+0x98/0xc0 fs/ext4/inode.c:6103
....
value changed: 0x00000000 -> 0x00000001
==================================================================
This issue is caused by missing data-race annotation for jh->b_modified.
Therefore, the missing annotation needs to be added.
Reported-by: syzbot+de24c3fe3c4091051710@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=de24c3fe3c4091051710
Fixes: 6e06ae88ed ("jbd2: speedup jbd2_journal_dirty_metadata()")
Signed-off-by: Jeongjun Park <aha310510@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250514130855.99010-1-aha310510@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
This patch reworks fast commit's commit path to remove locking the
journal for the entire duration of a fast commit. Instead, we only lock
the journal while marking all the eligible inodes as "committing". This
allows handles to make progress in parallel with the fast commit.
Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://patch.msgid.link/20250508175908.1004880-5-harshadshirwadkar@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
timer_delete[_sync]() replaces del_timer[_sync](). Convert the whole tree
over and remove the historical wrapper inlines.
Conversion was done with coccinelle plus manual fixups where necessary.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When the filesystem performs file or filesystem synchronization (e.g.,
ext4_sync_file()), it queries the journal to determine whether to flush
the file device through jbd2_trans_will_send_data_barrier(). If the
target transaction has not started committing, it assumes that the
journal will submit the flush command, allowing the filesystem to bypass
a redundant flush command. However, this assumption is not always valid.
If the journal is not located on the filesystem device, the journal
commit thread will not submit the flush command unless the variable
->t_need_data_flush is set to 1. Consequently, the flush may be missed,
and data may be lost following a power failure or system crash, even if
the synchronization appears to succeed.
Unfortunately, we cannot determine with certainty whether the target
transaction will flush to the filesystem device before it commits.
However, if it has not started committing, it must be the running
transaction. Therefore, fix it by always set its t_need_data_flush to 1,
ensuring that the committing thread will flush the filesystem device.
Fixes: bbd2be3691 ("jbd2: Add function jbd2_trans_will_send_data_barrier()")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20241206111327.4171337-1-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In __jbd2_journal_erase(), the block_stop parameter includes the last
block of a contiguous region; however, the calculation of byte_stop is
incorrect, as it does not account for the bytes in that last block.
Consequently, the page cache is not cleared properly, which occasionally
causes the ext4/050 test to fail.
Since block_stop operates on inclusion semantics, it involves repeated
increments and decrements by 1, significantly increasing the complexity
of the calculations. Optimize the calculation and fix the incorrect
byte_stop by make both block_stop and byte_stop to use exclusion
semantics.
This fixes a failure in fstests ext4/050.
Fixes: 01d5d96542 ("ext4: add discard/zeroout flags to journal flush")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250217065955.3829229-1-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If a journal is wiped, we will set journal->j_tail to 0. However if
'write' argument is not set (as it happens for read-only device or for
ocfs2), the on-disk superblock is not updated accordingly and thus
jbd2_journal_recover() cat try to recover the wiped journal. Fix the
check in jbd2_journal_recover() to use journal->j_tail for checking
empty journal instead.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250206094657.20865-4-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Pull CRC updates from Eric Biggers:
- Reorganize the architecture-optimized CRC32 and CRC-T10DIF code to be
directly accessible via the library API, instead of requiring the
crypto API. This is much simpler and more efficient.
- Convert some users such as ext4 to use the CRC32 library API instead
of the crypto API. More conversions like this will come later.
- Add a KUnit test that tests and benchmarks multiple CRC variants.
Remove older, less-comprehensive tests that are made redundant by
this.
- Add an entry to MAINTAINERS for the kernel's CRC library code. I'm
volunteering to maintain it. I have additional cleanups and
optimizations planned for future cycles.
* tag 'crc-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux: (31 commits)
MAINTAINERS: add entry for CRC library
powerpc/crc: delete obsolete crc-vpmsum_test.c
lib/crc32test: delete obsolete crc32test.c
lib/crc16_kunit: delete obsolete crc16_kunit.c
lib/crc_kunit.c: add KUnit test suite for CRC library functions
powerpc/crc-t10dif: expose CRC-T10DIF function through lib
arm64/crc-t10dif: expose CRC-T10DIF function through lib
arm/crc-t10dif: expose CRC-T10DIF function through lib
x86/crc-t10dif: expose CRC-T10DIF function through lib
crypto: crct10dif - expose arch-optimized lib function
lib/crc-t10dif: add support for arch overrides
lib/crc-t10dif: stop wrapping the crypto API
scsi: target: iscsi: switch to using the crc32c library
f2fs: switch to using the crc32 library
jbd2: switch to using the crc32c library
ext4: switch to using the crc32c library
lib/crc32: make crc32c() go directly to lib
bcachefs: Explicitly select CRYPTO from BCACHEFS_FS
x86/crc32: expose CRC32 functions through lib
x86/crc32: update prototype for crc32_pclmul_le_16()
...
When committing transaction in jbd2_journal_commit_transaction(), the
disk caches for the filesystem device should be flushed before updating
the journal tail sequence. However, this step is missed if the journal
is not located on the filesystem device. As a result, the filesystem may
become inconsistent following a power failure or system crash. Fix it by
ensuring that the filesystem device is flushed appropriately.
Fixes: 3339578f05 ("jbd2: cleanup journal tail after transaction commit")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://lore.kernel.org/r/20241203014407.805916-3-yi.zhang@huaweicloud.com
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
The code indicates that journal_init_common() fills the journal_t object
it returns while the comment incorrectly states that only a few fields are
initialised. Also, the comment claims that journal structures could be
created from scratch which isn't possible as journal_init_common() calls
journal_load_superblock() which loads and checks journal superblock from
disk.
Signed-off-by: Daniel Martín Gómez <dalme@riseup.net>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20241107144538.3544-1-dalme@riseup.net
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The b_frozen_data allocation should not be failed during journal
committing process, otherwise jbd2 will abort.
Since commit 490c1b444ce653d("jbd2: do not fail journal because of
frozen_buffer allocation failure") already added '__GFP_NOFAIL' flag
in do_get_write_access(), just add '__GFP_NOFAIL' flag for all allocations
in jbd2_journal_write_metadata_buffer(), like 'new_bh' allocation does.
Besides, remove all error handling branches for do_get_write_access().
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20241012085530.2147846-1-chengzhihao@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
