When orphan_file feature is enabled, ext4_orphan_cleanup() was always
walking through the orphan file looking for orphan inodes. This is
mostly harmless but for read-only filesystem it results in spurious
"orphan cleanup on readonly fs" message and in other cornercases it
could result in similar somewhat misleading messages. Skip orphan
cleanup if the orphan file is empty to avoid confusing messages.
Fixes: 02f310fcf4 ("ext4: Speedup ext4 orphan inode handling")
Reported-by: Tigran Aivazian <aivazian.tigran@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Baokun Li <libaokun@linux.alibaba.com>
Link: https://patch.msgid.link/20260803160037.64285-2-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_xattr_block_set() retries when a cache entry selected for reuse
has a saturated reference count after taking the buffer lock. The retry
returns to the mbcache lookup without making that entry ineligible, so
it can select the same unusable entry indefinitely. A task spinning
there can hold the parent directory's i_rwsem and leave concurrent
rmdir callers blocked.
Normally a reusable entry has a reference count below
EXT4_XATTR_REFCOUNT_MAX because the count and MBE_REUSABLE_B are
updated under the same buffer lock. A corrupted filesystem can violate
that invariant. The syzbot reproducer reports allocator and xattr
corruption before triggering this retry loop.
Check the untrusted on-disk count before incrementing it, avoiding
overflow, and clear MBE_REUSABLE_B when it is already saturated. The
next lookup then skips the entry that was just proven unusable. This
mirrors the normal transition at EXT4_XATTR_REFCOUNT_MAX; the release
path marks the entry reusable again on the exact 1024-to-1023
transition.
Using the same QEMU harness and guest parameters, current unpatched
Linux hung in 6 of 8 420-second trials with the do_rmdir signature;
representative NMI backtraces caught the owner spinning in
ext4_xattr_block_set(). The patched kernel completed 28 of 28 trials
without a hung-task report; the final twelve trials exercised the
reviewed overflow-safe form of the change. syzbot's patch testing also
completed without reproducing the hang.
Reported-and-tested-by: syzbot+e68dbebd9617a9250e8d@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=e68dbebd9617a9250e8d
Fixes: 65f8b80053 ("ext4: fix race when reusing xattr blocks")
Cc: stable@vger.kernel.org
Signed-off-by: Matthias Goergens <matthias.goergens@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reported-by: syzbot+e68dbebd9617a9250e8d@syzkaller.appspotmail.com
Tested-by: syzbot+e68dbebd9617a9250e8d@syzkaller.appspotmail.com
Link: https://patch.msgid.link/20260802065941.1726052-1-matthias.goergens@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, when a new encrypted regular file is created, the call to
ext4_set_inode_flags(inode, init=true) in __ext4_new_inode() is made
before EXT4_INODE_ENCRYPT is set. As a result, it can set S_DAX if the
filesystem is mounted with "-o dax=always".
EXT4_INODE_ENCRYPT then actually gets set a bit later in
__ext4_new_inode(), when it calls fscrypt_set_context() which calls
ext4_set_context(). ext4_set_context() sets EXT4_INODE_ENCRYPT and
calls ext4_set_inode_flags(inode, init=false) to set S_ENCRYPTED too.
This was intended to clear S_DAX as well. However, this was broken by
commit 043546e46d ("fs/ext4: Only change S_DAX on inode load"). This
causes data written to the file to bypass encryption, also causing
xfstests failures such as generic/548 (when "-o dax=always" is used).
Fix this by simplifying the flow by making __ext4_new_inode() set
EXT4_INODE_ENCRYPT earlier. This makes it take effect in
ext4_set_inode_flags(inode, init=true), making S_DAX never be set.
Similarly, make EXT4_STATE_MAY_INLINE_DATA never be set in the first
place on new encrypted inodes. Then it doesn't need to be cleared.
As a result of these simplifications, ext4_set_context() no longer needs
to change inode flags or state when 'handle != NULL'. Remove that too.
Reported-by: Disha Goel <disgoel@linux.ibm.com>
Reported-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Closes: https://lore.kernel.org/r/20260723085648.1500357-1-ojaswin@linux.ibm.com
Fixes: 043546e46d ("fs/ext4: Only change S_DAX on inode load")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Tested-by: Disha Goel <disgoel@linux.ibm.com>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260730175212.36923-1-ebiggers@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_alloc_file_blocks(), the WRITE_ZEROES path converts unwritten
extents to written in one transaction, while i_disksize is updated to
cover them only in a later transaction. A crash in between leaves
written extents beyond i_disksize on disk, which fsck will complain
about.
To fix this, add the inode to the orphan list in the same handle that
does the conversion, and remove it once i_disksize has caught up.
Also add a sanity check to ensure conversion does not extend beyond EOF.
Since ext4_alloc_file_blocks() is called from the fallocate() path,
partial allocation is safe. On partial conversion failure, advance
i_disksize only up to the boundary of successfully converted blocks, so
that orphan cleanup sees a consistent state. Document this behavior in
the function comment.
Reported-by: Jan Kara <jack@suse.cz>
Closes: https://lore.kernel.org/linux-ext4/3f6ao5amv7glbgigndtegcucgo3n34ij3lau6l3da3hgdxgn3v@ev66wv3r5umt/
Fixes: f4265b8d32 ("ext4: add FALLOC_FL_WRITE_ZEROES support")
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260729085918.3336221-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_convert_unwritten_extents() currently returns only a success or a
failure indication. A zero return means all requested blocks were
converted, and a negative value means the conversion failed. However,
some blocks may have already been converted when the function fails
partway through, and callers have no way to learn how many were done.
The WRITE_ZEROES caller in ext4_alloc_file_blocks() needs this
information to decide whether to add the inode to the orphan list before
updating i_disksize to cover the already-converted written extents, so
that a crash before i_disksize catches up can be recovered via orphan
truncation.
Switch the function to pass out the number of converted blocks through
the new output parameter @converted, which will be used by later
patches.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260729085918.3336221-2-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Validate EA inode state in ext4_xattr_inode_iget() to reject corrupted
EA inodes early, before they trigger WARN_ONCE in
ext4_xattr_inode_update_ref().
When a corrupted ext4 image has an EA inode with inconsistent i_nlink
and ref_count values (e.g. i_nlink=65535), the code currently allows it
through and later hits WARN_ONCE when ref_count transitions cross the
0/1 boundary. This is better handled as an early sanity check that
returns -EFSCORRUPTED, consistent with how ext4 treats other on-disk
corruption.
Since ext4_xattr_inode_iget() resolves references from active xattr
entries, the target EA inode must be in active state (i_nlink=1,
ref_count>0). Reject any inode that does not satisfy this.
Reported-by: syzbot+76916a45d2294b551fd9@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=76916a45d2294b551fd9
Fixes: dec214d00e ("ext4: xattr inode deduplication")
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260724100255.144768-1-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This makes starting handles a little more efficient, since it avoids
requiring bitshifts when setting or getting the h_type and h_line_no
fields in the jbd2_journal_handle structure.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The jbd2 layer enables NOFS mode using memalloc_nofs_{save,restore}()
while a handle is active. We need to do the same in nojournal mode so
that it is safe to remove GFP_NOFS flags while a jbd2 handle is
active.
This will require that we actually allocate a real handle, but with an
h_invalid flag set, so there is a place to put the saved memalloc
context.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
FALLOC_FL_WRITE_ZEROES requires that all blocks in the requested range
end up as written extents with zeroed content. For unaligned edges that
were partial-zeroed in dirty unwritten or delalloc state, the buffer
is left dirty while the underlying extent may not yet be converted to
written. As a result, a subsequent SYNC write to this range would still
trigger metadata changes, which violates the semantics of WRITE_ZEROES.
Fix this by calling filemap_write_and_wait_range() for partial-zeroed
edges to flush out the zeroed data and ensure the extent conversion
is complete.
Fixes: f4265b8d32 ("ext4: add FALLOC_FL_WRITE_ZEROES support")
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-9-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
FALLOC_FL_WRITE_ZEROES requires that all blocks in the requested range
end up as written extents with zeroed content. For unaligned edges that
were already allocated, ext4_zero_partial_blocks() zeros them directly.
However, for unaligned edges whose underlying extent is a clean
unwritten extent or a hole, the extent type remains unwritten after
partial zeroing, which does not align with the semantics of
WRITE_ZEROES.
Therefore, when ext4_zero_partial_blocks() skips partial zeroing, it
indicates that the corresponding edges are clean unwritten extents or
holes. In this case, we need to expand the aligned allocation range
outward to cover such edges, so that ext4_alloc_file_blocks() can
correctly allocate blocks for the unaligned range. Edges that were
partial-zeroed (i.e., written or dirty) are left untouched.
Fixes: f4265b8d32 ("ext4: add FALLOC_FL_WRITE_ZEROES support")
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Replace the single bool did_zero output of ext4_zero_partial_blocks()
with a bitmask that records which edge (start, end, or both in the
single-block case) was actually partial-zeroed. This allows callers to
distinguish which edges have been zeroed, preparing for unaligned
FALLOC_FL_WRITE_ZEROES handling in later patches.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-7-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_load_tail_bh() returns NULL for both holes and clean unwritten
buffers, but the conditions that lead to this are not obvious from the
code alone. Document this behavior to clarify the return value, so that
readers do not mistakenly assume that only holes result in a NULL
return.
Also update the inline comment following the ext4_get_block() call to
reflect this, and note that a lookup-only get_block (without
EXT4_GET_BLOCKS_CREATE) never sets BH_Mapped for clean unwritten
extents, which is why a clean unwritten bh falls through to the
"nothing to do" path.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_zero_range(), move the ext4_zero_partial_blocks() call, which
handles unaligned edges, into the same branch where the unaligned range
is preallocated, immediately after ext4_alloc_file_blocks(). This is
safe because there is no dependency between partial block handling and
the subsequent full block handling.
This change will be used by later patches that handle unaligned
FALLOC_FL_WRITE_ZEROES operations, which will need to check the partial
zeroed result.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_load_tail_bh() ignores the return value of ext4_get_block(), so an
I/O or allocation failure is silently discarded. buffer_mapped(bh) stays
false and the function returns NULL, which callers such as
ext4_block_do_zero_range() treat as "nothing to do" and return success.
This can mask real failures during zero-range, truncate, or punch-hole
operations, potentially exposing stale data if the block was not
actually a hole and needed zeroing. So propagate the error to the
callers.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_block_zero_eof() is called from ext4_write_checks() on every
append write beyond EOF. For inline data files, ext4_get_block()
returns -ERANGE when ext4_load_tail_bh() looks up the tail block.
However, this error is currently ignored because the return value
of ext4_get_block() in ext4_load_tail_bh() is discarded.
Before we fix ext4_load_tail_bh() to properly propagate the error,
skip the zeroing for inline data inodes to avoid unnecessary
failures or confusion.
Fixes: 3f60efd654 ("ext4: zero post-EOF partial block before appending write")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_load_tail_bh() returns a locked folio that callers immediately
mutate through folio_zero_range() and mark_buffer_dirty(). Use
FGP_WRITEBEGIN so that, on backing devices that require stable writes,
__filemap_get_folio() waits for writeback to finish before returning
the folio; on regular devices the wait is a no-op.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260714080044.4038124-2-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The jbd2 shrinker currently accounts only checkpoint buffers that it
successfully releases against nr_to_scan. Busy buffers therefore do not
consume the scan budget.
If a checkpoint transaction contains mostly busy buffers, the shrinker
can scan its entire checkpoint list while holding journal->j_list_lock.
Large checkpoint lists can result in excessive lock hold times and leave
other CPUs spinning on j_list_lock, causing soft lockups or RCU stalls.
Pass nr_to_scan into journal_shrink_one_cp_list() and decrement it for
every buffer examined, including busy buffers. Pass NULL from checkpoint
cleanup paths so their existing full-list behavior is preserved.
This restores the scan-budget semantics that existed before
journal_shrink_one_cp_list() was changed to always scan a complete
checkpoint list.
Fixes: b98dba273a ("jbd2: remove journal_clean_one_cp_list()")
Cc: stable@vger.kernel.org
Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260713102229.1598812-3-max.kellermann@ionos.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
journal_shrink_one_cp_list() skips busy checkpoint buffers when called
with JBD2_SHRINK_BUSY_SKIP. The continue statement on this path also
skips the need_resched() check at the end of the loop body.
Consequently, when a checkpoint list contains mostly busy buffers, the
shrinker can walk the entire list while holding journal->j_list_lock,
even when a reschedule has been requested. Large checkpoint lists under
memory pressure can therefore cause long lock hold times and leave other
CPUs spinning on j_list_lock, resulting in soft lockups or RCU stalls.
Route the busy-buffer path through the need_resched() check so that the
shrinker can release j_list_lock and reschedule promptly, restoring
parity with the clean-buffer path, which already checks need_resched().
This does not change which checkpoint buffers are eligible for removal.
Fixes: b98dba273a ("jbd2: remove journal_clean_one_cp_list()")
Cc: stable@vger.kernel.org
Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260713102229.1598812-2-max.kellermann@ionos.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Add a helper that drops an inode reference only if the caller does not
hold the last one. Returns true if the reference was dropped, false
otherwise.
This is useful for filesystems that need to release inode references
in contexts where triggering final iput (and thus eviction) would be
unsafe due to lock ordering constraints. The caller can check the
return value and defer the final iput to a safe context.
Unlike iput_not_last() which BUG_ON's if called with the last ref,
this variant is designed to be called speculatively.
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Suggested-by: Jan Kara <jack@suse.cz>
Suggested-by: Mateusz Guzik <mjguzik@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christian Brauner (Amutable) <brauner@kernel.org>
Tested-by: syzbot@syzkaller.appspotmail.com
Link: https://patch.msgid.link/20260710030851.2791589-2-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Now that ext4_put_ea_inode() handles deferred iput safely for all cases
(using iput_if_not_last + embedded llist_node), the ea_inode_array
mechanism for batching deferred iputs is redundant.
Remove:
- ext4_expand_inode_array() and ext4_xattr_inode_array_free()
- struct ext4_xattr_inode_array and EIA_INCR/EIA_MASK defines
- ea_inode_array parameter from ext4_xattr_inode_dec_ref_all(),
ext4_xattr_release_block(), and ext4_xattr_delete_inode()
- ea_inode_array variable from ext4_evict_inode()
Instead, ext4_xattr_inode_dec_ref_all() now calls ext4_put_ea_inode()
directly after processing each EA inode. This simplifies the code
by eliminating multi-layer parameter threading and removes the need
for callers to manage array lifetime.
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: syzbot@syzkaller.appspotmail.com
Link: https://patch.msgid.link/20260710030851.2791589-5-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Convert all iput() calls on EA inodes in xattr code paths to use
ext4_put_ea_inode(). This establishes a uniform rule: every EA inode
reference release in ext4 xattr code goes through ext4_put_ea_inode(),
eliminating the need to analyze each call site individually for lock
safety.
Converted sites:
- ext4_xattr_inode_get() read path
- ext4_xattr_inode_inc_ref_all() main loop and cleanup path
- ext4_xattr_inode_dec_ref_all() error paths
- ext4_xattr_inode_create() error path
- ext4_xattr_inode_cache_find() mismatch path
- ext4_xattr_inode_lookup_create() out_err
- ext4_xattr_set_entry() old_ea_inode
- ext4_xattr_block_set() new block path, cleanup, and tmp_inode
- ext4_xattr_ibody_set() error and success paths
- ext4_xattr_delete_inode() quota loop
For most of these, iput_if_not_last() will succeed (the EA inode has
other references) making the overhead a single atomic operation.
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: syzbot@syzkaller.appspotmail.com
Link: https://patch.msgid.link/20260710030851.2791589-4-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Calling iput() on EA inodes while holding xattr_sem or a jbd2 handle
can trigger write_inode_now() -> ext4_writepages() -> s_writepages_rwsem,
creating a lock ordering issue during mount (!SB_ACTIVE).
Add ext4_put_ea_inode() which uses iput_if_not_last() as a fast path.
If this is not the last reference, it is dropped immediately. If this
is the last reference, the inode is linked onto a per-sb lock-free llist
via i_ea_iput_node (embedded in ext4_inode_info, sharing space with the
unused xattr_sem of EA inodes via a union) and a delayed worker
(1 jiffie) performs the final iput() in a clean context. This avoids
per-iput memory allocation.
Flush points ensure all pending EA inode evictions complete before
dependent resources become unavailable:
- ext4_put_super / failed_mount9: before quota shutdown
- failed_mount_wq: before freeing xattr caches
- failed_mount3a: before freeing shrinker (journal replay case)
- ext4_sync_fs: before remount-ro, freeze, or sync completes
Initialization is placed before journal loading since fast commit
replay may trigger evictions that call ext4_put_ea_inode().
Also moves init_rwsem(xattr_sem) from init_once to ext4_alloc_inode to
handle slab object reuse after the union field has been overwritten.
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260710030851.2791589-3-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For casefolded encrypted directories ext4 stores an 8-byte hash trailer
after the name (EXT4_DIRENT_HASHES()), at an offset derived from
de->name_len. On the sb_no_casefold_compat_fallback() path ext4_match()
reads that trailer, but ext4_search_dir()'s by-hand pre-check only tests
de->name + de->name_len <= dlimit, which proves the name fits, not the
rounded trailer. A crafted entry whose name ends at the block boundary
passes the check while EXT4_DIRENT_HASHES(de) lands past the block end,
so ext4_match() reads out of bounds on an ordinary lookup. KASAN reports
it as a use-after-free when the page after the directory block holds a
freed object:
BUG: KASAN: use-after-free in ext4_match (fs/ext4/namei.c:1435)
Read of size 4 at addr ffff888010458000 by task exploit
Call Trace:
ext4_match (fs/ext4/namei.c:1435)
ext4_search_dir (fs/ext4/namei.c:1470)
__ext4_find_entry (fs/ext4/namei.c:1268 fs/ext4/namei.c:1632)
ext4_lookup (fs/ext4/namei.c:1703 fs/ext4/namei.c:1769)
...
filename_lookup (fs/namei.c:2842)
vfs_statx (fs/stat.c:353)
__do_sys_newfstatat (fs/stat.c:538)
do_syscall_64 (arch/x86/entry/syscall_64.c:94)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
Require, for hash-in-dirent directories, that the whole entry including
the rounded trailer fits before calling ext4_match(). This is the same
bound ext4_check_dir_entry() already enforces via ext4_dir_rec_len(), so
no well-formed entry is rejected. The other caller, ext4_find_dest_de(),
runs ext4_check_dir_entry() first and is unaffected.
Fixes: 471fbbea7f ("ext4: handle casefolding with encryption")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260709184101.441348-1-xmei5@asu.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_init_orphan_info() reads orphan file blocks with ext4_bread()
and stores the returned buffer_head in oi->of_binfo[i].ob_bh.
If ext4_bread() succeeds but the orphan block magic or checksum
validation fails, the function jumps to out_free. However, the old
out_free loop starts releasing buffers from i - 1, so the current
buffer_head at index i is skipped.
This leaks the buffer_head reference obtained by ext4_bread() on the
bad magic and bad checksum error paths.
Fix this by tracking the number of successfully read buffer_heads and
releasing exactly those buffer_heads on the error path.
Fixes: 02f310fcf4 ("ext4: Speedup ext4 orphan inode handling")
Signed-off-by: Guanghui Yang <3497809730@qq.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/tencent_B38798612A159E21450ECF959016371B0807@qq.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When ext4_xattr_make_inode_space() returns -ENOSPC,
ext4_expand_extra_isize_ea() can retry the expansion with
s_min_extra_isize. If that retry succeeds by finding enough ibody free
space, control jumps directly to the shift label.
The previous -ENOSPC is still stored in error in that path, so the
function can update i_extra_isize but still return -ENOSPC to the
caller. Clear error before retrying so a successful fallback expansion
returns success.
Reproduced with an ext4 image using 1 KiB blocks, project quota support,
256-byte inodes, and min_extra_isize/want_extra_isize set to 32.
FS_IOC_FSSETXATTR failures dropped from 802 to 86 after the fix.
Fixes: 69f3a3039b ("ext4: introduce ITAIL helper")
Cc: stable@vger.kernel.org
Signed-off-by: Guanghui Yang <3497809730@qq.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/tencent_192F8A699EFD21126E02101131C9546F3C08@qq.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_fc_replay() stops replaying fast commit tags only when a tag
handler returns a negative error. However, ext4_fc_replay_add_range()
and ext4_fc_replay_del_range() currently return 0 from their common
exit paths even after internal failures.
This hides errors from ext4_fc_record_modified_inode(),
ext4_map_blocks(), ext4_find_extent(), ext4_ext_insert_extent(),
ext4_ext_replay_update_ex(), and ext4_ext_remove_space(). As a result,
a failed ADD_RANGE or DEL_RANGE replay can be treated as successful and
the replay code may continue with subsequent fast commit tags.
This is particularly problematic for DEL_RANGE because it may already
have marked blocks as free before ext4_ext_remove_space() fails. If the
error is swallowed, replay may continue from a partially applied range
operation.
Return the saved error from the common exit paths and make the
ERR_PTR() cases in ADD_RANGE store PTR_ERR() before jumping to out.
Fixes: 8016e29f43 ("ext4: fast commit recovery path")
Cc: stable@vger.kernel.org
Signed-off-by: Guanghui Yang <3497809730@qq.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/tencent_E3622146846A84C75C31C7D32AC4D5AD0605@qq.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
A corrupted directory can trigger the following KASAN report when
ext4_readdir() resumes from an invalid position:
BUG: KASAN: use-after-free in __ext4_check_dir_entry+0x5ef/0x820
Read of size 2 at addr ffff88810a646000 by task repro_linear/509
Call Trace:
<TASK>
dump_stack_lvl+0x53/0x70
print_report+0xd0/0x630
kasan_report+0xce/0x100
__ext4_check_dir_entry+0x5ef/0x820
ext4_readdir+0xcde/0x2b70
iterate_dir+0x1a1/0x520
__x64_sys_getdents64+0x12b/0x220
do_syscall_64+0xf9/0x540
entry_SYSCALL_64_after_hwframe+0x77/0x7f
</TASK>
KASAN reports use-after-free because the out-of-bounds access lands in an
adjacent freed page. The directory buffer itself is still referenced.
ext4_dir_llseek() invalidates the directory cookie so that ext4_readdir()
rescans directory entries from the start of the block. The rescan checks
only the lower bound of rec_len before advancing. A corrupted rec_len can
therefore place the offset where the block has insufficient space for a
complete directory entry. The rescan itself may dereference that truncated
entry, or the main loop may pass it to __ext4_check_dir_entry(). The latter
reads de->rec_len before validating the range. For example:
block offset 0 4092 4096
|---- de1.rec_len = 4092 -----|----|
de2.inode
| de2.rec_len
^ OOB, reported as UAF
de2 starts at offset 4092 in this 4 KiB block. Its four-byte inode fits in
the block, but its rec_len starts at offset 4096 and crosses the boundary.
The minimum safe length is inode-dependent. Encrypted and casefolded
directory entries need eight additional hash bytes, while a valid metadata
checksum tail is only 12 bytes.
Cache the metadata checksum feature state and derive the minimum directory
entry length from the on-disk format. Use it to bound both the rescan and
the offset passed to the main loop. Report an offset in a truncated block
tail and skip the remainder of the block, while continuing to accept an
offset exactly at the block boundary.
Reported-by: syzbot+5322c5c260eb44d209ed@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=5322c5c260eb44d209ed
Fixes: ac27a0ec11 ("[PATCH] ext4: initial copy of files from ext3")
Signed-off-by: Yao Kai <yaokai34@huawei.com>
Reviewed-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/20260706041313.708346-1-yaokai34@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After implementing bitwise flags for tracking the inline data write
state in the address space fsdata parameter, the CONVERT_INLINE_DATA
state flag is left unused and can be removed.
Perform this clean-up by:
1) Deleting the CONVERT_INLINE_DATA definition from ext4.h.
2) Removing the void **fsdata argument from both the forward
declaration and the definition of the internal helper
ext4_da_convert_inline_data_to_extent().
3) Removing the void **fsdata argument from the declaration and
definition of ext4_generic_write_inline_data() and updating
the caller ext4_try_to_write_inline_data() and the internal
re-alloc retry logic accordingly.
4) Updating ext4_da_write_begin() to call
ext4_generic_write_inline_data() without the fsdata parameter.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Aditya Prakash Srivastava <aditya.ansh182@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260703045414.1768-2-aditya.ansh182@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Instead of checking the live inode state (ext4_has_inline_data(inode)
and ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) in the
write_end handlers, use the fsdata parameter of the address space
operations to explicitly pass down the state in which write_begin
prepared the write.
A concurrent thread (such as ext4_page_mkwrite()) can convert the
inline data to an extent between write_begin and write_end. If this
happens, the write_end handlers would previously miss the inline
write_end path and fall through to extent-based write_end logic.
However, since block buffers were never allocated in write_begin,
this resulted in NULL pointer dereferences or data loss because
folio_buffers(folio) was NULL.
Define EXT4_WRITE_DATA_INLINE (4) as a bit flag (Bit 2), treating
fsdata as bitwise flags rather than mutually exclusive enums to keep
states of the write path independent. Communicate this state via
fsdata:
1) ext4_write_begin() and ext4_da_write_begin() set the
EXT4_WRITE_DATA_INLINE bit in *fsdata via bitwise OR when an inline
write is successfully prepared.
2) On entry, ext4_write_begin() clears the EXT4_WRITE_DATA_INLINE bit
to safely handle VFS retries (where generic_perform_write() bypasses
the fsdata initialization on its retry jump).
3) The write_end handlers perform a bitwise AND to check if the
EXT4_WRITE_DATA_INLINE bit is set and invoke the inline write_end
helper accordingly.
Furthermore, during a buffered write, ext4_write_inline_data_end()
acquires the xattr lock after preparing the write. If a concurrent
page fault (ext4_page_mkwrite()) converts the inline data to an extent
after the write_end handlers check the state but before
ext4_write_inline_data_end() acquires the xattr write lock, the
subsequent check will trigger a kernel panic via
BUG_ON(!ext4_has_inline_data(inode)).
To keep git history working and bisectability clean, replace the
BUG_ON check in ext4_write_inline_data_end() with a graceful error-
handling retry path in this same commit. If the inline data is cleared
after locking the xattr, we safely release all resources (releasing
iloc.bh, unlocking/putting the folio, stopping the active journal
transaction handle) and return 0 (VFS retry) to let the generic write
path retry the operation safely.
Reported-by: syzbot+0c89d865531d053abb2d@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=0c89d865531d053abb2d
Fixes: 3fdcfb668f ("ext4: add journalled write support for inline data")
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Aditya Prakash Srivastava <aditya.ansh182@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260703045414.1768-1-aditya.ansh182@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When a DAX write starts before EOF but extends past i_disksize,
ext4_write_checks() skips the IOCB_NOWAIT check because
iocb->ki_pos <= old_size. However, ext4_dax_write_iter() later calls
ext4_journal_start() to prepare for inode extension, which can sleep
waiting for journal space or transaction commit.
This violates NOWAIT semantics and can stall asynchronous I/O frameworks
like io_uring that rely on non-blocking behavior.
Fix this by checking IOCB_NOWAIT before calling ext4_journal_start()
in the extending write path. If NOWAIT is set and extension is needed,
return -EAGAIN so the caller can retry in blocking context.
Example scenario:
- File: i_size = 1000, i_disksize = 1000
- DAX NOWAIT write: offset = 500, count = 2000
- ext4_write_checks(): ki_pos (500) <= old_size (1000), skip NOWAIT check
- ext4_dax_write_iter(): offset + count (2500) > i_disksize (1000)
- ext4_journal_start() → may sleep → violates NOWAIT
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260618125735.4156639-1-libaokun@linux.alibaba.com?part=5
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20260629113827.4074335-7-libaokun@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
file_modified() passes flags=0 which drops IOCB_NOWAIT, causing
file_update_time() to sleep in ext4_journal_start() via
ext4_dirty_inode() even in non-blocking contexts.
kiocb_modified(iocb) propagates iocb->ki_flags so that
generic_update_time() correctly returns -EAGAIN when IOCB_NOWAIT
is set and ->dirty_inode could block, matching the behavior
already adopted by XFS, FUSE, and ext2.
Affected paths:
- ext4_dio_write_checks(): DIO NOWAIT write
- ext4_write_checks(): shared by buffered (rejects NOWAIT upfront)
and DAX write (supports NOWAIT)
ext4_fallocate() in extents.c is not affected as it has no kiocb.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
Link: https://patch.msgid.link/20260629113827.4074335-6-libaokun@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For unaligned DIO writes, the previous ext4_overwrite_io() required the
entire range to fall within a single written extent. This was overly
conservative: the DIO layer only performs partial block zeroing for the
head and tail blocks when they are partially covered by the write.
Middle blocks that are fully covered are written as whole blocks
without any zeroing, so they are safe regardless of extent state.
Therefore exclusive lock is only required when partial block zeroing
will actually happen:
- The head partial block (if any) lands on a hole or unwritten extent.
- The tail partial block (if any) lands on a hole or unwritten extent.
Middle full-cover blocks can be in any state (hole, unwritten, or
written) - block allocation under shared lock is safe per the previous
patch's analysis (inode_dio_begin + i_data_sem protection).
Replace ext4_overwrite_io() with ext4_dio_needs_zeroing(), which
directly answers the question driving the lock decision. It uses at
most two ext4_map_blocks() calls: one for the head partial block (also
catching the case where it spans through the tail), and one for the
tail partial block if not already covered.
This enables shared lock for previously-rejected scenarios such as:
- Unaligned write spanning written extent + mid-range hole + written
extent at the tail.
- Unaligned write where the partial blocks land on written extents but
the middle has unwritten extents.
Performance:
Hardware: /dev/sda (rotational disk, ~1 GB/s sustained write)
Filesystem: ext4 default mkfs
Unaligned DIO writes (14336 bytes at +512 within each 16K stripe).
Each stripe is laid out as [written][unwritten][unwritten][written],
so the head and tail partial blocks land on written extents but the
middle is unwritten. Metric: IOPS.
JOBS Before After speedup
---- -------- --------- -------
1 15,547 17,381 1.12x
2 15,910 34,172 2.15x
4 15,014 57,567 3.83x
8 15,022 81,947 5.46x
16 14,586 99,126 6.80x
32 14,047 92,519 6.59x
Wall time at JOBS=32: 149.3s (Before) -> 22.7s (After), 6.58x faster.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
Link: https://patch.msgid.link/20260629113827.4074335-5-libaokun@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, ext4_dio_write_checks() calls ext4_overwrite_io() to
determine if a write is a pure overwrite, and upgrades to exclusive
i_rwsem if not. However, ext4_overwrite_io() uses a single
ext4_map_blocks() call which only returns the first contiguous extent of
the same type. A write spanning multiple pre-allocated extents (e.g.
written + unwritten, or two physically discontiguous written extents)
produces a false negative, forcing an unnecessary exclusive lock upgrade.
After commit 5d87c7fca2 ("ext4: avoid starting handle when dio
writing an unwritten extent") and commit 012924f0ee ("ext4: remove
useless ext4_iomap_overwrite_ops"), ext4_iomap_begin()'s fast path
accepts both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN without starting a
journal transaction. The iomap iteration naturally handles multi-extent
ranges: each call returns the mapping for the current segment, and
unwritten-to-written conversion is deferred to ext4_dio_write_end_io().
This means the common case of mixed written/unwritten extents never
reaches ext4_iomap_alloc() at all.
Even for the less common case where the range contains a hole and
ext4_iomap_alloc() is needed, exclusive i_rwsem is still unnecessary for
aligned non-extending writes:
- truncate/punch_hole are kept out: they require exclusive i_rwsem
(blocked by our shared lock during allocation), and inode_dio_begin()
keeps their inode_dio_wait() blocked until in-flight bios complete.
- i_data_sem write-lock inside ext4_map_blocks() serializes concurrent
extent tree modifications (parallel writers to the same hole).
- The journal handle is per-thread and does not require i_rwsem
exclusion.
- i_disksize and orphan list are not involved in non-extending writes.
Skip the ext4_overwrite_io() check entirely for aligned writes by
initializing overwrite to true and only calling ext4_overwrite_io() for
unaligned writes. Unaligned writes still need the extent state check
because concurrent partial block zeroing in the DIO layer requires
exclusive serialization unless the range is a pure written-extent
overwrite.
Performance:
Hardware: /dev/sda (rotational disk, ~1 GB/s sustained write)
Filesystem: ext4 default mkfs
Aligned 8K DIO writes spanning written+unwritten extent boundaries.
Each thread writes its own 1G region sequentially; the file is rebuilt
between runs so every block is written exactly once. Metric: IOPS.
JOBS Before After speedup
---- -------- --------- -------
1 42,322 43,329 1.02x
2 68,516 70,677 1.03x
4 62,489 97,072 1.55x
8 58,701 110,819 1.89x
16 58,569 116,392 1.99x
32 60,860 117,244 1.93x
Wall time at JOBS=32: 69.2s (Before) -> 35.4s (After), 1.96x faster.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
Link: https://patch.msgid.link/20260629113827.4074335-4-libaokun@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
generic/746 started failing intermittently on ext3 (no-extent inodes).
The test triggers 'Page cache invalidation failure on direct I/O'
warnings and subsequent fsync returns -EIO. Adding a 50ms delay
between ext4_buffered_write_iter() and filemap_write_and_wait_range()
in ext4_dio_write_iter() makes the race almost always reproducible.
On no-extent inodes, DIO writes to holes cannot use unwritten extents,
so ext4_iomap_alloc() leaves m_flags=0 and ext4_map_blocks() returns 0.
The iomap layer then returns -ENOTBLK, causing fallback to buffered I/O.
The fallback path in ext4_dio_write_iter() calls
ext4_buffered_write_iter() which dirties pages, then does flush and
invalidate. However, there's an unprotected window between
ext4_buffered_write_iter() returning (with inode lock released) and
the subsequent flush+invalidate.
Concurrent async DIO completions from other threads can run
kiocb_invalidate_post_direct_write() during this window. If pages have
been re-dirtied, post-invalidation finds dirty pages and triggers the
warning, setting -EIO in the error sequence.
Consider a file with two 4k extents: [hole][written]. Thread A does
DIO to the written extent, while thread B does DIO spanning both:
kworker A (4k DIO, allocated block) kworker B (8k DIO, fallback)
----------------------------------- ----------------------------
inode_lock_shared() inode_lock_shared()
iomap_dio_rw(): iomap_dio_rw():
kiocb_invalidate_pages -> clean iomap_begin -> -ENOTBLK
submit_bio (async) dio->size = 0
inode_unlock_shared() inode_unlock_shared()
[bio pending in block layer] /* fallback: lock released */
ext4_buffered_write_iter()
inode_lock(exclusive)
generic_perform_write()
-> dirty pages [0, 8k]
inode_unlock(exclusive)
/* pages dirty, no lock */
[bio completes] filemap_write_and_wait_range()
iomap_dio_complete() -> flush dirty pages
kiocb_invalidate_post_direct_write() invalidate_mapping_pages()
invalidate_inode_pages2_range()
-> finds dirty page!
-> dio_warn_stale_pagecache()
-> errseq_set(-EIO)
This issue can be triggered through normal I/O paths, not just
intentionally overlapping DIO writes from userspace. For example,
generic/746 uses a loop device where multiple kworkers issue concurrent
I/O to the backing file. Additionally, when block_size < folio_size,
non-overlapping DIO writes that share a large folio can also trigger
the race.
Add inode_dio_wait() in ext4_buffered_write_iter() before
ext4_write_checks() to drain all in-flight DIO. This ensures that
all DIO clears existing pages before submitting IO (via
kiocb_invalidate_pages()), all BIO waits for all DIO to complete
(via inode_dio_wait()), and ext4_write_checks() observes the inode
size after all completed DIO so that ext4_block_zero_eof() does not
race with in-flight DIO, thus eliminating the race.
Fixes: 378f32bab3 ("ext4: introduce direct I/O write using iomap infrastructure")
Suggested-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/d1adcf7c-c276-458d-9cac-68a4410f7626@gmail.com
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
Link: https://patch.msgid.link/20260629113827.4074335-3-libaokun@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In data=journal mode, the writeback thread can hit the
WARN_ON_ONCE(sb_rdonly(sb)) in ext4_journal_check_start() while the
superblock is being remounted read-only during reboot:
Workqueue: writeback wb_workfn (flush-253:0)
RIP: 0010:ext4_journal_check_start+0x8b/0xd0
Call Trace:
__ext4_journal_start_sb+0x3c/0x1e0
mpage_prepare_extent_to_map+0x4af/0x580
ext4_do_writepages+0x3c0/0x1080
ext4_writepages+0xc8/0x1a0
do_writepages+0xc4/0x180
__writeback_single_inode+0x45/0x2f0
writeback_sb_inodes+0x26b/0x5d0
__writeback_inodes_wb+0x54/0x100
wb_writeback+0x1ac/0x320
wb_workfn+0x394/0x470
And followed by the warning:
EXT4-fs warning (device vda1): ext4_evict_inode:195: inode #6263:
comm (sd-umount): data will be lost
This issue is not reproduced every time, but frequently.
The reproduction step is to create a VM with 8 CPUs, 16G memory and
setup data=journal:
sudo tune2fs -o journal_data /dev/vda1
Run fio:
rm -f fiotest
fio --name=fiotest --rw=randwrite --bs=4k --runtime=6 --ioengine=libaio
--iodepth=256 --numjobs=8 --filename=fiotest --filesize=30G
--group_reporting
Reboot the VM, and check the console output from:
virsh console testvm
But there is no dirty inode, folio_clear_dirty_for_io clears PG_dirty
but leaves tags PAGECACHE_TAG_DIRTY and PAGECACHE_TAG_TOWRITE set which
are only cleared by __folio_start_writeback.
In data=journal mode, jbd2 checkpoints the journalled data to its final
location and clears its own dirty flag without touching folio PG_dirty
or xarray dirty flags.
The commit f4a2b42e78 ("ext4: fix stale xarray tags after writeback")
fixes when PG_dirty is still set but there is no dirty page.
Another case is PG_dirty is cleared, but PAGECACHE_TAG_DIRTY and
PAGECACHE_TAG_TOWRITE is still set. In this case, writeback thread
checks clean folio and skips it in mpage_prepare_extent_to_map:
if (!folio_test_dirty(folio) ||
...
folio_unlcok(folio);
continue
And never reaches ext4_bio_write_folio where the commit f4a2b42e78
clears the stale xarray tags. Print debug logs after the filesystem
is remounted read-only:
writepages RDONLY nrpages=2048 dirtytag=1 wbtag=0 towrite=1 sync=0
And all folios are actually clean:
folio idx=3 dirty=0 wb=0 checked=0 dirtybuf=0 jbddirty=0 mapped=1
...
We need to clear the xarray stale tags for such clean folios by
cycling them through writeback in the skip path, the same way
f4a2b42e78 does in ext4_bio_write_folio.
Fixes: dff4ac75ee ("ext4: move keep_towrite handling to ext4_bio_write_page()")
Signed-off-by: Gerald Yang <gerald.yang@canonical.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260625160127.162272-1-gerald.yang@canonical.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Syzbot/stress-ng reported an ABBA deadlock in ext4 when exercising
concurrent xattr workloads (using the ea_inode mount/format option).
The deadlock occurs between the running transaction and the eviction
thread:
- Task 1 (stress-ng): Holds a reference to a shared mbcache_entry (ce)
and calls ext4_xattr_inode_cache_find() -> ext4_iget() to retrieve
the corresponding EA inode. Since the EA inode is currently being
evicted, ext4_iget() blocks in __wait_on_freeing_inode() waiting for
eviction to complete.
- Task 2 (eviction thread): Currently evicting the same EA inode in
ext4_evict_ea_inode(). It calls mb_cache_entry_wait_unused(oe) which
blocks waiting for Task 1 to release the reference to the mbcache_entry.
To break this deadlock, implement a new ext4_iget() configuration flag
named EXT4_IGET_NOWAIT. When set, perform a non-blocking lookup of the
inode via VFS's find_inode_nowait() API.
If the inode is currently being evicted (marked with I_FREEING or
I_WILL_FREE) or created (I_CREATING), or if it is not present in the VFS
inode cache (cache miss), simply skip it (returning -ENOENT) rather than
waiting for eviction/creation to complete, breaking the ABBA cycle.
Since we return -ENOENT immediately on a cache miss, we never attempt to
allocate a new inode or call iget_locked(), completely eliminating any
TOCTOU race window.
If the returned inode is I_NEW, wait for its initialization to clear via
wait_on_new_inode(). If initialization fails and the inode is unhashed
during wait_on_new_inode() waking up (e.g., due to an I/O read error in
another thread), safely drop the reference and return -ENOENT. This
unhashed check is executed unconditionally on all cache-hit pathways to
properly handle concurrent initialization failures.
Finally, standard validation checks (including is_bad_inode,
EXT4_EA_INODE_FL, file_acl, and xattr flags) are executed as normal inside
check_igot_inode() to fully guarantee VFS-layer safety.
In ext4_xattr_inode_cache_find(), invoke ext4_iget() with the new
EXT4_IGET_NOWAIT flag to perform the non-blocking cache search.
Suggested-by: Jan Kara <jack@suse.cz>
Reported-by: Colin Ian King <colin.i.king@gmail.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=219283
Fixes: 0a46ef2347 ("ext4: do not create EA inode under buffer lock")
Signed-off-by: Aditya Prakash Srivastava <aditya.ansh182@gmail.com>
Tested-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260626054821.1729-1-aditya.ansh182@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
An inode being evicted will never need its extra isize expanded. Set
EXT4_STATE_NO_EXPAND before ext4_mark_inode_dirty() in ext4_evict_inode()
to make this explicit and prevent any unnecessary work in
ext4_try_to_expand_extra_isize().
This also provides defense-in-depth for the s_writepages_rwsem deadlock
during mount-time orphan cleanup, ensuring the expand path is blocked
for inodes under eviction regardless of how they are reached.
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260623061903.2148767-2-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_try_to_expand_extra_isize() is called from __ext4_mark_inode_dirty()
while holding an active jbd2 handle. During mount (!SB_ACTIVE), the
expand path may move xattrs to external blocks and release ea_inodes via
iput(). When !SB_ACTIVE, iput() calls write_inode_now() which acquires
s_writepages_rwsem, creating a circular lock dependency:
s_writepages_rwsem --> jbd2_handle --> xattr_sem --> s_writepages_rwsem
This can be triggered via:
ext4_process_orphan() -> ext4_truncate() -> ext4_mark_inode_dirty()
-> ext4_try_to_expand_extra_isize()
or:
ext4_evict_inode() -> ext4_mark_inode_dirty()
-> ext4_try_to_expand_extra_isize()
Skip expansion when !SB_ACTIVE. This is a minor loss of functionality
(extra isize won't grow for these inodes during mount), which e2fsck
can resolve later if needed.
Reported-by: syzbot+5d19358d7eb30ffb0cc5@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=5d19358d7eb30ffb0cc5
Fixes: c8585c6fca ("ext4: fix races between changing inode journal mode and ext4_writepages")
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260623061903.2148767-1-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_read_inline_dir() can read a dirent header past the end of its inline
buffer, triggering a slab-out-of-bounds read during getdents64():
BUG: KASAN: slab-out-of-bounds in __ext4_check_dir_entry
Read of size 2 at addr ffff88800f3dd23c by task exploit/148
...
__ext4_check_dir_entry
ext4_read_inline_dir
iterate_dir
The dirent payload lives in a buffer of exactly inline_size bytes:
dir_buf = kmalloc(inline_size, GFP_NOFS);
but iteration runs in a position space extra_offset bytes larger
(extra_size = extra_offset + inline_size) so the synthetic "." and ".."
land at their block-dir offsets. A dirent is formed at "dir_buf + pos -
extra_offset", yet the ext4_check_dir_entry() length argument uses the
larger extra_size. A position whose dirent header would extend past
extra_size is therefore accepted, and the rescan loop's rec_len probe and
ext4_check_dir_entry() dereference de->rec_len before the entry is rejected.
Reject a position whose minimum-size dirent header would not fit within
extra_size before forming de, in both the rescan and main loops, and pass
inline_size rather than extra_size to ext4_check_dir_entry() so the length
check matches the physical buffer.
Fixes: c4d8b0235a ("ext4: fix readdir error in case inline_data+^dir_index.")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260615190519.946736-1-xmei5@asu.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Getting ext4_group_desc structure can contribute to the cost of
ext4_mb_prefetch() without any need, as most groups fail the
!EXT4_MB_GRP_TEST_AND_SET_READ check.
Optimize ext4_mb_prefetch by getting the group description only when
necessary.
The result is further increase in performance of fallocate() system call
path that triggers ext4_mb_prefetch() via a linear group scan.
Signed-off-by: Bohdan Trach <bohdan.trach@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Link: https://patch.msgid.link/20260615100331.163997-3-bohdan.trach@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
EXT4_MB_GRP_TEST_AND_SET_READ uses test_and_set_bit function which
issues an atomic write. This can cause high overhead due to cache
contention when multiple threads iterate over groups in a tight loop,
as is the case for ext4_mb_prefetch(). We have seen this to be a
problem for Kunpeng 920b CPUs which uses a single ARM LSE instruction
for this purpose.
Avoid this unconditional atomic write by testing the bit first without
changing its value. This is OK for this use case as this bit is never
unset.
This change significantly reduces costs of fallocate() operations which
trigger linear group scans on large multicore machines where
test_and_set_bit issues an atomic write operation unconditionally.
Signed-off-by: Bohdan Trach <bohdan.trach@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260615100331.163997-2-bohdan.trach@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Move iput(tmp_inode) after ext4_writepages_up_write() to avoid a
circular lock dependency between s_writepages_rwsem and sb_internal
(freeze protection).
The deadlock scenario:
CPU0 (EXT4_IOC_MIGRATE) CPU1 (orphan cleanup during mount)
---- ----
ext4_ext_migrate()
ext4_writepages_down_write()
s_writepages_rwsem (write)
ext4_evict_inode()
sb_start_intwrite() [sb_internal]
...
ext4_writepages()
s_writepages_rwsem (read) [BLOCKED]
iput(tmp_inode)
ext4_evict_inode()
sb_start_intwrite() [BLOCKED]
The tmp_inode is a temporary inode with nlink=0 created solely for
building the extent tree. Its eviction does not require
s_writepages_rwsem protection, so deferring iput() until after
releasing the rwsem is safe.
Reported-by: syzbot+212e8f62790f8e0bc63b@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=212e8f62790f8e0bc63b
Fixes: cb85f4d23f ("ext4: fix race between writepages and enabling EXT4_EXTENTS_FL")
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260612005330.1930804-1-yun.zhou@windriver.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Ext4 buffered writes into large folios also pay a full buffer_head
walk in ext4_block_write_begin(). For a small overwrite of an existing
cached folio, the folio is already uptodate and the write only needs to
prepare the buffers through the written range. Walking the suffix still
makes the write_begin cost proportional to the folio size.
Before ext4 enabled large folios for regular files, the same loop was
bounded by a single page of buffers. That commit made the existing
full-folio walk visible as a regression for cached small overwrites.
The suffix walk is needed for non-uptodate folios, where ext4 may have
to submit reads for partial blocks, preserve new-buffer cleanup, and run
error zeroing. Keep those folios on the old full walk.
For already-uptodate folios, keep the walk starting at the first buffer
rather than seeking directly to from. This preserves the existing prefix
buffer state handling. Stop once block_start reaches the end of the
write range, because the skipped suffix would only repeat the
outside-range uptodate handling for buffers beyond @to.
On current master, the libMicro ext4 large-folio overwrite test shows
the following full-series result. Results are median usecs/call over 10
runs, lower is better:
case nofix this series improvement
write_u1k 1.418 0.3405 76.0%
write_u10k 1.887 0.4175 77.9%
pwrite_u1k 1.6775 0.3390 79.8%
pwrite_u10k 1.9035 0.4130 78.3%
Fixes: 7ac67301e8 ("ext4: enable large folio for regular file")
Cc: stable@vger.kernel.org # v6.16+
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jia Zhu <zhujia.zj@bytedance.com>
Link: https://patch.msgid.link/20260609035202.90669-3-zhujia.zj@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
block_commit_write() always walks every buffer_head attached to the
folio. That was cheap for order-0 folios, but large folios can contain
hundreds of buffer_heads. For a small buffered overwrite of an
already-uptodate large folio, the commit work is therefore proportional
to the folio size rather than the copied range.
This became visible with ext4 regular-file large folios, where cached
small overwrites reach block_commit_write() through block_write_end().
Before ext4 enabled large folios for regular files, this path was only
hit with order-0 folios for normal ext4 buffered writes, so the full walk
was bounded. The ext4 large-folio commit is therefore the regression
point for this generic helper cost.
The full walk is still needed when the folio is not uptodate, because
block_commit_write() uses per-buffer uptodate state to decide whether
the whole folio can be marked uptodate. Keep those folios on the old
full-buffer path.
For a folio that was already uptodate on entry, the commit no longer
needs tail buffers for folio-uptodate discovery. The copied range has
already been processed once block_start reaches @to, so stop there and
avoid the suffix walk.
Fixes: 7ac67301e8 ("ext4: enable large folio for regular file")
Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: stable@vger.kernel.org # v6.16+
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jia Zhu <zhujia.zj@bytedance.com>
Link: https://patch.msgid.link/20260609035202.90669-2-zhujia.zj@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
