With bigalloc enabled, the logic to report last extent has a bug since
we try to use cluster units instead of block units. This can cause an
issue where extra incorrect entries might be returned back to the
user. This was flagged by generic/365 with 64k bs and -O bigalloc.
** Details of issue **
The issue was noticed on 5G 64k blocksize FS with -O bigalloc which has
only 1 bg.
$ xfs_io -c "fsmap -d" /mnt/scratch
0: 253:48 [0..127]: static fs metadata 128 /* sb */
1: 253:48 [128..255]: special 102:1 128 /* gdt */
3: 253:48 [256..383]: special 102:3 128 /* block bitmap */
4: 253:48 [384..2303]: unknown 1920 /* flex bg empty space */
5: 253:48 [2304..2431]: special 102:4 128 /* inode bitmap */
6: 253:48 [2432..4351]: unknown 1920 /* flex bg empty space */
7: 253:48 [4352..6911]: inodes 2560
8: 253:48 [6912..538623]: unknown 531712
9: 253:48 [538624..10485759]: free space 9947136
The issue can be seen with:
$ xfs_io -c "fsmap -d 0 3" /mnt/scratch
0: 253:48 [0..127]: static fs metadata 128
1: 253:48 [384..2047]: unknown 1664
Only the first entry was expected to be returned but we get 2. This is
because:
ext4_getfsmap_datadev()
first_cluster, last_cluster = 0
...
info->gfi_last = true;
ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster + 1, 0, info);
fsb = C2B(1) = 16
fslen = 0
...
/* Merge in any relevant extents from the meta_list */
list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
...
// since fsb = 16, considers all metadata which starts before 16 blockno
iter 1: error = ext4_getfsmap_helper(sb, info, p); // p = sb (0,1), nop
info->gfi_next_fsblk = 1
iter 2: error = ext4_getfsmap_helper(sb, info, p); // p = gdt (1,2), nop
info->gfi_next_fsblk = 2
iter 3: error = ext4_getfsmap_helper(sb, info, p); // p = blk bitmap (2,3), nop
info->gfi_next_fsblk = 3
iter 4: error = ext4_getfsmap_helper(sb, info, p); // p = ino bitmap (18,19)
if (rec_blk > info->gfi_next_fsblk) { // (18 > 3)
// emits an extra entry ** BUG **
}
}
Fix this by directly calling ext4_getfsmap_datadev() with a dummy
record that has fmr_physical set to (end_fsb + 1) instead of
last_cluster + 1. By using the block instead of cluster we get the
correct behavior.
Replacing ext4_getfsmap_datadev_helper() with ext4_getfsmap_helper()
is okay since the gfi_lastfree and metadata checks in
ext4_getfsmap_datadev_helper() are anyways redundant when we only want
to emit the last allocated block of the range, as we have already
taken care of emitting metadata and any last free blocks.
Cc: stable@kernel.org
Reported-by: Disha Goel <disgoel@linux.ibm.com>
Fixes: 4a622e4d47 ("ext4: fix FS_IOC_GETFSMAP handling")
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Link: https://patch.msgid.link/e7472c8535c9c5ec10f425f495366864ea12c9da.1754377641.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The check for a fast symlink in the presence of only an
external xattr inode is incorrect. If a fast symlink does
not have an xattr block (i_file_acl == 0), but does have
an external xattr inode that increases inode i_blocks, then
the check for a fast symlink will incorrectly fail and
__ext4_iget()->ext4_ind_check_inode() will report the inode
is corrupt when it "validates" i_data[] on the next read:
# ln -s foo /mnt/tmp/bar
# setfattr -h -n trusted.test \
-v "$(yes | head -n 4000)" /mnt/tmp/bar
# umount /mnt/tmp
# mount /mnt/tmp
# ls -l /mnt/tmp
ls: cannot access '/mnt/tmp/bar': Structure needs cleaning
total 4
? l?????????? ? ? ? ? ? bar
# dmesg | tail -1
EXT4-fs error (device dm-8): __ext4_iget:5098:
inode #24578: block 7303014: comm ls: invalid block
(note that "block 7303014" = 0x6f6f66 = "foo" in LE order).
ext4_inode_is_fast_symlink() should check the superblock
EXT4_FEATURE_INCOMPAT_EA_INODE feature flag, not the inode
EXT4_EA_INODE_FL, since the latter is only set on the xattr
inode itself, and not on the inode that uses this xattr.
Cc: stable@vger.kernel.org
Fixes: fc82228a5e ("ext4: support fast symlinks from ext3 file systems")
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Li Dongyang <dongyangli@ddn.com>
Reviewed-by: Alex Zhuravlev <bzzz@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Reviewed-on: https://review.whamcloud.com/59879
Lustre-bug-id: https://jira.whamcloud.com/browse/LU-19121
Link: https://patch.msgid.link/20250717063709.757077-1-adilger@dilger.ca
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
IMA testing revealed that after an ext4 remount, file accesses triggered
full measurements even without modifications, instead of skipping as
expected when i_version is unchanged.
Debugging showed `SB_I_VERSION` was cleared in reconfigure_super() during
remount due to commit 1ff2030739 ("ext4: unconditionally enable the
i_version counter") removing the fix from commit 960e0ab63b ("ext4: fix
i_version handling on remount").
To rectify this, `SB_I_VERSION` is always set for `fc->sb_flags` in
ext4_init_fs_context(), instead of `sb->s_flags` in __ext4_fill_super(),
ensuring it persists across all mounts.
Cc: stable@kernel.org
Fixes: 1ff2030739 ("ext4: unconditionally enable the i_version counter")
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250703073903.6952-2-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
A syzbot fuzzed image triggered a BUG_ON in ext4_update_inline_data()
when an inode had the INLINE_DATA_FL flag set but was missing the
system.data extended attribute.
Since this can happen due to a maiciouly fuzzed file system, we
shouldn't BUG, but rather, report it as a corrupted file system.
Add similar replacements of BUG_ON with EXT4_ERROR_INODE() ii
ext4_create_inline_data() and ext4_inline_data_truncate().
Reported-by: syzbot+544248a761451c0df72f@syzkaller.appspotmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Although we now perform ordered traversal within an xarray, this is
currently limited to a single xarray. However, we have multiple such
xarrays, which prevents us from guaranteeing a linear-like traversal
where all groups on the right are visited before all groups on the left.
For example, suppose we have 128 block groups, with a target group of 64,
a target length corresponding to an order of 1, and available free groups
of 16 (order 1) and group 65 (order 8):
For linear traversal, when no suitable free block is found in group 64, it
will search in the next block group until group 127, then start searching
from 0 up to block group 63. It ensures continuous forward traversal, which
is consistent with the unidirectional rotation behavior of HDD platters.
Additionally, the block group lock contention during freeing block is
unavoidable. The goal increasing from 0 to 64 indicates that previously
scanned groups (which had no suitable free space and are likely to free
blocks later) and skipped groups (which are currently in use) have newly
freed some used blocks. If we allocate blocks in these groups, the
probability of competing with other processes increases.
For non-linear traversal, we first traverse all groups in order_1. If only
group 16 has free space in this list, we first traverse [63, 128), then
traverse [0, 64) to find the available group 16, and then allocate blocks
in group 16. Therefore, it cannot guarantee continuous traversal in one
direction, thus increasing the probability of contention.
So refactor ext4_mb_scan_groups_xarray() to ext4_mb_scan_groups_xa_range()
to only traverse a fixed range of groups, and move the logic for handling
wrap around to the caller. The caller first iterates through all xarrays
in the range [start, ngroups) and then through the range [0, start). This
approach simulates a linear scan, which reduces contention between freeing
blocks and allocating blocks.
Assume we have the following groups, where "|" denotes the xarray traversal
start position:
order_1_groups: AB | CD
order_2_groups: EF | GH
Traversal order:
Before: C > D > A > B > G > H > E > F
After: C > D > G > H > A > B > E > F
Performance test data follows:
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 19555 | 20049 (+2.5%) | 315636 | 316724 (-0.3%) |
|mb_optimize_scan=1 | 15496 | 19342 (+24.8%) | 323569 | 328324 (+1.4%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 53192 | 52125 (-2.0%) | 212678 | 215136 (+1.1%) |
|mb_optimize_scan=1 | 37636 | 50331 (+33.7%) | 214189 | 209431 (-2.2%) |
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-18-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This commit converts the `choose group` logic to `scan group` using
previously prepared helper functions. This allows us to leverage xarrays
for ordered non-linear traversal, thereby mitigating the "bouncing" issue
inherent in the `choose group` mechanism.
This also decouples linear and non-linear traversals, leading to cleaner
and more readable code.
Key changes:
* ext4_mb_choose_next_group() is refactored to ext4_mb_scan_groups().
* Replaced ext4_mb_good_group() with ext4_mb_scan_group() in non-linear
traversals, and related functions now return error codes instead of
group info.
* Added ext4_mb_scan_groups_linear() for performing linear scans starting
from a specific group for a set number of times.
* Linear scans now execute up to sbi->s_mb_max_linear_groups times,
so ac_groups_linear_remaining is removed as it's no longer used.
* ac->ac_criteria is now used directly instead of passing cr around.
Also, ac->ac_criteria is incremented directly after groups scan fails
for the corresponding criteria.
* Since we're now directly scanning groups instead of finding a good group
then scanning, the following variables and flags are no longer needed,
s_bal_cX_groups_considered is sufficient.
s_bal_p2_aligned_bad_suggestions
s_bal_goal_fast_bad_suggestions
s_bal_best_avail_bad_suggestions
EXT4_MB_CR_POWER2_ALIGNED_OPTIMIZED
EXT4_MB_CR_GOAL_LEN_FAST_OPTIMIZED
EXT4_MB_CR_BEST_AVAIL_LEN_OPTIMIZED
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-17-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
While traversing the list, holding a spin_lock prevents load_buddy, making
direct use of ext4_try_lock_group impossible. This can lead to a bouncing
scenario where spin_is_locked(grp_A) succeeds, but ext4_try_lock_group()
fails, forcing the list traversal to repeatedly restart from grp_A.
In contrast, linear traversal directly uses ext4_try_lock_group(),
avoiding this bouncing. Therefore, we need a lockless, ordered traversal
to achieve linear-like efficiency.
Therefore, this commit converts both average fragment size lists and
largest free order lists into ordered xarrays.
In an xarray, the index represents the block group number and the value
holds the block group information; a non-empty value indicates the block
group's presence.
While insertion and deletion complexity remain O(1), lookup complexity
changes from O(1) to O(nlogn), which may slightly reduce single-threaded
performance.
Additionally, xarray insertions might fail, potentially due to memory
allocation issues. However, since we have linear traversal as a fallback,
this isn't a major problem. Therefore, we've only added a warning message
for insertion failures here.
A helper function ext4_mb_find_good_group_xarray() is added to find good
groups in the specified xarray starting at the specified position start,
and when it reaches ngroups-1, it wraps around to 0 and then to start-1.
This ensures an ordered traversal within the xarray.
Performance test results are as follows: Single-process operations
on an empty disk show negligible impact, while multi-process workloads
demonstrate a noticeable performance gain.
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 20097 | 19555 (-2.6%) | 316141 | 315636 (-0.2%) |
|mb_optimize_scan=1 | 13318 | 15496 (+16.3%) | 325273 | 323569 (-0.5%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 53603 | 53192 (-0.7%) | 214243 | 212678 (-0.7%) |
|mb_optimize_scan=1 | 20887 | 37636 (+80.1%) | 213632 | 214189 (+0.2%) |
[ Applied spelling fixes per discussion on the ext4-list see thread
referened in the Link tag. --tytso]
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-16-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The grp->bb_largest_free_order is updated regardless of whether
mb_optimize_scan is enabled. This can lead to inconsistencies between
grp->bb_largest_free_order and the actual s_mb_largest_free_orders list
index when mb_optimize_scan is repeatedly enabled and disabled via remount.
For example, if mb_optimize_scan is initially enabled, largest free
order is 3, and the group is in s_mb_largest_free_orders[3]. Then,
mb_optimize_scan is disabled via remount, block allocations occur,
updating largest free order to 2. Finally, mb_optimize_scan is re-enabled
via remount, more block allocations update largest free order to 1.
At this point, the group would be removed from s_mb_largest_free_orders[3]
under the protection of s_mb_largest_free_orders_locks[2]. This lock
mismatch can lead to list corruption.
To fix this, whenever grp->bb_largest_free_order changes, we now always
attempt to remove the group from its old order list. However, we only
insert the group into the new order list if `mb_optimize_scan` is enabled.
This approach helps prevent lock inconsistencies and ensures the data in
the order lists remains reliable.
Fixes: 196e402adf ("ext4: improve cr 0 / cr 1 group scanning")
CC: stable@vger.kernel.org
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-12-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Groups with no free blocks shouldn't be in any average fragment size list.
However, when all blocks in a group are allocated(i.e., bb_fragments or
bb_free is 0), we currently skip updating the average fragment size, which
means the group isn't removed from its previous s_mb_avg_fragment_size[old]
list.
This created "zombie" groups that were always skipped during traversal as
they couldn't satisfy any block allocation requests, negatively impacting
traversal efficiency.
Therefore, when a group becomes completely full, bb_avg_fragment_size_order
is now set to -1. If the old order was not -1, a removal operation is
performed; if the new order is not -1, an insertion is performed.
Fixes: 196e402adf ("ext4: improve cr 0 / cr 1 group scanning")
CC: stable@vger.kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-11-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Attempt to merge ext4_free_data with already inserted free extents prior
to adding new ones. This strategy drastically cuts down the number of
times locks are held.
For example, if prev, new, and next extents are all mergeable, the existing
code (before this patch) requires acquiring the s_md_lock three times:
prev merge into new and free prev // hold lock
next merge into new and free next // hold lock
insert new // hold lock
After the patch, it only needs to be acquired once:
new merge into next and free new // no lock
next merge into prev and free next // hold lock
Performance test data follows:
Test: Running will-it-scale/fallocate2 on CPU-bound containers.
Observation: Average fallocate operations per container per second.
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 20043 | 20097 (+0.2%) | 314331 | 316141 (+0.5%) |
|mb_optimize_scan=1 | 7290 | 13318 (+87.4%) | 324226 | 325273 (+0.3%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 54999 | 53603 (-2.5%) | 214380 | 214243 (-0.06%)|
|mb_optimize_scan=1 | 13497 | 20887 (+54.6%) | 216276 | 213632 (-1.2%) |
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-10-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Previously, s_md_lock was used to protect s_mb_free_pending during
modifications, while smp_mb() ensured fresh reads, so s_md_lock just
guarantees the atomicity of s_mb_free_pending. Thus we optimized it by
converting s_mb_free_pending into an atomic variable, thereby eliminating
s_md_lock and minimizing lock contention. This also prepares for future
lockless merging of free extents.
Following this modification, s_md_lock is exclusively responsible for
managing insertions and deletions within s_freed_data_list, along with
operations involving list_splice.
Performance test data follows:
Test: Running will-it-scale/fallocate2 on CPU-bound containers.
Observation: Average fallocate operations per container per second.
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 19628 | 20043 (+2.1%) | 320885 | 314331 (-2.0%) |
|mb_optimize_scan=1 | 7129 | 7290 (+2.2%) | 321275 | 324226 (+0.9%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 53760 | 54999 (+2.3%) | 213145 | 214380 (+0.5%) |
|mb_optimize_scan=1 | 12716 | 13497 (+6.1%) | 215262 | 216276 (+0.4%) |
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-9-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When allocating data blocks, if the first try (goal allocation) fails and
stream allocation is on, it tries a global goal starting from the last
group we used (s_mb_last_group). This helps cluster large files together
to reduce free space fragmentation, and the data block contiguity also
accelerates write-back to disk.
However, when multiple processes allocate blocks, having just one global
goal means they all fight over the same group. This drastically lowers
the chances of extents merging and leads to much worse file fragmentation.
To mitigate this multi-process contention, we now employ multiple global
goals, with the number of goals being the minimum between the number of
possible CPUs and one-quarter of the filesystem's total block group count.
To ensure a consistent goal for each inode, we select the corresponding
goal by taking the inode number modulo the total number of goals.
Performance test data follows:
Test: Running will-it-scale/fallocate2 on CPU-bound containers.
Observation: Average fallocate operations per container per second.
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 9636 | 19628 (+103%) | 337597 | 320885 (-4.9%) |
|mb_optimize_scan=1 | 4834 | 7129 (+47.4%) | 341440 | 321275 (-5.9%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 22341 | 53760 (+140%) | 219707 | 213145 (-2.9%) |
|mb_optimize_scan=1 | 9177 | 12716 (+38.5%) | 215732 | 215262 (+0.2%) |
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-6-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After we optimized the block group lock, we found another lock
contention issue when running will-it-scale/fallocate2 with multiple
processes. The fallocate's block allocation and the truncate's block
release were fighting over the s_md_lock. The problem is, this lock
protects totally different things in those two processes: the list of
freed data blocks (s_freed_data_list) when releasing, and where to start
looking for new blocks (mb_last_group) when allocating.
Now we only need to track s_mb_last_group and no longer need to track
s_mb_last_start, so we don't need the s_md_lock lock to ensure that the
two are consistent. Since s_mb_last_group is merely a hint and doesn't
require strong synchronization, READ_ONCE/WRITE_ONCE is sufficient.
Besides, the s_mb_last_group data type only requires ext4_group_t
(i.e., unsigned int), rendering unsigned long superfluous.
Performance test data follows:
Test: Running will-it-scale/fallocate2 on CPU-bound containers.
Observation: Average fallocate operations per container per second.
|CPU: Kunpeng 920 | P80 | P1 |
|Memory: 512GB |------------------------|-------------------------|
|960GB SSD (0.5GB/s)| base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 4821 | 9636 (+99.8%) | 314065 | 337597 (+7.4%) |
|mb_optimize_scan=1 | 4784 | 4834 (+1.04%) | 316344 | 341440 (+7.9%) |
|CPU: AMD 9654 * 2 | P96 | P1 |
|Memory: 1536GB |------------------------|-------------------------|
|960GB SSD (1GB/s) | base | patched | base | patched |
|-------------------|-------|----------------|--------|----------------|
|mb_optimize_scan=0 | 15371 | 22341 (+45.3%) | 205851 | 219707 (+6.7%) |
|mb_optimize_scan=1 | 6101 | 9177 (+50.4%) | 207373 | 215732 (+4.0%) |
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-5-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When ext4 allocates blocks, we used to just go through the block groups
one by one to find a good one. But when there are tons of block groups
(like hundreds of thousands or even millions) and not many have free space
(meaning they're mostly full), it takes a really long time to check them
all, and performance gets bad. So, we added the "mb_optimize_scan" mount
option (which is on by default now). It keeps track of some group lists,
so when we need a free block, we can just grab a likely group from the
right list. This saves time and makes block allocation much faster.
But when multiple processes or containers are doing similar things, like
constantly allocating 8k blocks, they all try to use the same block group
in the same list. Even just two processes doing this can cut the IOPS in
half. For example, one container might do 300,000 IOPS, but if you run two
at the same time, the total is only 150,000.
Since we can already look at block groups in a non-linear way, the first
and last groups in the same list are basically the same for finding a block
right now. Therefore, add an ext4_try_lock_group() helper function to skip
the current group when it is locked by another process, thereby avoiding
contention with other processes. This helps ext4 make better use of having
multiple block groups.
Also, to make sure we don't skip all the groups that have free space
when allocating blocks, we won't try to skip busy groups anymore when
ac_criteria is CR_ANY_FREE.
Performance test data follows:
Test: Running will-it-scale/fallocate2 on CPU-bound containers.
Observation: Average fallocate operations per container per second.
|CPU: Kunpeng 920 | P80 |
|Memory: 512GB |-------------------------|
|960GB SSD (0.5GB/s)| base | patched |
|-------------------|-------|-----------------|
|mb_optimize_scan=0 | 2667 | 4821 (+80.7%) |
|mb_optimize_scan=1 | 2643 | 4784 (+81.0%) |
|CPU: AMD 9654 * 2 | P96 |
|Memory: 1536GB |-------------------------|
|960GB SSD (1GB/s) | base | patched |
|-------------------|-------|-----------------|
|mb_optimize_scan=0 | 3450 | 15371 (+345%) |
|mb_optimize_scan=1 | 3209 | 6101 (+90.0%) |
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250714130327.1830534-2-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
There was a lot of common code in the codepaths used to convert an
inline directory and to creaet a new directory. To address this,
rename ext4_init_dot_dotdot() to ext4_init_dirblock() and then move
common code into that function.
This reduces the lines of code count in fs/ext4/inline.c and
fs/ext4/namei.c, as well as reducing the size of their object files.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://patch.msgid.link/20250712181249.434530-3-tytso@mit.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The strcpy() function is considered dangerous and eeeevil by people
who are using sophisticated code analysis tools such as "grep". This
is true even when a quick inspection would show that the source is a
constant string ("." or "..") and the destination is a fixed array
which is guaranteed to have enough space. Make the "grep" code
analysis tool happy by using memcpy() isstead of strcpy(). :-)
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://patch.msgid.link/20250712181249.434530-2-tytso@mit.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In a discussion over a proposed patch, "ext4: replace strcpy() with
'.' assignment"[1], I had asserted that directory entries in ext4 were
not NUL terminated, and hence it was safe to replace strcpy() with a
direct assignment. As it turns out, this was incorrect. It's true
for all all directory entries *except* for '.' and '..' where the
kernel was using strcmp() and where e2fsck actually checks and offers
to fix things if '.' and '..' are not NUL terminated.
[1] https://lore.kernel.org/r/202505191316.JJMnPobO-lkp@intel.com
We can't change this without breaking old kernel versions, but in the
spirit of "be liberal in what you receive", use direct comparison of
de->name_len and de->name[0,1] instead of strcmp(). This has the side
benefit of reducing the compiled text size by 96 bytes on x86_64.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Link: https://patch.msgid.link/20250712181249.434530-1-tytso@mit.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently we clear BH_New bit in case of error and also in the standard
ext4_write_end() handler (in block_commit_write()). However
ext4_journalled_write_end() misses this clearing and thus we are leaving
stale BH_New bits behind. Generally ext4_block_write_begin() clears
these bits before any harm can be done but in case blocksize < pagesize
and we hit some error when processing a page with these stale bits,
we'll try to zero buffers with these stale BH_New bits and jbd2 will
complain (as buffers were not prepared for writing in this transaction).
Fix the problem by clearing BH_New bits in ext4_journalled_write_end()
and WARN if ext4_block_write_begin() sees stale BH_New bits.
Reported-by: Baolin Liu <liubaolin12138@163.com>
Reported-by: Zhi Long <longzhi@sangfor.com.cn>
Fixes: 3910b513fc ("ext4: persist the new uptodate buffers in ext4_journalled_zero_new_buffers")
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250709084831.23876-2-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_io_end_defer_completion(), check if io_end->list_vec is empty to
avoid adding an io_end that requires no conversion to the
i_rsv_conversion_list, which in turn prevents starting an unnecessary
worker. An ext4_emergency_state() check is also added to avoid attempting
to abort the journal in an emergency state.
Additionally, ext4_put_io_end_defer() is refactored to call
ext4_io_end_defer_completion() directly instead of being open-coded.
This also prevents starting an unnecessary worker when EXT4_IO_END_FAILED
is set but data_err=abort is not enabled.
This ensures that the check in ext4_put_io_end_defer() is consistent with
the check in ext4_end_bio(). Otherwise, we might add an io_end to the
i_rsv_conversion_list and then call ext4_finish_bio(), after which the
inode could be freed before ext4_end_io_rsv_work() is called, triggering
a use-after-free issue.
Fixes: ce51afb8cc ("ext4: abort journal on data writeback failure if in data_err=abort mode")
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250708111504.3208660-1-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Refactor the condition for breaking the loop within xattr_find_entry().
Elimate the usage of "<=" and take condition shortcut when "!cmp" is
true.
Originally, the condition was "(cmp <= 0 && (sorted || cmp == 0))", which
means after it knows "cmp <= 0" is true, it has to check the value of
"sorted" and "cmp". The checking of "cmp" here would be redundant since
it has already checked it.
Observing from the logic, when "cmp == 0" the branch is going to be true,
no need to check "cmp == 0" again, so we only need to take shortcut when
"cmp == 0", on the other hand, we'll check "sorted" when "cmp < 0".
The refactor can shrink the generated code size by 44 bytes. Numerous
instructions can be saved thus should also benefit execution efficiency
as well.
$ ./scripts/bloat-o-meter vmlinux_old vmlinux_new
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-44 (-44)
Function old new delta
xattr_find_entry 300 256 -44
Total: Before=22989434, After=22989390, chg -0.00%
The test is done on kernel version 6.16 with x86_64 defconfig
and gcc 13.3.0.
Signed-off-by: I Hsin Cheng <richard120310@gmail.com>
Link: https://patch.msgid.link/20250708020013.175728-1-richard120310@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After ext4 supports large folios, the semantics of reserving credits in
pages is no longer applicable. In most scenarios, reserving credits in
extents is sufficient. Therefore, introduce ext4_chunk_trans_extent()
to replace ext4_writepage_trans_blocks(). move_extent_per_page() is the
only remaining location where we are still processing extents in pages.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-10-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After ext4 supports large folios, reserving journal credits for one
maximum-ordered folio based on the worst case cenario during the
writeback process can easily exceed the maximum transaction credits.
Additionally, reserving journal credits for one page is also no
longer appropriate.
Currently, the folio writeback process can either extend the journal
credits or initiate a new transaction if the currently reserved journal
credits are insufficient. Therefore, it can be modified to reserve
credits for only one extent at the outset. In most cases involving
continuous mapping, these credits are generally adequate, and we may
only need to perform some basic credit expansion. However, in extreme
cases where the block size and folio size differ significantly, or when
the folios are sufficiently discontinuous, it may be necessary to
restart a new transaction and resubmit the folios.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-9-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Now, we reserve journal credits for converting extents in only one page
to written state when the I/O operation is complete. This is
insufficient when large folio is enabled.
Fix this by reserving credits for converting up to one extent per block in
the largest 2MB folio, this calculation should only involve extents index
and leaf blocks, so it should not estimate too many credits.
Fixes: 7ac67301e8 ("ext4: enable large folio for regular file")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Baokun Li <libaokun1@huawei.com>
Link: https://patch.msgid.link/20250707140814.542883-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After mpage_map_and_submit_extent() supports restarting handle if
credits are insufficient during allocating blocks, it is more likely to
exit the current mapping iteration and continue to process the current
processing partially mapped folio again. The existing tracepoints are
not sufficient to track this situation, so enhance the tracepoints to
track the writeback position and the return value before and after
submitting the folios.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-7-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After large folios are supported on ext4, writing back a sufficiently
large and discontinuous folio may consume a significant number of
journal credits, placing considerable strain on the journal. For
example, in a 20GB filesystem with 1K block size and 1MB journal size,
writing back a 2MB folio could require thousands of credits in the
worst-case scenario (when each block is discontinuous and distributed
across different block groups), potentially exceeding the journal size.
This issue can also occur in ext4_write_begin() and ext4_page_mkwrite()
when delalloc is not enabled.
Fix this by ensuring that there are sufficient journal credits before
allocating an extent in mpage_map_one_extent() and
ext4_block_write_begin(). If there are not enough credits, return
-EAGAIN, exit the current mapping loop, restart a new handle and a new
transaction, and allocating blocks on this folio again in the next
iteration.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The block allocation process and error handling in ext4_page_mkwrite()
is complex now. Refactor it by introducing a new helper function,
ext4_block_page_mkwrite(). It will call ext4_block_write_begin() to
allocate blocks instead of directly calling block_page_mkwrite().
Preparing to implement retry logic in a subsequent patch to address
situations where the reserved journal credits are insufficient.
Additionally, this modification will help prevent potential deadlocks
that may occur when waiting for folio writeback while holding the
transaction handle.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
During the process of writing back folios, if
mpage_map_and_submit_extent() exits the extent mapping loop due to an
ENOSPC or ENOMEM error, it may result in stale data or filesystem
inconsistency in environments where the block size is smaller than the
folio size.
When mapping a discontinuous folio in mpage_map_and_submit_extent(),
some buffers may have already be mapped. If we exit the mapping loop
prematurely, the folio data within the mapped range will not be written
back, and the file's disk size will not be updated. Once the transaction
that includes this range of extents is committed, this can lead to stale
data or filesystem inconsistency.
Fix this by submitting the current processing partially mapped folio.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250707140814.542883-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The EXT_STATS macro in fs/ext4/ext4_extents.h has been defined
but never used in the codebase since its introduction. This patch
removes it.
Analysis:
1. No references found in fs/ext4/ or other kernel code.
2. No impact on compilation or functionality.
3. Git history shows it was never utilized.
Signed-off-by: Baolin Liu <liubaolin@kylinos.cn>
Reviewed-by: Baokun Li <libaokun1@huawei.com>
Link: https://patch.msgid.link/20250527053805.1550912-1-liubaolin12138@163.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Pull smb client fixes from Steve French:
- Multichannel reconnect lock ordering deadlock fix
- Fix for regression in handling native Windows symlinks
- Three smbdirect fixes:
- oops in RDMA response processing
- smbdirect memcpy issue
- fix smbdirect regression with large writes (smbdirect test cases
now all passing)
- Fix for "FAILED_TO_PARSE" warning in trace-cmd report output
* tag 'v6.16-rc3-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
cifs: Fix reading into an ITER_FOLIOQ from the smbdirect code
cifs: Fix the smbd_response slab to allow usercopy
smb: client: fix potential deadlock when reconnecting channels
smb: client: remove \t from TP_printk statements
smb: client: let smbd_post_send_iter() respect the peers max_send_size and transmit all data
smb: client: fix regression with native SMB symlinks
Pull misc fixes from Andrew Morton:
"16 hotfixes.
6 are cc:stable and the remainder address post-6.15 issues or aren't
considered necessary for -stable kernels. 5 are for MM"
* tag 'mm-hotfixes-stable-2025-06-27-16-56' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
MAINTAINERS: add Lorenzo as THP co-maintainer
mailmap: update Duje Mihanović's email address
selftests/mm: fix validate_addr() helper
crashdump: add CONFIG_KEYS dependency
mailmap: correct name for a historical account of Zijun Hu
mailmap: add entries for Zijun Hu
fuse: fix runtime warning on truncate_folio_batch_exceptionals()
scripts/gdb: fix dentry_name() lookup
mm/damon/sysfs-schemes: free old damon_sysfs_scheme_filter->memcg_path on write
mm/alloc_tag: fix the kmemleak false positive issue in the allocation of the percpu variable tag->counters
lib/group_cpus: fix NULL pointer dereference from group_cpus_evenly()
mm/hugetlb: remove unnecessary holding of hugetlb_lock
MAINTAINERS: add missing files to mm page alloc section
MAINTAINERS: add tree entry to mm init block
mm: add OOM killer maintainer structure
fs/proc/task_mmu: fix PAGE_IS_PFNZERO detection for the huge zero folio
Pull bcachefs fixes from Kent Overstreet:
- Lots of small check/repair fixes, primarily in subvol loop and
directory structure loop (when involving snapshots).
- Fix a few 6.16 regressions: rare UAF in the foreground allocator path
when taking a transaction restart from the transaction bump
allocator, and some small fallout from the change to log the error
being corrected in the journal when repairing errors, also some
fallout from the btree node read error logging improvements.
(Alan, Bharadwaj)
- New option: journal_rewind
This lets the entire filesystem be reset to an earlier point in time.
Note that this is only a disaster recovery tool, and right now there
are major caveats to using it (discards should be disabled, in
particular), but it successfully restored the filesystem of one of
the users who was bit by the subvolume deletion bug and didn't have
backups. I'll likely be making some changes to the discard path in
the future to make this a reliable recovery tool.
- Some new btree iterator tracepoints, for tracking down some
livelock-ish behaviour we've been seeing in the main data write path.
* tag 'bcachefs-2025-06-26' of git://evilpiepirate.org/bcachefs: (51 commits)
bcachefs: Plumb correct ip to trans_relock_fail tracepoint
bcachefs: Ensure we rewind to run recovery passes
bcachefs: Ensure btree node scan runs before checking for scanned nodes
bcachefs: btree_root_unreadable_and_scan_found_nothing should not be autofix
bcachefs: fix bch2_journal_keys_peek_prev_min() underflow
bcachefs: Use wait_on_allocator() when allocating journal
bcachefs: Check for bad write buffer key when moving from journal
bcachefs: Don't unlock the trans if ret doesn't match BCH_ERR_operation_blocked
bcachefs: Fix range in bch2_lookup_indirect_extent() error path
bcachefs: fix spurious error_throw
bcachefs: Add missing bch2_err_class() to fileattr_set()
bcachefs: Add missing key type checks to check_snapshot_exists()
bcachefs: Don't log fsck err in the journal if doing repair elsewhere
bcachefs: Fix *__bch2_trans_subbuf_alloc() error path
bcachefs: Fix missing newlines before ero
bcachefs: fix spurious error in read_btree_roots()
bcachefs: fsck: Fix oops in key_visible_in_snapshot()
bcachefs: fsck: fix unhandled restart in topology repair
bcachefs: fsck: Fix check_directory_structure when no check_dirents
bcachefs: Fix restart handling in btree_node_scrub_work()
...
When performing a file read from RDMA, smbd_recv() prints an "Invalid msg
type 4" error and fails the I/O. This is due to the switch-statement there
not handling the ITER_FOLIOQ handed down from netfslib.
Fix this by collapsing smbd_recv_buf() and smbd_recv_page() into
smbd_recv() and just using copy_to_iter() instead of memcpy(). This
future-proofs the function too, in case more ITER_* types are added.
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Tom Talpey <tom@talpey.com>
cc: Paulo Alcantara (Red Hat) <pc@manguebit.com>
cc: Matthew Wilcox <willy@infradead.org>
cc: linux-cifs@vger.kernel.org
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Steve French <stfrench@microsoft.com>
