| Age | Commit message (Collapse) | Author |
|---|
|
Fix a bug in gfs2's asynchronous glock handling for rename and exchange
operations. The original async implementation from commit ad26967b9afa
("gfs2: Use async glocks for rename") mentioned that retries were needed
but never implemented them, causing operations to fail with -ESTALE
instead of retrying on timeout.
Also makes the waiting interruptible.
In addition, the timeouts used were too high for situations in which
timing out is a rare but expected scenario. Switch to shorter timeouts
with randomization and exponentional backoff.
Fixes: ad26967b9afa ("gfs2: Use async glocks for rename")
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
|
|
When an asynchronous glock holder is dequeued that hasn't been granted
yet (HIF_HOLDER not set) and no dlm operation is in progress on behalf
of that holder (GLF_LOCK not set), the dequeuing takes place in
__gfs2_glock_dq(). There, we are not clearing the HIF_WAIT flag and
waking up waiters. Fix that.
This bug prevents the same holder from being enqueued later (gfs2_glock_nq())
without first reinitializing it (gfs2_holder_reinit()). The code
doesn't currently use this pattern, but this will change in the next
commit.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
|
|
The free space and inode btree repair functions will rebuild both btrees
at the same time, after which it needs to evaluate both btrees to
confirm that the corruptions are gone.
However, Jiaming Zhang ran syzbot and produced a crash in the second
xchk_allocbt call. His root-cause analysis is as follows (with minor
corrections):
In xrep_revalidate_allocbt(), xchk_allocbt() is called twice (first
for BNOBT, second for CNTBT). The cause of this issue is that the
first call nullified the cursor required by the second call.
Let's first enter xrep_revalidate_allocbt() via following call chain:
xfs_file_ioctl() ->
xfs_ioc_scrubv_metadata() ->
xfs_scrub_metadata() ->
`sc->ops->repair_eval(sc)` ->
xrep_revalidate_allocbt()
xchk_allocbt() is called twice in this function. In the first call:
/* Note that sc->sm->sm_type is XFS_SCRUB_TYPE_BNOPT now */
xchk_allocbt() ->
xchk_btree() ->
`bs->scrub_rec(bs, recp)` ->
xchk_allocbt_rec() ->
xchk_allocbt_xref() ->
xchk_allocbt_xref_other()
since sm_type is XFS_SCRUB_TYPE_BNOBT, pur is set to &sc->sa.cnt_cur.
Kernel called xfs_alloc_get_rec() and returned -EFSCORRUPTED. Call
chain:
xfs_alloc_get_rec() ->
xfs_btree_get_rec() ->
xfs_btree_check_block() ->
(XFS_IS_CORRUPT || XFS_TEST_ERROR), the former is false and the latter
is true, return -EFSCORRUPTED. This should be caused by
ioctl$XFS_IOC_ERROR_INJECTION I guess.
Back to xchk_allocbt_xref_other(), after receiving -EFSCORRUPTED from
xfs_alloc_get_rec(), kernel called xchk_should_check_xref(). In this
function, *curpp (points to sc->sa.cnt_cur) is nullified.
Back to xrep_revalidate_allocbt(), since sc->sa.cnt_cur has been
nullified, it then triggered null-ptr-deref via xchk_allocbt() (second
call) -> xchk_btree().
So. The bnobt revalidation failed on a cross-reference attempt, so we
deleted the cntbt cursor, and then crashed when we tried to revalidate
the cntbt. Therefore, check for a null cntbt cursor before that
revalidation, and mark the repair incomplete. Also we can ignore the
second tree entirely if the first tree was rebuilt but is already
corrupt.
Apply the same fix to xrep_revalidate_iallocbt because it has the same
problem.
Cc: r772577952@gmail.com
Link: https://lore.kernel.org/linux-xfs/CANypQFYU5rRPkTy=iG5m1Lp4RWasSgrHXAh3p8YJojxV0X15dQ@mail.gmail.com/T/#m520c7835fad637eccf843c7936c200589427cc7e
Cc: <stable@vger.kernel.org> # v6.8
Fixes: dbfbf3bdf639a2 ("xfs: repair inode btrees")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jiaming Zhang <r772577952@gmail.com>
|
|
We cannot dereference bs->cur when trying to determine if bs->cur
aliases bs->sc->sa.{bno,rmap}_cur after the latter has been freed.
Fix this by sampling before type before any freeing could happen.
The correct temporal ordering was broken when we removed xfs_btnum_t.
Cc: r772577952@gmail.com
Cc: <stable@vger.kernel.org> # v6.9
Fixes: ec793e690f801d ("xfs: remove xfs_btnum_t")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jiaming Zhang <r772577952@gmail.com>
|
|
Fix this function to return NULL instead of a mangled ENOMEM, then fix
the callers to actually check for a null pointer and return ENOMEM.
Most of the corrections here are for code merged between 6.2 and 6.10.
Cc: r772577952@gmail.com
Cc: <stable@vger.kernel.org> # v6.12
Fixes: 1a5f6e08d4e379 ("xfs: create subordinate scrub contexts for xchk_metadata_inode_subtype")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jiaming Zhang <r772577952@gmail.com>
|
|
Only call the xfarray and xfblob destructor if we have a valid pointer,
and be sure to null out that pointer afterwards. Note that this patch
fixes a large number of commits, most of which were merged between 6.9
and 6.10.
Cc: r772577952@gmail.com
Cc: <stable@vger.kernel.org> # v6.12
Fixes: ab97f4b1c03075 ("xfs: repair AGI unlinked inode bucket lists")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jiaming Zhang <r772577952@gmail.com>
|
|
Currently, when we are doing an extent split and convert operation of
written to unwritten extent (example, as done by ZERO_RANGE), we don't
allow the zeroout fallback in case the extent tree manipulation fails.
This is mostly because zeroout might take unsually long and the fact that
this code path is more tolerant to failures than endio.
Since we have zeroout machinery in place, we might as well use it hence
lift this restriction. To mitigate zeroout taking too long respect the
max zeroout limit here so that the operation finishes relatively fast.
Also, add kunit tests for this case.
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/1c3349020b8e098a63f293b84bc8a9b56011cef4.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
ext4_split_convert_extents() has been historically prone to subtle
bugs and inconsistent behavior due to the way all the various flags
interact with the extent split and conversion process. For example,
callers like ext4_convert_unwritten_extents_endio() and
convert_initialized_extents() needed to open code extent conversion
despite passing CONVERT or CONVERT_UNWRITTEN flags because
ext4_split_convert_extents() wasn't performing the conversion.
Hence, refactor ext4_split_convert_extents() to clearly enforce the
semantics of each flag. The major changes here are:
* Clearly separate the split and convert process:
* ext4_split_extent() and ext4_split_extent_at() are now only
responsible to perform the split.
* ext4_split_convert_extents() is now responsible to perform extent
conversion after calling ext4_split_extent() for splitting.
* This helps get rid of all the MARK_UNWRIT* flags.
* Clearly enforce the semantics of flags passed to
ext4_split_convert_extents():
* EXT4_GET_BLOCKS_CONVERT: Will convert the split extent to written
* EXT4_GET_BLOCKS_CONVERT_UNWRITTEN: Will convert the split extent to
unwritten
* Modify all callers to enforce the above semantics.
* Use ext4_split_convert_extents() instead of ext4_split_extents()
in ext4_ext_convert_to_initialized() for uniformity.
* Now that ext4_split_convert_extents() is handling caching to es, we
dont need to do it in ext4_split_extent_zeroout().
* Cleanup all callers open coding the conversion logic. Further, modify
kuniy tests to pass flags based on the new semantics.
>From an end user point of view, we should not see any changes in
behavior of ext4.
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/2084a383d69ceefbaa293b8fcf725365eca0a349.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Currently, zeroout is used as a fallback in case we fail to
split/convert extents in the "traditional" modify-the-extent-tree way.
This is essential to mitigate failures in critical paths like extent
splitting during endio. However, the logic is very messy and not easy to
follow. Further, the fragile use of various flags has made it prone to
errors.
Refactor zeroout out logic by moving it up to ext4_split_extents().
Further, zeroout correctly based on the type of conversion we want, ie:
- unwritten to written: Zeroout everything around the mapped range.
- written to unwritten: Zeroout only the mapped range.
Also, ext4_ext_convert_to_initialized() now passes
EXT4_GET_BLOCKS_CONVERT to make the intention clear.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/e1b51dedeca7c0b1f702141d91edfe4230560e7b.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Currently, callers like ext4_convert_unwritten_extents() pass
EXT4_EX_NOCACHE flag to avoid caching extents however this is not
respected by ext4_convert_unwritten_extents_endio(). Hence, modify it to
accept flags from the caller and to pass the flags on to other extent
manipulation functions it calls. This makes sure the NOCACHE flag is
respected throughout the code path.
Also, since the caller already passes METADATA_NOFAIL and CONVERT flags
we don't need to explicitly pass it anymore.
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/7c2139e0ad32c49c19b194f72219e15d613de284.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Currently, ext4_zero_range passes EXT4_EX_NOCACHE flag to avoid caching
extents however this is not respected by convert_initialized_extent().
Hence, modify it to accept flags from the caller and to pass the flags
on to other extent manipulation functions it calls. This makes
sure the NOCACHE flag is respected throughout the code path.
Also, we no longer explicitly pass CONVERT_UNWRITTEN as the caller takes
care of this.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/07008fbb14db727fddcaf4c30e2346c49f6c8fe0.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Add support in Kunit tests to ensure that the extent status cache is
also in sync after the extent split and conversion operations.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/5f9d2668feeb89a3f3e9d03dadab8c10cbea3741.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Add more kunit tests to cover the high level caller
ext4_map_create_blocks(). We pass flags in a manner that covers
the below function:
1. ext4_ext_handle_unwritten_extents()
1.1 - Split/Convert unwritten extent to written in endio convtext.
1.2 - Split/Convert unwritten extent to written in non endio context.
1.3 - Zeroout tests for the above 2 cases
2. convert_initialized_extent() - Convert written extent to unwritten
during zero range
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/9d8ad32cb62f44999c0fe3545b44fc3113546c70.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Add multiple KUnit tests to test various permutations of extent
splitting and conversion.
We test the following cases:
1. Split of unwritten extent into 2 parts and convert 1 part to written
2. Split of unwritten extent into 3 parts and convert 1 part to written
3. Split of written extent into 2 parts and convert 1 part to unwritten
4. Split of written extent into 3 parts and convert 1 part to unwritten
5. Zeroout fallback for all the above cases except 3-4 because zeroout
is not supported for written to unwritten splits
The main function we test here is ext4_split_convert_extents().
Currently some of the tests are failing due to issues in implementation.
All failures are mitigated at other layers in ext4 [1] but still point
out the mismatch in expectation of what the caller wants vs what the
function does.
The aim is to eventually fix all the failures we see here. More detailed
implementation notes can be found in the topmost commit in the test
file.
[1] for example, EXT4_GET_BLOCKS_CONVERT doesn't really convert the
split extent to written, but rather the callers end up doing the
conversion.
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/22bb9d17cd88c1318a2edde48887ca7488cb8a13.1769149131.git.ojaswin@linux.ibm.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
|
|
Pull smb server fixes from Steve French:
- Use the original nents value for ib_dma_unmap_sg(), preventing
potential memory corruption in the RDMA transport layer
- Fix a naming discrepancy in the kernel-doc for
ksmbd_vfs_kern_path_start_removing() as identified by sparse static
analysis
- Reset smb_direct_port to its default value during initialization to
ensure the correct port is used when switching between different RDMA
device types without module reload
* tag 'v6.19-rc6-server-fixes' of git://git.samba.org/ksmbd:
smb: server: reset smb_direct_port = SMB_DIRECT_PORT_INFINIBAND on init
smb: server: fix comment for ksmbd_vfs_kern_path_start_removing()
ksmbd: smbd: fix dma_unmap_sg() nents
|
|
The xchk_xfile_*_descr macros call kasprintf, which can fail to allocate
memory if the formatted string is larger than 16 bytes (or whatever the
nofail guarantees are nowadays). Some of them could easily exceed that,
and Jiaming Zhang found a few places where that can happen with syzbot.
The descriptions are debugging aids and aren't required to be unique, so
let's just pass in static strings and eliminate this path to failure.
Note this patch touches a number of commits, most of which were merged
between 6.6 and 6.14.
Cc: r772577952@gmail.com
Cc: <stable@vger.kernel.org> # v6.12
Fixes: ab97f4b1c03075 ("xfs: repair AGI unlinked inode bucket lists")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jiaming Zhang <r772577952@gmail.com>
|
|
If we're trying to replace an xattr in a shortform attr structure and
the old entry fits the new entry, we can just memcpy and exit without
having to delete, compact, and re-add the entry (or worse use the attr
intent machinery). For parent pointers this only advantages renaming
where the filename length stays the same (e.g. mv autoexec.bat
scandisk.exe) but for regular xattrs it might be useful for updating
security labels and the like.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
After a recent fsmark benchmarking run, I observed that the overhead of
parent pointers on file creation and deletion can be a bit high. On a
machine with 20 CPUs, 128G of memory, and an NVME SSD capable of pushing
750000iops, I see the following results:
$ mkfs.xfs -f -l logdev=/dev/nvme1n1,size=1g /dev/nvme0n1 -n parent=0
meta-data=/dev/nvme0n1 isize=512 agcount=40, agsize=9767586 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=1, sparse=1, rmapbt=1
= reflink=1 bigtime=1 inobtcount=1 nrext64=1
= exchange=0 metadir=0
data = bsize=4096 blocks=390703440, imaxpct=5
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0, ftype=1, parent=0
log =/dev/nvme1n1 bsize=4096 blocks=262144, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
= rgcount=0 rgsize=0 extents
= zoned=0 start=0 reserved=0
So we created 40 AGs, one per CPU. Now we create 40 directories and run
fsmark:
$ time fs_mark -D 10000 -S 0 -n 100000 -s 0 -L 8 -d ...
# Version 3.3, 40 thread(s) starting at Wed Dec 10 14:22:07 2025
# Sync method: NO SYNC: Test does not issue sync() or fsync() calls.
# Directories: Time based hash between directories across 10000 subdirectories with 180 seconds per subdirectory.
# File names: 40 bytes long, (16 initial bytes of time stamp with 24 random bytes at end of name)
# Files info: size 0 bytes, written with an IO size of 16384 bytes per write
# App overhead is time in microseconds spent in the test not doing file writing related system calls.
parent=0 parent=1
================== ==================
real 0m57.573s real 1m2.934s
user 3m53.578s user 3m53.508s
sys 19m44.440s sys 25m14.810s
$ time rm -rf ...
parent=0 parent=1
================== ==================
real 0m59.649s real 1m12.505s
user 0m41.196s user 0m47.489s
sys 13m9.566s sys 20m33.844s
Parent pointers increase the system time by 28% overhead to create 32
million files that are totally empty. Removing them incurs a system
time increase of 56%. Wall time increases by 9% and 22%.
For most filesystems, each file tends to have a single owner and not
that many xattrs. If the xattr structure is shortform, then all xattr
changes are logged with the inode and do not require the the xattr
intent mechanism to persist the parent pointer.
Therefore, we can speed up parent pointer operations by calling the
shortform xattr functions directly if the child's xattr is in short
format. Now the overhead looks like:
$ time fs_mark -D 10000 -S 0 -n 100000 -s 0 -L 8 -d ...
parent=0 parent=1
================== ==================
real 0m58.030s real 1m0.983s
user 3m54.141s user 3m53.758s
sys 19m57.003s sys 21m30.605s
$ time rm -rf ...
parent=0 parent=1
================== ==================
real 0m58.911s real 1m4.420s
user 0m41.329s user 0m45.169s
sys 13m27.857s sys 15m58.564s
Now parent pointers only increase the system time by 8% for creation and
19% for deletion. Wall time increases by 5% and 9% now.
Close the performance gap by creating helpers for the attr set, remove,
and replace operations that will try to make direct shortform updates,
and fall back to the attr intent machinery if that doesn't work. This
works for regular xattrs and for parent pointers.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
The dp parameter to this function is an alias of args->dp, so remove it
for clarity before we go adding new callers.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
In debugging other problems with generic/753, it turns out that it's
possible for the system go to down in the middle of a remote xattr set
operation such that the leaf block entry is marked incomplete and
valueblk is set to zero. Make this no longer a failure.
Cc: <stable@vger.kernel.org> # v4.15
Fixes: 13791d3b833428 ("xfs: scrub extended attribute leaf space")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
In the previous patches, we observed that it's possible for there to be
freemap entries with zero size but a nonzero base. This isn't an
inconsistency per se, but older kernels can get confused by this and
corrupt the block, leading to corruption.
If we see this, flag the xattr structure for optimization so that it
gets rebuilt.
Cc: <stable@vger.kernel.org> # v4.15
Fixes: 13791d3b833428 ("xfs: scrub extended attribute leaf space")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
Check for erroneous overlapping freemap regions and collisions between
freemap regions and the xattr leaf entry array.
Note that we must explicitly zero out the extra freemaps in
xfs_attr3_leaf_compact so that the in-memory buffer has a correctly
initialized freemap array to satisfy the new verification code, even if
subsequent code changes the contents before unlocking the buffer.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
Replace all the open-coded callsites with a single static inline helper.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
xfs/592 and xfs/794 both trip this assertion in the leaf block freemap
adjustment code after ~20 minutes of running on my test VMs:
ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
+ xfs_attr3_leaf_hdr_size(leaf));
Upon enabling quite a lot more debugging code, I narrowed this down to
fsstress trying to set a local extended attribute with namelen=3 and
valuelen=71. This results in an entry size of 80 bytes.
At the start of xfs_attr3_leaf_add_work, the freemap looks like this:
i 0 base 448 size 0 rhs 448 count 46
i 1 base 388 size 132 rhs 448 count 46
i 2 base 2120 size 4 rhs 448 count 46
firstused = 520
where "rhs" is the first byte past the end of the leaf entry array.
This is inconsistent -- the entries array ends at byte 448, but
freemap[1] says there's free space starting at byte 388!
By the end of the function, the freemap is in worse shape:
i 0 base 456 size 0 rhs 456 count 47
i 1 base 388 size 52 rhs 456 count 47
i 2 base 2120 size 4 rhs 456 count 47
firstused = 440
Important note: 388 is not aligned with the entries array element size
of 8 bytes.
Based on the incorrect freemap, the name area starts at byte 440, which
is below the end of the entries array! That's why the assertion
triggers and the filesystem shuts down.
How did we end up here? First, recall from the previous patch that the
freemap array in an xattr leaf block is not intended to be a
comprehensive map of all free space in the leaf block. In other words,
it's perfectly legal to have a leaf block with:
* 376 bytes in use by the entries array
* freemap[0] has [base = 376, size = 8]
* freemap[1] has [base = 388, size = 1500]
* the space between 376 and 388 is free, but the freemap stopped
tracking that some time ago
If we add one xattr, the entries array grows to 384 bytes, and
freemap[0] becomes [base = 384, size = 0]. So far, so good. But if we
add a second xattr, the entries array grows to 392 bytes, and freemap[0]
gets pushed up to [base = 392, size = 0]. This is bad, because
freemap[1] hasn't been updated, and now the entries array and the free
space claim the same space.
The fix here is to adjust all freemap entries so that none of them
collide with the entries array. Note that this fix relies on commit
2a2b5932db6758 ("xfs: fix attr leaf header freemap.size underflow") and
the previous patch that resets zero length freemap entries to have
base = 0.
Cc: <stable@vger.kernel.org> # v2.6.12
Fixes: 1da177e4c3f415 ("Linux-2.6.12-rc2")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
Back in commit 2a2b5932db6758 ("xfs: fix attr leaf header freemap.size
underflow"), Brian Foster observed that it's possible for a small
freemap at the end of the end of the xattr entries array to experience
a size underflow when subtracting the space consumed by an expansion of
the entries array. There are only three freemap entries, which means
that it is not a complete index of all free space in the leaf block.
This code can leave behind a zero-length freemap entry with a nonzero
base. Subsequent setxattr operations can increase the base up to the
point that it overlaps with another freemap entry. This isn't in and of
itself a problem because the code in _leaf_add that finds free space
ignores any freemap entry with zero size.
However, there's another bug in the freemap update code in _leaf_add,
which is that it fails to update a freemap entry that begins midway
through the xattr entry that was just appended to the array. That can
result in the freemap containing two entries with the same base but
different sizes (0 for the "pushed-up" entry, nonzero for the entry
that's actually tracking free space). A subsequent _leaf_add can then
allocate xattr namevalue entries on top of the entries array, leading to
data loss. But fixing that is for later.
For now, eliminate the possibility of confusion by zeroing out the base
of any freemap entry that has zero size. Because the freemap is not
intended to be a complete index of free space, a subsequent failure to
find any free space for a new xattr will trigger block compaction, which
regenerates the freemap.
It looks like this bug has been in the codebase for quite a long time.
Cc: <stable@vger.kernel.org> # v2.6.12
Fixes: 1da177e4c3f415 ("Linux-2.6.12-rc2")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
|
|
While setting file attributes, the read-only flags are reset
for ->xflags, but not for ->flags if flag is shared between both. This
is fine for now as all read-only xflags don't overlap with flags.
However, for any read-only shared flag this will create inconsistency
between xflags and flags. The non-shared flag will be reset in
vfs_fileattr_set() to the current value, but shared one is past further
to ->fileattr_set.
Reported-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Andrey Albershteyn <aalbersh@kernel.org>
Link: https://patch.msgid.link/20260121193645.3611716-1-aalbersh@kernel.org
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
|
|
Currently fsnotify_sb_delete() was called after we have evicted
superblock's dcache and inode cache. This was done mainly so that we
iterate as few inodes as possible when removing inode marks. However, as
Jakub reported, this is problematic because for some filesystems
encoding of file handles uses sb->s_root which gets cleared as part of
dcache eviction. And either delayed fsnotify events or reading fdinfo
for fsnotify group with marks on fs being unmounted may trigger encoding
of file handles during unmount. So move shutdown of fsnotify subsystem
before shrinking of dcache.
Link: https://lore.kernel.org/linux-fsdevel/CAOQ4uxgXvwumYvJm3cLDFfx-TsU3g5-yVsTiG=6i8KS48dn0mQ@mail.gmail.com/
Reported-by: Jakub Acs <acsjakub@amazon.de>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
|
|
Instead of iterating all inodes belonging to a superblock to find inode
marks and remove them on umount, iterate all inode connectors for the
superblock. This may be substantially faster since there are generally
much less inodes with fsnotify marks than all inodes. It also removes
one use of sb->s_inodes list which we strive to ultimately remove.
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
|
|
Introduce a linked list tracking all inode connectors for a superblock.
We will use this list when the superblock is getting shutdown to
properly clean up all the inode marks instead of relying on scanning all
inodes in the superblock which can get rather slow.
Suggested-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
|
|
This patch implements the .fadvise interface for page cache share.
Similar to overlayfs, it drops those clean, unused pages through
vfs_fadvise().
Signed-off-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
This patch adds page cache sharing functionality for compressed inodes.
Signed-off-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
This patch adds inode page cache sharing functionality for unencoded
files.
I conducted experiments in the container environment. Below is the
memory usage for reading all files in two different minor versions
of container images:
+-------------------+------------------+-------------+---------------+
| Image | Page Cache Share | Memory (MB) | Memory |
| | | | Reduction (%) |
+-------------------+------------------+-------------+---------------+
| | No | 241 | - |
| redis +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 163 | 33% |
+-------------------+------------------+-------------+---------------+
| | No | 872 | - |
| postgres +------------------+-------------+---------------+
| 16.1 & 16.2 | Yes | 630 | 28% |
+-------------------+------------------+-------------+---------------+
| | No | 2771 | - |
| tensorflow +------------------+-------------+---------------+
| 2.11.0 & 2.11.1 | Yes | 2340 | 16% |
+-------------------+------------------+-------------+---------------+
| | No | 926 | - |
| mysql +------------------+-------------+---------------+
| 8.0.11 & 8.0.12 | Yes | 735 | 21% |
+-------------------+------------------+-------------+---------------+
| | No | 390 | - |
| nginx +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 219 | 44% |
+-------------------+------------------+-------------+---------------+
| tomcat | No | 924 | - |
| 10.1.25 & 10.1.26 +------------------+-------------+---------------+
| | Yes | 474 | 49% |
+-------------------+------------------+-------------+---------------+
Additionally, the table below shows the runtime memory usage of the
container:
+-------------------+------------------+-------------+---------------+
| Image | Page Cache Share | Memory (MB) | Memory |
| | | | Reduction (%) |
+-------------------+------------------+-------------+---------------+
| | No | 35 | - |
| redis +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 28 | 20% |
+-------------------+------------------+-------------+---------------+
| | No | 149 | - |
| postgres +------------------+-------------+---------------+
| 16.1 & 16.2 | Yes | 95 | 37% |
+-------------------+------------------+-------------+---------------+
| | No | 1028 | - |
| tensorflow +------------------+-------------+---------------+
| 2.11.0 & 2.11.1 | Yes | 930 | 10% |
+-------------------+------------------+-------------+---------------+
| | No | 155 | - |
| mysql +------------------+-------------+---------------+
| 8.0.11 & 8.0.12 | Yes | 132 | 15% |
+-------------------+------------------+-------------+---------------+
| | No | 25 | - |
| nginx +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 20 | 20% |
+-------------------+------------------+-------------+---------------+
| tomcat | No | 186 | - |
| 10.1.25 & 10.1.26 +------------------+-------------+---------------+
| | Yes | 98 | 48% |
+-------------------+------------------+-------------+---------------+
Co-developed-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
The trace_erofs_read_folio accesses inode information through folio,
but this method fails if the real inode is not associated with the
folio(such as in the upcoming page cache sharing case). Therefore,
we pass the real inode to it so that the inode information can be
printed out in that case.
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
Currently, reading files with different paths (or names) but the same
content will consume multiple copies of the page cache, even if the
content of these page caches is the same. For example, reading
identical files (e.g., *.so files) from two different minor versions of
container images will cost multiple copies of the same page cache,
since different containers have different mount points. Therefore,
sharing the page cache for files with the same content can save memory.
This introduces the page cache share feature in erofs. It allocate a
shared inode and use its page cache as shared. Reads for files
with identical content will ultimately be routed to the page cache of
the shared inode. In this way, a single page cache satisfies
multiple read requests for different files with the same contents.
We introduce new mount option `inode_share` to enable the page
sharing mode during mounting. This option is used in conjunction
with `domain_id` to share the page cache within the same trusted
domain.
Signed-off-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
Either the existing fscache usecase or the upcoming page
cache sharing case, the `domain_id` should be protected as
sensitive information, so we use the safer helpers to allocate,
free and display domain_id.
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
Add erofs_inode_set_aops helper to set the inode->i_mapping->a_ops
and use IS_ENABLED to make it cleaner.
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
When creating the EROFS image, users can specify the fingerprint name.
This is to prepare for the upcoming inode page cache share.
Signed-off-by: Hongzhen Luo <hongzhen@linux.alibaba.com>
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
- Move the `struct erofs_anon_fs_type` to super.c and expose it
in preparation for the upcoming page cache share feature;
- Remove the `.owner` field, as they are all internal mounts and
fully managed by EROFS. Retaining `.owner` would unnecessarily
increment module reference counts, preventing the EROFS kernel
module from being unloaded.
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
There is no need to open nonexistent real files if backing files
couldn't be backed by real files (e.g., EROFS page cache sharing
doesn't need typical real files to open again).
Therefore, we export the alloc_empty_backing_file() helper, allowing
filesystems to dynamically set the backing file without real file
open. This is particularly useful for obtaining the correct @path
and @inode when calling file_user_path() and file_user_inode().
Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Acked-by: Amir Goldstein <amir73il@gmail.com>
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull 'vfs-7.0.iomap' to allow iomap page cache users to set
`iomap_iter::private` for the upcoming page cache sharing support.
It also includes a patch to avoid triggering inline data reads for
the FIEMAP operation.
Signed-off-by: Gao Xiang <xiang@kernel.org>
|
|
In close_range(), the kernel traditionally performs a linear scan over the
[fd, max_fd] range, resulting in O(N) complexity where N is the range size.
For processes with sparse FD tables, this is inefficient as it checks many
unallocated slots.
This patch optimizes __range_close() by using find_next_bit() on the
open_fds bitmap to skip holes. This shifts the algorithmic complexity from
O(Range Size) to O(Active FDs), providing a significant performance boost
for large-range close operations on sparse file descriptor tables.
Signed-off-by: Qiliang Yuan <realwujing@gmail.com>
Signed-off-by: Qiliang Yuan <yuanql9@chinatelecom.cn>
Link: https://patch.msgid.link/20260123081221.659125-1-realwujing@gmail.com
Signed-off-by: Christian Brauner <brauner@kernel.org>
|
|
This allows testing with different devices (iwrap vs. non-iwarp) without
'rmmod ksmbd && modprobe ksmbd', but instead
'ksmbd.control -s && ksmbd.mountd' is enough.
In the long run we want to listen on iwarp and non-iwarp at the same time,
but requires more changes, most likely also in the rdma layer.
Cc: Namjae Jeon <linkinjeon@kernel.org>
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
|
|
This was found by sparse...
Fixes: 1ead2213dd7d ("smb/server: use end_removing_noperm for for target of smb2_create_link()")
Cc: Namjae Jeon <linkinjeon@kernel.org>
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: NeilBrown <neil@brown.name>
Cc: Christian Brauner <brauner@kernel.org>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
|
|
The dma_unmap_sg() functions should be called with the same nents as the
dma_map_sg(), not the value the map function returned.
Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
Cc: <stable@vger.kernel.org>
Signed-off-by: Thomas Fourier <fourier.thomas@gmail.com>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
|
|
Mainly get rid of the use of `struct erofs_xattr_iter`, as it is
no longer needed now that meta buffers are used.
This also simplifies the code and uses an early return when there
are no xattrs.
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
- Get rid of `sbi->opt.max_sync_decompress_pages` since it's fixed as
3 all the time;
- Add Z_EROFS_MAX_SYNC_DECOMPRESS_BYTES in bytes instead of in pages,
since for non-4K pages, 3-page limitation makes no sense;
- Move `sync_decompress` to sbi to avoid unexpected remount impact;
- Fold z_erofs_is_sync_decompress() into its caller;
- Better description of sysfs entry `sync_decompress`.
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
Use it->kaddr directly.
Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
Reviewed-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
It can be simply in xattr.c due to no external users.
Reviewed-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
No external users other than those in xattr.c.
Reviewed-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
|
Crafted EROFS images containing valid volume labels can trigger
incorrect early returns, leading to folio reference leaks.
However, this does not cause system crashes or other severe issues.
Fixes: 1cf12c717741 ("erofs: Add support for FS_IOC_GETFSLABEL")
Cc: stable@kernel.org
Reviewed-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
|
