Josef Bacik [Wed, 13 Nov 2024 16:05:13 +0000 (11:05 -0500)]
btrfs: fix incorrect comparison for delayed refs
When I reworked delayed ref comparison in cf4f04325b2b ("btrfs: move
->parent and ->ref_root into btrfs_delayed_ref_node"), I made a mistake
and returned -1 for the case where ref1->ref_root was > than
ref2->ref_root. This is a subtle bug that can result in improper
delayed ref running order, which can result in transaction aborts.
Haisu Wang [Fri, 25 Oct 2024 06:54:40 +0000 (14:54 +0800)]
btrfs: fix the length of reserved qgroup to free
The dealloc flag may be cleared and the extent won't reach the disk in
cow_file_range when errors path. The reserved qgroup space is freed in
commit 30479f31d44d ("btrfs: fix qgroup reserve leaks in
cow_file_range"). However, the length of untouched region to free needs
to be adjusted with the correct remaining region size.
Filipe Manana [Mon, 4 Nov 2024 12:11:15 +0000 (12:11 +0000)]
btrfs: reinitialize delayed ref list after deleting it from the list
At insert_delayed_ref() if we need to update the action of an existing
ref to BTRFS_DROP_DELAYED_REF, we delete the ref from its ref head's
ref_add_list using list_del(), which leaves the ref's add_list member
not reinitialized, as list_del() sets the next and prev members of the
list to LIST_POISON1 and LIST_POISON2, respectively.
If later we end up calling drop_delayed_ref() against the ref, which can
happen during merging or when destroying delayed refs due to a transaction
abort, we can trigger a crash since at drop_delayed_ref() we call
list_empty() against the ref's add_list, which returns false since
the list was not reinitialized after the list_del() and as a consequence
we call list_del() again at drop_delayed_ref(). This results in an
invalid list access since the next and prev members are set to poison
pointers, resulting in a splat if CONFIG_LIST_HARDENED and
CONFIG_DEBUG_LIST are set or invalid poison pointer dereferences
otherwise.
So fix this by deleting from the list with list_del_init() instead.
But this leads to a new problem, that per-subvolume RO/RW mount no
longer works, if the initial mount is RO:
# mount -o subvol=subv1,ro /dev/test/scratch1 /mnt/test
# mount -o rw,subvol=subv2 /dev/test/scratch1 /mnt/scratch
# mount | grep mnt
/dev/mapper/test-scratch1 on /mnt/test type btrfs (ro,relatime,discard=async,space_cache=v2,subvolid=256,subvol=/subv1)
/dev/mapper/test-scratch1 on /mnt/scratch type btrfs (ro,relatime,discard=async,space_cache=v2,subvolid=257,subvol=/subv2)
# touch /mnt/scratch/foobar
touch: cannot touch '/mnt/scratch/foobar': Read-only file system
This is a common use cases on distros.
[CAUSE]
We have a workaround for remount to handle the RO->RW change, but if the
mount is using the new mount API, we do not do that, and rely on the
mount tool NOT to set the ro flag.
But that's not how the mount tool is doing for the new API:
This means we will set the super block RO at the first mount.
Later RW mount will not try to reconfigure the fs to RW because the
mount tool is already using the new API.
This totally breaks the per-subvolume RO/RW mount behavior.
[FIX]
Do not skip the reconfiguration even if using the new API. The old
comments are just expecting any mount tool to properly skip the RO flag
set even if we specify "ro", which is not the reality.
Update the comments regarding the backward compatibility on the kernel
level so it works with old and new mount utilities.
Filipe Manana [Tue, 29 Oct 2024 15:18:45 +0000 (15:18 +0000)]
btrfs: fix defrag not merging contiguous extents due to merged extent maps
When running defrag (manual defrag) against a file that has extents that
are contiguous and we already have the respective extent maps loaded and
merged, we end up not defragging the range covered by those contiguous
extents. This happens when we have an extent map that was the result of
merging multiple extent maps for contiguous extents and the length of the
merged extent map is greater than or equals to the defrag threshold
length.
$ ./test.sh
Initial number of file extent items: 4
Number of file extent items after defrag with 128K threshold: 4
Number of file extent items after defrag with 256K threshold: 4
The 4 extents don't get merged because we have an extent map with a size
of 256K that is the result of merging the individual extent maps for each
of the four 64K extents and at defrag_lookup_extent() we have a value of
zero for the generation threshold ('newer_than' argument) since this is a
manual defrag. As a consequence we don't call defrag_get_extent() to get
an extent map representing a single file extent item in the inode's
subvolume tree, so we end up using the merged extent map at
defrag_collect_targets() and decide not to defrag.
Fix this by updating defrag_lookup_extent() to always discard extent maps
that were merged and call defrag_get_extent() regardless of the minimum
generation threshold ('newer_than' argument).
Filipe Manana [Mon, 28 Oct 2024 16:23:00 +0000 (16:23 +0000)]
btrfs: fix extent map merging not happening for adjacent extents
If we have 3 or more adjacent extents in a file, that is, consecutive file
extent items pointing to adjacent extents, within a contiguous file range
and compatible flags, we end up not merging all the extents into a single
extent map.
For example:
$ mkfs.btrfs -f /dev/sdc
$ mount /dev/sdc /mnt/sdc
After all the ordered extents complete we unpin the extent maps and try
to merge them, but instead of getting a single extent map we get two
because:
1) When the first ordered extent completes (file range [0, 64K)) we
unpin its extent map and attempt to merge it with the extent map for
the range [64K, 128K), but we can't because that extent map is still
pinned;
2) When the second ordered extent completes (file range [64K, 128K)), we
unpin its extent map and merge it with the previous extent map, for
file range [0, 64K), but we can't merge with the next extent map, for
the file range [128K, 192K), because this one is still pinned.
The merged extent map for the file range [0, 128K) gets the flag
EXTENT_MAP_MERGED set;
3) When the third ordered extent completes (file range [128K, 192K)), we
unpin its extent map and attempt to merge it with the previous extent
map, for file range [0, 128K), but we can't because that extent map
has the flag EXTENT_MAP_MERGED set (mergeable_maps() returns false
due to different flags) while the extent map for the range [128K, 192K)
doesn't have that flag set.
We also can't merge it with the next extent map, for file range
[192K, 256K), because that one is still pinned.
At this moment we have 3 extent maps:
One for file range [0, 128K), with the flag EXTENT_MAP_MERGED set.
One for file range [128K, 192K).
One for file range [192K, 256K) which is still pinned;
4) When the fourth and final extent completes (file range [192K, 256K)),
we unpin its extent map and attempt to merge it with the previous
extent map, for file range [128K, 192K), which succeeds since none
of these extent maps have the EXTENT_MAP_MERGED flag set.
So we end up with 2 extent maps:
One for file range [0, 128K), with the flag EXTENT_MAP_MERGED set.
One for file range [128K, 256K), with the flag EXTENT_MAP_MERGED set.
Since after merging extent maps we don't attempt to merge again, that
is, merge the resulting extent map with the one that is now preceding
it (and the one following it), we end up with those two extent maps,
when we could have had a single extent map to represent the whole file.
Fix this by making mergeable_maps() ignore the EXTENT_MAP_MERGED flag.
While this doesn't present any functional issue, it prevents the merging
of extent maps which allows to save memory, and can make defrag not
merging extents too (that will be addressed in the next patch).
Zhihao Cheng [Mon, 21 Oct 2024 14:02:15 +0000 (22:02 +0800)]
btrfs: fix use-after-free of block device file in __btrfs_free_extra_devids()
Mounting btrfs from two images (which have the same one fsid and two
different dev_uuids) in certain executing order may trigger an UAF for
variable 'device->bdev_file' in __btrfs_free_extra_devids(). And
following are the details:
1. Attach image_1 to loop0, attach image_2 to loop1, and scan btrfs
devices by ioctl(BTRFS_IOC_SCAN_DEV):
3. mount /dev/loop1 /mnt
btrfs_open_devices
btrfs_get_bdev_and_sb(&bdev_file)
// EIO, btrfs_device_1->bdev_file is not assigned,
// which points to a freed memory area
btrfs_device_2->bdev_file = btrfs_get_bdev_and_sb(loop1)
btrfs_fill_super
open_ctree
btrfs_free_extra_devids
if (btrfs_device_1->bdev_file)
fput(btrfs_device_1->bdev_file) // UAF !
Fix it by setting 'device->bdev_file' as 'NULL' after closing the
btrfs_device in btrfs_close_one_device().
Naohiro Aota [Wed, 9 Oct 2024 13:52:06 +0000 (22:52 +0900)]
btrfs: fix error propagation of split bios
The purpose of btrfs_bbio_propagate_error() shall be propagating an error
of split bio to its original btrfs_bio, and tell the error to the upper
layer. However, it's not working well on some cases.
* Case 1. Immediate (or quick) end_bio with an error
When btrfs sends btrfs_bio to mirrored devices, btrfs calls
btrfs_bio_end_io() when all the mirroring bios are completed. If that
btrfs_bio was split, it is from btrfs_clone_bioset and its end_io function
is btrfs_orig_write_end_io. For this case, btrfs_bbio_propagate_error()
accesses the orig_bbio's bio context to increase the error count.
That works well in most cases. However, if the end_io is called enough
fast, orig_bbio's (remaining part after split) bio context may not be
properly set at that time. Since the bio context is set when the orig_bbio
(the last btrfs_bio) is sent to devices, that might be too late for earlier
split btrfs_bio's completion. That will result in NULL pointer
dereference.
That bug is easily reproducible by running btrfs/146 on zoned devices [1]
and it shows the following trace.
[1] You need raid-stripe-tree feature as it create "-d raid0 -m raid1" FS.
* Case 2. Earlier completion of orig_bbio for mirrored btrfs_bios
btrfs_bbio_propagate_error() assumes the end_io function for orig_bbio is
called last among split bios. In that case, btrfs_orig_write_end_io() sets
the bio->bi_status to BLK_STS_IOERR by seeing the bioc->error [2].
Otherwise, the increased orig_bio's bioc->error is not checked by anyone
and return BLK_STS_OK to the upper layer.
[2] Actually, this is not true. Because we only increases orig_bioc->errors
by max_errors, the condition "atomic_read(&bioc->error) > bioc->max_errors"
is still not met if only one split btrfs_bio fails.
* Case 3. Later completion of orig_bbio for un-mirrored btrfs_bios
In contrast to the above case, btrfs_bbio_propagate_error() is not working
well if un-mirrored orig_bbio is completed last. It sets
orig_bbio->bio.bi_status to the btrfs_bio's error. But, that is easily
over-written by orig_bbio's completion status. If the status is BLK_STS_OK,
the upper layer would not know the failure.
* Solution
Considering the above cases, we can only save the error status in the
orig_bbio (remaining part after split) itself as it is always
available. Also, the saved error status should be propagated when all the
split btrfs_bios are finished (i.e, bbio->pending_ios == 0).
This commit introduces "status" to btrfs_bbio and saves the first error of
split bios to original btrfs_bio's "status" variable. When all the split
bios are finished, the saved status is loaded into original btrfs_bio's
status.
With this commit, btrfs/146 on zoned devices does not hit the NULL pointer
dereference anymore.
Yue Haibing [Tue, 22 Oct 2024 09:52:08 +0000 (17:52 +0800)]
btrfs: fix passing 0 to ERR_PTR in btrfs_search_dir_index_item()
The ret may be zero in btrfs_search_dir_index_item() and should not
passed to ERR_PTR(). Now btrfs_unlink_subvol() is the only caller to
this, reconstructed it to check ERR_PTR(-ENOENT) while ret >= 0.
This fixes smatch warnings:
fs/btrfs/dir-item.c:353
btrfs_search_dir_index_item() warn: passing zero to 'ERR_PTR'
btrfs: reject ro->rw reconfiguration if there are hard ro requirements
[BUG]
Syzbot reports the following crash:
BTRFS info (device loop0 state MCS): disabling free space tree
BTRFS info (device loop0 state MCS): clearing compat-ro feature flag for FREE_SPACE_TREE (0x1)
BTRFS info (device loop0 state MCS): clearing compat-ro feature flag for FREE_SPACE_TREE_VALID (0x2)
Oops: general protection fault, probably for non-canonical address 0xdffffc0000000003: 0000 [#1] PREEMPT SMP KASAN NOPTI
KASAN: null-ptr-deref in range [0x0000000000000018-0x000000000000001f]
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:backup_super_roots fs/btrfs/disk-io.c:1691 [inline]
RIP: 0010:write_all_supers+0x97a/0x40f0 fs/btrfs/disk-io.c:4041
Call Trace:
<TASK>
btrfs_commit_transaction+0x1eae/0x3740 fs/btrfs/transaction.c:2530
btrfs_delete_free_space_tree+0x383/0x730 fs/btrfs/free-space-tree.c:1312
btrfs_start_pre_rw_mount+0xf28/0x1300 fs/btrfs/disk-io.c:3012
btrfs_remount_rw fs/btrfs/super.c:1309 [inline]
btrfs_reconfigure+0xae6/0x2d40 fs/btrfs/super.c:1534
btrfs_reconfigure_for_mount fs/btrfs/super.c:2020 [inline]
btrfs_get_tree_subvol fs/btrfs/super.c:2079 [inline]
btrfs_get_tree+0x918/0x1920 fs/btrfs/super.c:2115
vfs_get_tree+0x90/0x2b0 fs/super.c:1800
do_new_mount+0x2be/0xb40 fs/namespace.c:3472
do_mount fs/namespace.c:3812 [inline]
__do_sys_mount fs/namespace.c:4020 [inline]
__se_sys_mount+0x2d6/0x3c0 fs/namespace.c:3997
do_syscall_x64 arch/x86/entry/common.c:52 [inline]
do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
entry_SYSCALL_64_after_hwframe+0x77/0x7f
[CAUSE]
To support mounting different subvolume with different RO/RW flags for
the new mount APIs, btrfs introduced two workaround to support this feature:
- Skip mount option/feature checks if we are mounting a different
subvolume
- Reconfigure the fs to RW if the initial mount is RO
Combining these two, we can have the following sequence:
- Mount the fs ro,rescue=all,clear_cache,space_cache=v1
rescue=all will mark the fs as hard read-only, so no v2 cache clearing
will happen.
- Mount a subvolume rw of the same fs.
We go into btrfs_get_tree_subvol(), but fc_mount() returns EBUSY
because our new fc is RW, different from the original fs.
Now we enter btrfs_reconfigure_for_mount(), which switches the RO flag
first so that we can grab the existing fs_info.
Then we reconfigure the fs to RW.
- During reconfiguration, option/features check is skipped
This means we will restart the v2 cache clearing, and convert back to
v1 cache.
This will trigger fs writes, and since the original fs has "rescue=all"
option, it skips the csum tree read.
And eventually causing NULL pointer dereference in super block
writeback.
[FIX]
For reconfiguration caused by different subvolume RO/RW flags, ensure we
always run btrfs_check_options() to ensure we have proper hard RO
requirements met.
In fact the function btrfs_check_options() doesn't really do many
complex checks, but hard RO requirement and some feature dependency
checks, thus there is no special reason not to do the check for mount
reconfiguration.
Boris Burkov [Fri, 18 Oct 2024 22:44:34 +0000 (15:44 -0700)]
btrfs: fix read corruption due to race with extent map merging
In debugging some corrupt squashfs files, we observed symptoms of
corrupt page cache pages but correct on-disk contents. Further
investigation revealed that the exact symptom was a correct page
followed by an incorrect, duplicate, page. This got us thinking about
extent maps.
commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them")
enforces a reference count on the primary `em` extent_map being merged,
as that one gets modified.
However, since,
commit 3d2ac9922465 ("btrfs: introduce new members for extent_map")
both 'em' and 'merge' get modified, which started modifying 'merge'
and thus introduced the same race.
We were able to reproduce this by looping the affected squashfs workload
in parallel on a bunch of separate btrfs-es while also dropping caches.
We are still working on a simple enough reproducer to make into an fstest.
The simplest fix is to stop modifying 'merge', which is not essential,
as it is dropped immediately after the merge. This behavior is simply
a consequence of the order of the two extent maps being important in
computing the new values. Modify merge_ondisk_extents to take prev and
next by const* and also take a third merged parameter that it puts the
results in. Note that this introduces the rather odd behavior of passing
'em' to merge_ondisk_extents as a const * and as a regular ptr.
Qu Wenruo [Tue, 8 Oct 2024 23:07:03 +0000 (09:37 +1030)]
btrfs: fix the delalloc range locking if sector size < page size
Inside lock_delalloc_folios(), there are several problems related to
sector size < page size handling:
- Set the writer locks without checking if the folio is still valid
We call btrfs_folio_start_writer_lock() just like it's folio_lock().
But since the folio may not even be the folio of the current mapping,
we can easily screw up the folio->private.
- The range is not clamped inside the page
This means we can over write other bitmaps if the start/len is not
properly handled, and trigger the btrfs_subpage_assert().
- @processed_end is always rounded up to page end
If the delalloc range is not page aligned, and we need to retry
(returning -EAGAIN), then we will unlock to the page end.
Thankfully this is not a huge problem, as now
btrfs_folio_end_writer_lock() can handle range larger than the locked
range, and only unlock what is already locked.
Fix all these problems by:
- Lock and check the folio first, then call
btrfs_folio_set_writer_lock()
So that if we got a folio not belonging to the inode, we won't
touch folio->private.
btrfs: qgroup: set a more sane default value for subtree drop threshold
Since commit 011b46c30476 ("btrfs: skip subtree scan if it's too high to
avoid low stall in btrfs_commit_transaction()"), btrfs qgroup can
automatically skip large subtree scan at the cost of marking qgroup
inconsistent.
It's designed to address the final performance problem of snapshot drop
with qgroup enabled, but to be safe the default value is
BTRFS_MAX_LEVEL, requiring a user space daemon to set a different value
to make it work.
I'd say it's not a good idea to rely on user space tool to set this
default value, especially when some operations (snapshot dropping) can
be triggered immediately after mount, leaving a very small window to
that that sysfs interface.
So instead of disabling this new feature by default, enable it with a
low threshold (3), so that large subvolume tree drop at mount time won't
cause huge qgroup workload.
Filipe Manana [Mon, 14 Oct 2024 15:14:18 +0000 (16:14 +0100)]
btrfs: clear force-compress on remount when compress mount option is given
After the migration to use fs context for processing mount options we had
a slight change in the semantics for remounting a filesystem that was
mounted with compress-force. Before we could clear compress-force by
passing only "-o compress[=algo]" during a remount, but after that change
that does not work anymore, force-compress is still present and one needs
to pass "-o compress-force=no,compress[=algo]" to the mount command.
Example, when running on a kernel 6.8+:
$ mount -o compress-force=zlib:9 /dev/sdi /mnt/sdi
$ mount | grep sdi
/dev/sdi on /mnt/sdi type btrfs (rw,relatime,compress-force=zlib:9,discard=async,space_cache=v2,subvolid=5,subvol=/)
$ mount -o remount,compress=zlib:5 /mnt/sdi
$ mount | grep sdi
/dev/sdi on /mnt/sdi type btrfs (rw,relatime,compress-force=zlib:5,discard=async,space_cache=v2,subvolid=5,subvol=/)
On a 6.7 kernel (or older):
$ mount -o compress-force=zlib:9 /dev/sdi /mnt/sdi
$ mount | grep sdi
/dev/sdi on /mnt/sdi type btrfs (rw,relatime,compress-force=zlib:9,discard=async,space_cache=v2,subvolid=5,subvol=/)
$ mount -o remount,compress=zlib:5 /mnt/sdi
$ mount | grep sdi
/dev/sdi on /mnt/sdi type btrfs (rw,relatime,compress=zlib:5,discard=async,space_cache=v2,subvolid=5,subvol=/)
So update btrfs_parse_param() to clear "compress-force" when "compress" is
given, providing the same semantics as kernel 6.7 and older.
Naohiro Aota [Tue, 1 Oct 2024 08:03:32 +0000 (17:03 +0900)]
btrfs: zoned: fix zone unusable accounting for freed reserved extent
When btrfs reserves an extent and does not use it (e.g, by an error), it
calls btrfs_free_reserved_extent() to free the reserved extent. In the
process, it calls btrfs_add_free_space() and then it accounts the region
bytes as block_group->zone_unusable.
However, it leaves the space_info->bytes_zone_unusable side not updated. As
a result, ENOSPC can happen while a space_info reservation succeeded. The
reservation is fine because the freed region is not added in
space_info->bytes_zone_unusable, leaving that space as "free". OTOH,
corresponding block group counts it as zone_unusable and its allocation
pointer is not rewound, we cannot allocate an extent from that block group.
That will also negate space_info's async/sync reclaim process, and cause an
ENOSPC error from the extent allocation process.
Fix that by returning the space to space_info->bytes_zone_unusable.
Ideally, since a bio is not submitted for this reserved region, we should
return the space to free space and rewind the allocation pointer. But, it
needs rework on extent allocation handling, so let it work in this way for
now.
Roi Martin [Thu, 10 Oct 2024 19:47:17 +0000 (21:47 +0200)]
btrfs: fix uninitialized pointer free on read_alloc_one_name() error
The function read_alloc_one_name() does not initialize the name field of
the passed fscrypt_str struct if kmalloc fails to allocate the
corresponding buffer. Thus, it is not guaranteed that
fscrypt_str.name is initialized when freeing it.
This is a follow-up to the linked patch that fixes the remaining
instances of the bug introduced by commit e43eec81c516 ("btrfs: use
struct qstr instead of name and namelen pairs").
Roi Martin [Wed, 9 Oct 2024 08:08:33 +0000 (10:08 +0200)]
btrfs: fix uninitialized pointer free in add_inode_ref()
The add_inode_ref() function does not initialize the "name" struct when
it is declared. If any of the following calls to "read_one_inode()
returns NULL,
dir = read_one_inode(root, parent_objectid);
if (!dir) {
ret = -ENOENT;
goto out;
}
inode = read_one_inode(root, inode_objectid);
if (!inode) {
ret = -EIO;
goto out;
}
then "name.name" would be freed on "out" before being initialized.
out:
...
kfree(name.name);
This issue was reported by Coverity with CID 1526744.
Filipe Manana [Tue, 24 Sep 2024 14:58:31 +0000 (15:58 +0100)]
btrfs: use sector numbers as keys for the dirty extents xarray
We are using the logical address ("bytenr") of an extent as the key for
qgroup records in the dirty extents xarray. This is a problem because the
xarrays use "unsigned long" for keys/indices, meaning that on a 32 bits
platform any extent starting at or beyond 4G is truncated, which is a too
low limitation as virtually everyone is using storage with more than 4G of
space. This means a "bytenr" of 4G gets truncated to 0, and so does 8G and
16G for example, resulting in incorrect qgroup accounting.
Fix this by using sector numbers as keys instead, that is, using keys that
match the logical address right shifted by fs_info->sectorsize_bits, which
is what we do for the fs_info->buffer_radix that tracks extent buffers
(radix trees also use an "unsigned long" type for keys). This also makes
the index space more dense which helps optimize the xarray (as mentioned
at Documentation/core-api/xarray.rst).
Fixes: 3cce39a8ca4e ("btrfs: qgroup: use xarray to track dirty extents in transaction") Reviewed-by: Qu Wenruo <[email protected]> Signed-off-by: Filipe Manana <[email protected]> Reviewed-by: David Sterba <[email protected]> Signed-off-by: David Sterba <[email protected]>
Naohiro Aota [Fri, 4 Oct 2024 04:53:35 +0000 (13:53 +0900)]
btrfs: fix clear_dirty and writeback ordering in submit_one_sector()
This commit is a replay of commit 6252690f7e1b ("btrfs: fix invalid
mapping of extent xarray state"). We need to call
btrfs_folio_clear_dirty() before btrfs_set_range_writeback(), so that
xarray DIRTY tag is cleared.
With a refactoring commit 8189197425e7 ("btrfs: refactor
__extent_writepage_io() to do sector-by-sector submission"), it screwed
up and the order is reversed and causing the same hang. Fix the ordering
now in submit_one_sector().
Fixes: 8189197425e7 ("btrfs: refactor __extent_writepage_io() to do sector-by-sector submission") Reviewed-by: Qu Wenruo <[email protected]> Reviewed-by: Johannes Thumshirn <[email protected]> Signed-off-by: Naohiro Aota <[email protected]> Signed-off-by: David Sterba <[email protected]>
Filipe Manana [Wed, 2 Oct 2024 14:02:56 +0000 (15:02 +0100)]
btrfs: zoned: fix missing RCU locking in error message when loading zone info
At btrfs_load_zone_info() we have an error path that is dereferencing
the name of a device which is a RCU string but we are not holding a RCU
read lock, which is incorrect.
Fix this by using btrfs_err_in_rcu() instead of btrfs_err().
The problem is there since commit 08e11a3db098 ("btrfs: zoned: load zone's
allocation offset"), back then at btrfs_load_block_group_zone_info() but
then later on that code was factored out into the helper
btrfs_load_zone_info() by commit 09a46725cc84 ("btrfs: zoned: factor out
per-zone logic from btrfs_load_block_group_zone_info").
Filipe Manana [Tue, 24 Sep 2024 13:39:19 +0000 (14:39 +0100)]
btrfs: fix missing error handling when adding delayed ref with qgroups enabled
When adding a delayed ref head, at delayed-ref.c:add_delayed_ref_head(),
if we fail to insert the qgroup record we don't error out, we ignore it.
In fact we treat it as if there was no error and there was already an
existing record - we don't distinguish between the cases where
btrfs_qgroup_trace_extent_nolock() returns 1, meaning a record already
existed and we can free the given record, and the case where it returns
a negative error value, meaning the insertion into the xarray that is
used to track records failed.
Effectively we end up ignoring that we are lacking qgroup record in the
dirty extents xarray, resulting in incorrect qgroup accounting.
Fix this by checking for errors and return them to the callers.
Fixes: 3cce39a8ca4e ("btrfs: qgroup: use xarray to track dirty extents in transaction") Reviewed-by: Qu Wenruo <[email protected]> Signed-off-by: Filipe Manana <[email protected]> Reviewed-by: David Sterba <[email protected]> Signed-off-by: David Sterba <[email protected]>
There are reports that system cannot suspend due to running trim because
the task responsible for trimming the device isn't able to finish in
time, especially since we have a free extent discarding phase, which can
trim a lot of unallocated space. There are no limits on the trim size
(unlike the block group part).
Since trime isn't a critical call it can be interrupted at any time,
in such cases we stop the trim, report the amount of discarded bytes and
return an error.
Per Qu Wenruo in case we have a very large disk, e.g. 8TiB device,
mostly empty although we will do the split according to our super block
locations, the last super block ends at 256G, we can submit a huge
discard for the range [256G, 8T), causing a large delay.
Split the space left to discard based on BTRFS_MAX_DISCARD_CHUNK_SIZE in
preparation of introduction of cancellation points to trim. The value
of the chunk size is arbitrary, it can be higher or derived from actual
device capabilities but we can't easily read that using
bio_discard_limit().
Leo Martins [Tue, 24 Sep 2024 23:42:29 +0000 (16:42 -0700)]
btrfs: disable rate limiting when debug enabled
Disable ratelimiting for btrfs_printk when CONFIG_BTRFS_DEBUG is
enabled. This allows for more verbose output which is often needed by
functions like btrfs_dump_space_info().
Filipe Manana [Tue, 1 Oct 2024 10:06:52 +0000 (11:06 +0100)]
btrfs: wait for fixup workers before stopping cleaner kthread during umount
During unmount, at close_ctree(), we have the following steps in this order:
1) Park the cleaner kthread - this doesn't destroy the kthread, it basically
halts its execution (wake ups against it work but do nothing);
2) We stop the cleaner kthread - this results in freeing the respective
struct task_struct;
3) We call btrfs_stop_all_workers() which waits for any jobs running in all
the work queues and then free the work queues.
Syzbot reported a case where a fixup worker resulted in a crash when doing
a delayed iput on its inode while attempting to wake up the cleaner at
btrfs_add_delayed_iput(), because the task_struct of the cleaner kthread
was already freed. This can happen during unmount because we don't wait
for any fixup workers still running before we call kthread_stop() against
the cleaner kthread, which stops and free all its resources.
Fix this by waiting for any fixup workers at close_ctree() before we call
kthread_stop() against the cleaner and run pending delayed iputs.
The stack traces reported by syzbot were the following:
BUG: KASAN: slab-use-after-free in __lock_acquire+0x77/0x2050 kernel/locking/lockdep.c:5065
Read of size 8 at addr ffff8880272a8a18 by task kworker/u8:3/52
Last potentially related work creation:
kasan_save_stack+0x3f/0x60 mm/kasan/common.c:47
__kasan_record_aux_stack+0xac/0xc0 mm/kasan/generic.c:541
__call_rcu_common kernel/rcu/tree.c:3086 [inline]
call_rcu+0x167/0xa70 kernel/rcu/tree.c:3190
context_switch kernel/sched/core.c:5318 [inline]
__schedule+0x184b/0x4ae0 kernel/sched/core.c:6675
schedule_idle+0x56/0x90 kernel/sched/core.c:6793
do_idle+0x56a/0x5d0 kernel/sched/idle.c:354
cpu_startup_entry+0x42/0x60 kernel/sched/idle.c:424
start_secondary+0x102/0x110 arch/x86/kernel/smpboot.c:314
common_startup_64+0x13e/0x147
The buggy address belongs to the object at ffff8880272a8000
which belongs to the cache task_struct of size 7424
The buggy address is located 2584 bytes inside of
freed 7424-byte region [ffff8880272a8000, ffff8880272a9d00)
[CAUSE]
The allocation failure happens at the start_transaction() inside
prepare_to_relocate(), and during the error handling we call
unset_reloc_control(), which makes fs_info->balance_ctl to be NULL.
Then we continue the error path cleanup in btrfs_balance() by calling
reset_balance_state() which will call del_balance_item() to fully delete
the balance item in the root tree.
However during the small window between set_reloc_contrl() and
unset_reloc_control(), we can have a subvolume tree update and created a
reloc_root for that subvolume.
Then we go into the final btrfs_commit_transaction() of
del_balance_item(), and into btrfs_update_reloc_root() inside
commit_fs_roots().
That function checks if fs_info->reloc_ctl is in the merge_reloc_tree
stage, but since fs_info->reloc_ctl is NULL, it results a NULL pointer
dereference.
[FIX]
Just add extra check on fs_info->reloc_ctl inside
btrfs_update_reloc_root(), before checking
fs_info->reloc_ctl->merge_reloc_tree.
That DEAD_RELOC_TREE handling is to prevent further modification to the
reloc tree during merge stage, but since there is no reloc_ctl at all,
we do not need to bother that.
Filipe Manana [Fri, 27 Sep 2024 09:50:12 +0000 (10:50 +0100)]
btrfs: send: fix invalid clone operation for file that got its size decreased
During an incremental send we may end up sending an invalid clone
operation, for the last extent of a file which ends at an unaligned offset
that matches the final i_size of the file in the send snapshot, in case
the file had its initial size (the size in the parent snapshot) decreased
in the send snapshot. In this case the destination will fail to apply the
clone operation because its end offset is not sector size aligned and it
ends before the current size of the file.
Sending the truncate operation always happens when we finish processing an
inode, after we process all its extents (and xattrs, names, etc). So fix
this by ensuring the file has a valid size before we send a clone
operation for an unaligned extent that ends at the final i_size of the
file. The size we truncate to matches the start offset of the clone range
but it could be any value between that start offset and the final size of
the file since the clone operation will expand the i_size if the current
size is smaller than the end offset. The start offset of the range was
chosen because it's always sector size aligned and avoids a truncation
into the middle of a page, which results in dirtying the page due to
filling part of it with zeroes and then making the clone operation at the
receiver trigger IO.
The following test reproduces the issue:
$ cat test.sh
#!/bin/bash
DEV=/dev/sdi
MNT=/mnt/sdi
mkfs.btrfs -f $DEV
mount $DEV $MNT
# Create a file with a size of 256K + 5 bytes, having two extents, one
# with a size of 128K and another one with a size of 128K + 5 bytes.
last_ext_size=$((128 * 1024 + 5))
xfs_io -f -d -c "pwrite -S 0xab -b 128K 0 128K" \
-c "pwrite -S 0xcd -b $last_ext_size 128K $last_ext_size" \
$MNT/foo
# Another file which we will later clone foo into, but initially with
# a larger size than foo.
xfs_io -f -c "pwrite -S 0xef 0 1M" $MNT/bar
btrfs subvolume snapshot -r $MNT/ $MNT/snap1
# Now resize bar and clone foo into it.
xfs_io -c "truncate 0" \
-c "reflink $MNT/foo" $MNT/bar
Filipe Manana [Wed, 25 Sep 2024 10:38:55 +0000 (11:38 +0100)]
btrfs: tracepoints: end assignment with semicolon at btrfs_qgroup_extent event class
While running checkpatch.pl against a patch that modifies the
btrfs_qgroup_extent event class, it complained about using a comma instead
of a semicolon:
$ ./scripts/checkpatch.pl qgroups/0003-btrfs-qgroups-remove-bytenr-field-from-struct-btrfs_.patch
WARNING: Possible comma where semicolon could be used
#215: FILE: include/trace/events/btrfs.h:1720:
+ __entry->bytenr = bytenr,
__entry->num_bytes = rec->num_bytes;
total: 0 errors, 1 warnings, 184 lines checked
So replace the comma with a semicolon to silence checkpatch and possibly
other tools. It also makes the code consistent with the rest.
Josef Bacik [Tue, 24 Sep 2024 20:50:22 +0000 (16:50 -0400)]
btrfs: drop the backref cache during relocation if we commit
Since the inception of relocation we have maintained the backref cache
across transaction commits, updating the backref cache with the new
bytenr whenever we COWed blocks that were in the cache, and then
updating their bytenr once we detected a transaction id change.
This works as long as we're only ever modifying blocks, not changing the
structure of the tree.
However relocation does in fact change the structure of the tree. For
example, if we are relocating a data extent, we will look up all the
leaves that point to this data extent. We will then call
do_relocation() on each of these leaves, which will COW down to the leaf
and then update the file extent location.
But, a key feature of do_relocation() is the pending list. This is all
the pending nodes that we modified when we updated the file extent item.
We will then process all of these blocks via finish_pending_nodes, which
calls do_relocation() on all of the nodes that led up to that leaf.
The purpose of this is to make sure we don't break sharing unless we
absolutely have to. Consider the case that we have 3 snapshots that all
point to this leaf through the same nodes, the initial COW would have
created a whole new path. If we did this for all 3 snapshots we would
end up with 3x the number of nodes we had originally. To avoid this we
will cycle through each of the snapshots that point to each of these
nodes and update their pointers to point at the new nodes.
Once we update the pointer to the new node we will drop the node we
removed the link for and all of its children via btrfs_drop_subtree().
This is essentially just btrfs_drop_snapshot(), but for an arbitrary
point in the snapshot.
The problem with this is that we will never reflect this in the backref
cache. If we do this btrfs_drop_snapshot() for a node that is in the
backref tree, we will leave the node in the backref tree. This becomes
a problem when we change the transid, as now the backref cache has
entire subtrees that no longer exist, but exist as if they still are
pointed to by the same roots.
In the best case scenario you end up with "adding refs to an existing
tree ref" errors from insert_inline_extent_backref(), where we attempt
to link in nodes on roots that are no longer valid.
Worst case you will double free some random block and re-use it when
there's still references to the block.
This is extremely subtle, and the consequences are quite bad. There
isn't a way to make sure our backref cache is consistent between
transid's.
In order to fix this we need to simply evict the entire backref cache
anytime we cross transid's. This reduces performance in that we have to
rebuild this backref cache every time we change transid's, but fixes the
bug.
This has existed since relocation was added, and is a pretty critical
bug. There's a lot more cleanup that can be done now that this
functionality is going away, but this patch is as small as possible in
order to fix the problem and make it easy for us to backport it to all
the kernels it needs to be backported to.
Followup series will dismantle more of this code and simplify relocation
drastically to remove this functionality.
We have a reproducer that reproduced the corruption within a few minutes
of running. With this patch it survives several iterations/hours of
running the reproducer.
NOCOW writes do not generate stripe_extent entries in the RAID stripe
tree, as the RAID stripe-tree feature initially was designed with a
zoned filesystem in mind and on a zoned filesystem, we do not allow NOCOW
writes. But the RAID stripe-tree feature is independent from the zoned
feature, so we must also do NOCOW writes for RAID stripe-tree filesystems.
Filipe Manana [Thu, 19 Sep 2024 21:20:34 +0000 (22:20 +0100)]
btrfs: send: fix buffer overflow detection when copying path to cache entry
Starting with commit c0247d289e73 ("btrfs: send: annotate struct
name_cache_entry with __counted_by()") we annotated the variable length
array "name" from the name_cache_entry structure with __counted_by() to
improve overflow detection. However that alone was not correct, because
the length of that array does not match the "name_len" field - it matches
that plus 1 to include the NUL string terminator, so that makes a
fortified kernel think there's an overflow and report a splat like this:
Fix this by not storing the NUL string terminator since we don't actually
need it for name cache entries, this way "name_len" corresponds to the
actual size of the "name" array. This requires marking the "name" array
field with __nonstring and using memcpy() instead of strcpy() as
recommended by the guidelines at:
Filipe Manana [Sun, 15 Sep 2024 19:52:53 +0000 (20:52 +0100)]
btrfs: fix use-after-free on rbtree that tracks inodes for auto defrag
When cleaning up defrag inodes at btrfs_cleanup_defrag_inodes(), called
during remount and unmount, we are freeing every node from the rbtree
that tracks inodes for auto defrag using
rbtree_postorder_for_each_entry_safe(), which doesn't modify the tree
itself. So once we unlock the lock that protects the rbtree, we have a
tree pointing to a root that was freed (and a root pointing to freed
nodes, and their children pointing to other freed nodes, and so on).
This makes further access to the tree result in a use-after-free with
unpredictable results.
Fix this by initializing the rbtree to an empty root after the call to
rbtree_postorder_for_each_entry_safe() and before unlocking.
btrfs: tree-checker: fix the wrong output of data backref objectid
[BUG]
There are some reports about invalid data backref objectids, the report
looks like this:
BTRFS critical (device sda): corrupt leaf: block=333654787489792 slot=110 extent bytenr=333413935558656 len=65536 invalid data ref objectid value 2543
The data ref objectid is the inode number inside the subvolume.
But in above case, the value is completely sane, not really showing the
problem.
[CAUSE]
The root cause of the problem is the deprecated feature, inode cache.
This feature results a special inode number, -12ULL, and it's no longer
recognized by tree-checker, triggering the error.
The direct problem here is the output of data ref objectid. The value
shown is in fact the dref_root (subvolume id), not the dref_objectid
(inode number).
[FIX]
Fix the output to use dref_objectid instead.
Filipe Manana [Tue, 3 Sep 2024 09:55:36 +0000 (10:55 +0100)]
btrfs: fix race setting file private on concurrent lseek using same fd
When doing concurrent lseek(2) system calls against the same file
descriptor, using multiple threads belonging to the same process, we have
a short time window where a race happens and can result in a memory leak.
The race happens like this:
1) A program opens a file descriptor for a file and then spawns two
threads (with the pthreads library for example), lets call them
task A and task B;
2) Task A calls lseek with SEEK_DATA or SEEK_HOLE and ends up at
file.c:find_desired_extent() while holding a read lock on the inode;
3) At the start of find_desired_extent(), it extracts the file's
private_data pointer into a local variable named 'private', which has
a value of NULL;
4) Task B also calls lseek with SEEK_DATA or SEEK_HOLE, locks the inode
in shared mode and enters file.c:find_desired_extent(), where it also
extracts file->private_data into its local variable 'private', which
has a NULL value;
5) Because it saw a NULL file private, task A allocates a private
structure and assigns to the file structure;
6) Task B also saw a NULL file private so it also allocates its own file
private and then assigns it to the same file structure, since both
tasks are using the same file descriptor.
At this point we leak the private structure allocated by task A.
Besides the memory leak, there's also the detail that both tasks end up
using the same cached state record in the private structure (struct
btrfs_file_private::llseek_cached_state), which can result in a
use-after-free problem since one task can free it while the other is
still using it (only one task took a reference count on it). Also, sharing
the cached state is not a good idea since it could result in incorrect
results in the future - right now it should not be a problem because it
end ups being used only in extent-io-tree.c:count_range_bits() where we do
range validation before using the cached state.
Fix this by protecting the private assignment and check of a file while
holding the inode's spinlock and keep track of the task that allocated
the private, so that it's used only by that task in order to prevent
user-after-free issues with the cached state record as well as potentially
using it incorrectly in the future.
btrfs: only unlock the to-be-submitted ranges inside a folio
[SUBPAGE COMPRESSION LIMITS]
Currently inside writepage_delalloc(), if a delalloc range is going to
be submitted asynchronously (inline or compression, the page
dirty/writeback/unlock are all handled in at different time, not at the
submission time), then we return 1 and extent_writepage() will skip the
submission.
This is fine if every sector matches page size, but if a sector is
smaller than page size (aka, subpage case), then it can be very
problematic, for example for the following 64K page:
0 16K 32K 48K 64K
|/| |///////| |/|
| |
4K 52K
Where |/| is the dirty range we need to submit.
In the above case, we need the following different handling for the 3
ranges:
- [0, 4K) needs to be submitted for regular write
A single sector cannot be compressed.
- [16K, 32K) needs to be submitted for compressed write
- [48K, 52K) needs to be submitted for regular write.
Above, if we try to submit [16K, 32K) for compressed write, we will
return 1 and immediately, and without submitting the remaining
[48K, 52K) range.
Furthermore, since extent_writepage() will exit without unlocking any
sectors, the submitted range [0, 4K) will not have sector unlocked.
That's the reason why for now subpage is only allowed for full page
range.
[ENHANCEMENT]
- Introduce a submission bitmap at btrfs_bio_ctrl::submit_bitmap
This records which sectors will be submitted by extent_writepage_io().
This allows us to track which sectors needs to be submitted thus later
to be properly unlocked.
For asynchronously submitted range (inline/compression), the
corresponding bits will be cleared from that bitmap.
- Only return 1 if no sector needs to be submitted in
writepage_delalloc()
- Only submit sectors marked by submission bitmap inside
extent_writepage_io()
So we won't touch the asynchronously submitted part.
- Introduce btrfs_folio_end_writer_lock_bitmap() helper
This will only unlock the involved sectors specified by @bitmap
parameter, to avoid touching the range asynchronously submitted.
Please note that, since subpage compression is still limited to page
aligned range, this change is only a preparation for future sector
perfect compression support for subpage.
btrfs: merge btrfs_folio_unlock_writer() into btrfs_folio_end_writer_lock()
The function btrfs_folio_unlock_writer() is already calling
btrfs_folio_end_writer_lock() to do the heavy lifting work, the only
missing 0 writer check.
Thus there is no need to keep two different functions, move the 0 writer
check into btrfs_folio_end_writer_lock(), and remove
btrfs_folio_unlock_writer().
Leo Martins [Tue, 3 Sep 2024 18:19:05 +0000 (11:19 -0700)]
btrfs: DEFINE_FREE for struct btrfs_path
Add a DEFINE_FREE for struct btrfs_path. This defines a function that
can be called using the __free attribute. Define a macro
BTRFS_PATH_AUTO_FREE to make the declaration of an auto freeing path
very clear.
The intended use is to define the auto free of path in cases where the
path is allocated somewhere at the beginning and freed either on all
error paths or at the end of the function.
Qu Wenruo [Fri, 30 Aug 2024 07:18:20 +0000 (16:48 +0930)]
btrfs: remove btrfs_folio_end_all_writers()
The function btrfs_folio_end_all_writers() is only utilized in
extent_writepage() as a way to unlock all subpage range (for both
successful submission and error handling).
Meanwhile we have a similar function, btrfs_folio_end_writer_lock().
The difference is, btrfs_folio_end_writer_lock() expects a range that is
a subset of the already locked range.
This limit on btrfs_folio_end_writer_lock() is a little overkilled,
preventing it from being utilized for error paths.
So here we enhance btrfs_folio_end_writer_lock() to accept a superset of
the locked range, and only end the locked subset.
This means we can replace btrfs_folio_end_all_writers() with
btrfs_folio_end_writer_lock() instead.
David Sterba [Wed, 26 Jun 2024 21:39:11 +0000 (23:39 +0200)]
btrfs: constify more pointer parameters
Continue adding const to parameters. This is for clarity and minor
addition to safety. There are some minor effects, in the assembly code
and .ko measured on release config.
David Sterba [Wed, 5 Jun 2024 00:28:30 +0000 (02:28 +0200)]
btrfs: rework BTRFS_I as macro to preserve parameter const
Currently BTRFS_I is a static inline function that takes a const inode
and returns btrfs inode, dropping the 'const' qualifier. This can break
assumptions of compiler though it seems there's no real case.
To make the parameter and return type consistent regardint const we can
use the container_of_const() that preserves it. However this would not
check the parameter type. To fix that use the same _Generic construct
but implement only the two expected types.
Filipe Manana [Thu, 29 Aug 2024 18:03:52 +0000 (19:03 +0100)]
btrfs: add and use helper to verify the calling task has locked the inode
We have a few places that check if we have the inode locked by doing:
ASSERT(inode_is_locked(vfs_inode));
This actually proved to be useful several times as if assertions are
enabled (and by default they are in many distros) it immediately triggers
a crash which is impossible for users to miss.
However that doesn't check if the lock is held by the calling task, so
the check passes if some other task locked the inode.
Using one of the lockdep functions to check the lock is held, like
lockdep_assert_held() for example, does check that the calling task
holds the lock, and if that's not the case it produces a warning and
stack trace in dmesg. However, despite the misleading "assert" in the
name of the lockdep helpers, it does not trigger a crash/BUG_ON(), just
a warning and splat in dmesg, which is easy to get unnoticed by users
who may have lockdep enabled.
So add a helper that does the ASSERT() and calls lockdep_assert_held()
immediately after and use it every where we check the inode is locked.
Like this if the lock is held by some other task we get the warning
in dmesg which is caught by fstests, very helpful during development,
and may also be occassionaly noticed by users with lockdep enabled.
Li Zetao [Wed, 28 Aug 2024 18:29:08 +0000 (02:29 +0800)]
btrfs: convert copy_inline_to_page() to use folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Moreover find_or_create_page() is compatible API, and it can
replaced with __filemap_get_folio(). Some interfaces have been converted
to use folio before, so the conversion operation from page can be
eliminated here.
Li Zetao [Wed, 28 Aug 2024 18:29:07 +0000 (02:29 +0800)]
btrfs: convert btrfs_decompress() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Based on the previous patch, the compression path can be
directly used in folio without converting to page.
Li Zetao [Wed, 28 Aug 2024 18:29:06 +0000 (02:29 +0800)]
btrfs: convert zstd_decompress() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. And memcpy_to_page() can be replaced with memcpy_to_folio().
But there is no memzero_folio(), but it can be replaced equivalently by
folio_zero_range().
Li Zetao [Wed, 28 Aug 2024 18:29:05 +0000 (02:29 +0800)]
btrfs: convert lzo_decompress() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. And memcpy_to_page() can be replaced with memcpy_to_folio().
But there is no memzero_folio(), but it can be replaced equivalently by
folio_zero_range().
Li Zetao [Wed, 28 Aug 2024 18:29:04 +0000 (02:29 +0800)]
btrfs: convert zlib_decompress() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. And memcpy_to_page() can be replaced with memcpy_to_folio().
But there is no memzero_folio(), but it can be replaced equivalently by
folio_zero_range().
Li Zetao [Wed, 28 Aug 2024 18:29:03 +0000 (02:29 +0800)]
btrfs: convert try_release_extent_mapping() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. And page_to_inode() can be replaced with folio_to_inode() now.
Li Zetao [Wed, 28 Aug 2024 18:29:02 +0000 (02:29 +0800)]
btrfs: convert try_release_extent_state() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Moreover, use folio_pos() instead of page_offset(),
which is more consistent with folio usage.
Li Zetao [Wed, 28 Aug 2024 18:29:01 +0000 (02:29 +0800)]
btrfs: convert submit_eb_page() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio.
Li Zetao [Wed, 28 Aug 2024 18:29:00 +0000 (02:29 +0800)]
btrfs: convert submit_eb_subpage() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Moreover, use folio_pos() instead of page_offset(),
which is more consistent with folio usage.
Li Zetao [Wed, 28 Aug 2024 18:28:59 +0000 (02:28 +0800)]
btrfs: convert read_key_bytes() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Moreover, use kmap_local_folio() instead of kmap_local_page(),
which is more consistent with folio usage.
Li Zetao [Wed, 28 Aug 2024 18:28:58 +0000 (02:28 +0800)]
btrfs: convert try_release_extent_buffer() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio.
Li Zetao [Wed, 28 Aug 2024 18:28:57 +0000 (02:28 +0800)]
btrfs: convert try_release_subpage_extent_buffer() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. And use folio_pos instead of page_offset, which is more
consistent with folio usage. At the same time, folio_test_private() can
handle folio directly without converting from page to folio first.
Li Zetao [Wed, 28 Aug 2024 18:28:56 +0000 (02:28 +0800)]
btrfs: convert get_next_extent_buffer() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Use folio_pos instead of page_offset, which is more
consistent with folio usage.
Li Zetao [Wed, 28 Aug 2024 18:28:55 +0000 (02:28 +0800)]
btrfs: convert clear_page_extent_mapped() to take a folio
The old page API is being gradually replaced and converted to use folio
to improve code readability and avoid repeated conversion between page
and folio. Now clear_page_extent_mapped() can deal with a folio
directly, so change its name to clear_folio_extent_mapped().
Qu Wenruo [Wed, 29 May 2024 07:33:47 +0000 (17:03 +0930)]
btrfs: make compression path to be subpage compatible
Currently btrfs compression path is not really subpage compatible, every
thing is still done in page unit.
That's fine for regular sector size and subpage routine. As even for
subpage routine compression is only enabled if the whole range is page
aligned, so reading the page cache in page unit is totally fine.
However in preparation for the future subpage perfect compression
support, we need to change the compression routine to properly handle a
subpage range.
This patch would prepare both zlib and zstd to only read the subpage
range for compression.
Lzo is already doing subpage aware read, as lzo's on-disk format is
already sectorsize dependent.
Qu Wenruo [Sat, 24 Aug 2024 10:06:43 +0000 (19:36 +0930)]
btrfs: merge btrfs_orig_bbio_end_io() into btrfs_bio_end_io()
There are only two differences between the two functions:
- btrfs_orig_bbio_end_io() does extra error propagation
This is mostly to allow tolerance for write errors.
- btrfs_orig_bbio_end_io() does extra pending_ios check
This check can handle both the original bio, or the cloned one.
(All accounting happens in the original one).
This makes btrfs_orig_bbio_end_io() a much safer call.
In fact we already had a double freeing error due to usage of
btrfs_bio_end_io() in the error path of btrfs_submit_chunk().
So just move the whole content of btrfs_orig_bbio_end_io() into
btrfs_bio_end_io().
For normal paths this brings no change, because they are already calling
btrfs_orig_bbio_end_io() in the first place.
For error paths (not only inside bio.c but also external callers), this
change will introduce extra checks, especially for external callers, as
they will error out without submitting the btrfs bio.
But considering it's already in the error path, such slower but much
safer checks are still an overall win.
Josef Bacik [Fri, 16 Aug 2024 19:16:24 +0000 (15:16 -0400)]
btrfs: do not hold the extent lock for entire read
Historically we've held the extent lock throughout the entire read.
There's been a few reasons for this, but it's mostly just caused us
problems. For example, this prevents us from allowing page faults
during direct io reads, because we could deadlock. This has forced us
to only allow 4k reads at a time for io_uring NOWAIT requests because we
have no idea if we'll be forced to page fault and thus have to do a
whole lot of work.
On the buffered side we are protected by the page lock, as long as we're
reading things like buffered writes, punch hole, and even direct IO to a
certain degree will get hung up on the page lock while the page is in
flight.
On the direct side we have the dio extent lock, which acts much like the
way the extent lock worked previously to this patch, however just for
direct reads. This protects direct reads from concurrent direct writes,
while we're protected from buffered writes via the inode lock.
Now that we're protected in all cases, narrow the extent lock to the
part where we're getting the extent map to submit the reads, no longer
holding the extent lock for the entire read operation. Push the extent
lock down into do_readpage() so that we're only grabbing it when looking
up the extent map. This portion was contributed by Goldwyn.
Josef Bacik [Mon, 19 Aug 2024 20:50:01 +0000 (16:50 -0400)]
btrfs: take the dio extent lock during O_DIRECT operations
Currently we hold the extent lock for the entire duration of a read.
This isn't really necessary in the buffered case, we're protected by the
page lock, however it's necessary for O_DIRECT.
For O_DIRECT reads, if we only locked the extent for the part where we
get the extent, we could potentially race with an O_DIRECT write in the
same region. This isn't really a problem, unless the read is delayed so
much that the write does the COW, unpins the old extent, and some other
application re-allocates the extent before the read is actually able to
be submitted. At that point at best we'd have a checksum mismatch, but
at worse we could read data that doesn't belong to us.
To address this potential race we need to make sure we don't have
overlapping, concurrent direct io reads and writes.
To accomplish this use the new EXTENT_DIO_LOCKED bit in the direct IO
case in the same spot as the current extent lock. The writes will take
this while they're creating the ordered extent, which is also used to
make sure concurrent buffered reads or concurrent direct reads are not
allowed to occur, and drop it after the ordered extent is taken. For
reads it will act as the current read behavior for the EXTENT_LOCKED
bit, we set it when we're starting the read, we clear it in the end_io
to allow other direct writes to continue.
This still has the drawback of disallowing concurrent overlapping direct
reads from occurring, but that exists with the current extent locking.
Josef Bacik [Mon, 19 Aug 2024 18:15:52 +0000 (14:15 -0400)]
btrfs: introduce EXTENT_DIO_LOCKED
In order to support dropping the extent lock during a read we need a way
to make sure that direct reads and direct writes for overlapping ranges
are protected from each other. To accomplish this introduce another
lock bit specifically for direct io. Subsequent patches will utilize
this to protect direct IO operations.
David Sterba [Tue, 27 Aug 2024 02:26:51 +0000 (04:26 +0200)]
btrfs: always pass readahead state to defrag
Defrag ioctl passes readahead from the file, but autodefrag does not
have a file so the readahead state is allocated when needed.
The autodefrag loop in cleaner thread iterates over inodes so we can
simply provide an on-stack readahead state and will not need to allocate
it in btrfs_defrag_file(). The size is 32 bytes which is acceptable.
David Sterba [Tue, 27 Aug 2024 02:10:11 +0000 (04:10 +0200)]
btrfs: return void from btrfs_add_inode_defrag()
The potential memory allocation failure is not a fatal error, skipping
autodefrag is fine and the caller inode_should_defrag() does not care
about the errors. Further writes can attempt to add the inode back to
the defragmentation list again.
David Sterba [Tue, 27 Aug 2024 02:05:48 +0000 (04:05 +0200)]
btrfs: clear defragmented inodes using postorder in btrfs_cleanup_defrag_inodes()
btrfs_cleanup_defrag_inodes() is not called frequently, only in remount
or unmount, but the way it frees the inodes in fs_info->defrag_inodes
is inefficient. Each time it needs to locate first node, remove it,
potentially rebalance tree until it's done. This allows to do a
conditional reschedule.
For cleanups the rbtree_postorder_for_each_entry_safe() iterator is
convenient but we can't reschedule and restart iteration because some of
the tree nodes would be already freed.
The cleanup operation is kmem_cache_free() which will likely take the
fast path for most objects so rescheduling should not be necessary.
Qu Wenruo [Wed, 14 Aug 2024 01:20:21 +0000 (10:50 +0930)]
btrfs: remove the nr_ret parameter from __extent_writepage_io()
The parameter @nr_ret is used to tell the caller how many sectors have
been submitted for IO.
Then callers check @nr_ret value to determine if we need to manually
clear the PAGECACHE_TAG_DIRTY, as if we submitted no sector (e.g. all
sectors are beyond i_size) there is no folio_start_writeback() called thus
PAGECACHE_TAG_DIRTY tag will not be cleared.
Remove this parameter by:
- Moving the btrfs_folio_clear_writeback() call into
__extent_writepage_io()
So that if we didn't submit any IO, then manually call
btrfs_folio_set_writeback() to clear PAGECACHE_TAG_DIRTY when
the page is no longer dirty.
- Use a bool to record if we have submitted any sector
Instead of an int.
- Use subpage compatible helpers to end folio writeback.
This brings no change to the behavior, just for the sake of consistency.
As for the call site inside __extent_writepage(), we're always called
for the whole page, so the existing full page helper
folio_(start|end)_writeback() is totally fine.
For the call site inside extent_write_locked_range(), although we can
have subpage range, folio_start_writeback() will only clear
PAGECACHE_TAG_DIRTY if the page is no longer dirty, and the full folio
will still be dirty if there is any subpage dirty range.
Only when the last dirty subpage sector is cleared, the
folio_start_writeback() will clear PAGECACHE_TAG_DIRTY.
So no matter if we call the full page or subpage helper, the result
is still the same, then just use the subpage helpers for consistency.
Junchao Sun [Fri, 7 Jun 2024 14:30:21 +0000 (22:30 +0800)]
btrfs: qgroup: use xarray to track dirty extents in transaction
Use xarray to track dirty extents to reduce the size of the struct
btrfs_qgroup_extent_record from 64 bytes to 40 bytes. The xarray is
more cache line friendly, it also reduces the complexity of insertion
and search code compared to rb tree.
Another change introduced is about error handling. Before this patch,
the result of btrfs_qgroup_trace_extent_nolock() is always a success. In
this patch, because of this function calls the function xa_store() which
has the possibility to fail, so mark qgroup as inconsistent if error
happened and then free preallocated memory. Also we preallocate memory
before spin_lock(), if memory preallcation failed, error handling is the
same the existing code.
Qu Wenruo [Wed, 7 Aug 2024 05:01:54 +0000 (14:31 +0930)]
btrfs: refactor __extent_writepage_io() to do sector-by-sector submission
Unlike the bitmap usage inside raid56, for __extent_writepage_io() we
handle the subpage submission not sector-by-sector, but for each dirty
range we found.
This is not a big deal normally, as the subpage complex code is already
mostly optimized out by the compiler for x86_64.
However for the sake of consistency and for the future of subpage
sector-perfect compression support, this patch does:
- Extract the sector submission code into submit_one_sector()
- Add the needed code to extract the dirty bitmap for subpage case
There is a small pitfall for non-subpage case, as we cleared page
dirty before starting writeback, so we have to manually set
the default dirty_bitmap to 1 for such case.
- Use bitmap_and() to calculate the target sectors we need to submit
This is done for both subpage and non-subpage cases, and will later
be expanded to skip inline/compression ranges.
For x86_64, the dirty bitmap will be fixed to 1, with the length of 1,
so we're still doing the same workload per sector.
For larger page sizes, the overhead will be a little larger, as previous
we only need to do one extent_map lookup per-dirty-range, but now it
will be one extent_map lookup per-sector.
But that is the same frequency as x86_64, so we're just aligning the
behavior to x86_64.
Qu Wenruo [Fri, 30 Aug 2024 07:05:48 +0000 (16:35 +0930)]
btrfs: subpage: fix the bitmap dump which can cause bitmap corruption
In commit 75258f20fb70 ("btrfs: subpage: dump extra subpage bitmaps for
debug") an internal macro GET_SUBPAGE_BITMAP() is introduced to grab the
bitmap of each attribute.
But that commit is using bitmap_cut() which will do the left shift of
the larger bitmap, causing incorrect values.
Thankfully this bitmap_cut() is only called for debug usage, and so far
it's not yet causing problem.
Fix it to use bitmap_read() to only grab the desired sub-bitmap.
btrfs: reduce chunk_map lookups in btrfs_map_block()
Currently we're calling btrfs_num_copies() before btrfs_get_chunk_map() in
btrfs_map_block(). But btrfs_num_copies() itself does a chunk map lookup
to be able to calculate the number of copies.
So split out the code getting the number of copies from btrfs_num_copies()
into a helper called btrfs_chunk_map_num_copies() and directly call it
from btrfs_map_block() and btrfs_num_copies().
This saves us one rbtree lookup per btrfs_map_block() invocation.
Filipe Manana [Wed, 7 Aug 2024 16:13:58 +0000 (17:13 +0100)]
btrfs: directly wake up cleaner kthread in the BTRFS_IOC_SYNC ioctl
The BTRFS_IOC_SYNC ioctl wants to wake up the cleaner kthread so that it
does any pending work (subvolume deletion, delayed iputs, etc), however
it is waking up the transaction kthread, which in turn wakes up the
cleaner. Since we don't have any transaction to commit, as any ongoing
transaction was already committed when it called btrfs_sync_fs() and
the goal is just to wake up the cleaner thread, directly wake up the
cleaner instead of the transaction kthread.
Qu Wenruo [Mon, 5 Aug 2024 05:32:54 +0000 (15:02 +0930)]
btrfs: make btrfs_is_subpage() to return false directly for 4K page size
Btrfs only supports sectorsize 4K, 8K, 16K, 32K, 64K for now, thus for
systems with 4K page size, there is no way the fs is subpage (sectorsize
< PAGE_SIZE).
So here we define btrfs_is_subpage() different according to the
PAGE_SIZE:
- PAGE_SIZE > 4K
We may hit real subpage cases, define btrfs_is_subpage() as a regular
function and do the usual checks.
- PAGE_SIZE == 4K (no smaller PAGE_SIZE support AFAIK)
There is no way the fs is subpage, so just define btrfs_is_subpage()
as an inline function which always return false.
This saves about 7K bytes for x86_64 debug builds:
text data bss dec hex filename
Before: 1484452 168693 25776 1678921 199e49 fs/btrfs/btrfs.ko
After: 1476605 168445 25776 1670826 197eaa fs/btrfs/btrfs.ko
btrfs: don't readahead the relocation inode on RST
On relocation we're doing readahead on the relocation inode, but if the
filesystem is backed by a RAID stripe tree we can get ENOENT (e.g. due to
preallocated extents not being mapped in the RST) from the lookup.
But readahead doesn't handle the error and submits invalid reads to the
device, causing an assertion in the scatter-gather list code:
btrfs: rename btrfs_io_stripe::is_scrub to rst_search_commit_root
Rename 'btrfs_io_stripe::is_scrub' to 'rst_search_commit_root'. While
'is_scrub' describes the state of the io_stripe (it is a stripe submitted
by scrub) it does not describe the purpose, namely looking at the commit
root when searching RAID stripe-tree entries.
Renaming the stripe to rst_search_commit_root describes this purpose.
Boris Burkov [Wed, 31 Jul 2024 19:41:06 +0000 (12:41 -0700)]
btrfs: add comment about locking in cow_file_range_inline()
Add a comment to document the complicated locked_page unlock logic in
cow_file_range_inline. The specifically tricky part is that a caller
just up the stack converts ret == 0 to ret == 1 and then another
caller far up the callstack handles ret == 1 as a success, AND returns
without cleanup in that case, both of which "feel" unnatural and led to
the original bug.
Try to document that somewhat specific callstack logic here to explain
the weird un-setting of locked_folio on success.
Filipe Manana [Thu, 25 Jul 2024 10:48:10 +0000 (11:48 +0100)]
btrfs: more efficient chunk map iteration when device replace finishes
When iterating the chunk maps when a device replace finishes we are doing
a full rbtree search for each chunk map, which is not the most efficient
thing to do, wasting CPU time. As we are holding a write lock on the tree
during the whole iteration, we can simply start from the first node in the
tree and then move to the next chunk map by doing a rb_next() call - the
only exception is when we need to reschedule, in which case we have to do
a full rbtree search since we dropped the write lock and the tree may have
changed (chunk maps may have been removed and the tree got rebalanced).
So just do that.
Filipe Manana [Thu, 25 Jul 2024 09:46:01 +0000 (10:46 +0100)]
btrfs: reschedule when updating chunk maps at the end of a device replace
At the end of a device replace we must go over all the chunk maps and
update their stripes to point to the target device instead of the source
device. We iterate over the chunk maps while holding a write lock and
we never reschedule, which can result in monopolizing a CPU for too long
and blocking readers for too long (it's a rw lock, non-blocking).
So improve on this by rescheduling if necessary. This is safe because at
this point we are holding the chunk mutex, which means no new chunks can
be allocated and therefore we don't risk missing a new chunk map that
covers a range behind the last one we processed before rescheduling.
Josef Bacik [Thu, 25 Jul 2024 00:22:32 +0000 (20:22 -0400)]
btrfs: convert btrfs_set_range_writeback() to use a folio
We already use a lot of functions here that use folios, update the
function to use __filemap_get_folio instead of find_get_page and then
use the folio directly.
projects / linux.git / log