At did_overwrite_ref() we always call get_inode_gen() to find out the
generation of the inode 'ow_inode'. However we don't always need to use
that generation, and in fact it's very common to not use it, so we end
up doing a b+tree search on the send root, allocating a path, etc, for
nothing. So improve on this by getting the generation only if we need
to use it.
This patch is part of a larger patchset and the changelog of the last
patch in the series contains a sample performance test and results.
The patches that comprise the patchset are the following:
btrfs: send: directly return from did_overwrite_ref() and simplify it
btrfs: send: avoid unnecessary generation search at did_overwrite_ref()
btrfs: send: directly return from will_overwrite_ref() and simplify it
btrfs: send: avoid extra b+tree searches when checking reference overrides
btrfs: send: remove send_progress argument from can_rmdir()
btrfs: send: avoid duplicated orphan dir allocation and initialization
btrfs: send: avoid unnecessary orphan dir rbtree search at can_rmdir()
btrfs: send: reduce searches on parent root when checking if dir can be removed
btrfs: send: iterate waiting dir move rbtree only once when processing refs
btrfs: send: initialize all the red black trees earlier
btrfs: send: genericize the backref cache to allow it to be reused
btrfs: adapt lru cache to allow for 64 bits keys on 32 bits systems
btrfs: send: cache information about created directories
btrfs: allow a generation number to be associated with lru cache entries
btrfs: add an api to delete a specific entry from the lru cache
btrfs: send: use the lru cache to implement the name cache
btrfs: send: update size of roots array for backref cache entries
btrfs: send: cache utimes operations for directories if possible
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are no resources to release before did_overwrite_ref() returns, so
we don't really need the 'out' label and jumping to it when conditions are
met - we can directly return and get rid of the label and jumps. Also we
can deal with -ENOENT and other errors in a single if-else logic, as it's
more straightforward.
This helps the next patch in the series to be more simple as well.
This patch is part of a larger patchset and the changelog of the last
patch in the series contains a sample performance test and results.
The patches that comprise the patchset are the following:
btrfs: send: directly return from did_overwrite_ref() and simplify it
btrfs: send: avoid unnecessary generation search at did_overwrite_ref()
btrfs: send: directly return from will_overwrite_ref() and simplify it
btrfs: send: avoid extra b+tree searches when checking reference overrides
btrfs: send: remove send_progress argument from can_rmdir()
btrfs: send: avoid duplicated orphan dir allocation and initialization
btrfs: send: avoid unnecessary orphan dir rbtree search at can_rmdir()
btrfs: send: reduce searches on parent root when checking if dir can be removed
btrfs: send: iterate waiting dir move rbtree only once when processing refs
btrfs: send: initialize all the red black trees earlier
btrfs: send: genericize the backref cache to allow it to be reused
btrfs: adapt lru cache to allow for 64 bits keys on 32 bits systems
btrfs: send: cache information about created directories
btrfs: allow a generation number to be associated with lru cache entries
btrfs: add an api to delete a specific entry from the lru cache
btrfs: send: use the lru cache to implement the name cache
btrfs: send: update size of roots array for backref cache entries
btrfs: send: cache utimes operations for directories if possible
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
Since the introduction of per-fs feature sysfs interface
(/sys/fs/btrfs/<UUID>/features/), the content of that directory is never
updated.
Thus for the following case, that directory will not show the new
features like RAID56:
# mkfs.btrfs -f $dev1 $dev2 $dev3
# mount $dev1 $mnt
# btrfs balance start -f -mconvert=raid5 $mnt
# ls /sys/fs/btrfs/$uuid/features/
extended_iref free_space_tree no_holes skinny_metadata
While after unmount and mount, we got the correct features:
# umount $mnt
# mount $dev1 $mnt
# ls /sys/fs/btrfs/$uuid/features/
extended_iref free_space_tree no_holes raid56 skinny_metadata
[CAUSE]
Because we never really try to update the content of per-fs features/
directory.
We had an attempt to update the features directory dynamically in commit
14e46e0495 ("btrfs: synchronize incompat feature bits with sysfs
files"), but unfortunately it get reverted in commit e410e34fad
("Revert "btrfs: synchronize incompat feature bits with sysfs files"").
The problem in the original patch is, in the context of
btrfs_create_chunk(), we can not afford to update the sysfs group.
The exported but never utilized function, btrfs_sysfs_feature_update()
is the leftover of such attempt. As even if we go sysfs_update_group(),
new files will need extra memory allocation, and we have no way to
specify the sysfs update to go GFP_NOFS.
[FIX]
This patch will address the old problem by doing asynchronous sysfs
update in the cleaner thread.
This involves the following changes:
- Make __btrfs_(set|clear)_fs_(incompat|compat_ro) helpers to set
BTRFS_FS_FEATURE_CHANGED flag when needed
- Update btrfs_sysfs_feature_update() to use sysfs_update_group()
And drop unnecessary arguments.
- Call btrfs_sysfs_feature_update() in cleaner_kthread
If we have the BTRFS_FS_FEATURE_CHANGED flag set.
- Wake up cleaner_kthread in btrfs_commit_transaction if we have
BTRFS_FS_FEATURE_CHANGED flag
By this, all the previously dangerous call sites like
btrfs_create_chunk() need no new changes, as above helpers would
have already set the BTRFS_FS_FEATURE_CHANGED flag.
The real work happens at cleaner_kthread, thus we pay the cost of
delaying the update to sysfs directory, but the delayed time should be
small enough that end user can not distinguish though it might get
delayed if the cleaner thread is busy with removing subvolumes or
defrag.
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
extent-tree.h is included more than once, added in a0231804af ("btrfs:
move extent-tree helpers into their own header file").
Signed-off-by: ye xingchen <ye.xingchen@zte.com.cn>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
When debugging a scrub related metadata error, it turns out that our
metadata error reporting is not ideal.
The only 3 error messages are:
- BTRFS error (device dm-2): bdev /dev/mapper/test-scratch1 errs: wr 0, rd 0, flush 0, corrupt 0, gen 1
Showing we have metadata generation mismatch errors.
- BTRFS error (device dm-2): unable to fixup (regular) error at logical 7110656 on dev /dev/mapper/test-scratch1
Showing which tree blocks are corrupted.
- BTRFS warning (device dm-2): checksum/header error at logical 24772608 on dev /dev/mapper/test-scratch2, physical 3801088: metadata node (level 1) in tree 5
Showing which physical range the corrupted metadata is at.
We have to combine the above 3 to know we have a corrupted metadata with
generation mismatch.
And this is already the better case, if we have other problems, like
fsid mismatch, we can not even know the cause.
[CAUSE]
The problem is caused by the fact that, scrub_checksum_tree_block()
never outputs any error message.
It just return two bits for scrub: sblock->header_error, and
sblock->generation_error.
And later we report error in scrub_print_warning(), but unfortunately we
only have two bits, there is not really much thing we can done to print
any detailed errors.
[FIX]
This patch will do the following to enhance the error reporting of
metadata scrub:
- Add extra warning (ratelimited) for every error we hit
This can help us to distinguish the different types of errors.
Some errors can help us to know what's going wrong immediately,
like bytenr mismatch.
- Re-order the checks
Currently we check bytenr first, then immediately generation.
This can lead to false generation mismatch reports, while the fsid
mismatches.
Here is the new output for the bug I'm debugging (we forgot to
writeback tree blocks for commit roots):
BTRFS warning (device dm-2): tree block 24117248 mirror 1 has bad fsid, has b77cd862-f150-4c71-90ec-7baf0544d83f want 17df6abf-23cd-445f-b350-5b3e40bfd2fc
BTRFS warning (device dm-2): tree block 24117248 mirror 0 has bad fsid, has b77cd862-f150-4c71-90ec-7baf0544d83f want 17df6abf-23cd-445f-b350-5b3e40bfd2fc
Now we can immediately know it's some tree blocks didn't even get written
back, other than the original confusing generation mismatch.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When a file system has ZNS devices which are constrained by a maximum
number of active block groups, then not being able to use all the block
groups for every allocation is not ideal, and could cause us to loop a
ton with mixed size allocations.
In general, since zoned doesn't write into gaps behind where block
groups are writing, it is not susceptible to the same sort of
fragmentation that size classes are designed to solve, so we can skip
size classes for zoned file systems in general, even though there would
probably be no harm for SMR devices.
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
Since the size class is an artifact of an arbitrary anti fragmentation
strategy, it doesn't really make sense to persist it. Furthermore, most
of the size class logic assumes fresh block groups. That is of course
not a reasonable assumption -- we will be upgrading kernels with
existing filesystems whose block groups are not classified.
To work around those issues, implement logic to compute the size class
of the block groups as we cache them in. To perfectly assess the state
of a block group, we would have to read the entire extent tree (since
the free space cache mashes together contiguous extent items) which
would be prohibitively expensive for larger file systems with more
extents.
We can do it relatively cheaply by implementing a simple heuristic of
sampling a handful of extents and picking the smallest one we see. In
the happy case where the block group was classified, we will only see
extents of the correct size. In the unhappy case, we will hopefully find
one of the smaller extents, but there is no perfect answer anyway.
Autorelocation will eventually churn up the block group if there is
significant freeing anyway.
There was no regression in mount performance at end state of the fsperf
test suite, and the delay until the block group is marked cached is
minimized by the constant number of extent samples.
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
The aim of this patch is to reduce the fragmentation of block groups
under certain unhappy workloads. It is particularly effective when the
size of extents correlates with their lifetime, which is something we
have observed causing fragmentation in the fleet at Meta.
This patch categorizes extents into size classes:
- x < 128KiB: "small"
- 128KiB < x < 8MiB: "medium"
- x > 8MiB: "large"
and as much as possible reduces allocations of extents into block groups
that don't match the size class. This takes advantage of any (possible)
correlation between size and lifetime and also leaves behind predictable
re-usable gaps when extents are freed; small writes don't gum up bigger
holes.
Size classes are implemented in the following way:
- Mark each new block group with a size class of the first allocation
that goes into it.
- Add two new passes to ffe: "unset size class" and "wrong size class".
First, try only matching block groups, then try unset ones, then allow
allocation of new ones, and finally allow mismatched block groups.
- Filtering is done just by skipping inappropriate ones, there is no
special size class indexing.
Other solutions I considered were:
- A best fit allocator with an rb-tree. This worked well, as small
writes didn't leak big holes from large freed extents, but led to
regressions in ffe and write performance due to lock contention on
the rb-tree with every allocation possibly updating it in parallel.
Perhaps something clever could be done to do the updates in the
background while being "right enough".
- A fixed size "working set". This prevents freeing an extent
drastically changing where writes currently land, and seems like a
good option too. Doesn't take advantage of size in any way.
- The same size class idea, but implemented with xarray marks. This
turned out to be slower than looping the linked list and skipping
wrong block groups, and is also less flexible since we must have only
3 size classes (max #marks). With the current approach we can have as
many as we like.
Performance testing was done via: https://github.com/josefbacik/fsperf
Of particular relevance are the new fragmentation specific tests.
A brief summary of the testing results:
- Neutral results on existing tests. There are some minor regressions
and improvements here and there, but nothing that truly stands out as
notable.
- Improvement on new tests where size class and extent lifetime are
correlated. Fragmentation in these cases is completely eliminated
and write performance is generally a little better. There is also
significant improvement where extent sizes are just a bit larger than
the size class boundaries.
- Regression on one new tests: where the allocations are sized
intentionally a hair under the borders of the size classes. Results
are neutral on the test that intentionally attacks this new scheme by
mixing extent size and lifetime.
The full dump of the performance results can be found here:
https://bur.io/fsperf/size-class-2022-11-15.txt
(there are ANSI escape codes, so best to curl and view in terminal)
Here is a snippet from the full results for a new test which mixes
buffered writes appending to a long lived set of files and large short
lived fallocates:
bufferedappendvsfallocate results
metric baseline current stdev diff
======================================================================================
avg_commit_ms 31.13 29.20 2.67 -6.22%
bg_count 14 15.60 0 11.43%
commits 11.10 12.20 0.32 9.91%
elapsed 27.30 26.40 2.98 -3.30%
end_state_mount_ns 11122551.90 10635118.90 851143.04 -4.38%
end_state_umount_ns 1.36e+09 1.35e+09 12248056.65 -1.07%
find_free_extent_calls 116244.30 114354.30 964.56 -1.63%
find_free_extent_ns_max 599507.20 1047168.20 103337.08 74.67%
find_free_extent_ns_mean 3607.19 3672.11 101.20 1.80%
find_free_extent_ns_min 500 512 6.67 2.40%
find_free_extent_ns_p50 2848 2876 37.65 0.98%
find_free_extent_ns_p95 4916 5000 75.45 1.71%
find_free_extent_ns_p99 20734.49 20920.48 1670.93 0.90%
frag_pct_max 61.67 0 8.05 -100.00%
frag_pct_mean 43.59 0 6.10 -100.00%
frag_pct_min 25.91 0 16.60 -100.00%
frag_pct_p50 42.53 0 7.25 -100.00%
frag_pct_p95 61.67 0 8.05 -100.00%
frag_pct_p99 61.67 0 8.05 -100.00%
fragmented_bg_count 6.10 0 1.45 -100.00%
max_commit_ms 49.80 46 5.37 -7.63%
sys_cpu 2.59 2.62 0.29 1.39%
write_bw_bytes 1.62e+08 1.68e+08 17975843.50 3.23%
write_clat_ns_mean 57426.39 54475.95 2292.72 -5.14%
write_clat_ns_p50 46950.40 42905.60 2101.35 -8.62%
write_clat_ns_p99 148070.40 143769.60 2115.17 -2.90%
write_io_kbytes 4194304 4194304 0 0.00%
write_iops 2476.15 2556.10 274.29 3.23%
write_lat_ns_max 2101667.60 2251129.50 370556.59 7.11%
write_lat_ns_mean 59374.91 55682.00 2523.09 -6.22%
write_lat_ns_min 17353.10 16250 1646.08 -6.36%
There are some mixed improvements/regressions in most metrics along with
an elimination of fragmentation in this workload.
On the balance, the drastic 1->0 improvement in the happy cases seems
worth the mix of regressions and improvements we do observe.
Some considerations for future work:
- Experimenting with more size classes
- More hinting/search ordering work to approximate a best-fit allocator
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
find_free_extent is a complicated function. It consists (at least) of:
- a hint that jumps into the middle of a for loop macro
- a middle loop trying every raid level
- an outer loop ascending through ffe loop levels
- complicated logic for skipping some of those ffe loop levels
- multiple underlying in-bg allocators (zoned, cluster, no cluster)
Which is all to say that more tracing is helpful for debugging its
behavior. Add two new tracepoints: at the entrance to the block_groups
loop (hit for every raid level and every ffe_ctl loop) and at the point
we seriously consider a block_group for allocation. This way we can see
the whole path through the algorithm, including hints, multiple loops,
etc.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The allocator tracepoints currently have a pile of values from ffe_ctl.
In modifying the allocator and adding more tracepoints, I found myself
adding to the already long argument list of the tracepoints. It makes it
a lot simpler to just send in the ffe_ctl itself.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Given that wait is always set to 1, so remove the argument.
Last use of wait with 0 was in 0c304304fe ("Btrfs: remove
csum_bytes_left").
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We currently use 'ret' and 'err' to track the return value for
log_dir_items(), which is confusing and likely the cause for previous
bugs where log_dir_items() did not return an error when it should, fixed
in previous patches.
So change this and use only a single variable, 'ret', to track the return
value. This is simpler and makes it similar to most of the existing code.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently we use the value 1 for BTRFS_LOG_FORCE_COMMIT, but that value
has a few inconveniences:
1) If it's ever used by btrfs_log_inode(), or any function down the call
chain, we have to remember to btrfs_set_log_full_commit(), which is
repetitive and has a chance to be forgotten in future use cases.
btrfs_log_inode_parent() only calls btrfs_set_log_full_commit() when
it gets a negative value from btrfs_log_inode();
2) Down the call chain of btrfs_log_inode(), we may have functions that
need to force a log commit, but can return either an error (negative
value), false (0) or true (1). So they are forced to return some
random negative to force a log commit - using BTRFS_LOG_FORCE_COMMIT
would make the intention more clear. Currently the only example is
flush_dir_items_batch().
So turn BTRFS_LOG_FORCE_COMMIT into a negative value. The chosen value
is -(MAX_ERRNO + 1), so that it does not overlap any errno value and makes
it easier to debug.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The header file linux/mm.h provides PAGE_ALIGN, PAGE_ALIGNED,
PAGE_ALIGN_DOWN macros. Use these macros to make code more
concise.
Signed-off-by: Yushan Zhou <katrinzhou@tencent.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When btrfs_get_chunk_map fails to allocate a new em the cleanup does not
need to be done so the goto target is out_err, which is consistent with
current coding style.
Signed-off-by: Peng Hao <flyingpeng@tencent.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
We had a recent bug that would have been caught by a newer compiler with
-Wmaybe-uninitialized and would have saved us a month of failing tests
that I didn't have time to investigate.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
With -Wmaybe-uninitialized compiler complains about ret being possibly
uninitialized, which isn't possible as the WQ_ constants are set only
from our code, however we can handle the default case and get rid of the
warning.
The value is set to BLK_STS_IOERR so it does not issue any IO and could
be potentially detected, but this is basically a "cannot happen" error.
To catch any problems during development use the assert.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ set the error in default: ]
Signed-off-by: David Sterba <dsterba@suse.com>
Fix an uninitialized warning we get with -Wmaybe-uninitialized where it
thought zno may have been uninitialized, in both cases it depends on
zinfo->zone_cache but we know the value won't change between checks.
Reported-by: Josef Bacik <josef@toxicpanda.com>
Link: https://lore.kernel.org/linux-btrfs/af6c527cbd8bdc782e50bd33996ee83acc3a16fb.1671221596.git.josef@toxicpanda.com/
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We only have 3 possible mirrors, and we have ASSERT()'s to make sure
we're not passing in an invalid super mirror into this function, so
technically this value isn't uninitialized. However
-Wmaybe-uninitialized will complain, so set it to U64_MAX so if we don't
have ASSERT()'s turned on it'll error out later on when it see's the
zone is beyond our maximum zones.
Reviewed-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We will pass in the parent and p pointer into our tree_search function
to avoid doing a second search when inserting a new extent state into
the tree. However because this is conditional upon passing in these
pointers the compiler seems to think these values can be uninitialized
if we're using -Wmaybe-uninitialized. Fix this by initializing these
values.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
reclaim isn't set in the alloc case, however we only care about
reclaim in the !alloc case. This isn't an actual problem, however
-Wmaybe-uninitialized will complain, so initialize reclaim to quiet the
compiler.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Anybody that calls get_inode_gen() can have an uninitialized gen if
there's an error. This isn't a big deal because all the users just exit
if they get an error, however it makes -Wmaybe-uninitialized complain,
so fix this up to always initialize the passed in gen, this quiets all
of the uninitialized warnings in send.c.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We can conditionally pass in a locked page, and then we'll use that page
range to skip marking errors as that will happen in another layer.
However this causes the compiler to complain because it doesn't
understand we only use these values when we have the page. Make the
compiler stop complaining by setting these values to 0.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
While trying to sync messages.[ch] I ended up with this dependency on
messages.h in the rest of btrfs-progs code base because it's where
btrfs_abort_transaction() was now held. We want to keep messages.[ch]
limited to the kernel code, and the btrfs_abort_transaction() code
better fits in the transaction code and not in messages.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ move the __cold attributes ]
Signed-off-by: David Sterba <dsterba@suse.com>
Now that none of the functions called by btrfs_merge_delayed_refs() needs
a btrfs_trans_handle, directly pass in a btrfs_fs_info to
btrfs_merge_delayed_refs().
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that drop_delayed_ref() doesn't need a btrfs_trans_handle, drop it
from insert_delayed_ref() as well.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that drop_delayed_ref() doesn't get the btrfs_trans_handle passed in
anymore, we can get rid of it in merge_ref() as well.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
drop_delayed_ref() doesn't use the btrfs_trans_handle it gets passed in,
so remove it.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=KGNk
-----END PGP SIGNATURE-----
Merge tag 'for-6.2-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
- one more fix for a tree-log 'write time corruption' report, update
the last dir index directly and don't keep in the log context
- do VFS-level inode lock around FIEMAP to prevent a deadlock with
concurrent fsync, the extent-level lock is not sufficient
- don't cache a single-device filesystem device to avoid cases when a
loop device is reformatted and the entry gets stale
* tag 'for-6.2-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: free device in btrfs_close_devices for a single device filesystem
btrfs: lock the inode in shared mode before starting fiemap
btrfs: simplify update of last_dir_index_offset when logging a directory
We have this check to make sure we don't accidentally add older devices
that may have disappeared and re-appeared with an older generation from
being added to an fs_devices (such as a replace source device). This
makes sense, we don't want stale disks in our file system. However for
single disks this doesn't really make sense.
I've seen this in testing, but I was provided a reproducer from a
project that builds btrfs images on loopback devices. The loopback
device gets cached with the new generation, and then if it is re-used to
generate a new file system we'll fail to mount it because the new fs is
"older" than what we have in cache.
Fix this by freeing the cache when closing the device for a single device
filesystem. This will ensure that the mount command passed device path is
scanned successfully during the next mount.
CC: stable@vger.kernel.org # 5.10+
Reported-by: Daan De Meyer <daandemeyer@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently fiemap does not take the inode's lock (VFS lock), it only locks
a file range in the inode's io tree. This however can lead to a deadlock
if we have a concurrent fsync on the file and fiemap code triggers a fault
when accessing the user space buffer with fiemap_fill_next_extent(). The
deadlock happens on the inode's i_mmap_lock semaphore, which is taken both
by fsync and btrfs_page_mkwrite(). This deadlock was recently reported by
syzbot and triggers a trace like the following:
task:syz-executor361 state:D stack:20264 pid:5668 ppid:5119 flags:0x00004004
Call Trace:
<TASK>
context_switch kernel/sched/core.c:5293 [inline]
__schedule+0x995/0xe20 kernel/sched/core.c:6606
schedule+0xcb/0x190 kernel/sched/core.c:6682
wait_on_state fs/btrfs/extent-io-tree.c:707 [inline]
wait_extent_bit+0x577/0x6f0 fs/btrfs/extent-io-tree.c:751
lock_extent+0x1c2/0x280 fs/btrfs/extent-io-tree.c:1742
find_lock_delalloc_range+0x4e6/0x9c0 fs/btrfs/extent_io.c:488
writepage_delalloc+0x1ef/0x540 fs/btrfs/extent_io.c:1863
__extent_writepage+0x736/0x14e0 fs/btrfs/extent_io.c:2174
extent_write_cache_pages+0x983/0x1220 fs/btrfs/extent_io.c:3091
extent_writepages+0x219/0x540 fs/btrfs/extent_io.c:3211
do_writepages+0x3c3/0x680 mm/page-writeback.c:2581
filemap_fdatawrite_wbc+0x11e/0x170 mm/filemap.c:388
__filemap_fdatawrite_range mm/filemap.c:421 [inline]
filemap_fdatawrite_range+0x175/0x200 mm/filemap.c:439
btrfs_fdatawrite_range fs/btrfs/file.c:3850 [inline]
start_ordered_ops fs/btrfs/file.c:1737 [inline]
btrfs_sync_file+0x4ff/0x1190 fs/btrfs/file.c:1839
generic_write_sync include/linux/fs.h:2885 [inline]
btrfs_do_write_iter+0xcd3/0x1280 fs/btrfs/file.c:1684
call_write_iter include/linux/fs.h:2189 [inline]
new_sync_write fs/read_write.c:491 [inline]
vfs_write+0x7dc/0xc50 fs/read_write.c:584
ksys_write+0x177/0x2a0 fs/read_write.c:637
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f7d4054e9b9
RSP: 002b:00007f7d404fa2f8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 00007f7d405d87a0 RCX: 00007f7d4054e9b9
RDX: 0000000000000090 RSI: 0000000020000000 RDI: 0000000000000006
RBP: 00007f7d405a51d0 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 61635f65646f6e69
R13: 65646f7475616f6e R14: 7261637369646f6e R15: 00007f7d405d87a8
</TASK>
INFO: task syz-executor361:5697 blocked for more than 145 seconds.
Not tainted 6.2.0-rc3-syzkaller-00376-g7c6984405241 #0
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:syz-executor361 state:D stack:21216 pid:5697 ppid:5119 flags:0x00004004
Call Trace:
<TASK>
context_switch kernel/sched/core.c:5293 [inline]
__schedule+0x995/0xe20 kernel/sched/core.c:6606
schedule+0xcb/0x190 kernel/sched/core.c:6682
rwsem_down_read_slowpath+0x5f9/0x930 kernel/locking/rwsem.c:1095
__down_read_common+0x54/0x2a0 kernel/locking/rwsem.c:1260
btrfs_page_mkwrite+0x417/0xc80 fs/btrfs/inode.c:8526
do_page_mkwrite+0x19e/0x5e0 mm/memory.c:2947
wp_page_shared+0x15e/0x380 mm/memory.c:3295
handle_pte_fault mm/memory.c:4949 [inline]
__handle_mm_fault mm/memory.c:5073 [inline]
handle_mm_fault+0x1b79/0x26b0 mm/memory.c:5219
do_user_addr_fault+0x69b/0xcb0 arch/x86/mm/fault.c:1428
handle_page_fault arch/x86/mm/fault.c:1519 [inline]
exc_page_fault+0x7a/0x110 arch/x86/mm/fault.c:1575
asm_exc_page_fault+0x22/0x30 arch/x86/include/asm/idtentry.h:570
RIP: 0010:copy_user_short_string+0xd/0x40 arch/x86/lib/copy_user_64.S:233
Code: 74 0a 89 (...)
RSP: 0018:ffffc9000570f330 EFLAGS: 00050202
RAX: ffffffff843e6601 RBX: 00007fffffffefc8 RCX: 0000000000000007
RDX: 0000000000000000 RSI: ffffc9000570f3e0 RDI: 0000000020000120
RBP: ffffc9000570f490 R08: 0000000000000000 R09: fffff52000ae1e83
R10: fffff52000ae1e83 R11: 1ffff92000ae1e7c R12: 0000000000000038
R13: ffffc9000570f3e0 R14: 0000000020000120 R15: ffffc9000570f3e0
copy_user_generic arch/x86/include/asm/uaccess_64.h:37 [inline]
raw_copy_to_user arch/x86/include/asm/uaccess_64.h:58 [inline]
_copy_to_user+0xe9/0x130 lib/usercopy.c:34
copy_to_user include/linux/uaccess.h:169 [inline]
fiemap_fill_next_extent+0x22e/0x410 fs/ioctl.c:144
emit_fiemap_extent+0x22d/0x3c0 fs/btrfs/extent_io.c:3458
fiemap_process_hole+0xa00/0xad0 fs/btrfs/extent_io.c:3716
extent_fiemap+0xe27/0x2100 fs/btrfs/extent_io.c:3922
btrfs_fiemap+0x172/0x1e0 fs/btrfs/inode.c:8209
ioctl_fiemap fs/ioctl.c:219 [inline]
do_vfs_ioctl+0x185b/0x2980 fs/ioctl.c:810
__do_sys_ioctl fs/ioctl.c:868 [inline]
__se_sys_ioctl+0x83/0x170 fs/ioctl.c:856
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f7d4054e9b9
RSP: 002b:00007f7d390d92f8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 00007f7d405d87b0 RCX: 00007f7d4054e9b9
RDX: 0000000020000100 RSI: 00000000c020660b RDI: 0000000000000005
RBP: 00007f7d405a51d0 R08: 00007f7d390d9700 R09: 0000000000000000
R10: 00007f7d390d9700 R11: 0000000000000246 R12: 61635f65646f6e69
R13: 65646f7475616f6e R14: 7261637369646f6e R15: 00007f7d405d87b8
</TASK>
What happens is the following:
1) Task A is doing an fsync, enters btrfs_sync_file() and flushes delalloc
before locking the inode and the i_mmap_lock semaphore, that is, before
calling btrfs_inode_lock();
2) After task A flushes delalloc and before it calls btrfs_inode_lock(),
another task dirties a page;
3) Task B starts a fiemap without FIEMAP_FLAG_SYNC, so the page dirtied
at step 2 remains dirty and unflushed. Then when it enters
extent_fiemap() and it locks a file range that includes the range of
the page dirtied in step 2;
4) Task A calls btrfs_inode_lock() and locks the inode (VFS lock) and the
inode's i_mmap_lock semaphore in write mode. Then it tries to flush
delalloc by calling start_ordered_ops(), which will block, at
find_lock_delalloc_range(), when trying to lock the range of the page
dirtied at step 2, since this range was locked by the fiemap task (at
step 3);
5) Task B generates a page fault when accessing the user space fiemap
buffer with a call to fiemap_fill_next_extent().
The fault handler needs to call btrfs_page_mkwrite() for some other
page of our inode, and there we deadlock when trying to lock the
inode's i_mmap_lock semaphore in read mode, since the fsync task locked
it in write mode (step 4) and the fsync task can not progress because
it's waiting to lock a file range that is currently locked by us (the
fiemap task, step 3).
Fix this by taking the inode's lock (VFS lock) in shared mode when
entering fiemap. This effectively serializes fiemap with fsync (except the
most expensive part of fsync, the log sync), preventing this deadlock.
Reported-by: syzbot+cc35f55c41e34c30dcb5@syzkaller.appspotmail.com
Link: https://lore.kernel.org/linux-btrfs/00000000000032dc7305f2a66f46@google.com/
CC: stable@vger.kernel.org # 6.1+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When logging a directory, we always set the inode's last_dir_index_offset
to the offset of the last dir index item we found. This is using an extra
field in the log context structure, and it makes more sense to update it
only after we insert dir index items, and we could directly update the
inode's last_dir_index_offset field instead.
So make this simpler by updating the inode's last_dir_index_offset only
when we actually insert dir index keys in the log tree, and getting rid
of the last_dir_item_offset field in the log context structure.
Reported-by: David Arendt <admin@prnet.org>
Link: https://lore.kernel.org/linux-btrfs/ae169fc6-f504-28f0-a098-6fa6a4dfb612@leemhuis.info/
Reported-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Link: https://lore.kernel.org/linux-btrfs/Y8voyTXdnPDz8xwY@mail.gmail.com/
Reported-by: Hunter Wardlaw <wardlawhunter@gmail.com>
Link: https://bugzilla.suse.com/show_bug.cgi?id=1207231
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=216851
CC: stable@vger.kernel.org # 6.1+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=5fdr
-----END PGP SIGNATURE-----
Merge tag 'for-6.2-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
- explicitly initialize zlib work memory to fix a KCSAN warning
- limit number of send clones by maximum memory allocated
- limit device size extent in case it device shrink races with chunk
allocation
- raid56 fixes:
- fix copy&paste error in RAID6 stripe recovery
- make error bitmap update atomic
* tag 'for-6.2-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: raid56: make error_bitmap update atomic
btrfs: send: limit number of clones and allocated memory size
btrfs: zlib: zero-initialize zlib workspace
btrfs: limit device extents to the device size
btrfs: raid56: fix stripes if vertical errors are found
In the rework of raid56 code, there is very limited concurrency in the
endio context.
Most of the work is done inside the sectors arrays, which different bios
will never touch the same sector.
But there is a concurrency here for error_bitmap. Both read and write
endio functions need to touch them, and we can have multiple write bios
touching the same error bitmap if they all hit some errors.
Here we fix the unprotected bitmap operation by going set_bit() in a
loop.
Since we have a very small ceiling of the sectors (at most 16 sectors),
such set_bit() in a loop should be very acceptable.
Fixes: 2942a50dea ("btrfs: raid56: introduce btrfs_raid_bio::error_bitmap")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The arg->clone_sources_count is u64 and can trigger a warning when a
huge value is passed from user space and a huge array is allocated.
Limit the allocated memory to 8MiB (can be increased if needed), which
in turn limits the number of clone sources to 8M / sizeof(struct
clone_root) = 8M / 40 = 209715. Real world number of clones is from
tens to hundreds, so this is future proof.
Reported-by: syzbot+4376a9a073770c173269@syzkaller.appspotmail.com
Signed-off-by: David Sterba <dsterba@suse.com>
KMSAN reports uses of uninitialized memory in zlib's longest_match()
called on memory originating from zlib_alloc_workspace().
This issue is known by zlib maintainers and is claimed to be harmless,
but to be on the safe side we'd better initialize the memory.
Link: https://zlib.net/zlib_faq.html#faq36
Reported-by: syzbot+14d9e7602ebdf7ec0a60@syzkaller.appspotmail.com
CC: stable@vger.kernel.org # 5.4+
Signed-off-by: Alexander Potapenko <glider@google.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There was a recent regression in btrfs/177 that started happening with
the size class patches ("btrfs: introduce size class to block group
allocator"). This however isn't a regression introduced by those
patches, but rather the bug was uncovered by a change in behavior in
these patches. The patches triggered more chunk allocations in the
^free-space-tree case, which uncovered a race with device shrink.
The problem is we will set the device total size to the new size, and
use this to find a hole for a device extent. However during shrink we
may have device extents allocated past this range, so we could
potentially find a hole in a range past our new shrink size. We don't
actually limit our found extent to the device size anywhere, we assume
that we will not find a hole past our device size. This isn't true with
shrink as we're relocating block groups and thus creating holes past the
device size.
Fix this by making sure we do not search past the new device size, and
if we wander into any device extents that start after our device size
simply break from the loop and use whatever hole we've already found.
CC: stable@vger.kernel.org # 4.14+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We take two stripe numbers if vertical errors are found. In case it is
just a pstripe it does not matter but in case of raid 6 it matters as
both stripes need to be fixed.
Fixes: 7a31507230 ("btrfs: raid56: do data csum verification during RMW cycle")
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Tanmay Bhushan <007047221b@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=Kqrg
-----END PGP SIGNATURE-----
Merge tag 'for-6.2-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
- fix potential out-of-bounds access to leaf data when seeking in an
inline file
- fix potential crash in quota when rescan races with disable
- reimplement super block signature scratching by marking page/folio
dirty and syncing block device, allow removing write_one_page
* tag 'for-6.2-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: fix race between quota rescan and disable leading to NULL pointer deref
btrfs: fix invalid leaf access due to inline extent during lseek
btrfs: stop using write_one_page in btrfs_scratch_superblock
btrfs: factor out scratching of one regular super block
If we have one task trying to start the quota rescan worker while another
one is trying to disable quotas, we can end up hitting a race that results
in the quota rescan worker doing a NULL pointer dereference. The steps for
this are the following:
1) Quotas are enabled;
2) Task A calls the quota rescan ioctl and enters btrfs_qgroup_rescan().
It calls qgroup_rescan_init() which returns 0 (success) and then joins a
transaction and commits it;
3) Task B calls the quota disable ioctl and enters btrfs_quota_disable().
It clears the bit BTRFS_FS_QUOTA_ENABLED from fs_info->flags and calls
btrfs_qgroup_wait_for_completion(), which returns immediately since the
rescan worker is not yet running.
Then it starts a transaction and locks fs_info->qgroup_ioctl_lock;
4) Task A queues the rescan worker, by calling btrfs_queue_work();
5) The rescan worker starts, and calls rescan_should_stop() at the start
of its while loop, which results in 0 iterations of the loop, since
the flag BTRFS_FS_QUOTA_ENABLED was cleared from fs_info->flags by
task B at step 3);
6) Task B sets fs_info->quota_root to NULL;
7) The rescan worker tries to start a transaction and uses
fs_info->quota_root as the root argument for btrfs_start_transaction().
This results in a NULL pointer dereference down the call chain of
btrfs_start_transaction(). The stack trace is something like the one
reported in Link tag below:
general protection fault, probably for non-canonical address 0xdffffc0000000041: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000208-0x000000000000020f]
CPU: 1 PID: 34 Comm: kworker/u4:2 Not tainted 6.1.0-syzkaller-13872-gb6bb9676f216 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/26/2022
Workqueue: btrfs-qgroup-rescan btrfs_work_helper
RIP: 0010:start_transaction+0x48/0x10f0 fs/btrfs/transaction.c:564
Code: 48 89 fb 48 (...)
RSP: 0018:ffffc90000ab7ab0 EFLAGS: 00010206
RAX: 0000000000000041 RBX: 0000000000000208 RCX: ffff88801779ba80
RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000000
RBP: dffffc0000000000 R08: 0000000000000001 R09: fffff52000156f5d
R10: fffff52000156f5d R11: 1ffff92000156f5c R12: 0000000000000000
R13: 0000000000000001 R14: 0000000000000001 R15: 0000000000000003
FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f2bea75b718 CR3: 000000001d0cc000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
btrfs_qgroup_rescan_worker+0x3bb/0x6a0 fs/btrfs/qgroup.c:3402
btrfs_work_helper+0x312/0x850 fs/btrfs/async-thread.c:280
process_one_work+0x877/0xdb0 kernel/workqueue.c:2289
worker_thread+0xb14/0x1330 kernel/workqueue.c:2436
kthread+0x266/0x300 kernel/kthread.c:376
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308
</TASK>
Modules linked in:
So fix this by having the rescan worker function not attempt to start a
transaction if it didn't do any rescan work.
Reported-by: syzbot+96977faa68092ad382c4@syzkaller.appspotmail.com
Link: https://lore.kernel.org/linux-btrfs/000000000000e5454b05f065a803@google.com/
Fixes: e804861bd4 ("btrfs: fix deadlock between quota disable and qgroup rescan worker")
CC: stable@vger.kernel.org # 5.4+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
During lseek, for SEEK_DATA and SEEK_HOLE modes, we access the disk_bytenr
of an extent without checking its type. However inline extents have their
data starting the offset of the disk_bytenr field, so accessing that field
when we have an inline extent can result in either of the following:
1) Interpret the inline extent's data as a disk_bytenr value;
2) In case the inline data is less than 8 bytes, we access part of some
other item in the leaf, or unused space in the leaf;
3) In case the inline data is less than 8 bytes and the extent item is
the first item in the leaf, we can access beyond the leaf's limit.
So fix this by not accessing the disk_bytenr field if we have an inline
extent.
Fixes: b6e833567e ("btrfs: make hole and data seeking a lot more efficient")
Reported-by: Matthias Schoepfer <matthias.schoepfer@googlemail.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=216908
Link: https://lore.kernel.org/linux-btrfs/7f25442f-b121-2a3a-5a3d-22bcaae83cd4@leemhuis.info/
CC: stable@vger.kernel.org # 6.1
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
write_one_page is an awkward interface that expects the page locked and
->writepage to be implemented. Replace that by zeroing the signature
bytes and synchronize the block device page using the proper bdev
helpers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_scratch_superblocks open codes scratching super block of a
non-zoned super block. Split the code to read, zero and write the
superblock for regular devices into a separate helper.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=vcwq
-----END PGP SIGNATURE-----
Merge tag 'for-6.2-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
"Another batch of fixes, dealing with fallouts from 6.1 reported by
users:
- tree-log fixes:
- fix directory logging due to race with concurrent index key
deletion
- fix missing error handling when logging directory items
- handle case of conflicting inodes being added to the log
- remove transaction aborts for not so serious errors
- fix qgroup accounting warning when rescan can be started at time
with temporarily disable accounting
- print more specific errors to system log when device scan ioctl
fails
- disable space overcommit for ZNS devices, causing heavy performance
drop"
* tag 'for-6.2-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: do not abort transaction on failure to update log root
btrfs: do not abort transaction on failure to write log tree when syncing log
btrfs: add missing setup of log for full commit at add_conflicting_inode()
btrfs: fix directory logging due to race with concurrent index key deletion
btrfs: fix missing error handling when logging directory items
btrfs: zoned: enable metadata over-commit for non-ZNS setup
btrfs: qgroup: do not warn on record without old_roots populated
btrfs: add extra error messages to cover non-ENOMEM errors from device_add_list()
When syncing a log, if we fail to update a log root in the log root tree,
we are aborting the transaction if the failure was not -ENOSPC. This is
excessive because there is a chance that a transaction commit can succeed,
and therefore avoid to turn the filesystem into RO mode. All we need to be
careful about is to mark the log for a full commit, which we already do,
to make sure no one commits a super block pointing to an outdated log root
tree.
So don't abort the transaction if we fail to update a log root in the log
root tree, and log an error if the failure is not -ENOSPC, so that it does
not go completely unnoticed.
CC: stable@vger.kernel.org # 6.0+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When syncing the log, if we fail to write log tree extent buffers, we mark
the log for a full commit and abort the transaction. However we don't need
to abort the transaction, all we really need to do is to make sure no one
can commit a superblock pointing to new log tree roots. Just because we
got a failure writing extent buffers for a log tree, it does not mean we
will also fail to do a transaction commit.
One particular case is if due to a bug somewhere, when writing log tree
extent buffers, the tree checker detects some corruption and the writeout
fails because of that. Aborting the transaction can be very disruptive for
a user, specially if the issue happened on a root filesystem. One example
is the scenario in the Link tag below, where an isolated corruption on log
tree leaves was causing transaction aborts when syncing the log.
Link: https://lore.kernel.org/linux-btrfs/ae169fc6-f504-28f0-a098-6fa6a4dfb612@leemhuis.info/
CC: stable@vger.kernel.org # 5.15+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When logging conflicting inodes, if we reach the maximum limit of inodes,
we return BTRFS_LOG_FORCE_COMMIT to force a transaction commit. However
we don't mark the log for full commit (with btrfs_set_log_full_commit()),
which means that once we leave the log transaction and before we commit
the transaction, some other task may sync the log, which is incomplete
as we have not logged all conflicting inodes, leading to some inconsistent
in case that log ends up being replayed.
So also call btrfs_set_log_full_commit() at add_conflicting_inode().
Fixes: e09d94c9e4 ("btrfs: log conflicting inodes without holding log mutex of the initial inode")
CC: stable@vger.kernel.org # 6.1
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Sometimes we log a directory without holding its VFS lock, so while we
logging it, dir index entries may be added or removed. This typically
happens when logging a dentry from a parent directory that points to a
new directory, through log_new_dir_dentries(), or when while logging
some other inode we also need to log its parent directories (through
btrfs_log_all_parents()).
This means that while we are at log_dir_items(), we may not find a dir
index key we found before, because it was deleted in the meanwhile, so
a call to btrfs_search_slot() may return 1 (key not found). In that case
we return from log_dir_items() with a success value (the variable 'err'
has a value of 0). This can lead to a few problems, specially in the case
where the variable 'last_offset' has a value of (u64)-1 (and it's
initialized to that when it was declared):
1) By returning from log_dir_items() with success (0) and a value of
(u64)-1 for '*last_offset_ret', we end up not logging any other dir
index keys that follow the missing, just deleted, index key. The
(u64)-1 value makes log_directory_changes() not call log_dir_items()
again;
2) Before returning with success (0), log_dir_items(), will log a dir
index range item covering a range from the last old dentry index
(stored in the variable 'last_old_dentry_offset') to the value of
'last_offset'. If 'last_offset' has a value of (u64)-1, then it means
if the log is persisted and replayed after a power failure, it will
cause deletion of all the directory entries that have an index number
between last_old_dentry_offset + 1 and (u64)-1;
3) We can end up returning from log_dir_items() with
ctx->last_dir_item_offset having a lower value than
inode->last_dir_index_offset, because the former is set to the current
key we are processing at process_dir_items_leaf(), and at the end of
log_directory_changes() we set inode->last_dir_index_offset to the
current value of ctx->last_dir_item_offset. So if for example a
deletion of a lower dir index key happened, we set
ctx->last_dir_item_offset to that index value, then if we return from
log_dir_items() because btrfs_search_slot() returned 1, we end up
returning from log_dir_items() with success (0) and then
log_directory_changes() sets inode->last_dir_index_offset to a lower
value than it had before.
This can result in unpredictable and unexpected behaviour when we
need to log again the directory in the same transaction, and can result
in ending up with a log tree leaf that has duplicated keys, as we do
batch insertions of dir index keys into a log tree.
So fix this by making log_dir_items() move on to the next dir index key
if it does not find the one it was looking for.
Reported-by: David Arendt <admin@prnet.org>
Link: https://lore.kernel.org/linux-btrfs/ae169fc6-f504-28f0-a098-6fa6a4dfb612@leemhuis.info/
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When logging a directory, at log_dir_items(), if we get an error when
attempting to search the subvolume tree for a dir index item, we end up
returning 0 (success) from log_dir_items() because 'err' is left with a
value of 0.
This can lead to a few problems, specially in the case the variable
'last_offset' has a value of (u64)-1 (and it's initialized to that when
it was declared):
1) By returning from log_dir_items() with success (0) and a value of
(u64)-1 for '*last_offset_ret', we end up not logging any other dir
index keys that follow the missing, just deleted, index key. The
(u64)-1 value makes log_directory_changes() not call log_dir_items()
again;
2) Before returning with success (0), log_dir_items(), will log a dir
index range item covering a range from the last old dentry index
(stored in the variable 'last_old_dentry_offset') to the value of
'last_offset'. If 'last_offset' has a value of (u64)-1, then it means
if the log is persisted and replayed after a power failure, it will
cause deletion of all the directory entries that have an index number
between last_old_dentry_offset + 1 and (u64)-1;
3) We can end up returning from log_dir_items() with
ctx->last_dir_item_offset having a lower value than
inode->last_dir_index_offset, because the former is set to the current
key we are processing at process_dir_items_leaf(), and at the end of
log_directory_changes() we set inode->last_dir_index_offset to the
current value of ctx->last_dir_item_offset. So if for example a
deletion of a lower dir index key happened, we set
ctx->last_dir_item_offset to that index value, then if we return from
log_dir_items() because btrfs_search_slot() returned an error, we end up
returning without any error from log_dir_items() and then
log_directory_changes() sets inode->last_dir_index_offset to a lower
value than it had before.
This can result in unpredictable and unexpected behaviour when we
need to log again the directory in the same transaction, and can result
in ending up with a log tree leaf that has duplicated keys, as we do
batch insertions of dir index keys into a log tree.
Fix this by setting 'err' to the value of 'ret' in case
btrfs_search_slot() or btrfs_previous_item() returned an error. That will
result in falling back to a full transaction commit.
Reported-by: David Arendt <admin@prnet.org>
Link: https://lore.kernel.org/linux-btrfs/ae169fc6-f504-28f0-a098-6fa6a4dfb612@leemhuis.info/
Fixes: e02119d5a7 ("Btrfs: Add a write ahead tree log to optimize synchronous operations")
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The commit 79417d040f ("btrfs: zoned: disable metadata overcommit for
zoned") disabled the metadata over-commit to track active zones properly.
However, it also introduced a heavy overhead by allocating new metadata
block groups and/or flushing dirty buffers to release the space
reservations. Specifically, a workload (write only without any sync
operations) worsen its performance from 343.77 MB/sec (v5.19) to 182.89
MB/sec (v6.0).
The performance is still bad on current misc-next which is 187.95 MB/sec.
And, with this patch applied, it improves back to 326.70 MB/sec (+73.82%).
This patch introduces a new fs_info->flag BTRFS_FS_NO_OVERCOMMIT to
indicate it needs to disable the metadata over-commit. The flag is enabled
when a device with max active zones limit is loaded into a file-system.
Fixes: 79417d040f ("btrfs: zoned: disable metadata overcommit for zoned")
CC: stable@vger.kernel.org # 6.0+
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
