kernel/git/stable/linux.git/fs/btrfs/direct-io.c, branch master

Merge tag 'for-7.3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

2026-08-20T20:00:40+00:00

Pull btrfs updates from David Sterba: "This is the summer edition of btrfs changes, smaller than usual. Yet, there are performance improvements in various areas or for specific workloads and some notable changes like removing space cache v1 code or mount option reduction. User visible changes: - free space v1 disabled by default; the v2 (free space tree) is mkfs default since 5.15, filesystems with v1 still work but could be slightly slower due to lack of block group caching - mount option 'rescue=usebackuproot' requires read-only mount, it's too risky to allow writable mount - remove standalone mount option 'usebackuproot', deprecated in 5.9 - preserve constraints of NODATASUM and NODATACOW when chattr and mount options may change the attributes - remove arbitrary limitation of 4KiB for page size when allowing block sizes smaller than page - print messages when pinned block groups affect swap activation Performance improvements: - use iomap bounce buffer for direct io instead of a fall back to buffered io; past correctness vs speed trade-offs dropped performance to ~50% of theoretical maximum, now it's ~95%, effectively doubled - replace xarray with local LRU list for tracking inhibited extent buffers, restored performance to pre-inhibition state (relatively ~3x) - remove unnecessary 1 jiffy delay in "non-SSD" mode with multiple logging tasks, decrease latency, throughput increased ~5x on sample workload - skip hole detection during full fsync for files without holes and lots of extents, reduce run time ~5x on sample workload (microsecond ranges) - reduce locking around extent readahead so it does not slow down other tasks using an overlapping range - enhance extent buffer allocation modes to allow NOWAIT semantics in some cases Notable fixes: - write-protect folios during writeback, prevent concurrent mmap and compress/checksumming/etc undesired interactions - in zoned mode, handle transient overcommit full instead of going read-only - fix possible deadlock between defragmentation and delayed allocation reservations - handle remaining iputs at umount time - fix lockdep warning between device scan locking and log mutex - add workaround for degenerate RAID56 device count modes (2 and 3) not supported by the parity calculation library - restore check that subvolume is not read-only when changing ACLs - retry reading verity data colliding with up-to-date status changes Core: - simplify raid56 stripe handling by using contiguous virtual allocations - in zoned mode, fix various metadata write issues in writeback or unmount - space reservation fixes - remove unused data structure members - more auto-freeing conversions - error pointer values are printed using %pe format" * tag 'for-7.3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (72 commits) btrfs: skip hole detection during full fsync for files without holes btrfs: add extra ASSERT()s to make sure the folio size is correct btrfs: use GFP_NOWAIT for tree block readahead btrfs: enable unlocked NOFAIL retry for eb allocations btrfs: add struct btrfs_eb_prealloc btrfs: factor init_extent_buffer from __alloc_extent_buffer btrfs: qgroup: fix a wrong length calculation in qgroup_free_reserved_data() btrfs: add validation for extent states btrfs: use aligned range for locking in reflink btrfs: use aligned range for locking in extent_fiemap() btrfs: zoned: don't clobber the extent buffer when zeroing it out btrfs: zoned: drop stranded dirty metadata buffers at unmount btrfs: zoned: drop stranded dirty metadata on transaction abort btrfs: zoned: flush active metadata block group at btree_writepages() start btrfs: convert reflink.c to use btrfs_inode as parameters btrfs: use simple booleans for log_commit field in struct btrfs_root btrfs: check for exit condition after waking in wait_log_commit() btrfs: move condition for log commit wait into wait_log_commit() btrfs: remove log batch counter use for fsync btrfs: stop sleeping for one jiffy in non-ssd mounts during log commit ...

btrfs: get rid of useless label in btrfs_create_dio_extent()

2026-08-07T17:17:16+00:00

There's no point in having a label where under it we do nothing but return a variable. So remove it and directly return where we used to goto. Reviewed-by: Johannes Thumshirn Reviewed-by: Qu Wenruo Signed-off-by: Filipe Manana Reviewed-by: David Sterba Signed-off-by: David Sterba

btrfs: fix extent map leak in NOCOW direct I/O write

2026-08-07T17:17:16+00:00

btrfs_dio_iomap_begin() calls btrfs_get_extent(), which returns an extent map reference that must be dropped on all exit paths. For direct writes into a NOCOW range, btrfs_get_blocks_direct_write() keeps using that extent map and asks btrfs_create_dio_extent() to allocate the ordered extent. If that fails, for example because btrfs_alloc_ordered_extent() fails, the function returns the error without dropping the input extent map. The PREALLOC path avoided this by dropping the input extent map before replacing it with the newly created one. Check the error from btrfs_create_dio_extent() before replacing the map and drop the input extent map on failure. Fixes: 5f9a8a51d8b9 ("Btrfs: add semaphore to synchronize direct IO writes with fsync") CC: stable@vger.kernel.org Reviewed-by: Qu Wenruo Reviewed-by: Filipe Manana Signed-off-by: Shuangpeng Bai Signed-off-by: Filipe Manana Signed-off-by: David Sterba

btrfs: use IOMAP_DIO_BOUNCE flag instead of falling back to buffered IO

2026-08-07T17:17:16+00:00

Previously btrfs forces direct writes to fall back to buffered ones if the inode has data checksum or the profile has duplication. That fallback is to avoid the content being modified that the final content may mismatch with the checksum or the other mirrors. That brings a pretty huge performance cost, which already caused some concern at that time. But later upstream commit c9d114846b38 ("iomap: add a flag to bounce buffer direct I/O") introduced a new method by copying the content into new pages, and do all the operations based on the newly allocated pages. So let btrfs to utilize the new flag for direct writes if we require stable folios. There is a quick benchmark, using the following fio setup: fio --name=randwrite --filename $mnt/foobar --ioengine=libaio --size=4G \ --rw=randwrite --iodepth=64 --runtime=60 --time_based --direct=1 \ --bs=$blocksize Unit is MiB/s. Blocksize | Zero-copy (*) | Buffered | Bounce -----------+---------------+----------+----------- 4K | 35.1 | 17.1 | 33.8 64K | 522 | 251 | 492 *: This is done by reverting the commit 968f19c5b1b7 ("btrfs: always fallback to buffered write if the inode requires checksum") Although with page bouncing the performance is only around 95% of true-zero copy, it's still almost double the performance of buffered fallback. There will be a small change in behavior, since we're using IOMAP_DIO_BOUNCE flag to allocate new folios, NOWAIT flag will immediately fail. So for true NOWAIT direct IOs, NODATASUM and RAID0/SINGLE profiles are still required. Signed-off-by: Qu Wenruo Signed-off-by: David Sterba

btrfs: refactor btrfs_dio_iomap_end()

2026-08-07T17:16:27+00:00

That function has the following problems: - Read/write handling scattered across different locations E.g. At the beginning there is a dedicated hole read handling, but later short read handling is at an if() branch. - Modifying of @pos and @length parameter for short read Although it's completely fine to modify those parameters as they are passed by value, but it can still be confusing to read. As normally we would assume @pos and @length to be the original range. But for short IO handling we modify @pos/@length, and completely ignore @written. - Unnecessary split for ordered extent and changeset handling Both OE and changeset are only for writes, but they are handled in two different if (write) {} blocks. Refactor the function so that: - Handling of reads and writes are concentrated in their code block Now the handling of reads are in its own small if () branch. Leaving the more complex writes handling to take the remaining function, and reduce the indent level. This also removes all unnecessary "if (write)" checks. - Do not modify @pos and @length Let short IO handling to manually calculate the remaining range. Reviewed-by: Filipe Manana Signed-off-by: Qu Wenruo Signed-off-by: David Sterba

btrfs: remove btrfs_dio_data::submitted

2026-08-07T17:16:27+00:00

This member records how many bytes are submitted for a direct read/write, utilized by iomap_end() callback to handle short IO cases. However iomap_end() callback is already providing an internally tracked @written member, which is doing the same accounting and providing the same value as btrfs_dio_data::submitted. There is no need to duplicate the work, just remove btrfs_dio_data::submitted. Reviewed-by: Filipe Manana Signed-off-by: Qu Wenruo Reviewed-by: David Sterba Signed-off-by: David Sterba

btrfs: convert iomap ops to ->iomap_next()

2026-07-31T10:28:49+00:00

Convert btrfs iomap_ops to the new ->iomap_next() callback. The callback is generated with the DEFINE_IOMAP_ITER_NEXT_END() macro, which wraps the iomap_iter_next() helper to finish the previous mapping if needed and produce the next one. No functional changes are intended. Acked-by: David Sterba Signed-off-by: Joanne Koong Link: https://patch.msgid.link/20260729192737.3190206-8-joannelkoong@gmail.com Signed-off-by: Christian Brauner (Amutable)

btrfs: retry faulting in the pages after a zero sized short direct write

2026-06-09T16:22:46+00:00

Currently btrfs_direct_write() will not try to fault in the pages, but directly fall back to buffered writes, if the first page of the buffer can not be faulted in. For example, during generic/362 with nodatasum mount option, there is a write at file offset 0, length PAGE_SIZE, and the page is not faulted in. Then we go the following callchain and directly fall back to buffered IO: btrfs_direct_write() |- btrfs_dio_write() |- __iomap_dio_rw() | |- iomap_iter() | | |- btrfs_dio_iomap_begin() | | Now an ordered extent is allocated for the 4K write. | | | |- iomi.status = iomap_dio_iter() | | Where iomap_dio_iter() returned -EFAULT. | | | |- ret = iomap_iter() | | |- btrfs_dio_iomap_end() | | | | return -ENOTBLK | | |- return -ENOTBLK | |- if (ret == -ENOTBLK) { ret = 0; } | Now the return value is reset to 0. | |- ret = iomap_dio_complete() | Since no byte is submitted, @ret is now zero. | |- if (iov_iter_count() > 0 && (ret == -EFAULT || ret > 0)) | @ret is zero, thus not meeting the above retry condition | |- Fallback to buffered Just slightly loosen the condition to allow retry faulting in pages after a zero sized short write. Unlike the previous two bug fixes, this one is not really cause any real bug, but only reducing the chance to do zero-copy direct IO. Thus it doesn't really require stable-CC nor fixes-tag. Reviewed-by: Boris Burkov Signed-off-by: Qu Wenruo Signed-off-by: David Sterba

btrfs: fix incorrect buffered IO fallback for append direct writes

2026-06-09T16:22:46+00:00

[BUG] With the previous bug of short direct writes fixed, test case generic/362 (*) still fails with the following error with nodatasum mount option: generic/362 0s ... - output mismatch (see /home/adam/xfstests/results//generic/362.out.bad) - output mismatch (see /home/adam/xfstests/results//generic/362.out.bad) --- tests/generic/362.out 2024-08-24 15:31:37.200000000 +0930 +++ /home/adam/xfstests/results//generic/362.out.bad 2026-05-27 10:13:09.072485767 +0930 @@ -1,2 +1,3 @@ QA output created by 362 +Wrong file size after first write, got 8192 expected 4096 Silence is golden ... *: If the test case has been executed before with default data checksum, the failure will not reproduce. Need the following fix to make it reliably reproducible: https://lore.kernel.org/linux-btrfs/20260528111659.87113-1-wqu@suse.com/ [CAUSE] Inside btrfs_dio_iomap_begin() for a direct write, we increase the isize if it's beyond the current isize. But if the direct io finished short, we do not revert the isize to the previous value nor to the short write end. Then if we need to fall back to buffered writes, and the write has IOCB_APPEND flag, then the buffered write will be positioned at the incorrect isize. The call chain looks like this: btrfs_direct_write(pos=0, length=4K) |- __iomap_dio_rw() | |- iomap_iter() | | |- btrfs_dio_iomap_begin() | | |- btrfs_get_blocks_direct_write() | | |- i_size_write() | | Which updates the isize to the write end (4K). | | | |- iomap_dio_iter() | | Failed with -EFAULT on the first page. | | | |- iomap_iter() | | |- btrfs_dio_iomap_end() | | Detects a short write, return -ENOTBLK | |- if (ret == -ENOTBLK) { ret = 0;} | Which resets the return value. | |- ret = iomap_dio_complet() | Which returns 0. | |- btrfs_buffered_write(iocb, from); |- generic_write_checks() |- iocb->ki_pos = i_size_read() Which is still the new size (4K), other than the original isize 0. [FIX] Introduce the following btrfs_dio_data members: - old_isize - updated_isize If the direct write has enlarged the isize. Then if we got a short write, and btrfs_dio_data::updated_isize is set, revert to the correct isize based on old_isize and current file position. And here we call i_size_write() without holding an extent lock, which is a very special case that we're safe to do: - Only a single writer can be enlarging isize Enlarging isize will take the exclusive inode lock. - Buffered readers need to wait for the OE we're holding Buffered readers will lock extent and wait for OE of the folio range. Sometimes we can skip the OE wait, but since all page cache is invalidated, the OE wait can not be skipped. But I do not think this is the most elegant solution, nor covers all cases. E.g. if the bio is submitted but IO failed, we are unable to do the revert. I believe the more elegant one would be extend the EXTENT_DIO_LOCKED lifespan for direct writes, so that we can update the isize when a write beyond EOF finished successfully. However that change is too huge for a small bug fix. So only implement the minimal partial fix for now. [REASON FOR NO FIXES TAG] The bug is again very old, before commit f85781fb505e ("btrfs: switch to iomap for direct IO") we are already increasing isize without a proper rollback for short writes. Thus only a CC to stable. CC: stable@vger.kernel.org # 5.15+ Reviewed-by: Boris Burkov Signed-off-by: Qu Wenruo Signed-off-by: David Sterba

btrfs: fix false IO failure after falling back to buffered write

2026-06-09T16:22:46+00:00

[BUG] The test case generic/362 will fail with "nodatasum" mount option (*): MOUNT_OPTIONS -- -o nodatasum /dev/mapper/test-scratch1 /mnt/scratch generic/362 0s ... - output mismatch (see /home/adam/xfstests/results//generic/362.out.bad) --- tests/generic/362.out 2024-08-24 15:31:37.200000000 +0930 +++ /home/adam/xfstests/results//generic/362.out.bad 2026-05-27 10:21:17.574771567 +0930 @@ -1,2 +1,3 @@ QA output created by 362 +First write failed: Input/output error Silence is golden ... *: If the test case has been executed before with default data checksum, the failure will not reproduce. Need the following fix to make it reliably reproducible: https://lore.kernel.org/linux-btrfs/20260528111659.87113-1-wqu@suse.com/ [CAUSE] Inside __iomap_dio_rw(), the -EFAULT/-ENOTBLK error is not directly returned. Thus we never got an error pointer from __iomap_dio_rw(). The call chain looks like this: btrfs_direct_write() |- btrfs_dio_write() |- __iomap_dio_rw() | |- iomap_iter() | | |- btrfs_dio_iomap_begin() | | Now an ordered extent is allocated for the 4K write. | | | |- iomi.status = iomap_dio_iter() | | Where iomap_dio_iter() returned -EFAULT. | | | |- ret = iomap_iter() | | |- btrfs_dio_iomap_end() | | | |- btrfs_finish_ordered_extent(uptodate = false) | | | | |- can_finish_ordered_extent() | | | | |- btrfs_mark_ordered_extent_error() | | | | |- mapping_set_error() | | | | Now the address space is marked error. | | | | return -ENOTBLK | | |- return -ENOTBLK | |- if (ret == -ENOTBLK) { ret = 0; } | Now the return value is reset to 0. | Thus no error pointer will be returned. | |- ret = iomap_dio_complete() | Since no byte is submitted, @ret is 0. | |- Fallback to buffered IO | And the buffered write finished without error | |- filemap_fdatawait_range() |- filemap_check_errors() The previous error is recorded, thus an error is returned However the buffered write is properly submitted and finished, the error is from the btrfs_finish_ordered_extent() call with @uptodate = false. [FIX] When a short dio write happened, any range that is submitted will have btrfs_extract_ordered_extent() to be called, thus the submitted range will always have an OE just covering the submitted range. The remaining OE range is never submitted, thus they should be treated as truncated, not an error. So that we can properly reclaim and not insert an unnecessary file extent item, without marking the mapping as error. Extract a helper, btrfs_mark_ordered_extent_truncated(), and utilize that helper to mark the direct IO ordered extent as truncated, so it won't cause failure for the later buffered fallback. [REASON FOR NO FIXES TAG] The bug itself is pretty old, at commit f85781fb505e ("btrfs: switch to iomap for direct IO") we're already passing @uptodate=false finishing the OE. But at that time OE with IOERR won't call mapping_set_error(), so it's not exposed. Later commit d61bec08b904 ("btrfs: mark ordered extent and inode with error if we fail to finish") finally exposed the bug, but that commit is doing a correct job, not the root cause. Anyway the bug is very old, dating back to 5.1x days, thus only CC to stable. CC: stable@vger.kernel.org # 5.15+ Reviewed-by: Boris Burkov Signed-off-by: Qu Wenruo Signed-off-by: David Sterba