[RFC PATCH v5 0/4] migrates ocfs2 DIO from buffer_head to iomap
From: Heming Zhao
Date: Thu Nov 05 2026 - 02:19:56 EST
This patch set migrates ocfs2 Direct I/O (DIO) from buffer_head mode
to iomap mode. And the code base is commit 6628e7d4c6b2 (v7.4-rc1),
and also merge Joseph's code and ideas from [1]:
The key for the patch set
To simplify iomap development, we currently adopt the simplest approach:
retaining the buffer_head style, where only one cluster is allocated
per iomap_begin() call.
Introduce OCFS2_MAX_LOGICAL_BLOCK to cap the logical offset in the iomap
code path.
Put zeroing and orphan inode handling in ocfs2_file_write_iter(), the
buffer_head style is handled in ocfs2_dio_wr_get_block() with dwc
helper.
I attempted to implement .submit_io/.end_io in "struct iomap_dio_ops" to
free dwc inside end_io. However, because ->bi_end_io runs in an interrupt
context and ocfs2_iomap_dio_end_io_write() contains sleepable functions,
this approach required scheduling a workqueue directly from the
->bi_end_io callback and calling ocfs2_iomap_dio_end_io_write() within
the workqueue thread context. Furthermore, since iomap_dio_ops.end_io
still cannot access any private data, it cannot accurately wait for the
work to finish. Testing with xfstests triggered hangs and crashes.
In iomap end_io, there is no dwc. ocfs2_iomap_dio_end_io_write() calls
ocfs2_lock_allocators() with a roughly calculated extents_to_split value.
v5 changelog:
Fix several issues found during the sashiko review[2].
Remove "map->flags = 0" in ocfs2_map_blocks().
Remove some useless/unnecessary OCFS2_IOMAP_* definitions.
When iomap falls back to buffered reads, add a check to ensure ki_pos is
strictly less than i_size.
In ocfs2_iomap_begin():
- Remove redundant code for the case where blocks are already allocated.
- Remove the IOMAP_ATOMIC code, which is not currently supported.
In ocfs2_file_write_iter():
- call kiocb_modified() after ocfs2_prepare_inode_for_write().
- use direct_write_fallback() to handle fallback to the buffered write case.
[1]:
https://gitee.com/anolis/storage-cloud-kernel/commit/ea3fe75e2060118ba093e90b1cd34fa382914b29
[2]:
https://sashiko.dev/#/patchset/20261103135605.22943-1-heming.zhao%40suse.com
v1 patch set:
https://lore.kernel.org/ocfs2-devel/20260724052559.15409-1-heming.zhao@xxxxxxxx/
v2 patch set:
https://lore.kernel.org/ocfs2-devel/20260727061802.18485-1-heming.zhao@xxxxxxxx/
v3 patch set:
https://lore.kernel.org/ocfs2-devel/20261018143319.22329-1-heming.zhao@xxxxxxxx/
v3 sashiko review:
https://sashiko.dev/#/patchset/20261018143319.22329-1-heming.zhao%40suse.com
v4 patch set:
https://lore.kernel.org/ocfs2-devel/20261103135605.22943-1-heming.zhao@xxxxxxxx/
v4 sashiko review:
https://sashiko.dev/#/patchset/20261103135605.22943-1-heming.zhao%40suse.com
--------------
# test report
This patch set has been tested on 7.3.0-rc7 kernel with the following test
cases.
## ltp
runltp -f dio -d /mnt/ocfs2/
The test results: passed
## xfstests
./check -g quick -T -b -s ocfs2 -e generic/081 \
-e generic/266 -e generic/272 -e generic/281 \
-e generic/338 -e generic/347 -e generic/361 \
-e generic/628 -e generic/629 -e generic/648 -e generic/650
The test results are the same before and after the patch set.
```v7.4-rc1
OCFS2_MKFS_LOCAL_OPT="-b 2K -C 8K --fs-features=local --fs-feature-level=max-features"
Failures: generic/003 generic/056 generic/075 generic/091 generic/094
generic/112 generic/177 generic/211 generic/225 generic/263 generic/321
generic/322 generic/329 generic/331 generic/370 generic/376 generic/424
generic/448 generic/467 generic/477 generic/481 generic/498 generic/510
generic/512 generic/513 generic/534 generic/537 generic/552 generic/557
generic/563 generic/564 generic/578 generic/588 generic/607 generic/620
generic/630 generic/755 generic/759 generic/760 generic/771 generic/777
generic/779 generic/784 ocfs2/001
Failed 44 of 632 tests
```
Please note v7.3 kernel only has 35 cases failed, and v7.4-rc1 has 44
cases failed, so the failures are not caused by this patch set.
OCFS2_MKFS_LOCAL_OPT="-b 2K -C 8K --fs-features=local --fs-feature-level=max-features"
Failures: generic/003 generic/075 generic/091 generic/094 generic/112
generic/177 generic/211 generic/225 generic/263 generic/322 generic/329
generic/331 generic/370 generic/376 generic/424 generic/448 generic/467
generic/477 generic/481 generic/510 generic/513 generic/537 generic/552
generic/563 generic/564 generic/578 generic/607 generic/620 generic/630
generic/755 generic/759 generic/760 generic/777 generic/784 ocfs2/001
Failed 35 of 632 tests
## fio on single node
fio --direct=1 --iodepth=32 --rw=write --ioengine=libaio --bs=128k --size=1g \
--numjobs=4 --runtime=100 --group_reporting --filename=/mnt/ocfs2/write.1g --name=tst
- before patch write: IOPS=31.5k, BW=3935MiB/s (4126MB/s)(4096MiB/1041msec)
- before path overwrite (don't remove write.1g, run above fio command):
write: IOPS=39.5k, BW=4935MiB/s (5175MB/s)(4096MiB/830msec)
- applied patch write: IOPS=37.4k, BW=4676MiB/s (4903MB/s)(4096MiB/876msec)
- applied patch overwrite (don't remove write.1g, run above fio command):
write: IOPS=39.4k, BW=4929MiB/s (5168MB/s)(4096MiB/831msec)
fio --direct=1 --iodepth=32 --rw=read --ioengine=libaio --bs=128k --size=1g \
--numjobs=4 --runtime=100 --group_reporting --filename=/mnt/ocfs2/write.1g --name=tst
- before patch : IOPS=47.7k, BW=5962MiB/s (6252MB/s)(4096MiB/687msec)
- applied patch: IOPS=48.2k, BW=6024MiB/s (6316MB/s)(4096MiB/680msec)
fio --direct=1 --iodepth=32 --rw=randwrite --ioengine=libaio --bs=4k --size=1g \
--numjobs=4 --runtime=100 --group_reporting --filename=/mnt/ocfs2/write.1g --name=tst
- before patch
- creating write : IOPS=26.9k, BW=105MiB/s (110MB/s)(4096MiB/38985msec)
- overwriting write.1g: IOPS=231k, BW=902MiB/s (946MB/s)(4096MiB/4539msec)
- applied patch
- creating write : IOPS=26.9k, BW=105MiB/s (110MB/s)(4096MiB/38987msec)
- overwriting write.1g: IOPS=228k, BW=891MiB/s (935MB/s)(4096MiB/4595msec)
fio --direct=1 --iodepth=32 --rw=randread --ioengine=libaio --bs=4k --size=1g \
--numjobs=4 --runtime=100 --group_reporting --filename=/mnt/ocfs2/write.1g --name=tst
- before patch read: IOPS=361k, BW=1409MiB/s (1477MB/s)(4096MiB/2907msec)
- applied patch read: IOPS=362k, BW=1416MiB/s (1485MB/s)(4096MiB/2893msec)
---------------------
Heming Zhao (4):
ocfs2: Add new ocfs2_map_blocks() to introduce iomap feature
ocfs2: switch DIO read path from buffer_head to iomap
ocfs2: switch dio write path from buffer_head to iomap
ocfs2: remove legacy blockdev direct-IO path APIs
fs/ocfs2/Kconfig | 2 +-
fs/ocfs2/aops.c | 539 ++++++++++++++++++++++++++++----------------
fs/ocfs2/aops.h | 38 +---
fs/ocfs2/file.c | 213 ++++++++++++++---
fs/ocfs2/ocfs2.h | 25 +-
fs/ocfs2/ocfs2_fs.h | 3 +
6 files changed, 565 insertions(+), 255 deletions(-)
--
2.55.0