[PATCH] xfs: wait for direct I/O before swapping extents

[PATCH] xfs: wait for direct I/O before swapping extents

[Daejun Park via B4 Relay]

From: Daejun Park <daejun7.park@samsung.com>

An AIO direct write drops the iolock before its bios complete.
XFS_IOC_SWAPEXT does not wait for it. The write can land in blocks that
the swap moved to the temporary file. Once xfs_fsr frees them, the
write corrupts whatever reuses them.

Wait for direct I/O first, as exchange-range does.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Daejun Park <daejun7.park@samsung.com>
---
Tested with an AIO direct write held back by cgroup io.max. Before, the
write reached files that reused the freed blocks in 7 of 7 runs. After,
those files were intact. The new fstests case "xfs: test that swapext
waits for in-flight direct I/O" fails before and passes after. Running
fstests -g fsr shows no change.

Applies with or without "xfs: break layouts before swapping extents".
---
 fs/xfs/xfs_bmap_util.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index 865aec3ae5a..4d214708f9c 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -1257,6 +1257,7 @@ xfs_swap_extent_flush(
 {
 	int	error;
 
+	inode_dio_wait(VFS_I(ip));
 	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
 	if (error)
 		return error;

---
base-commit: 1e17fc56469d62d1c1ac45c137a2456d14456f1d
change-id: 20261109-xfs-swapext-dio-wait-6b700c2a981e

Best regards,
-- 
Daejun Park <daejun7.park@samsung.com>

Re: [PATCH] xfs: wait for direct I/O before swapping extents

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>