-
Darrick J. Wong authored
If a blkdev_issue_flush fails, fsync needs to report that to upper levels. Modify xfs_file_fsync to capture the errors, while trying to flush as much data and log updates to disk as possible. If log writes cannot flush the data device, we need to shut down the log immediately because we've violated a log invariant. Modify this code to check the return value of blkdev_issue_flush as well. This behavior seems to go back to about 2.6.15 or so, which makes this fixes tag a bit misleading. Link: https://elixir.bootlin.com/linux/v2.6.15/source/fs/xfs/xfs_vnodeops.c#L1187 Fixes: b5071ada ("xfs: remove xfs_blkdev_issue_flush") Signed-off-by:
Darrick J. Wong <djwong@kernel.org> Reviewed-by:
Dave Chinner <dchinner@redhat.com>
7d839e32