]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commit
iomap: iomap that extends beyond EOF should be marked dirty
authorDave Chinner <dchinner@redhat.com>
Thu, 17 Oct 2019 20:12:01 +0000 (13:12 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 17 Oct 2019 20:12:01 +0000 (13:12 -0700)
commit7684e2c4384d5d1f884b01ab8bff2369e4db0bff
tree38603082924ece0934d128b7ed08be6a39fa3d23
parent906753befc4d2610194cd4d3d2ed15dff1ed1ca0
iomap: iomap that extends beyond EOF should be marked dirty

When doing a direct IO that spans the current EOF, and there are
written blocks beyond EOF that extend beyond the current write, the
only metadata update that needs to be done is a file size extension.

However, we don't mark such iomaps as IOMAP_F_DIRTY to indicate that
there is IO completion metadata updates required, and hence we may
fail to correctly sync file size extensions made in IO completion
when O_DSYNC writes are being used and the hardware supports FUA.

Hence when setting IOMAP_F_DIRTY, we need to also take into account
whether the iomap spans the current EOF. If it does, then we need to
mark it dirty so that IO completion will call generic_write_sync()
to flush the inode size update to stable storage correctly.

Fixes: 3460cac1ca76 ("iomap: Use FUA for pure data O_DSYNC DIO writes")
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
[darrick: removed the ext4 part; they'll handle it separately]
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/xfs/xfs_iomap.c
include/linux/iomap.h