summaryrefslogtreecommitdiff
path: root/fs/xfs/xfs_reflink.c
diff options
context:
space:
mode:
authorChristoph Hellwig <hch@lst.de>2019-02-18 20:38:46 +0300
committerDarrick J. Wong <darrick.wong@oracle.com>2019-02-21 18:55:07 +0300
commit60271ab79d40b99ce6cb28d8dc48aa5e9ffb6df3 (patch)
tree532b032a3c46bc1dbe5c6ba17e2e30edfb9b1996 /fs/xfs/xfs_reflink.c
parent16be1433737ee46f88da57d47f594c4fc1376538 (diff)
downloadlinux-60271ab79d40b99ce6cb28d8dc48aa5e9ffb6df3.tar.xz
xfs: fix SEEK_DATA for speculative COW fork preallocation
We speculatively allocate extents in the COW fork to reduce fragmentation. But when we write data into such COW fork blocks that do now shadow an allocation in the data fork SEEK_DATA will not correctly report it, as it only looks at the data fork extents. The only reason why that hasn't been an issue so far is because we even use these speculative COW fork preallocations over holes in the data fork at all for buffered writes, and blocks in the COW fork that are written by direct writes are moved into the data fork immediately at I/O completion time. Add a new set of iomap_ops for SEEK_HOLE/SEEK_DATA which looks into both the COW and data fork, and reports all COW extents as unwritten to the iomap layer. While this isn't strictly true for COW fork extents that were already converted to real extents, the practical semantics that you can't read data from them until they are moved into the data fork are very similar, and this will force the iomap layer into probing the extents for actually present data. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Diffstat (limited to 'fs/xfs/xfs_reflink.c')
0 files changed, 0 insertions, 0 deletions