xfs: don't free post-EOF blocks on read close

When we have a workload that does open/read/close in parallel with other
allocation, the file becomes rapidly fragmented. This is due to close()
calling xfs_file_release() and removing the speculative preallocation
beyond EOF.

Add a check for a writable context to xfs_file_release to skip the
post-EOF block freeing (an the similarly pointless flushing on truncate
down).

Before:

Test 1: sync write fragmentation counts

/mnt/scratch/file.0: 919
/mnt/scratch/file.1: 916
/mnt/scratch/file.2: 919
/mnt/scratch/file.3: 920
/mnt/scratch/file.4: 920
/mnt/scratch/file.5: 921
/mnt/scratch/file.6: 916
/mnt/scratch/file.7: 918

After:

Test 1: sync write fragmentation counts

/mnt/scratch/file.0: 24
/mnt/scratch/file.1: 24
/mnt/scratch/file.2: 11
/mnt/scratch/file.3: 24
/mnt/scratch/file.4: 3
/mnt/scratch/file.5: 24
/mnt/scratch/file.6: 24
/mnt/scratch/file.7: 23

Signed-off-by: Dave Chinner <dchinner@redhat.com>
[darrick: wordsmithing, fix commit message]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
[hch: ported to the new ->release code structure]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index dae8dd1223550d935252067aeb66c8e319547b3c..60424e64230743c7dedee67eb1d6b008976c5697 100644 (file)
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -1217,12 +1217,18 @@ xfs_file_release(
         * There is no point in freeing blocks here for open but unlinked files
         * as they will be taken care of by the inactivation path soon.
         *
+        * When releasing a read-only context, don't flush data or trim post-EOF
+        * blocks.  This avoids open/read/close workloads from removing EOF
+        * blocks that other writers depend upon to reduce fragmentation.
+        *
         * If we can't get the iolock just skip truncating the blocks past EOF
         * because we could deadlock with the mmap_lock otherwise. We'll get
         * another chance to drop them once the last reference to the inode is
         * dropped, so we'll never leak blocks permanently.
         */
-       if (inode->i_nlink && xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
+       if (inode->i_nlink &&
+           (file->f_mode & FMODE_WRITE) &&
+           xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
                if (xfs_can_free_eofblocks(ip) &&
                    !xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) {
                        /*