ocfs2: reflink: fix slow unlink for refcounted file
authorJunxiao Bi <junxiao.bi@oracle.com>
Fri, 19 Dec 2014 00:17:32 +0000 (16:17 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 19 Dec 2014 03:08:11 +0000 (19:08 -0800)
commitf62f12b3a426c8f65b10011b1ec40ba4277cbf5f
tree08e41966685e3bd305089e9a5aed257a2272461c
parentd82fa87d2b60e8affea3b244ad23c5d9a59c584a
ocfs2: reflink: fix slow unlink for refcounted file

When running ocfs2 test suite multiple nodes reflink stress test, for a
4 nodes cluster, every unlink() for refcounted file needs about 700s.

The slow unlink is caused by the contention of refcount tree lock since
all nodes are unlink files using the same refcount tree.  When the
unlinking file have many extents(over 1600 in our test), most of the
extents has refcounted flag set.  In ocfs2_commit_truncate(), it will
execute the following call trace for every extents.  This means it needs
get and released refcount tree lock about 1600 times.  And when several
nodes are do this at the same time, the performance will be very low.

  ocfs2_remove_btree_range()
  --  ocfs2_lock_refcount_tree()
  ----  ocfs2_refcount_lock()
  ------  __ocfs2_cluster_lock()

ocfs2_refcount_lock() is costly, move it to ocfs2_commit_truncate() to
do lock/unlock once can improve a lot performance.

Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Wengang <wen.gang.wang@oracle.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ocfs2/alloc.c
fs/ocfs2/alloc.h
fs/ocfs2/dir.c
fs/ocfs2/file.c