writeback: fix occasional slow sync(1)
authorJan Kara <jack@suse.cz>
Wed, 11 Sep 2013 21:22:22 +0000 (14:22 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 11 Sep 2013 22:57:55 +0000 (15:57 -0700)
commit47df3ddedd22c3f8e68aff831edb7921937674a2
tree3a8b82782ef6c90e7974063b642e5bbab04fd55e
parent7cb2ef56e6a8b7b368b2e883a0a47d02fed66911
writeback: fix occasional slow sync(1)

In case when system contains no dirty pages, wakeup_flusher_threads() will
submit WB_SYNC_NONE writeback for 0 pages so wb_writeback() exits
immediately without doing anything, even though there are dirty inodes in
the system.  Thus sync(1) will write all the dirty inodes from a
WB_SYNC_ALL writeback pass which is slow.

Fix the problem by using get_nr_dirty_pages() in wakeup_flusher_threads()
instead of calculating number of dirty pages manually.  That function also
takes number of dirty inodes into account.

Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Paul Taysom <taysom@chromium.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/fs-writeback.c