]> git.karo-electronics.de Git - mv-sheeva.git/commit
writeback: fix WB_SYNC_NONE writeback from umount
authorJens Axboe <jens.axboe@oracle.com>
Mon, 17 May 2010 10:55:07 +0000 (12:55 +0200)
committerJens Axboe <jens.axboe@oracle.com>
Mon, 17 May 2010 10:55:07 +0000 (12:55 +0200)
commite913fc825dc685a444cb4c1d0f9d32f372f59861
treee470697e43ffe4028ac81c17d3ef90ee9f30bcfb
parent69b62d01ec44fe0d505d89917392347732135a4d
writeback: fix WB_SYNC_NONE writeback from umount

When umount calls sync_filesystem(), we first do a WB_SYNC_NONE
writeback to kick off writeback of pending dirty inodes, then follow
that up with a WB_SYNC_ALL to wait for it. Since umount already holds
the sb s_umount mutex, WB_SYNC_NONE ends up doing nothing and all
writeback happens as WB_SYNC_ALL. This can greatly slow down umount,
since WB_SYNC_ALL writeback is a data integrity operation and thus
a bigger hammer than simple WB_SYNC_NONE. For barrier aware file systems
it's a lot slower.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
fs/fs-writeback.c
fs/sync.c
include/linux/backing-dev.h
include/linux/writeback.h
mm/page-writeback.c