From: Jan Kara <jack@ucw.cz>

I've attached a fix for a problem in ext3 journalled quota patch - the problem
is that detecting whether dqput() sleeps was wrong and so we could possibly
schedule when holding a spinlock.