In message <20080402081215.GA2716@xxxxxxxxxxxx>, Jan Kara writes: > On Tue 01-04-08 17:23:34, Erez Zadok wrote: > <snip> [...] > I'll try to explain the locking here: > 1) We are not allowed to call into __invalidate_mapping_pages() with > inode_lock held - that it the bug lockdep is complaining about. Moreover it > leads to rather long waiting times for inode_lock (quite possibly several > seconds). > 2) When we release inode_lock, we need to protect from inode going away, > thus we hold reference to it - that guarantees us inode stays in the list. > 3) When we continue scanning of the list we must get inode_lock before we > put the inode reference to avoid races. But we cannot do iput() when we > hold inode_lock. Thus we save pointer to inode and do iput() next time we > have released the inode_lock... > > > Peter's post: > > > > http://lkml.org/lkml/2008/3/23/202 > > > > includes a reference to a mail by Andrew which implies that the fix may be > > much more involved than what you outlined above, no? > Definitely we can do a more involved fix ;) What Andrew proposes would have > some other benefits as well. But until somebody gets to that, this slight > hack should work fine (Andrew has already merged my patch in -mm so I > guess he agrees ;). Thanks for the explanation. And good to know it's in -mm. I'll give it a try. Cheers, Erez. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
