On Tue, Apr 17, 2012 at 09:33:06AM -0700, Linus Torvalds wrote: > On Tue, Apr 17, 2012 at 9:22 AM, J. Bruce Fields <bfields@xxxxxxxxxxxx> wrote: > > > > Ugh, no, I think you're right: > > > > rename A/a->A/b > > rename B/b->B/b > > > > where A/a and B/a are the same file, and A/b and B/b are the same file, > > can result in the first rename holding the lock on A and a and waiting > > on b, and the second holding the lock on B and b and waiting on a. > > In fact I don't think you need even that much. Just a simple > > touch a > ln a b > mv a b > > looks like it should deadlock on itself, no? source and dest inodes > will be the same, so the mutex_lock() will just deadlock without even > any ABBA race. > > (I didn't really check - maybe there is some reason that doesn't happen). Yeah, rename has that funny exception that makes the above a no-op, so I think that's safe. But the patch is still wrong; back to the drawing board. Maybe a paper bag over my head will help my concentration.... --b. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html