Comment 210 for bug 330824

Revision history for this message
Theodore Ts'o (tytso) wrote :

We might have a break in this bug. For those people who can reliably reproduce the problem, are you using ecryptfs, possibly extensively? Roland Drier has reported a potential lockdep report that might explain why some of us have had extreme problems reproducing the problem; namely we might not be using ecryptfs. See:

http://lkml.org/lkml/2009/7/4/93

If so, there is a sample patch which **might*** fix this problem. See:

http://lkml.org/lkml/2009/7/5/79

Also, if people could try building their kernel with CONFIG_PROVE_LOCKING, that would be interesting to see if we get a lockdep report.

If so, it might be that this bug has been around all along, but it's something about Ubuntu patches that makes it 100,000 times more likely to trigger. (In practice, it looks like it should only trigger on a truncate to a size which is not a multiple of the filesystem blocksize, not on an unlink --- but maybe there was a bug in the Ubuntu backports that made this possible to trigger on an unlink. This is only a theory, but it's the first lockdep report I've gotten that could at least potentially be related to this problem that to date, only Ubuntu users have been complaining about, even though at this point we've got a huge number of Fedora 11 users using ext4 w/o any problems.)