fs/attr.c:notify_change locking warning.

[Christoph Hellwig]

On Wed, Oct 16, 2013 at 09:26:51PM +1100, Dave Chinner wrote:
> The killpriv calls? I couldn't find anything that implemented those
> security hooks nor any documentation about it, so I'm pretty much
> clueless about it. FWIW, ocfs2 doesn't implement them, either....

The killpriv code ends up doing xattr calls for per-file capabilities
(grep security/commoncap.c for killpriv).  Seems like ocfs2 is buggy in
that regard.

I suspect the easiest way to solve it properly in XFS is to simply
retake the iolock exclusive and get the i_mutex as part of it.  This
means direct I/O writes to files with the suid bit won't scale, but I
think we can live with that given that it avoids introducing special
cases that impact more code.