Re: VFS deadlock ?

[Al Viro <viro@ZenIV.linux.org.uk>]

On Thu, Mar 21, 2013 at 06:33:35PM -0700, Linus Torvalds wrote:
> On Thu, Mar 21, 2013 at 6:22 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > In theory, we can make vfs_rmdir() and vfs_unlink() check the presense of
> > the corresponding method before locking the victim; that would suffice to
> > kludge around that mess on procfs.  Along with ->d_inode comparison in
> > lock_rename() it *might* suffice.
> 
> Hmm, yes. Maybe we can do that as a stopgap, backport that, and leave
> any bigger changes for the development tree. That would make the issue
> less urgent, never mind all the other worries about backporting
> complicated patches for subtle issues.
> 
> I realize you aren't entirely thrilled about it, but we actually
> already seem to do that check in both vfs_rmdir().and vfs_unlink()
> before getting the child i_mutex.  I wonder if that is because we've
> already seen lockdep splats for this case...

Yeah, I went to do such patch after sending the previous mail and noticed
that we already did it that way.  Simplicity of error recovery was probably
more important consideration there - I honestly don't remember the reasoning
in such details; it had been a decade or so...  So lock_rename() doing
->d_inode comparison (with dire comment re not expecting that to be sufficient
for anything other than this bug in procfs) will probably suffice for fs/namei.c
part of it; I'm still looking at dcache.c side of things...