Re: BUG_ON(nd->inode != parent->d_inode);

[Dave Jones <davej@redhat.com>]

On Sat, Mar 09, 2013 at 12:13:16AM -0500, Sasha Levin wrote:
 > On 03/08/2013 11:39 PM, Dave Jones wrote:
 > > On Fri, Mar 08, 2013 at 08:31:48PM -0800, Linus Torvalds wrote:
 > >  > On Fri, Mar 8, 2013 at 7:50 PM, Dave Jones <davej@redhat.com> wrote:
 > >  > >  > >
 > >  > >  > > I have a feeling there were some sysfs ones that may still be unfixed.
 > >  > >
 > >  > > I was right..
 > >  > >
 > >  > > [  425.836722] general protection fault: 0000 [#1] PREEMPT SMP
 > >  > 
 > >  > You forgot to enable DEBUG_PAGE_ALLOC again, but I don't think it much
 > >  > matters. It's another slab free poison thing.
 > >  > 
 > >  > The faulting instruction is
 > >  > 
 > >  >     mov    0x28(%rbx),%ecx
 > >  > 
 > >  > with %rbx having the value 6b6b6b6b6b6b6b6b.
 > >  > 
 > >  > > [  425.847859] RIP: 0010:[<ffffffff8123f657>]  [<ffffffff8123f657>] sysfs_find_dirent+0x47/0xf0
 > >  > 
 > >  > That seems to be
 > >  > 
 > >  >     if (hash != sd->s_hash)
 > >  > 
 > >  > from sysfs_name_compare() that has been inlined into
 > >  > sysfs_find_dirent(). And where "sd" is the corrupted value. If I read
 > >  > things right.
 > >  > 
 > >  > So it looks like the sysfs rbtree is corrupted or something. Adding
 > >  > Greg to the cc.
 > >  
 > > oh , that rings a bell.  ISTR this had something to do with Sasha's idr changes.
 > 
 > I believe the issue you're referring to is my report of a panic in sysfs keys
 > where the device has gone away? If that is it, I don't think that this issue
 > is related to that one.
 > 
 > If not, I'm not sure which change you're referring to.

Hmm I might have been thinking of the hlist changes rather than idr.
Though looking at those oopses, they were in find_pid_ns, so different.

So many bugs.

	Dave