Re: [origin tree SLAB corruption] BUG kmalloc-64: Poison overwritten, INFO: Allocated in bdi_alloc_work+0x2b/0x100 age=175 cpu=1 pid=3514

[Jens Axboe <jens.axboe@oracle.com>]

On Mon, Sep 14 2009, Pekka Enberg wrote:
> * Eric Paris <eparis@redhat.com> wrote:
> >> On Sat, 2009-09-12 at 09:24 +0200, Ingo Molnar wrote:
> >> > James - i did not see a security pull request email from you in my
> >> > lkml folder so i created this new thread. -tip testing found the
> >> > easy crash below. It reverts cleanly so i went that easy route.
> >> >
> >> > At a really quick 10-seconds glance the crash happens because we
> >> > destroy the slab cache twice, if the sysctl is toggled twice?
> >>
> >> Something a lot worse than SELinux here.  I added this exact code and
> >> got this warning.  Something is wrong in the world of
> >> kmem_cache_destroy.....
> 
> Btw, the kmem_cache_destroy() bug Eric found is not in Linu's tree yet.
> 
> On Mon, Sep 14, 2009 at 10:16 AM, Ingo Molnar <mingo@elte.hu> wrote:
> > -tip testing just triggered another type of SLAB problem (this time
> > not apparently related to the security subsystem):
> >
> > BUG kmalloc-64: Poison overwritten
> > -----------------------------------------------------------------------------
> >
> > INFO: 0xf498f6a0-0xf498f6a7. First byte 0x90 instead of 0x6b
> > INFO: Allocated in bdi_alloc_work+0x2b/0x100 age=175 cpu=1 pid=3514
> > INFO: Freed in bdi_work_free+0x45/0x60 age=9 cpu=1 pid=3509
> > INFO: Slab 0xc3257d84 objects=36 used=11 fp=0xf498f690 flags=0x400000c3
> > INFO: Object 0xf498f690 @offset=1680 fp=0xf498fe00
> >
> > Bytes b4 0xf498f680:  ab 0d 00 00 9c 27 ff ff 5a 5a 5a 5a 5a 5a 5a 5a «....'ÿÿZZZZZZZZ
> >  Object 0xf498f690:  6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk
> >  Object 0xf498f6a0:  90 f3 98 f4 60 3c 11 c1 6b 6b 6b 6b 6b 6b 6b 6b .ó.ô`<.Ákkkkkkkk
> 
> This would be use-after-free in kmalloc-64 cache. Given the trace and
> the fact that bdi_work_alloc() got introduce recently, it seems more
> likely that fs/fs-writeback.c is to blame here. Jens, does the warning
> ring a bell to you?

No bells, the code seems right to me. I'll prod at it a bit more. I
haven't seen anything like this during testing.

-- 
Jens Axboe