Re: .17rc5 cfq slab corruption.

[Jens Axboe <axboe@suse.de>]

On Tue, May 30 2006, Dave Jones wrote:
> On Tue, May 30, 2006 at 03:17:28PM +0200, Jens Axboe wrote:
>  > On Sat, May 27 2006, Dave Jones wrote:
>  > > On Sat, May 27, 2006 at 09:07:24AM +0200, Jens Axboe wrote:
>  > >  > On Fri, May 26 2006, Andrew Morton wrote:
>  > >  > > Dave Jones <davej@redhat.com> wrote:
>  > >  > > >
>  > >  > > > Was playing with googles new picasa toy, which hammered the disks
>  > >  > > > hunting out every image file it could find, when this popped out:
>  > >  > > > 
>  > >  > > > Slab corruption: (Not tainted) start=ffff810012b998c8, len=168
>  > >  > > > Redzone: 0x5a2cf071/0x5a2cf071.
>  > >  > > > Last user: [<ffffffff8032c319>](cfq_free_io_context+0x2f/0x74)
>  > >  > > > 090: 10 bd 28 1b 00 81 ff ff 6b 6b 6b 6b 6b 6b 6b 6b
>  > >  > > > Prev obj: start=ffff810012b99808, len=168
>  > >  > > > Redzone: 0x5a2cf071/0x5a2cf071.
>  > >  > > > Last user: [<ffffffff8032c319>](cfq_free_io_context+0x2f/0x74)
>  > >  > > > 000: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
>  > >  > > > 010: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
>  > >  > > > Next obj: start=ffff810012b99988, len=168
>  > >  > > > Redzone: 0x5a2cf071/0x5a2cf071.
>  > >  > > > Last user: [<ffffffff8032c319>](cfq_free_io_context+0x2f/0x74)
>  > >  > > > 000: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
>  > >  > > > 010: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
>  > >  > 
>  > >  > Pretty baffling... cfq has been hammered pretty thoroughly over the
>  > >  > last months and _nothing_ has shown up except some performance anomalies
>  > >  > that are now fixed. Since daves case (at least) seems to be
>  > >  > use-after-free, I'll see if I can reproduce with some contrived case.
>  > >  > I'm asuming that picasa forks and exits a lot with submitted io in
>  > >  > between than may not have finished at exit.
>  > > 
>  > > The second time I hit it, was actually during boot up.
>  > 
>  > Dave, do you have any io scheduler switching going on?
> 
> Nope, everything set to use CFQ as default, and left that way.

Hrmpf ok, I had hoped perhaps something in your init scripts modified
the scheduler value.

-- 
Jens Axboe