Re: fs: dcookie: freeing active timer

[Sasha Levin <sasha.levin@oracle.com>]

On 04/24/2014 07:49 PM, Al Viro wrote:
> On Thu, Apr 24, 2014 at 10:55:58PM +0100, Al Viro wrote:
>> On Thu, Apr 24, 2014 at 01:34:14PM -0400, Sasha Levin wrote:
>>
>>>> Why does that code bother with destroying/creating that sucker dynamically?
>>>> Is there any point at all?
>>>
>>> I'm not sure about the dynamic allocation part, but I fear that if we just
>>> switch to using static allocations it'll hide the underlying issue that
>>> triggered this bug instead of fixing it.
>>
>> FWIW, slub.c variant of kmem_cache_destroy() is buggered - struct kobject
>> embedded into struct kmem_cache, its ktype is slab_ktype, which has
>> NULL ->release()...
> 
> BTW, if your config has CONFIG_DEBUG_KOBJECT_RELEASE, that's exactly where
> that warning comes from.  Got broken by commit b7454a,
> Author: Glauber Costa <glommer@parallels.com>
> Date:   Fri Oct 19 18:20:25 2012 +0400
> 
>     mm/sl[au]b: Move slabinfo processing to slab_common.c
> 
> We *do* need ->release().  Greg and guilty parties Cc'd...

We actually had that conversation a long time ago, and Christoph has
sent out a patch to fix that (http://www.spinics.net/lists/netdev/msg259431.html).

I was assuming that it was merged upstream and went straight to blaming
fs/ (and Greg's drivers/usb/ actually) without checking that first. Sorry!

Could someone pretty please merge that patch? Specially since Greg acked it?


Thanks,
Sasha