Re: [PING][PATCH][BZ #15073] Fix race in free.

[Maxim Kuvyrkov <maxim at kugelworks dot com>]

On 28/12/2013, at 12:15 pm, Carlos O'Donell <carlos@redhat.com> wrote:

> On 12/27/2013 06:05 PM, Maxim Kuvyrkov wrote:
>> Dave, Carlos, Andreas,
>> 
>> Consider backporting fix [*] for BZ #15073
>> (https://sourceware.org/bugzilla/show_bug.cgi?id=15073) to your
>> respective release branches.
>> 
>> The bug is present since 2.12, conditioned on ATOMIC_FASTBINS, which
>> became unconditional default in 2.15.  Therefore backporting to 2.12
>> and 2.14 is optional, but 2.15 onwards is very desirable.
>> 
>> The patch applies cleanly to 2.15 onwards.
>> 
>> The essence of the fix is to avoid "chunksize(old)" memory reference
>> since OLD can be deallocated if !HAVE_LOCK.
>> 
>> [*] 362b47fe09ca9a928d444c7e2f7992f7f61bfc3e
>> 
>> Let me know if you have any questions or if the fix does not appear
>> to be obvious/safe.
> 
> So the way this works is that the responsibility for backporting
> and testing the bacported patch falls to the developer making
> the backport request. The branch maintainer is only responsible
> for reviewing and ACK'ing or NACK'ing that request.
> 
> Thus if there are interested parties in getting this fixed for
> any non-trunk branches then those parties need to backport, test,
> and submit the patch as usual for ACK'ing by the branch maintainer.

Fair enough.  I will backport and test the patch for 2.15 onwards.  Do I have your ACK to commit to 2.15 and 2.16 provided there are no regressions?

Patch is attached.

Thanks, and Happy Holidays!

--
Maxim Kuvyrkov
www.kugelworks.com

Attachment: 0001-Fix-race-in-free-of-fastbin-chunk-BZ-15073.patch
Description: Binary data