Will backports to 3.10 kernel for avoiding IPv6 crashes be accepted?

Will backports to 3.10 kernel for avoiding IPv6 crashes be accepted?

[Mike Manning]

Hello,

Will you in principle accept submission of patches for backports to the 3.10 kernel?

If so, I will start submitting some patches on a one-by-one basis for the following.

We have an older product using the 3.10 kernel that has been difficult to upgrade due to other issues and that will be going EOL in the same timeframe.

This has been hitting kernel crashes in fib6_clean_all(), kernel panics due to NULL pointer deref, kernel panics in fib6_purge_rt() due to invalid ref count (which turn out to be due to rtnl locking issues) for some usecases, to the extent that the only reliable workaround was to disable IPv6.

These crashes have been resolved by backporting to 3.10 a set of 6 patches from kernel versions 3.11-3.14 (a couple of these backports whilst useful are only to make subsequent fixes cleaner to patch).


Thanks
Mike Manning

Re: Will backports to 3.10 kernel for avoiding IPv6 crashes be accepted?

[Willy Tarreau]

Hi Mike,

On Tue, Dec 13, 2016 at 03:42:36PM +0000, Mike Manning wrote:
> Hello,
> 
> Will you in principle accept submission of patches for backports to the 3.10 kernel?

We accept fixes as long as they :
  1) exist in mainline, or have a good reason for not being there (eg: bug was
     accidently fixed later)

  2) do not break existing setups or change any behaviour

  3) are auditable by people who know the code

  4) are not rejected by the relevant subsystem maintainers

> If so, I will start submitting some patches on a one-by-one basis for the
> following.
(...)
> These crashes have been resolved by backporting to 3.10 a set of 6 patches
> from kernel versions 3.11-3.14 (a couple of these backports whilst useful are
> only to make subsequent fixes cleaner to patch).

Well, let's see. I'm a bit late these days on 3.10 unfortunately but I still
have some fixes waiting in the queue for 3.10.105, so if your fixes are OK,
I'm fine with including them as well.

Thanks,
Willy