Re: Kernel crash due to huge pressure on neighbour

[Willy Tarreau <w@1wt.eu>]

Hi Jia,

On Sat, Sep 30, 2017 at 01:03:36PM +0800, ??(??) wrote:
> Hi David and Willy,
> 
> The following kernel crash is observed, guessing the neighbour code seems to
> have a race condition:
(...)
> I create a patch to fix this issus, but I cannot confirm whether it resolves
> the root cause. This patch equals to adding del_timer_sync() in
> neigh_destroy() before freeing struct neighbour.

Thanks for your detailed analysis, but I'm having a problem with the process
here :
  - if this bug is specific to 3.10 because I failed on a backport, it's
    important to spot the faulty commit and find why it was wrong before
    usng your patch to ensure that it's sufficient and doesn't just targets
    the visible part of the iceberg.

  - if this bug affects other branches (and possibly linus' tree), the fix
    must be applied upstream first then later backported to all relevant
    stable trees.

  - if it happens that this bug used to affect some older kernels and was
    accidently fixed some time ago during some code refactoring, at least
    it's important to know what commit killed it so that all stable
    maintainers know whether or not they have to pick the fix or not.

In any case, a patch touching anything under net/ must be discussed on the
netdev list, where the people who really know the code are.

Thanks,
Willy