Re: [PATCH 3/3 for 2.6.38] oom: oom_kill_process: fix the child_points logic

[Oleg Nesterov <oleg@redhat.com>]

On 03/14, David Rientjes wrote:
>
> On Mon, 14 Mar 2011, Oleg Nesterov wrote:
>
> > oom_kill_process() starts with victim_points == 0. This means that
> > (most likely) any child has more points and can be killed erroneously.
> >
> > Also, "children has a different mm" doesn't match the reality, we
> > should check child->mm != t->mm. This check is not exactly correct
> > if t->mm == NULL but this doesn't really matter, oom_kill_task()
> > will kill them anyway.
> >
> > Note: "Kill all processes sharing p->mm" in oom_kill_task() is wrong
> > too.
> >
>
> There're two issues you're addressing in this patch.  It only kills a 
> child in place of its selected parent when:
>
>  - the child has a higher badness score, and
>
>  - it has a different ->mm.
>
> In the former case, NACK, we always want to sacrifice children regardless
> of their badness score (as long as it is non-zero) if it has a separate
> ->mm in place of its parent,

Ah. So this was intentional?

OK. I was hypnotized by the security implications, and this looked so
"obviously wrong" to me.

But, of course I can't judge when it comes to oom's heuristic, and you
certainly know better.

So, thanks for correcting me.


Just a question... what about oom_kill_allocating_task? Probably
Documentation/sysctl/vm.txt should be updated.

Oleg.