Re: Re:[PATCH v2]oom-kill: CAP_SYS_RESOURCE should get bonus

[Figo.zhang]

On Wed, 2010-11-03 at 19:54 -0700, David Rientjes wrote:
> On Thu, 4 Nov 2010, Figo.zhang wrote:
> 
> > In your new heuristic, you also get CAP_SYS_RESOURCE to protection.
> > see fs/proc/base.c, line 1167:
> > 	if (oom_score_adj < task->signal->oom_score_adj &&
> > 			!capable(CAP_SYS_RESOURCE)) {
> > 		err = -EACCES;
> > 		goto err_sighand;
> > 	}
> 
> That's unchanged from the old behavior with oom_adj.
> 
> > so i want to protect some process like normal process not
> > CAP_SYS_RESOUCE, i set a small oom_score_adj , if new oom_score_adj is
> > small than now and it is not limited resource, it will not adjust, that
> > seems not right?
> > 
> 
> Tasks without CAP_SYS_RESOURCE cannot lower their own oom_score_adj, 

CAP_SYS_RESOURCE == 1 means without resource limits just like a
superuser,
CAP_SYS_RESOURCE == 0 means hold resource limits, like normal user,
right?

a new lower oom_score_adj will protect the process, right?

Tasks without CAP_SYS_RESOURCE, means that it is not a superuser, why
user canot protect it by oom_score_adj?

like i want to protect my program such as gnome-terminal which is
without CAP_SYS_RESOURCE (have resource limits), 

[figo@myhost ~]$ ps -ax | grep gnome-ter
Warning: bad ps syntax, perhaps a bogus '-'? See
http://procps.sf.net/faq.html
 2280 ?        Sl     0:01 gnome-terminal
 8839 pts/0    S+     0:00 grep gnome-ter
[figo@myhost ~]$ cat /proc/2280/oom_adj 
3
[figo@myhost ~]$ echo -17 >  /proc/2280/oom_adj 
bash: echo: write error: Permission denied
[figo@myhost ~]$ 

so, i canot protect my program.


> otherwise it can trivially kill other tasks.  They can, however, increase 
> their own oom_score_adj so the oom killer prefers to kill it first.
> 
> I think you may be confused: CAP_SYS_RESOURCE override resource limits.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/