Re: Improving OOM killer

[David Rientjes]

On Thu, 4 Feb 2010, Frans Pop wrote:

> Shouldn't fork bomb detection take into account the age of children?
> After all, long running processes with a lot of long running children are 
> rather unlikely to be runaway fork _bombs_.
> 

Yeah, Lubos mentioned using cpu time as a requirement, in addition to the 
already existing child->mm != parent->mm, as a prerequisite to be added 
into the tally to check the forkbomb threshold.  I think something like 
this would be appropriate:

	struct task_cputime task_time;
	int forkcount = 0;
	int child_rss = 0;

	...

	list_for_each_entry(child, &p->children, sibling) {
		unsigned long runtime;

		task_lock(child);
		if (!child->mm || child->mm == p->mm) {
			task_unlock(child);
			continue;
		}
		thread_group_cputime(child, &task_time);
		runtime = cputime_to_jiffies(task_time.utime) +
				cputime_to_jiffies(task_time.stime);

		/*
		 * Only threads that have run for less than a second are
		 * considered toward the forkbomb, these threads rarely
		 * get to execute at all in such cases anyway.
		 */
		if (runtime < HZ) {
			task_unlock(child);
			continue;
		}
		child_rss += get_mm_rss(child->mm);
		forkcount++;
	}

	if (forkcount > sysctl_oom_forkbomb_thres) {
		/*
		 * Penalize forkbombs by considering the average rss and
		 * how many factors we are over the threshold.
		 */
		points += child_rss / sysctl_oom_forkbomb_thres;
	}

I changed the calculation from lowest child rss to average child rss, so 
this is functionally equivalent to

(average rss size of children) * (# of first-generated execve children) /
			sysctl_oom_forkbomb_thres
--
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/