Re: hit a KASan bug related to Perf during stress test

[Oleg Nesterov]

On 10/24, Oleg Nesterov wrote:
>
> On 10/24, Peter Zijlstra wrote:
> >
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -1257,7 +1257,14 @@ static u32 perf_event_pid(struct perf_event *event, struct task_struct *p)
> >  	if (event->parent)
> >  		event = event->parent;
> >
> > -	return task_tgid_nr_ns(p, event->ns);
> > +	/*
> > +	 * It is possible the task already got unhashed, in which case we
> > +	 * cannot determine the current->group_leader/real_parent.
> > +	 *
> > +	 * Also, report -1 to indicate unhashed, so as not to confused with
> > +	 * 0 for the idle task.
> > +	 */
> > +	return pid_alive(p) ? task_tgid_nr_ns(p, event->ns) : ~0;
> >  }
>
> Yes, but this _looks_ racy unless p == current. I mean, pid_alive() makes
> task_tgid_nr_ns() safe, but task_tgid_nr_ns() still can return zero _if_
> it can race with the exiting task.
>
> >  static u32 perf_event_tid(struct perf_event *event, struct task_struct *p)
> > @@ -1268,7 +1275,7 @@ static u32 perf_event_tid(struct perf_event *event, struct task_struct *p)
> >  	if (event->parent)
> >  		event = event->parent;
> >
> > -	return task_pid_nr_ns(p, event->ns);
> > +	return pid_alive(p) ? task_pid_nr_ns(p, event->ns) : ~0;
>
> The same.
>
> However. At first glance the only case when p != current is copy_process(),
> right? And in this case the new child can't go away. So I think this patch
> is fine.

Actually there is another case, comm_write() -> perf_event_comm_output(). It
checks same_thread_group(current, p), so we can only race with the exiting
sub-thread. perf_event_pid() can't return zero, perf_event_tid() can.

And I personally think we do not care and your patch is fine anyway ;)

Oleg.