Re: BUG: KASAN: slab-out-of-bounds in perf_callchain_user+0x494/0x530

[Namhyung Kim <namhyung@kernel.org>]

On Wed, Dec 06, 2017 at 04:45:44PM +0100, Peter Zijlstra wrote:
> On Wed, Dec 06, 2017 at 11:31:30PM +0900, Namhyung Kim wrote:
> 
> > > There's also a race against put_callchain_buffers() there, consider:
> > > 
> > > 
> > > 	get_callchain_buffers()		put_callchain_buffers()
> > > 	  mutex_lock();
> > > 	  inc()
> > > 					  dec_and_test() // false
> > > 
> > > 	  dec() // 0
> > > 
> > > 
> > > And the buffers leak.
> > 
> > Hmm.. did you mean that get_callchain_buffers() returns an error?
> 
> Yes, get_callchain_buffers() fails, but while doing so it has a
> temporary increment on the count.
> 
> > AFAICS it cannot fail when it sees count > 1 (and callchain_cpus_
> > entries is allocated).  
> 
> It can with your patch. We only test event_max_stack against the sysctl
> after incrementing.

So, are you ok with this?

Thanks,
Namhyung


diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 1b2be63c8528..ee0ba22d3993 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -137,8 +137,11 @@ int get_callchain_buffers(int event_max_stack)
 
        err = alloc_callchain_buffers();
 exit:
-       if (err)
-               atomic_dec(&nr_callchain_events);
+       if (err) {
+               /* might race with put_callchain_buffers() */
+               if (atomic_dec_and_test(&nr_callchain_events))
+                       release_callchain_buffers();
+       }
 
        mutex_unlock(&callchain_mutex);