Re: perf: use after free in perf_remove_from_context

From: Sasha Levin
Date: Thu May 29 2014 - 10:48:01 EST


On 05/29/2014 03:57 AM, Peter Zijlstra wrote:
> On Wed, May 28, 2014 at 07:52:07PM -0400, Sasha Levin wrote:
>> On 05/14/2014 12:35 PM, Peter Zijlstra wrote:
>>> On Wed, May 14, 2014 at 12:32:26PM -0400, Sasha Levin wrote:
>>>>> -----BEGIN PGP SIGNED MESSAGE-----
>>>>> Hash: SHA1
>>>>>
>>>>> On 05/14/2014 12:29 PM, Peter Zijlstra wrote:
>>>>>>> On Mon, May 12, 2014 at 11:42:33AM -0400, Sasha Levin wrote:
>>>>>>>>> Hi all,
>>>>>>>>>
>>>>>>>>> While fuzzing with trinity inside a KVM tools guest running the latest -next kernel I've stumbled on the following spew. Maybe related to the very recent change in freeing on task exit?
>>>>>>>>>
>>>>>>>>> [ 2509.827261] general protection fault: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC [ 2509.830379] Dumping ftrace buffer: [ 2509.830379] (ftrace buffer empty) [ 2509.830379] Modules linked in: [ 2509.830379] CPU: 47 PID: 43306 Comm: trinity-c126 Tainted: G W 3.15.0-rc5-next-20140512-sasha-00019-ga20bc00-dirty #456
>>>>>>>
>>>>>>> Any particular trinity setup? And would you happen to have the seed of that run?
>>>>>
>>>>> Nothing special about trinity options. 400 threads and blacklisting some of the
>>>>> destructive syscalls (umount, reboot, etc).
>>>>>
>>>>> I don't have the seed, but that problem did reproduce again tonight so I can test
>>>>> out debug code if you have something in mind.
>>> Nah, I drew a pretty big blank, which is why I wanted to see if I could
>>> reproduce. If you could share your trinity cmdline I'd be much obliged.
>>> While I did manage to clone (the repo moved since last time) and build
>>> it, I'm not really that handy with it and want to avoid destroying my
>>> machine if possible ;-)
>>
>> Anything I could do to help out with this? It reproduces pretty easily on my
>> configuration so I'd be happy to test out whatever might help.
>
> Yeah, it takes me days to test anything, and my last guess panned out to
> nothing, at which point I decided I needed to look at the things I'd
> neglected for a bit :/
>
> Could you see if the below makes any difference? I'll try and get back
> to tracking this.
>
> ---
> kernel/events/core.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 9efb1e7858ac..851dc9dc5643 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -7497,8 +7497,10 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
> */
> mutex_lock(&child_ctx->mutex);
>
> + rcu_read_lock();
> list_for_each_entry_rcu(child_event, &child_ctx->event_list, event_entry)
> __perf_event_exit_task(child_event, child_ctx, child);
> + rcu_read_unlock();
>
> mutex_unlock(&child_ctx->mutex);
>
>

It doesn't work out well because we later lock a mutex in sync_child_event().


Thanks,
Sasha
--
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/