Re: [PATCH] signal: Prevent sigqueue caching after task got released

[Oleg Nesterov <oleg@redhat.com>]

On 06/22, Thomas Gleixner wrote:
>
> syzbot reported a memory leak related to sigqueue caching. This happens
> when a thread group leader with child tasks is reaped.
>
> The group leader's sigqueue_cache is correctly freed. The group leader then
> reaps the child tasks and if any of them has a signal pending it caches
> that signal.

I guess you mean the race with exit_notify() ? Could you spell please?
I am just curious how exactly this problem was found.

This doesn't really matter, because damn yes, a task T can call
release_task(another_task)->sigqueue_cache_or_free() after
exit_task_sigqueue_cache(T) was already called. For example, a last non-leader
thread exits and reaps a zombie leader.

Somehow I thought that exit_task_sigqueue_cache() at the end of __exit_signal()
should fix this problem, but this is obviously wrong.


> @@ -463,13 +469,18 @@ void exit_task_sigqueue_cache(struct tas
>  	struct sigqueue *q = tsk->sigqueue_cache;
>  
>  	if (q) {
> -		tsk->sigqueue_cache = NULL;
>  		/*
>  		 * Hand it back to the cache as the task might
>  		 * be self reaping which would leak the object.
>  		 */
>  		 kmem_cache_free(sigqueue_cachep, q);
>  	}
> +
> +	/*
> +	 * Set an error pointer to ensure that @tsk will not cache a
> +	 * sigqueue when it is reaping it's child tasks
> +	 */
> +	tsk->sigqueue_cache = ERR_PTR(-1);
>  }


Reviewed-by: Oleg Nesterov <oleg@redhat.com>