Re: [PATCH -mm 4/7][AIO] - Make good_sigevent() non-static

[Oleg Nesterov]

On 02/01, S?bastien Dugu? wrote:
> 
> +struct task_struct * sigevent_find_task(sigevent_t * event)
> +{
> +	struct task_struct *task = NULL;
> +
> +	if (event->sigev_signo <= 0 || event->sigev_signo > SIGRTMAX)
> +		return NULL;
> +
> +	if ((event->sigev_notify & SIGEV_THREAD_ID ) == SIGEV_THREAD_ID) {
> +		task = find_task_by_pid(event->sigev_notify_thread_id);
> +
> +		if (!task || task->tgid != current->tgid)
> +			task = NULL;
> +	} else if (event->sigev_notify == SIGEV_SIGNAL)
> +		task = current->group_leader;
> +
> +	return task;
> +}

I am afraid this is still not right. Consider

	->sigev_notify == SIGEV_THREAD_ID | RANDOM_BIT

Now, the second "if (SIGEV_THREAD_ID)" returns a valid task. However,

	really_put_req:

		if (notify == SIGEV_THREAD_ID || notify == SIGEV_SIGNAL)
			put_task_struct();

doesn't work, so we have task_struct leak.

Worse, this breaks posix-timers. Note that posix-timers allow SIGEV_NONE,
the timer is not queued in that case, we shouldn't do ->sigev_signo check.
This means that aio should check SIGEV_NONE itself.

Also, it is critical for posix-timers that SIGEV_THREAD_ID doesn't come
with another bit (like in the example below), note the code like

	if (sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID))
		...

IOW: good_sigevent() in its current form is very cryptic, and it _really_
needs a cleanup, but we should not change its behaviour.

Apart from this, I don't see other problems in the signal related code in
this series.

Oleg.

-
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/