Re: safety of *mutex_unlock() (Was: [BUG] signal: sighand unprotected when accessed by /proc)

[Oleg Nesterov <oleg@redhat.com>]

On 06/10, Thomas Gleixner wrote:
>
> On Mon, 9 Jun 2014, Steven Rostedt wrote:
> > I think rtmutex has an
> > issue with it too. Specifically in the slow_unlock case:
> >
> > 	if (!rt_mutex_has_waiters(lock)) {
> > 		lock->owner = NULL;
> > 		raw_spin_unlock(&lock->wait_lock);
> > 		return;
> > 	}
>
> Indeed. If the fast path is enabled we have that issue. Fortunately
> there is a halfways reasonable solution for this.

Ah, yes, I missed that,

> +	while (!rt_mutex_has_waiters(lock)) {
> +		/* Drops lock->wait_lock ! */
> +		if (unlock_rt_mutex_safe(lock) == true)
> +			return;
> +		/* Relock the rtmutex and try again */
> +		raw_spin_lock(&lock->wait_lock);
>  	}

OK...

wakeup_next_waiter() does rt_mutex_set_owner(NULL) before we drop ->wait_lock,
but this looks fine: we know that rt_mutex_has_waiters() can not become false
until waiter->task takes this lock and does rt_mutex_dequeue(), so ->owner
can't be NULL, right?

Perhaps it could simply do ->owner = RT_MUTEX_HAS_WAITERS to make this more
clear...


Off-topic question. I simply can't understand why rt_mutex_slowtrylock() checks
rt_mutex_owner(lock) != current. This looks pointless, try_to_take_rt_mutex()
always fails (correctly) if rt_mutex_owner() != NULL ? IOW, can't we simply
remove this check or turn it into "if (!rt_mutex_owner(lock))" ?

Oleg.