Re: [PATCH 2/3] Keys: Use RCU to manage session keyring pointer

[David Howells <dhowells@redhat.com>]

Paul E. McKenney <paulmck@us.ibm.com> wrote:

> >  	spin_lock_irqsave(&tsk->sighand->siglock, flags);
> > -	old = tsk->signal->session_keyring;
> > -	tsk->signal->session_keyring = keyring;
> > +	old = rcu_dereference(tsk->signal->session_keyring);
> 
> I don't understand why rcu_dereference() is needed in this case.
> Since we are holding the lock, it should not be possible for
> this to change, right?  Or am I missing something?  (Quite possible,
> am not all that familiar with this code.)

Erm... you're right. I stuck the rcu_dereference() in then added the locks
back in when I realised I still needed them.

> > +	synchronize_kernel();
> 
> This would want to become synchronize_rcu().

I think the deprecation happened since I wrote my patch.

> > +	if (tsk->signal->session_keyring) {
> > +		rcu_read_lock();
> > +		key = keyring_search_aux(
> > +			rcu_dereference(tsk->signal->session_keyring),
> > +			type, description, match);
> > +		rcu_read_unlock();
> > +	}
> > +	else {
> > +		key = keyring_search_aux(tsk->user->session_keyring,
> > +					 type, description, match);
> 
> This one is constant, right?  If not, I don't understand the locking design.

Which one? tsk->user->session_keyring is, tsk->signal->session_keyring is not.

Thanks for the review.

David