Re: [TOMOYO 05/15](repost) Domain transition handler functions.

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Wed, 2007-10-03 at 22:59 +0900, Tetsuo Handa wrote:
> Hello.
> 
> KaiGai Kohei wrote:
> > If so, you can apply RCU instead to avoid read lock
> > when scanning the list, like:
> > 
> > rcu_read_lock();
> > list_for_each_entry(...) {
> >      ....
> > }
> > rcu_read_unlock();
> 
> Can I sleep between rcu_read_lock() and rcu_read_unlock() ?
> As far as I saw, rcu_read_lock() makes in_atomic() true, so I think I can't sleep.
> 
> If I use RCU, I have to give up " [TOMOYO 13/15] Conditional permission support"
> because tmy_check_condition() can sleep.

You can indeed not sleep in an rcu_read_lock() section. However, you
could grab a reference on an item, stop the iteration, drop
rcu_read_lock. Do you thing, re-acquire rcu_read_lock(), drop the ref,
and continue the iteration.

Also, how do you avoid referencing dead data with your sll? I haven't
actually looked at your patches, but the simple scheme you outlined
didn't handle the iteration + concurrent removal scenario:

thread 1             thread 2

 foo = ptr->next

 < schedule >

                     killme = ptr->next (same item)
                     ptr->next = ptr->next->next;
                     kfree(killme)

                     < schedule >

 foo->bar <-- *BANG*