On Thu, Aug 21, 2008 at 10:01:12PM +1000, Herbert Xu wrote:

quoted text
> On Thu, Aug 21, 2008 at 11:55:22AM +0000, Jarek Poplawski wrote:

> >

> > Sure, here is a scenario:

> >

> > cpu1					cpu2

> > rtnl_lock()

> > qdisc_graft()

> > // parent != NULL

> > ->cops-graft()

> > notify_and_destroy()			qdisc_run()

> > 					spin_lock(root_lock)

> > qdisc_destroy(old)			dequeue_skb()

> > 					tbf_dequeue()

> > 					qdisc_tree_decrease_qlen()

> > 					qdisc_lookup()

> > //deleting from qdisc_sleeping->list	//walking qdisc_sleeping->list

> > //under rtnl_lock() only		//under qdisc root_lock only

> > list_del(qdisc->list)			list_for_each_entry(txq_root)

>

> Good catch.  Longer term we should fix it so that it doesn't do

> the silly lookup at run-time.  In fact we'll be getting rid of

> requeue so there will be no need to do this in TBF at all.

>

> However, for now please create a patch to put a pair of root locks

> around the list_del in qdisc_destroy.

Since qdisc_destroy() is used for destroying root qdisc too, isn't it

better to get this lock back in notify_and_destroy() like this:
if (old) {

	if (parent)

		sch_tree_lock()

	qdisc_destroy()

	if (parent)

		sch_tree_unlock()

}
Thanks,

Jarek P.

--

To unsubscribe from this list: send the line "unsubscribe netdev" in

the body of a message to majordomo@vger.kernel.org

More majordomo info at  http://vger.kernel.org/majordomo-info.html