[PATCH]: Kill qdisc rcu freeing...

[PATCH]: Kill qdisc rcu freeing...

[David Miller]

Ok, I flipped the switch.  I've checked in the following to
net-2.6 and it seems to pass a bunch of smoke tests I just
did (flood pings while adding and deleting various qdisc
setups, etc.)

Jarek, I think we can possibly now revert that locking change
you had to make because destroy was outside of RTNL?

pkt_sched: No longer destroy qdiscs from RCU.

We can now kill them synchronously with all of the
previous dev_deactivate() cures.

This makes netdev destruction and shutdown saner as
the qdiscs hold references to the device.

Signed-off-by: David S. Miller <davem@davemloft.net>
---
 include/net/sch_generic.h |    1 -
 net/sched/sch_generic.c   |   27 +++++++++------------------
 2 files changed, 9 insertions(+), 19 deletions(-)

diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 757ab08..84d25f2 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -61,7 +61,6 @@ struct Qdisc
 	struct gnet_stats_basic	bstats;
 	struct gnet_stats_queue	qstats;
 	struct gnet_stats_rate_est	rate_est;
-	struct rcu_head 	q_rcu;
 	int			(*reshape_fail)(struct sk_buff *skb,
 					struct Qdisc *q);
 
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 30b76ae..6f96b7b 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -518,14 +518,19 @@ void qdisc_reset(struct Qdisc *qdisc)
 }
 EXPORT_SYMBOL(qdisc_reset);
 
-/* this is the rcu callback function to clean up a qdisc when there
- * are no further references to it */
+/* Under qdisc_lock(qdisc) and BH! */
 
-static void __qdisc_destroy(struct rcu_head *head)
+void qdisc_destroy(struct Qdisc *qdisc)
 {
-	struct Qdisc *qdisc = container_of(head, struct Qdisc, q_rcu);
 	const struct Qdisc_ops  *ops = qdisc->ops;
 
+	if (qdisc->flags & TCQ_F_BUILTIN ||
+	    !atomic_dec_and_test(&qdisc->refcnt))
+		return;
+
+	if (qdisc->parent)
+		list_del(&qdisc->list);
+
 #ifdef CONFIG_NET_SCHED
 	qdisc_put_stab(qdisc->stab);
 #endif
@@ -542,20 +547,6 @@ static void __qdisc_destroy(struct rcu_head *head)
 
 	kfree((char *) qdisc - qdisc->padded);
 }
-
-/* Under qdisc_lock(qdisc) and BH! */
-
-void qdisc_destroy(struct Qdisc *qdisc)
-{
-	if (qdisc->flags & TCQ_F_BUILTIN ||
-	    !atomic_dec_and_test(&qdisc->refcnt))
-		return;
-
-	if (qdisc->parent)
-		list_del(&qdisc->list);
-
-	call_rcu(&qdisc->q_rcu, __qdisc_destroy);
-}
 EXPORT_SYMBOL(qdisc_destroy);
 
 static bool dev_all_qdisc_sleeping_noop(struct net_device *dev)
-- 
1.5.6.5.GIT

Re: [PATCH]: Kill qdisc rcu freeing...

[Jarek Poplawski]

On Sun, Aug 17, 2008 at 10:34:35PM -0700, David Miller wrote:
> 
> Ok, I flipped the switch.  I've checked in the following to
> net-2.6 and it seems to pass a bunch of smoke tests I just
> did (flood pings while adding and deleting various qdisc
> setups, etc.)
> 
> Jarek, I think we can possibly now revert that locking change
> you had to make because destroy was outside of RTNL?

Yes, I think there are these two patches only to revert:

----------
commit d4766692e72422f3b0f0e9ac6773d92baad07d51
Author: Jarek Poplawski <jarkao2@gmail.com>
Date:   Wed Aug 13 15:20:24 2008 -0700

    pkt_sched: Protect gen estimators under est_lock.

----------

commit 1cfa26661a85549063e369e2b40275eeaa7b923c
Author: Jarek Poplawski <jarkao2@gmail.com>
Date:   Mon Aug 11 18:11:06 2008 -0700

    pkt_sched: Add BH protection for qdisc_stab_lock.

----------

I guess, I don't need to send any patches, but let me know if I'm
wrong.

Thanks,
Jarek P.

> 
> pkt_sched: No longer destroy qdiscs from RCU.
> 
> We can now kill them synchronously with all of the
> previous dev_deactivate() cures.
> 
> This makes netdev destruction and shutdown saner as
> the qdiscs hold references to the device.
> 
> Signed-off-by: David S. Miller <davem@davemloft.net>
> ---
...

Re: [PATCH]: Kill qdisc rcu freeing...

[David Miller]

From: Jarek Poplawski <jarkao2@gmail.com>
Date: Mon, 18 Aug 2008 12:07:25 +0000

> On Sun, Aug 17, 2008 at 10:34:35PM -0700, David Miller wrote:
> > 
> > Ok, I flipped the switch.  I've checked in the following to
> > net-2.6 and it seems to pass a bunch of smoke tests I just
> > did (flood pings while adding and deleting various qdisc
> > setups, etc.)
> > 
> > Jarek, I think we can possibly now revert that locking change
> > you had to make because destroy was outside of RTNL?
> 
> Yes, I think there are these two patches only to revert:
> 
> ----------
 ...
> commit d4766692e72422f3b0f0e9ac6773d92baad07d51
 ...
> ----------
 ...
> commit 1cfa26661a85549063e369e2b40275eeaa7b923c
 ...
> ----------
> 
> I guess, I don't need to send any patches, but let me know if I'm
> wrong.

Thanks, I've reverted both commits.