[PATCH RFC] net_sched/codel: do not defer queue length update

[PATCH RFC] net_sched/codel: do not defer queue length update

[Konstantin Khlebnikov]

When codel wants to drop last packet in ->dequeue() it cannot call
qdisc_tree_reduce_backlog() right away - it will notify parent qdisc
about zero qlen and HTB/HFSC will deactivate class. The same class will
be deactivated second time by caller of ->dequeue(). Currently codel and
fq_codel defer update. This triggers warning in HFSC when it's qlen != 0
but there is no active classes.

This patch update parent queue length immediately: just temporary increase
qlen around qdisc_tree_reduce_backlog() to prevent first class deactivation
if we have skb to return.

This might open another problem in HFSC - now operation peek could fail and
deactivate parent class.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=109581
---
 net/sched/sch_codel.c    |   14 ++++++++++----
 net/sched/sch_fq_codel.c |   24 +++++++++++++++---------
 2 files changed, 25 insertions(+), 13 deletions(-)

diff --git a/net/sched/sch_codel.c b/net/sched/sch_codel.c
index c518a1efcb9d..e8194e455efe 100644
--- a/net/sched/sch_codel.c
+++ b/net/sched/sch_codel.c
@@ -95,11 +95,17 @@ static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
 			    &q->stats, qdisc_pkt_len, codel_get_enqueue_time,
 			    drop_func, dequeue_func);
 
-	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
-	 * or HTB crashes. Defer it for next round.
+	/* If our qlen is 0 qdisc_tree_reduce_backlog() will deactivate
+	 * parent class, dequeue in parent qdisc will do the same if we
+	 * return skb. Temporary increment qlen if we have skb.
 	 */
-	if (q->stats.drop_count && sch->q.qlen) {
-		qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
+	if (q->stats.drop_count) {
+		if (skb)
+			sch->q.qlen++;
+		qdisc_tree_reduce_backlog(sch, q->stats.drop_count,
+					  q->stats.drop_len);
+		if (skb)
+			sch->q.qlen--;
 		q->stats.drop_count = 0;
 		q->stats.drop_len = 0;
 	}
diff --git a/net/sched/sch_fq_codel.c b/net/sched/sch_fq_codel.c
index 337f2d6d81e4..4048790a8c99 100644
--- a/net/sched/sch_fq_codel.c
+++ b/net/sched/sch_fq_codel.c
@@ -316,6 +316,21 @@ static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch)
 	flow->dropped += q->cstats.drop_count - prev_drop_count;
 	flow->dropped += q->cstats.ecn_mark - prev_ecn_mark;
 
+	/* If our qlen is 0 qdisc_tree_reduce_backlog() will deactivate
+	 * parent class, dequeue in parent qdisc will do the same if we
+	 * return skb. Temporary increment qlen if we have skb.
+	 */
+	if (q->cstats.drop_count) {
+		if (skb)
+			sch->q.qlen++;
+		qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
+					  q->cstats.drop_len);
+		if (skb)
+			sch->q.qlen--;
+		q->cstats.drop_count = 0;
+		q->cstats.drop_len = 0;
+	}
+
 	if (!skb) {
 		/* force a pass through old_flows to prevent starvation */
 		if ((head == &q->new_flows) && !list_empty(&q->old_flows))
@@ -326,15 +341,6 @@ static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch)
 	}
 	qdisc_bstats_update(sch, skb);
 	flow->deficit -= qdisc_pkt_len(skb);
-	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
-	 * or HTB crashes. Defer it for next round.
-	 */
-	if (q->cstats.drop_count && sch->q.qlen) {
-		qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
-					  q->cstats.drop_len);
-		q->cstats.drop_count = 0;
-		q->cstats.drop_len = 0;
-	}
 	return skb;
 }
 

Re: [PATCH RFC] net_sched/codel: do not defer queue length update

[Felix Fietkau]

On 2017-08-21 10:14, Konstantin Khlebnikov wrote:
> When codel wants to drop last packet in ->dequeue() it cannot call
> qdisc_tree_reduce_backlog() right away - it will notify parent qdisc
> about zero qlen and HTB/HFSC will deactivate class. The same class will
> be deactivated second time by caller of ->dequeue(). Currently codel and
> fq_codel defer update. This triggers warning in HFSC when it's qlen != 0
> but there is no active classes.
> 
> This patch update parent queue length immediately: just temporary increase
> qlen around qdisc_tree_reduce_backlog() to prevent first class deactivation
> if we have skb to return.
> 
> This might open another problem in HFSC - now operation peek could fail and
> deactivate parent class.
> 
> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=109581
This has been tested by several people and it seems to be working well.
Please make sure this gets merged soon.

Thanks,

- Felix