netem: fix dequeue
authorEric Dumazet <eric.dumazet@gmail.com>
Wed, 15 Feb 2012 20:28:25 +0000 (20:28 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sun, 19 Feb 2012 23:57:50 +0000 (18:57 -0500)
commit 50612537e9 (netem: fix classful handling) added two errors in
netem_dequeue()

1) After checking skb at the head of tfifo queue for time constraints,
   it dequeues tail skb, thus adding unwanted reordering.

2) qdisc stats are updated twice per packet
   (one when packet dequeued from tfifo, once when delivered)

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_netem.c

index e83d61c..5da548f 100644 (file)
@@ -501,9 +501,8 @@ tfifo_dequeue:
 
                /* if more time remaining? */
                if (cb->time_to_send <= psched_get_time()) {
-                       skb = qdisc_dequeue_tail(sch);
-                       if (unlikely(!skb))
-                               goto qdisc_dequeue;
+                       __skb_unlink(skb, &sch->q);
+                       sch->qstats.backlog -= qdisc_pkt_len(skb);
 
 #ifdef CONFIG_NET_CLS_ACT
                        /*
@@ -539,7 +538,6 @@ deliver:
                qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send);
        }
 
-qdisc_dequeue:
        if (q->qdisc) {
                skb = q->qdisc->ops->dequeue(q->qdisc);
                if (skb)