net_sched: fix pfifo_head_drop behavior vs backlog
authorEric Dumazet <edumazet@google.com>
Mon, 13 Jun 2016 03:01:25 +0000 (20:01 -0700)
committerDavid S. Miller <davem@davemloft.net>
Tue, 14 Jun 2016 21:17:58 +0000 (17:17 -0400)
When the qdisc is full, we drop a packet at the head of the queue,
queue the current skb and return NET_XMIT_CN

Now we track backlog on upper qdiscs, we need to call
qdisc_tree_reduce_backlog(), even if the qlen did not change.

Fixes: 2ccccf5fb43f ("net_sched: update hierarchical backlog too")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_fifo.c

index 2177eac0a61ed00c6c60655f577e0dd816fd2c08..2e4bd2c0a50c497fbc6cbfe5338cab981559b491 100644 (file)
@@ -37,14 +37,18 @@ static int pfifo_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 
 static int pfifo_tail_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 {
+       unsigned int prev_backlog;
+
        if (likely(skb_queue_len(&sch->q) < sch->limit))
                return qdisc_enqueue_tail(skb, sch);
 
+       prev_backlog = sch->qstats.backlog;
        /* queue full, remove one skb to fulfill the limit */
        __qdisc_queue_drop_head(sch, &sch->q);
        qdisc_qstats_drop(sch);
        qdisc_enqueue_tail(skb, sch);
 
+       qdisc_tree_reduce_backlog(sch, 0, prev_backlog - sch->qstats.backlog);
        return NET_XMIT_CN;
 }