net: sched: do not requeue a NULL skb
authorLars Persson <lars.persson@axis.com>
Tue, 12 Apr 2016 06:45:52 +0000 (08:45 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 14 Apr 2016 05:28:51 +0000 (01:28 -0400)
A failure in validate_xmit_skb_list() triggered an unconditional call
to dev_requeue_skb with skb=NULL. This slowly grows the queue
discipline's qlen count until all traffic through the queue stops.

We take the optimistic approach and continue running the queue after a
failure since it is unknown if later packets also will fail in the
validate path.

Fixes: 55a93b3ea780 ("qdisc: validate skb without holding lock")
Signed-off-by: Lars Persson <larper@axis.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_generic.c

index f18c3502420730e87254a6b01f0a6c1e4fb96941..80742edea96fe8a237573c69d3a6349342e6db00 100644 (file)
@@ -159,12 +159,15 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
        if (validate)
                skb = validate_xmit_skb_list(skb, dev);
 
-       if (skb) {
+       if (likely(skb)) {
                HARD_TX_LOCK(dev, txq, smp_processor_id());
                if (!netif_xmit_frozen_or_stopped(txq))
                        skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 
                HARD_TX_UNLOCK(dev, txq);
+       } else {
+               spin_lock(root_lock);
+               return qdisc_qlen(q);
        }
        spin_lock(root_lock);