tcp: xps: fix reordering issues
authorEric Dumazet <edumazet@google.com>
Thu, 23 May 2013 07:44:20 +0000 (07:44 +0000)
committerDavid S. Miller <davem@davemloft.net>
Fri, 24 May 2013 01:29:20 +0000 (18:29 -0700)
commit 3853b5841c01a ("xps: Improvements in TX queue selection")
introduced ooo_okay flag, but the condition to set it is slightly wrong.

In our traces, we have seen ACK packets being received out of order,
and RST packets sent in response.

We should test if we have any packets still in host queue.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_output.c

index 536d40929ba6f809d5c79968c48053202dd3748f..ec335fabd5cc12daba8addf5de86adc90abbd561 100644 (file)
@@ -874,11 +874,13 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
                                                           &md5);
        tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
 
-       if (tcp_packets_in_flight(tp) == 0) {
+       if (tcp_packets_in_flight(tp) == 0)
                tcp_ca_event(sk, CA_EVENT_TX_START);
-               skb->ooo_okay = 1;
-       } else
-               skb->ooo_okay = 0;
+
+       /* if no packet is in qdisc/device queue, then allow XPS to select
+        * another queue.
+        */
+       skb->ooo_okay = sk_wmem_alloc_get(sk) == 0;
 
        skb_push(skb, tcp_header_size);
        skb_reset_transport_header(skb);