Message ID | 1337413904.7029.93.camel@edumazet-glaptop |
---|---|
State | Accepted, archived |
Delegated to: | David Miller |
Headers | show |
From: Eric Dumazet <eric.dumazet@gmail.com> Date: Sat, 19 May 2012 09:51:44 +0200 > From: Eric Dumazet <edumazet@google.com> > > If the allfrag feature has been set on a host route (due to an ICMPv6 > Packet Too Big received indicating a MTU of less than 1280), we hit a > very slow behavior in TCP stack, because all big packets are dropped and > only a retransmit timer is able to push one MSS frame every 200 ms. > > One way to handle this is to disable GSO on the socket the first time a > super packet is dropped. Adding a specific dst_allfrag() in the fast > path is probably overkill since the dst_allfrag() case almost never > happen. > > Result on netperf TCP_STREAM, one flow : > > Before : 60 kbit/sec > After : 1.6 Gbit/sec > > Reported-by: Tore Anderson <tore@fud.no> > Signed-off-by: Eric Dumazet <edumazet@google.com> > Tested-by: Tore Anderson <tore@fud.no> Applied. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c index 3dc633f..d99fdc6 100644 --- a/net/ipv6/ip6_output.c +++ b/net/ipv6/ip6_output.c @@ -643,7 +643,10 @@ int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) /* We must not fragment if the socket is set to force MTU discovery * or if the skb it not generated by a local socket. */ - if (!skb->local_df && skb->len > mtu) { + if (unlikely(!skb->local_df && skb->len > mtu)) { + if (skb->sk && dst_allfrag(skb_dst(skb))) + sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK); + skb->dev = skb_dst(skb)->dev; icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),