Re: [Bloat] a flood of Bufferbloat-related papers

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mardi 11 octobre 2011 à 14:13 +0200, Eric Dumazet a écrit :
> Le mardi 11 octobre 2011 à 13:58 +0200, Steinar H. Gunderson a écrit :
> > On Tue, Oct 11, 2011 at 12:27:17PM +0200, David Täht wrote:
> > > On my list already would be "an analysis of the effects of broken sack
> > > processing on linux 2.4.24-3.1", of which I *think* I've captured
> > > multiple examples of in the raw traces I've been collecting for
> > > months... (so if anyone is interested in the raw data, I can provide)
> > 
> > Do you have any more information? The only thing I could find online was that
> > there were SACK issues that were supposed to be fixed by 2.6.16; nothing
> > about a fix in 3.1 or post-3.1.
> > 
> 
> Of course, instead of discussing all linux stuff privately, it would be
> good to discuss with linux network maintainers.
> 
> Or is the goal is to provide nice papers only ?
> 
> commit f779b2d60ab95c17f1e025778ed0df3ec2f05d75
> Author: Zheng Yan <zheng.z.yan@intel.com>
> Date:   Sun Sep 18 22:37:34 2011 -0400

Another fix (scheduled for linux-3.2 available in net-next tree :)

[ And some work is currently in progress to support RFC 5562 ]

commit 7a269ffad72f3604b8982fa09c387670e0d2ee14
Author: Eric Dumazet <eric.dumazet@gmail.com>
Date:   Thu Sep 22 20:02:19 2011 +0000

    tcp: ECN blackhole should not force quickack mode
    
    While playing with a new ADSL box at home, I discovered that ECN
    blackhole can trigger suboptimal quickack mode on linux : We send one
    ACK for each incoming data frame, without any delay and eventual
    piggyback.
    
    This is because TCP_ECN_check_ce() considers that if no ECT is seen on a
    segment, this is because this segment was a retransmit.
    
    Refine this heuristic and apply it only if we seen ECT in a previous
    segment, to detect ECN blackhole at IP level.
    
    Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
    CC: Jamal Hadi Salim <jhs@mojatatu.com>
    CC: Jerry Chu <hkchu@google.com>
    CC: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
    CC: Jim Gettys <jg@freedesktop.org>
    CC: Dave Taht <dave.taht@gmail.com>
    Acked-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
    Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/net/tcp.h b/include/net/tcp.h
index f357bef..702aefc 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -356,6 +356,7 @@ static inline void tcp_dec_quickack_mode(struct sock *sk,
 #define	TCP_ECN_OK		1
 #define	TCP_ECN_QUEUE_CWR	2
 #define	TCP_ECN_DEMAND_CWR	4
+#define	TCP_ECN_SEEN		8
 
 static __inline__ void
 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index a5d01b1..5a4408c 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -217,16 +217,25 @@ static inline void TCP_ECN_withdraw_cwr(struct tcp_sock *tp)
 	tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR;
 }
 
-static inline void TCP_ECN_check_ce(struct tcp_sock *tp, struct sk_buff *skb)
+static inline void TCP_ECN_check_ce(struct tcp_sock *tp, const struct sk_buff *skb)
 {
-	if (tp->ecn_flags & TCP_ECN_OK) {
-		if (INET_ECN_is_ce(TCP_SKB_CB(skb)->flags))
-			tp->ecn_flags |= TCP_ECN_DEMAND_CWR;
+	if (!(tp->ecn_flags & TCP_ECN_OK))
+		return;
+
+	switch (TCP_SKB_CB(skb)->flags & INET_ECN_MASK) {
+	case INET_ECN_NOT_ECT:
 		/* Funny extension: if ECT is not set on a segment,
-		 * it is surely retransmit. It is not in ECN RFC,
-		 * but Linux follows this rule. */
-		else if (INET_ECN_is_not_ect((TCP_SKB_CB(skb)->flags)))
+		 * and we already seen ECT on a previous segment,
+		 * it is probably a retransmit.
+		 */
+		if (tp->ecn_flags & TCP_ECN_SEEN)
 			tcp_enter_quickack_mode((struct sock *)tp);
+		break;
+	case INET_ECN_CE:
+		tp->ecn_flags |= TCP_ECN_DEMAND_CWR;
+		/* fallinto */
+	default:
+		tp->ecn_flags |= TCP_ECN_SEEN;
 	}
 }