[TCP]: Add vegas congestion avoidance support.

A forward port of an old 2.3.x kernel hack done
years ago.  I (DaveM) did the first rough port,
Stephen Hemminger actually cleaned it up and
made it usable.

include/linux/sysctl.h

@@ -327,6 +327,10 @@ enum
 	NET_TCP_WESTWOOD=95,
 	NET_IPV4_IGMP_MAX_MSF=96,
 	NET_TCP_NO_METRICS_SAVE=97,
+	NET_TCP_VEGAS=98,
+	NET_TCP_VEGAS_ALPHA=99,
+	NET_TCP_VEGAS_BETA=100,
+	NET_TCP_VEGAS_GAMMA=101,
 };
 
 enum {

include/linux/tcp.h

@@ -388,6 +388,18 @@ struct tcp_opt {
                 __u32    rtt;
                 __u32    rtt_min;          /* minimum observed RTT */
         } westwood;
+
+/* Vegas variables */
+	struct {
+		__u32	beg_snd_nxt;	/* right edge during last RTT */
+		__u32	beg_snd_una;	/* left edge  during last RTT */
+		__u32	beg_snd_cwnd;	/* saves the size of the cwnd */
+		__u8	do_vegas;	/* do vegas for this connection */
+		__u8	doing_vegas_now;/* if true, do vegas for this RTT */
+		__u16	cntRTT;		/* # of RTTs measured within last RTT */
+		__u32	minRTT;		/* min of RTTs measured within last RTT (in usec) */
+		__u32	baseRTT;	/* the min of all Vegas RTT measurements seen (in usec) */
+	} vegas;
 };
 
 /* WARNING: don't change the layout of the members in tcp_sock! */

include/net/tcp.h

@@ -583,6 +583,10 @@ extern int sysctl_tcp_tw_reuse;
 extern int sysctl_tcp_frto;
 extern int sysctl_tcp_low_latency;
 extern int sysctl_tcp_westwood;
+extern int sysctl_tcp_vegas_cong_avoid;
+extern int sysctl_tcp_vegas_alpha;
+extern int sysctl_tcp_vegas_beta;
+extern int sysctl_tcp_vegas_gamma;
 extern int sysctl_tcp_nometrics_save;
 
 extern atomic_t tcp_memory_allocated;
@@ -1212,8 +1216,56 @@ static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
 	return max(tp->snd_cwnd >> 1U, 2U);
 }
 
+/* Stop taking Vegas samples for now. */
+#define tcp_vegas_disable(__tp)	((__tp)->vegas.doing_vegas_now = 0)
+
+/* Is this TCP connection using Vegas (regardless of whether it is taking
+ * Vegas measurements at the current time)?
+ */
+#define tcp_is_vegas(__tp)	((__tp)->vegas.do_vegas)
+    
+static inline void tcp_vegas_enable(struct tcp_opt *tp)
+{
+	/* There are several situations when we must "re-start" Vegas:
+	 *
+	 *  o when a connection is established
+	 *  o after an RTO
+	 *  o after fast recovery
+	 *  o when we send a packet and there is no outstanding
+	 *    unacknowledged data (restarting an idle connection)
+	 *
+	 * In these circumstances we cannot do a Vegas calculation at the
+	 * end of the first RTT, because any calculation we do is using
+	 * stale info -- both the saved cwnd and congestion feedback are
+	 * stale.
+	 *
+	 * Instead we must wait until the completion of an RTT during
+	 * which we actually receive ACKs.
+	 */
+    
+	/* Begin taking Vegas samples next time we send something. */
+	tp->vegas.doing_vegas_now = 1;
+     
+	/* Set the beginning of the next send window. */
+	tp->vegas.beg_snd_nxt = tp->snd_nxt;
+
+	tp->vegas.cntRTT = 0;
+	tp->vegas.minRTT = 0x7fffffff;
+}
+
+/* Should we be taking Vegas samples right now? */
+#define tcp_vegas_enabled(__tp)	((__tp)->vegas.doing_vegas_now)
+
+extern void tcp_vegas_init(struct tcp_opt *tp);
+
 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
 {
+	if (tcp_is_vegas(tp)) {
+		if (ca_state == TCP_CA_Open) 
+			tcp_vegas_enable(tp);
+		else
+			tcp_vegas_disable(tp);
+	}
 	tp->ca_state = ca_state;
 }
 

net/ipv4/sysctl_net_ipv4.c

@@ -609,7 +609,38 @@ ctl_table ipv4_table[] = {
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec,
 	},
-
+	{
+		.ctl_name	= NET_TCP_VEGAS,
+		.procname	= "tcp_vegas_cong_avoid",
+		.data		= &sysctl_tcp_vegas_cong_avoid,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_ALPHA,
+		.procname	= "tcp_vegas_alpha",
+		.data		= &sysctl_tcp_vegas_alpha,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_BETA,
+		.procname	= "tcp_vegas_beta",
+		.data		= &sysctl_tcp_vegas_beta,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_GAMMA,
+		.procname	= "tcp_vegas_gamma",
+		.data		= &sysctl_tcp_vegas_gamma,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
 	{ .ctl_name = 0 }
 };
 

net/ipv4/tcp_minisocks.c

@@ -841,6 +841,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct open_request *req,
 		if (newtp->ecn_flags&TCP_ECN_OK)
 			newsk->sk_no_largesend = 1;
 
+		tcp_vegas_init(newtp);
 		TCP_INC_STATS_BH(TcpPassiveOpens);
 	}
 	return newsk;

net/ipv4/tcp_output.c

@@ -106,6 +106,9 @@ static void tcp_cwnd_restart(struct tcp_opt *tp, struct dst_entry *dst)
 	u32 restart_cwnd = tcp_init_cwnd(tp, dst);
 	u32 cwnd = tp->snd_cwnd;
 
+	if (tcp_is_vegas(tp)) 
+		tcp_vegas_enable(tp);
+
 	tp->snd_ssthresh = tcp_current_ssthresh(tp);
 	restart_cwnd = min(restart_cwnd, cwnd);
 
@@ -225,6 +228,19 @@ int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 			tcp_header_size += (TCPOLEN_SACK_BASE_ALIGNED +
 					    (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
 		}
+		
+		/*
+		 * If the connection is idle and we are restarting,
+		 * then we don't want to do any Vegas calculations
+		 * until we get fresh RTT samples.  So when we
+		 * restart, we reset our Vegas state to a clean
+		 * slate. After we get acks for this flight of
+		 * packets, _then_ we can make Vegas calculations
+		 * again.
+		 */
+		if (tcp_is_vegas(tp) && tcp_packets_in_flight(tp) == 0)
+			tcp_vegas_enable(tp);
+
 		th = (struct tcphdr *) skb_push(skb, tcp_header_size);
 		skb->h.th = th;
 		skb_set_owner_w(skb, sk);
@@ -1268,6 +1284,7 @@ static inline void tcp_connect_init(struct sock *sk)
 		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
 	tp->advmss = dst_metric(dst, RTAX_ADVMSS);
 	tcp_initialize_rcv_mss(sk);
+	tcp_vegas_init(tp);
 
 	tcp_select_initial_window(tcp_full_space(sk),
 				  tp->advmss - (tp->ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
@@ -1318,6 +1335,7 @@ int tcp_connect(struct sock *sk)
 	TCP_SKB_CB(buff)->end_seq = tp->write_seq;
 	tp->snd_nxt = tp->write_seq;
 	tp->pushed_seq = tp->write_seq;
+	tcp_vegas_init(tp);
 
 	/* Send it off. */
 	TCP_SKB_CB(buff)->when = tcp_time_stamp;