[TCP]: Add pluggable congestion control algorithm infrastructure.

Allow TCP to have multiple pluggable congestion control algorithms.
Algorithms are defined by a set of operations and can be built in
or modules.  The legacy "new RENO" algorithm is used as a starting
point and fallback.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

10 files changed, 313 insertions, 790 deletions

diff --git a/net/ipv4/Makefile b/net/ipv4/Makefile
index 65d57d8e1ad..89c0b4cb470 100644
--- a/net/ipv4/Makefile
+++ b/net/ipv4/Makefile
@@ -5,7 +5,8 @@
 obj-y     := utils.o route.o inetpeer.o protocol.o \
 	     ip_input.o ip_fragment.o ip_forward.o ip_options.o \
 	     ip_output.o ip_sockglue.o \
-	     tcp.o tcp_input.o tcp_output.o tcp_timer.o tcp_ipv4.o tcp_minisocks.o \
+	     tcp.o tcp_input.o tcp_output.o tcp_timer.o tcp_ipv4.o \
+	     tcp_minisocks.o tcp_cong.o \
 	     datagram.o raw.o udp.o arp.o icmp.o devinet.o af_inet.o igmp.o \
 	     sysctl_net_ipv4.o fib_frontend.o fib_semantics.o
 
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 23068bddbf0..e3289453241 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -118,6 +118,45 @@ static int ipv4_sysctl_forward_strategy(ctl_table *table,
 	return 1;
 }
 
+static int proc_tcp_congestion_control(ctl_table *ctl, int write, struct file * filp,
+				       void __user *buffer, size_t *lenp, loff_t *ppos)
+{
+	char val[TCP_CA_NAME_MAX];
+	ctl_table tbl = {
+		.data = val,
+		.maxlen = TCP_CA_NAME_MAX,
+	};
+	int ret;
+
+	tcp_get_default_congestion_control(val);
+
+	ret = proc_dostring(&tbl, write, filp, buffer, lenp, ppos);
+	if (write && ret == 0)
+		ret = tcp_set_default_congestion_control(val);
+	return ret;
+}
+
+int sysctl_tcp_congestion_control(ctl_table *table, int __user *name, int nlen,
+				  void __user *oldval, size_t __user *oldlenp,
+				  void __user *newval, size_t newlen,
+				  void **context)
+{
+	char val[TCP_CA_NAME_MAX];
+	ctl_table tbl = {
+		.data = val,
+		.maxlen = TCP_CA_NAME_MAX,
+	};
+	int ret;
+
+	tcp_get_default_congestion_control(val);
+	ret = sysctl_string(&tbl, name, nlen, oldval, oldlenp, newval, newlen,
+			    context);
+	if (ret == 0 && newval && newlen)
+		ret = tcp_set_default_congestion_control(val);
+	return ret;
+}
+
+
 ctl_table ipv4_table[] = {
         {
 		.ctl_name	= NET_IPV4_TCP_TIMESTAMPS,
@@ -612,70 +651,6 @@ ctl_table ipv4_table[] = {
 		.proc_handler	= &proc_dointvec,
 	},
 	{
-		.ctl_name	= NET_TCP_WESTWOOD, 
-		.procname	= "tcp_westwood",
-		.data		= &sysctl_tcp_westwood,
-		.maxlen		= sizeof(int),
-		.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	= NET_TCP_BIC,
-		.procname	= "tcp_bic",
-		.data		= &sysctl_tcp_bic,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
-	},
-	{
-		.ctl_name	= NET_TCP_BIC_FAST_CONVERGENCE,
-		.procname	= "tcp_bic_fast_convergence",
-		.data		= &sysctl_tcp_bic_fast_convergence,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
-	},
-	{
-		.ctl_name	= NET_TCP_BIC_LOW_WINDOW,
-		.procname	= "tcp_bic_low_window",
-		.data		= &sysctl_tcp_bic_low_window,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
-	},
-	{
 		.ctl_name	= NET_TCP_MODERATE_RCVBUF,
 		.procname	= "tcp_moderate_rcvbuf",
 		.data		= &sysctl_tcp_moderate_rcvbuf,
@@ -692,13 +667,14 @@ ctl_table ipv4_table[] = {
 		.proc_handler	= &proc_dointvec,
 	},
 	{
-		.ctl_name	= NET_TCP_BIC_BETA,
-		.procname	= "tcp_bic_beta",
-		.data		= &sysctl_tcp_bic_beta,
-		.maxlen		= sizeof(int),
+		.ctl_name	= NET_TCP_CONG_CONTROL,
+		.procname	= "tcp_congestion_control",
 		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
+		.maxlen		= TCP_CA_NAME_MAX,
+		.proc_handler	= &proc_tcp_congestion_control,
+		.strategy	= &sysctl_tcp_congestion_control,
 	},
+
 	{ .ctl_name = 0 }
 };
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 674bbd8cfd3..f3dbc8dc126 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2333,6 +2333,8 @@ void __init tcp_init(void)
 	printk(KERN_INFO "TCP: Hash tables configured "
 	       "(established %d bind %d)\n",
 	       tcp_ehash_size << 1, tcp_bhash_size);
+
+	tcp_register_congestion_control(&tcp_reno);
 }
 
 EXPORT_SYMBOL(tcp_accept);
diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
new file mode 100644
index 00000000000..665394a63ae
--- /dev/null
+++ b/net/ipv4/tcp_cong.c
@@ -0,0 +1,195 @@
+/*
+ * Plugable TCP congestion control support and newReno
+ * congestion control.
+ * Based on ideas from I/O scheduler suport and Web100.
+ *
+ * Copyright (C) 2005 Stephen Hemminger <shemminger@osdl.org>
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/types.h>
+#include <linux/list.h>
+#include <net/tcp.h>
+
+static DEFINE_SPINLOCK(tcp_cong_list_lock);
+static LIST_HEAD(tcp_cong_list);
+
+/* Simple linear search, don't expect many entries! */
+static struct tcp_congestion_ops *tcp_ca_find(const char *name)
+{
+	struct tcp_congestion_ops *e;
+
+	list_for_each_entry(e, &tcp_cong_list, list) {
+		if (strcmp(e->name, name) == 0)
+			return e;
+	}
+
+	return NULL;
+}
+
+/*
+ * Attach new congestion control algorthim to the list
+ * of available options.
+ */
+int tcp_register_congestion_control(struct tcp_congestion_ops *ca)
+{
+	int ret = 0;
+
+	/* all algorithms must implement ssthresh and cong_avoid ops */
+	if (!ca->ssthresh || !ca->cong_avoid || !ca->min_cwnd) {
+		printk(KERN_ERR "TCP %s does not implement required ops\n",
+		       ca->name);
+		return -EINVAL;
+	}
+
+	spin_lock(&tcp_cong_list_lock);
+	if (tcp_ca_find(ca->name)) {
+		printk(KERN_NOTICE "TCP %s already registered\n", ca->name);
+		ret = -EEXIST;
+	} else {
+		list_add_rcu(&ca->list, &tcp_cong_list);
+		printk(KERN_INFO "TCP %s registered\n", ca->name);
+	}
+	spin_unlock(&tcp_cong_list_lock);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(tcp_register_congestion_control);
+
+/*
+ * Remove congestion control algorithm, called from
+ * the module's remove function.  Module ref counts are used
+ * to ensure that this can't be done till all sockets using
+ * that method are closed.
+ */
+void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca)
+{
+	spin_lock(&tcp_cong_list_lock);
+	list_del_rcu(&ca->list);
+	spin_unlock(&tcp_cong_list_lock);
+}
+EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control);
+
+/* Assign choice of congestion control. */
+void tcp_init_congestion_control(struct tcp_sock *tp)
+{
+	struct tcp_congestion_ops *ca;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
+		if (try_module_get(ca->owner)) {
+			tp->ca_ops = ca;
+			break;
+		}
+
+	}
+	rcu_read_unlock();
+
+	if (tp->ca_ops->init)
+		tp->ca_ops->init(tp);
+}
+
+/* Manage refcounts on socket close. */
+void tcp_cleanup_congestion_control(struct tcp_sock *tp)
+{
+	if (tp->ca_ops->release)
+		tp->ca_ops->release(tp);
+	module_put(tp->ca_ops->owner);
+}
+
+/* Used by sysctl to change default congestion control */
+int tcp_set_default_congestion_control(const char *name)
+{
+	struct tcp_congestion_ops *ca;
+	int ret = -ENOENT;
+
+	spin_lock(&tcp_cong_list_lock);
+	ca = tcp_ca_find(name);
+#ifdef CONFIG_KMOD
+	if (!ca) {
+		spin_unlock(&tcp_cong_list_lock);
+
+		request_module("tcp_%s", name);
+		spin_lock(&tcp_cong_list_lock);
+		ca = tcp_ca_find(name);
+	}
+#endif
+
+	if (ca) {
+		list_move(&ca->list, &tcp_cong_list);
+		ret = 0;
+	}
+	spin_unlock(&tcp_cong_list_lock);
+
+	return ret;
+}
+
+/* Get current default congestion control */
+void tcp_get_default_congestion_control(char *name)
+{
+	struct tcp_congestion_ops *ca;
+	/* We will always have reno... */
+	BUG_ON(list_empty(&tcp_cong_list));
+
+	rcu_read_lock();
+	ca = list_entry(tcp_cong_list.next, struct tcp_congestion_ops, list);
+	strncpy(name, ca->name, TCP_CA_NAME_MAX);
+	rcu_read_unlock();
+}
+
+/*
+ * TCP Reno congestion control
+ * This is special case used for fallback as well.
+ */
+/* This is Jacobson's slow start and congestion avoidance.
+ * SIGCOMM '88, p. 328.
+ */
+void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack, u32 rtt, u32 in_flight,
+			 int flag)
+{
+	if (in_flight < tp->snd_cwnd)
+		return;
+
+        if (tp->snd_cwnd <= tp->snd_ssthresh) {
+                /* In "safe" area, increase. */
+		if (tp->snd_cwnd < tp->snd_cwnd_clamp)
+			tp->snd_cwnd++;
+	} else {
+                /* In dangerous area, increase slowly.
+		 * In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
+		 */
+		if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
+			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
+				tp->snd_cwnd++;
+			tp->snd_cwnd_cnt = 0;
+		} else
+			tp->snd_cwnd_cnt++;
+	}
+}
+EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);
+
+/* Slow start threshold is half the congestion window (min 2) */
+u32 tcp_reno_ssthresh(struct tcp_sock *tp)
+{
+	return max(tp->snd_cwnd >> 1U, 2U);
+}
+EXPORT_SYMBOL_GPL(tcp_reno_ssthresh);
+
+/* Lower bound on congestion window. */
+u32 tcp_reno_min_cwnd(struct tcp_sock *tp)
+{
+	return tp->snd_ssthresh/2;
+}
+EXPORT_SYMBOL_GPL(tcp_reno_min_cwnd);
+
+struct tcp_congestion_ops tcp_reno = {
+	.name		= "reno",
+	.owner		= THIS_MODULE,
+	.ssthresh	= tcp_reno_ssthresh,
+	.cong_avoid	= tcp_reno_cong_avoid,
+	.min_cwnd	= tcp_reno_min_cwnd,
+};
+
+EXPORT_SYMBOL_GPL(tcp_reno);
diff --git a/net/ipv4/tcp_diag.c b/net/ipv4/tcp_diag.c
index 634befc0792..867acc0f79d 100644
--- a/net/ipv4/tcp_diag.c
+++ b/net/ipv4/tcp_diag.c
@@ -42,7 +42,6 @@ struct tcpdiag_entry
 
 static struct sock *tcpnl;
 
-
 #define TCPDIAG_PUT(skb, attrtype, attrlen) \
 ({ int rtalen = RTA_LENGTH(attrlen);        \
    struct rtattr *rta;                      \
@@ -61,7 +60,6 @@ static int tcpdiag_fill(struct sk_buff *skb, struct sock *sk,
 	struct nlmsghdr  *nlh;
 	struct tcp_info  *info = NULL;
 	struct tcpdiag_meminfo  *minfo = NULL;
-	struct tcpvegas_info *vinfo = NULL;
 	unsigned char	 *b = skb->tail;
 
 	nlh = NLMSG_PUT(skb, pid, seq, TCPDIAG_GETSOCK, sizeof(*r));
@@ -73,9 +71,6 @@ static int tcpdiag_fill(struct sk_buff *skb, struct sock *sk,
 		if (ext & (1<<(TCPDIAG_INFO-1)))
 			info = TCPDIAG_PUT(skb, TCPDIAG_INFO, sizeof(*info));
 		
-		if ((tcp_is_westwood(tp) || tcp_is_vegas(tp))
-		    && (ext & (1<<(TCPDIAG_VEGASINFO-1))))
-			vinfo = TCPDIAG_PUT(skb, TCPDIAG_VEGASINFO, sizeof(*vinfo));
 	}
 	r->tcpdiag_family = sk->sk_family;
 	r->tcpdiag_state = sk->sk_state;
@@ -166,19 +161,8 @@ static int tcpdiag_fill(struct sk_buff *skb, struct sock *sk,
 	if (info) 
 		tcp_get_info(sk, info);
 
-	if (vinfo) {
-		if (tcp_is_vegas(tp)) {
-			vinfo->tcpv_enabled = tp->vegas.doing_vegas_now;
-			vinfo->tcpv_rttcnt = tp->vegas.cntRTT;
-			vinfo->tcpv_rtt = jiffies_to_usecs(tp->vegas.baseRTT);
-			vinfo->tcpv_minrtt = jiffies_to_usecs(tp->vegas.minRTT);
-		} else {
-			vinfo->tcpv_enabled = 0;
-			vinfo->tcpv_rttcnt = 0;
-			vinfo->tcpv_rtt = jiffies_to_usecs(tp->westwood.rtt);
-			vinfo->tcpv_minrtt = jiffies_to_usecs(tp->westwood.rtt_min);
-		}
-	}
+	if (sk->sk_state < TCP_TIME_WAIT && tp->ca_ops->get_info)
+		tp->ca_ops->get_info(tp, ext, skb);
 
 	nlh->nlmsg_len = skb->tail - b;
 	return skb->len;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 5bad504630a..7bbbbc33eb4 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -61,7 +61,6 @@
  *		Panu Kuhlberg:		Experimental audit of TCP (re)transmission
  *					engine. Lots of bugs are found.
  *		Pasi Sarolahti:		F-RTO for dealing with spurious RTOs
- *		Angelo Dell'Aera:	TCP Westwood+ support
  */
 
 #include <linux/config.h>
@@ -88,23 +87,9 @@ int sysctl_tcp_rfc1337;
 int sysctl_tcp_max_orphans = NR_FILE;
 int sysctl_tcp_frto;
 int sysctl_tcp_nometrics_save;
-int sysctl_tcp_westwood;
-int sysctl_tcp_vegas_cong_avoid;
 
 int sysctl_tcp_moderate_rcvbuf = 1;
 
-/* Default values of the Vegas variables, in fixed-point representation
- * with V_PARAM_SHIFT bits to the right of the binary point.
- */
-#define V_PARAM_SHIFT 1
-int sysctl_tcp_vegas_alpha = 1<<V_PARAM_SHIFT;
-int sysctl_tcp_vegas_beta  = 3<<V_PARAM_SHIFT;
-int sysctl_tcp_vegas_gamma = 1<<V_PARAM_SHIFT;
-int sysctl_tcp_bic = 1;
-int sysctl_tcp_bic_fast_convergence = 1;
-int sysctl_tcp_bic_low_window = 14;
-int sysctl_tcp_bic_beta = 819;		/* = 819/1024 (BICTCP_BETA_SCALE) */
-
 #define FLAG_DATA		0x01 /* Incoming frame contained data.		*/
 #define FLAG_WIN_UPDATE		0x02 /* Incoming ACK was a window update.	*/
 #define FLAG_DATA_ACKED		0x04 /* This ACK acknowledged new data.		*/
@@ -333,15 +318,6 @@ static void tcp_init_buffer_space(struct sock *sk)
 	tp->snd_cwnd_stamp = tcp_time_stamp;
 }
 
-static void init_bictcp(struct tcp_sock *tp)
-{
-	tp->bictcp.cnt = 0;
-
-	tp->bictcp.last_max_cwnd = 0;
-	tp->bictcp.last_cwnd = 0;
-	tp->bictcp.last_stamp = 0;
-}
-
 /* 5. Recalculate window clamp after socket hit its memory bounds. */
 static void tcp_clamp_window(struct sock *sk, struct tcp_sock *tp)
 {
@@ -558,45 +534,6 @@ static void tcp_event_data_recv(struct sock *sk, struct tcp_sock *tp, struct sk_
 		tcp_grow_window(sk, tp, skb);
 }
 
-/* When starting a new connection, pin down the current choice of 
- * congestion algorithm.
- */
-void tcp_ca_init(struct tcp_sock *tp)
-{
-	if (sysctl_tcp_westwood) 
-		tp->adv_cong = TCP_WESTWOOD;
-	else if (sysctl_tcp_bic)
-		tp->adv_cong = TCP_BIC;
-	else if (sysctl_tcp_vegas_cong_avoid) {
-		tp->adv_cong = TCP_VEGAS;
-		tp->vegas.baseRTT = 0x7fffffff;
-		tcp_vegas_enable(tp);
-	} 
-}
-
-/* Do RTT sampling needed for Vegas.
- * Basically we:
- *   o min-filter RTT samples from within an RTT to get the current
- *     propagation delay + queuing delay (we are min-filtering to try to
- *     avoid the effects of delayed ACKs)
- *   o min-filter RTT samples from a much longer window (forever for now)
- *     to find the propagation delay (baseRTT)
- */
-static inline void vegas_rtt_calc(struct tcp_sock *tp, __u32 rtt)
-{
-	__u32 vrtt = rtt + 1; /* Never allow zero rtt or baseRTT */
-
-	/* Filter to find propagation delay: */
-	if (vrtt < tp->vegas.baseRTT) 
-		tp->vegas.baseRTT = vrtt;
-
-	/* Find the min RTT during the last RTT to find
-	 * the current prop. delay + queuing delay:
-	 */
-	tp->vegas.minRTT = min(tp->vegas.minRTT, vrtt);
-	tp->vegas.cntRTT++;
-}
-
 /* Called to compute a smoothed rtt estimate. The data fed to this
  * routine either comes from timestamps, or from segments that were
  * known _not_ to have been retransmitted [see Karn/Partridge
@@ -606,13 +543,10 @@ static inline void vegas_rtt_calc(struct tcp_sock *tp, __u32 rtt)
  * To save cycles in the RFC 1323 implementation it was better to break
  * it up into three procedures. -- erics
  */
-static void tcp_rtt_estimator(struct tcp_sock *tp, __u32 mrtt)
+static void tcp_rtt_estimator(struct tcp_sock *tp, __u32 mrtt, u32 *usrtt)
 {
 	long m = mrtt; /* RTT */
 
-	if (tcp_vegas_enabled(tp))
-		vegas_rtt_calc(tp, mrtt);
-
 	/*	The following amusing code comes from Jacobson's
 	 *	article in SIGCOMM '88.  Note that rtt and mdev
 	 *	are scaled versions of rtt and mean deviation.
@@ -670,7 +604,8 @@ static void tcp_rtt_estimator(struct tcp_sock *tp, __u32 mrtt)
 		tp->rtt_seq = tp->snd_nxt;
 	}
 
-	tcp_westwood_update_rtt(tp, tp->srtt >> 3);
+	if (tp->ca_ops->rtt_sample)
+		tp->ca_ops->rtt_sample(tp, *usrtt);
 }
 
 /* Calculate rto without backoff.  This is the second half of Van Jacobson's
@@ -1185,8 +1120,8 @@ void tcp_enter_frto(struct sock *sk)
             tp->snd_una == tp->high_seq ||
             (tp->ca_state == TCP_CA_Loss && !tp->retransmits)) {
 		tp->prior_ssthresh = tcp_current_ssthresh(tp);
-		if (!tcp_westwood_ssthresh(tp))
-			tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+		tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
+		tcp_ca_event(tp, CA_EVENT_FRTO);
 	}
 
 	/* Have to clear retransmission markers here to keep the bookkeeping
@@ -1252,8 +1187,6 @@ static void tcp_enter_frto_loss(struct sock *sk)
 	tcp_set_ca_state(tp, TCP_CA_Loss);
 	tp->high_seq = tp->frto_highmark;
 	TCP_ECN_queue_cwr(tp);
-
-	init_bictcp(tp);
 }
 
 void tcp_clear_retrans(struct tcp_sock *tp)
@@ -1283,7 +1216,8 @@ void tcp_enter_loss(struct sock *sk, int how)
 	if (tp->ca_state <= TCP_CA_Disorder || tp->snd_una == tp->high_seq ||
 	    (tp->ca_state == TCP_CA_Loss && !tp->retransmits)) {
 		tp->prior_ssthresh = tcp_current_ssthresh(tp);
-		tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+		tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
+		tcp_ca_event(tp, CA_EVENT_LOSS);
 	}
 	tp->snd_cwnd	   = 1;
 	tp->snd_cwnd_cnt   = 0;
@@ -1596,28 +1530,14 @@ static inline void tcp_moderate_cwnd(struct tcp_sock *tp)
 }
 
 /* Decrease cwnd each second ack. */
-
 static void tcp_cwnd_down(struct tcp_sock *tp)
 {
 	int decr = tp->snd_cwnd_cnt + 1;
-	__u32 limit;
-
-	/*
-	 * TCP Westwood
-	 * Here limit is evaluated as BWestimation*RTTmin (for obtaining it
-	 * in packets we use mss_cache). If sysctl_tcp_westwood is off
-	 * tcp_westwood_bw_rttmin() returns 0. In such case snd_ssthresh is
-	 * still used as usual. It prevents other strange cases in which
-	 * BWE*RTTmin could assume value 0. It should not happen but...
-	 */
-
-	if (!(limit = tcp_westwood_bw_rttmin(tp)))
-		limit = tp->snd_ssthresh/2;
 
 	tp->snd_cwnd_cnt = decr&1;
 	decr >>= 1;
 
-	if (decr && tp->snd_cwnd > limit)
+	if (decr && tp->snd_cwnd > tp->ca_ops->min_cwnd(tp))
 		tp->snd_cwnd -= decr;
 
 	tp->snd_cwnd = min(tp->snd_cwnd, tcp_packets_in_flight(tp)+1);
@@ -1654,8 +1574,8 @@ static void DBGUNDO(struct sock *sk, struct tcp_sock *tp, const char *msg)
 static void tcp_undo_cwr(struct tcp_sock *tp, int undo)
 {
 	if (tp->prior_ssthresh) {
-		if (tcp_is_bic(tp))
-			tp->snd_cwnd = max(tp->snd_cwnd, tp->bictcp.last_max_cwnd);
+		if (tp->ca_ops->undo_cwnd)
+			tp->snd_cwnd = tp->ca_ops->undo_cwnd(tp);
 		else
 			tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh<<1);
 
@@ -1767,11 +1687,9 @@ static int tcp_try_undo_loss(struct sock *sk, struct tcp_sock *tp)
 
 static inline void tcp_complete_cwr(struct tcp_sock *tp)
 {
-	if (tcp_westwood_cwnd(tp)) 
-		tp->snd_ssthresh = tp->snd_cwnd;
-	else
-		tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);
+	tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);
 	tp->snd_cwnd_stamp = tcp_time_stamp;
+	tcp_ca_event(tp, CA_EVENT_COMPLETE_CWR);
 }
 
 static void tcp_try_to_open(struct sock *sk, struct tcp_sock *tp, int flag)
@@ -1946,7 +1864,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
 		if (tp->ca_state < TCP_CA_CWR) {
 			if (!(flag&FLAG_ECE))
 				tp->prior_ssthresh = tcp_current_ssthresh(tp);
-			tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+			tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
 			TCP_ECN_queue_cwr(tp);
 		}
 
@@ -1963,7 +1881,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
 /* Read draft-ietf-tcplw-high-performance before mucking
  * with this code. (Superceeds RFC1323)
  */
-static void tcp_ack_saw_tstamp(struct tcp_sock *tp, int flag)
+static void tcp_ack_saw_tstamp(struct tcp_sock *tp, u32 *usrtt, int flag)
 {
 	__u32 seq_rtt;
 
@@ -1983,13 +1901,13 @@ static void tcp_ack_saw_tstamp(struct tcp_sock *tp, int flag)
 	 * in window is lost... Voila.	 			--ANK (010210)
 	 */
 	seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
-	tcp_rtt_estimator(tp, seq_rtt);
+	tcp_rtt_estimator(tp, seq_rtt, usrtt);
 	tcp_set_rto(tp);
 	tp->backoff = 0;
 	tcp_bound_rto(tp);
 }
 
-static void tcp_ack_no_tstamp(struct tcp_sock *tp, u32 seq_rtt, int flag)
+static void tcp_ack_no_tstamp(struct tcp_sock *tp, u32 seq_rtt, u32 *usrtt, int flag)
 {
 	/* We don't have a timestamp. Can only use
 	 * packets that are not retransmitted to determine
@@ -2003,338 +1921,29 @@ static void tcp_ack_no_tstamp(struct tcp_sock *tp, u32 seq_rtt, int flag)
 	if (flag & FLAG_RETRANS_DATA_ACKED)
 		return;
 
-	tcp_rtt_estimator(tp, seq_rtt);
+	tcp_rtt_estimator(tp, seq_rtt, usrtt);
 	tcp_set_rto(tp);
 	tp->backoff = 0;
 	tcp_bound_rto(tp);
 }
 
 static inline void tcp_ack_update_rtt(struct tcp_sock *tp,
-				      int flag, s32 seq_rtt)
+				      int flag, s32 seq_rtt, u32 *usrtt)
 {
 	/* Note that peer MAY send zero echo. In this case it is ignored. (rfc1323) */
 	if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
-		tcp_ack_saw_tstamp(tp, flag);
+		tcp_ack_saw_tstamp(tp, usrtt, flag);
 	else if (seq_rtt >= 0)
-		tcp_ack_no_tstamp(tp, seq_rtt, flag);
+		tcp_ack_no_tstamp(tp, seq_rtt, usrtt, flag);
 }
 
-/*
- * Compute congestion window to use.
- *
- * This is from the implementation of BICTCP in
- * Lison-Xu, Kahaled Harfoush, and Injog Rhee.
- *  "Binary Increase Congestion Control for Fast, Long Distance
- *  Networks" in InfoComm 2004
- * Available from:
- *  http://www.csc.ncsu.edu/faculty/rhee/export/bitcp.pdf
- *
- * Unless BIC is enabled and congestion window is large
- * this behaves the same as the original Reno.
- */
-static inline __u32 bictcp_cwnd(struct tcp_sock *tp)
-{
-	/* orignal Reno behaviour */
-	if (!tcp_is_bic(tp))
-		return tp->snd_cwnd;
-
-	if (tp->bictcp.last_cwnd == tp->snd_cwnd &&
-	   (s32)(tcp_time_stamp - tp->bictcp.last_stamp) <= (HZ>>5))
-		return tp->bictcp.cnt;
-
-	tp->bictcp.last_cwnd = tp->snd_cwnd;
-	tp->bictcp.last_stamp = tcp_time_stamp;
-      
-	/* start off normal */
-	if (tp->snd_cwnd <= sysctl_tcp_bic_low_window)
-		tp->bictcp.cnt = tp->snd_cwnd;
-
-	/* binary increase */
-	else if (tp->snd_cwnd < tp->bictcp.last_max_cwnd) {
-		__u32 	dist = (tp->bictcp.last_max_cwnd - tp->snd_cwnd)
-			/ BICTCP_B;
-
-		if (dist > BICTCP_MAX_INCREMENT)
-			/* linear increase */
-			tp->bictcp.cnt = tp->snd_cwnd / BICTCP_MAX_INCREMENT;
-		else if (dist <= 1U)
-			/* binary search increase */
-			tp->bictcp.cnt = tp->snd_cwnd * BICTCP_FUNC_OF_MIN_INCR
-				/ BICTCP_B;
-		else
-			/* binary search increase */
-			tp->bictcp.cnt = tp->snd_cwnd / dist;
-	} else {
-		/* slow start amd linear increase */
-		if (tp->snd_cwnd < tp->bictcp.last_max_cwnd + BICTCP_B)
-			/* slow start */
-			tp->bictcp.cnt = tp->snd_cwnd * BICTCP_FUNC_OF_MIN_INCR
-				/ BICTCP_B;
-		else if (tp->snd_cwnd < tp->bictcp.last_max_cwnd
-			 		+ BICTCP_MAX_INCREMENT*(BICTCP_B-1))
-			/* slow start */
-			tp->bictcp.cnt = tp->snd_cwnd * (BICTCP_B-1)
-				/ (tp->snd_cwnd-tp->bictcp.last_max_cwnd);
-		else
-			/* linear increase */
-			tp->bictcp.cnt = tp->snd_cwnd / BICTCP_MAX_INCREMENT;
-	}
-	return tp->bictcp.cnt;
-}
-
-/* This is Jacobson's slow start and congestion avoidance. 
- * SIGCOMM '88, p. 328.
- */
-static inline void reno_cong_avoid(struct tcp_sock *tp)
+static inline void tcp_cong_avoid(struct tcp_sock *tp, u32 ack, u32 rtt,
+				  u32 in_flight, int good)
 {
-        if (tp->snd_cwnd <= tp->snd_ssthresh) {
-                /* In "safe" area, increase. */
-		if (tp->snd_cwnd < tp->snd_cwnd_clamp)
-			tp->snd_cwnd++;
-	} else {
-                /* In dangerous area, increase slowly.
-		 * In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
-		 */
-		if (tp->snd_cwnd_cnt >= bictcp_cwnd(tp)) {
-			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
-				tp->snd_cwnd++;
-			tp->snd_cwnd_cnt=0;
-		} else
-			tp->snd_cwnd_cnt++;
-        }
+	tp->ca_ops->cong_avoid(tp, ack, rtt, in_flight, good);
 	tp->snd_cwnd_stamp = tcp_time_stamp;
 }
 
-/* This is based on the congestion detection/avoidance scheme described in
- *    Lawrence S. Brakmo and Larry L. Peterson.
- *    "TCP Vegas: End to end congestion avoidance on a global internet."
- *    IEEE Journal on Selected Areas in Communication, 13(8):1465--1480,
- *    October 1995. Available from:
- *	ftp://ftp.cs.arizona.edu/xkernel/Papers/jsac.ps
- *
- * See http://www.cs.arizona.edu/xkernel/ for their implementation.
- * The main aspects that distinguish this implementation from the
- * Arizona Vegas implementation are:
- *   o We do not change the loss detection or recovery mechanisms of
- *     Linux in any way. Linux already recovers from losses quite well,
- *     using fine-grained timers, NewReno, and FACK.
- *   o To avoid the performance penalty imposed by increasing cwnd
- *     only every-other RTT during slow start, we increase during
- *     every RTT during slow start, just like Reno.
- *   o Largely to allow continuous cwnd growth during slow start,
- *     we use the rate at which ACKs come back as the "actual"
- *     rate, rather than the rate at which data is sent.
- *   o To speed convergence to the right rate, we set the cwnd
- *     to achieve the right ("actual") rate when we exit slow start.
- *   o To filter out the noise caused by delayed ACKs, we use the
- *     minimum RTT sample observed during the last RTT to calculate
- *     the actual rate.
- *   o When the sender re-starts from idle, it waits until it has
- *     received ACKs for an entire flight of new data before making
- *     a cwnd adjustment decision. The original Vegas implementation
- *     assumed senders never went idle.
- */
-static void vegas_cong_avoid(struct tcp_sock *tp, u32 ack, u32 seq_rtt)
-{
-	/* The key players are v_beg_snd_una and v_beg_snd_nxt.
-	 *
-	 * These are so named because they represent the approximate values
-	 * of snd_una and snd_nxt at the beginning of the current RTT. More
-	 * precisely, they represent the amount of data sent during the RTT.
-	 * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
-	 * we will calculate that (v_beg_snd_nxt - v_beg_snd_una) outstanding
-	 * bytes of data have been ACKed during the course of the RTT, giving
-	 * an "actual" rate of:
-	 *
-	 *     (v_beg_snd_nxt - v_beg_snd_una) / (rtt duration)
-	 *
-	 * Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
-	 * because delayed ACKs can cover more than one segment, so they
-	 * don't line up nicely with the boundaries of RTTs.
-	 *
-	 * Another unfortunate fact of life is that delayed ACKs delay the
-	 * advance of the left edge of our send window, so that the number
-	 * of bytes we send in an RTT is often less than our cwnd will allow.
-	 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
-	 */
-
-	if (after(ack, tp->vegas.beg_snd_nxt)) {
-		/* Do the Vegas once-per-RTT cwnd adjustment. */
-		u32 old_wnd, old_snd_cwnd;
-
-		
-		/* Here old_wnd is essentially the window of data that was
-		 * sent during the previous RTT, and has all
-		 * been acknowledged in the course of the RTT that ended
-		 * with the ACK we just received. Likewise, old_snd_cwnd
-		 * is the cwnd during the previous RTT.
-		 */
-		old_wnd = (tp->vegas.beg_snd_nxt - tp->vegas.beg_snd_una) /
-			tp->mss_cache_std;
-		old_snd_cwnd = tp->vegas.beg_snd_cwnd;
-
-		/* Save the extent of the current window so we can use this
-		 * at the end of the next RTT.
-		 */
-		tp->vegas.beg_snd_una  = tp->vegas.beg_snd_nxt;
-		tp->vegas.beg_snd_nxt  = tp->snd_nxt;
-		tp->vegas.beg_snd_cwnd = tp->snd_cwnd;
-
-		/* Take into account the current RTT sample too, to
-		 * decrease the impact of delayed acks. This double counts
-		 * this sample since we count it for the next window as well,
-		 * but that's not too awful, since we're taking the min,
-		 * rather than averaging.
-		 */
-		vegas_rtt_calc(tp, seq_rtt);
-
-		/* We do the Vegas calculations only if we got enough RTT
-		 * samples that we can be reasonably sure that we got
-		 * at least one RTT sample that wasn't from a delayed ACK.
-		 * If we only had 2 samples total,
-		 * then that means we're getting only 1 ACK per RTT, which
-		 * means they're almost certainly delayed ACKs.
-		 * If  we have 3 samples, we should be OK.
-		 */
-
-		if (tp->vegas.cntRTT <= 2) {
-			/* We don't have enough RTT samples to do the Vegas
-			 * calculation, so we'll behave like Reno.
-			 */
-			if (tp->snd_cwnd > tp->snd_ssthresh)
-				tp->snd_cwnd++;
-		} else {
-			u32 rtt, target_cwnd, diff;
-
-			/* We have enough RTT samples, so, using the Vegas
-			 * algorithm, we determine if we should increase or
-			 * decrease cwnd, and by how much.
-			 */
-
-			/* Pluck out the RTT we are using for the Vegas
-			 * calculations. This is the min RTT seen during the
-			 * last RTT. Taking the min filters out the effects
-			 * of delayed ACKs, at the cost of noticing congestion
-			 * a bit later.
-			 */
-			rtt = tp->vegas.minRTT;
-
-			/* Calculate the cwnd we should have, if we weren't
-			 * going too fast.
-			 *
-			 * This is:
-			 *     (actual rate in segments) * baseRTT
-			 * We keep it as a fixed point number with
-			 * V_PARAM_SHIFT bits to the right of the binary point.
-			 */
-			target_cwnd = ((old_wnd * tp->vegas.baseRTT)
-				       << V_PARAM_SHIFT) / rtt;
-
-			/* Calculate the difference between the window we had,
-			 * and the window we would like to have. This quantity
-			 * is the "Diff" from the Arizona Vegas papers.
-			 *
-			 * Again, this is a fixed point number with
-			 * V_PARAM_SHIFT bits to the right of the binary
-			 * point.
-			 */
-			diff = (old_wnd << V_PARAM_SHIFT) - target_cwnd;
-
-			if (tp->snd_cwnd < tp->snd_ssthresh) {
-				/* Slow start.  */
-				if (diff > sysctl_tcp_vegas_gamma) {
-					/* Going too fast. Time to slow down
-					 * and switch to congestion avoidance.
-					 */
-					tp->snd_ssthresh = 2;
-
-					/* Set cwnd to match the actual rate
-					 * exactly:
-					 *   cwnd = (actual rate) * baseRTT
-					 * Then we add 1 because the integer
-					 * truncation robs us of full link
-					 * utilization.
-					 */
-					tp->snd_cwnd = min(tp->snd_cwnd,
-							   (target_cwnd >>
-							    V_PARAM_SHIFT)+1);
-
-				}
-			} else {
-				/* Congestion avoidance. */
-				u32 next_snd_cwnd;
-
-				/* Figure out where we would like cwnd
-				 * to be.
-				 */
-				if (diff > sysctl_tcp_vegas_beta) {
-					/* The old window was too fast, so
-					 * we slow down.
-					 */
-					next_snd_cwnd = old_snd_cwnd - 1;
-				} else if (diff < sysctl_tcp_vegas_alpha) {
-					/* We don't have enough extra packets
-					 * in the network, so speed up.
-					 */
-					next_snd_cwnd = old_snd_cwnd + 1;
-				} else {
-					/* Sending just as fast as we
-					 * should be.
-					 */
-					next_snd_cwnd = old_snd_cwnd;
-				}
-
-				/* Adjust cwnd upward or downward, toward the
-				 * desired value.
-				 */
-				if (next_snd_cwnd > tp->snd_cwnd)
-					tp->snd_cwnd++;
-				else if (next_snd_cwnd < tp->snd_cwnd)
-					tp->snd_cwnd--;
-			}
-		}
-
-		/* Wipe the slate clean for the next RTT. */
-		tp->vegas.cntRTT = 0;
-		tp->vegas.minRTT = 0x7fffffff;
-	}
-
-	/* The following code is executed for every ack we receive,
-	 * except for conditions checked in should_advance_cwnd()
-	 * before the call to tcp_cong_avoid(). Mainly this means that
-	 * we only execute this code if the ack actually acked some
-	 * data.
-	 */
-
-	/* If we are in slow start, increase our cwnd in response to this ACK.
-	 * (If we are not in slow start then we are in congestion avoidance,
-	 * and adjust our congestion window only once per RTT. See the code
-	 * above.)
-	 */
-	if (tp->snd_cwnd <= tp->snd_ssthresh) 
-		tp->snd_cwnd++;
-
-	/* to keep cwnd from growing without bound */
-	tp->snd_cwnd = min_t(u32, tp->snd_cwnd, tp->snd_cwnd_clamp);
-
-	/* Make sure that we are never so timid as to reduce our cwnd below
-	 * 2 MSS.
-	 *
-	 * Going below 2 MSS