[PATCH 0/4] TCP related patches for net-2.6.22

[PATCH 0/4] TCP related patches for net-2.6.22

[Stephen Hemminger]

A bunch of TCP congestion control updates for 2.6.22.

The first one might be a bug fix worth backporting.
It addresses a problem that causes vegas not to work right
when using the setsockopt() setup.

-- 

[PATCH 1/4] tcp: congestion control initialization

[Stephen Hemminger]

[-- Attachment #1: tcp-cong-start.patch --]
[-- Type: text/plain, Size: 1833 bytes --]

Change to defer congestion control initialization.

If setsockopt() was used to change TCP_CONGESTION before
connection is established, then protocols that use sequence numbers
to keep track of one RTT interval (vegas, illinois, ...) get confused.

Change the init hook to be called after handshake.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

---
 net/ipv4/tcp_cong.c |   23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

--- net-2.6.22.orig/net/ipv4/tcp_cong.c
+++ net-2.6.22/net/ipv4/tcp_cong.c
@@ -79,18 +79,19 @@ void tcp_init_congestion_control(struct 
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct tcp_congestion_ops *ca;
 
-	if (icsk->icsk_ca_ops != &tcp_init_congestion_ops)
-		return;
+	/* if no choice made yet assign the current value set as default */
+	if (icsk->icsk_ca_ops == &tcp_init_congestion_ops) {
+		rcu_read_lock();
+		list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
+			if (try_module_get(ca->owner)) {
+				icsk->icsk_ca_ops = ca;
+				break;
+			}
 
-	rcu_read_lock();
-	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
-		if (try_module_get(ca->owner)) {
-			icsk->icsk_ca_ops = ca;
-			break;
+			/* fallback to next available */
 		}
-
+		rcu_read_unlock();
 	}
-	rcu_read_unlock();
 
 	if (icsk->icsk_ca_ops->init)
 		icsk->icsk_ca_ops->init(sk);
@@ -238,6 +239,7 @@ int tcp_set_congestion_control(struct so
 
 	rcu_read_lock();
 	ca = tcp_ca_find(name);
+
 	/* no change asking for existing value */
 	if (ca == icsk->icsk_ca_ops)
 		goto out;
@@ -263,7 +265,8 @@ int tcp_set_congestion_control(struct so
 	else {
 		tcp_cleanup_congestion_control(sk);
 		icsk->icsk_ca_ops = ca;
-		if (icsk->icsk_ca_ops->init)
+
+		if (sk->sk_state != TCP_CLOSE && icsk->icsk_ca_ops->init)
 			icsk->icsk_ca_ops->init(sk);
 	}
  out:

-- 

[PATCH 2/4] TCP Illinois update

[Stephen Hemminger]

[-- Attachment #1: tcp-illinois-v4.patch --]
[-- Type: text/plain, Size: 11116 bytes --]

This version more closely matches the paper, and fixes several
math errors. The biggest difference is that it updates alpha/beta
once per RTT

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
---
 net/ipv4/tcp_illinois.c |  298 +++++++++++++++++++++++++++++-------------------
 1 file changed, 186 insertions(+), 112 deletions(-)

--- net-2.6.22.orig/net/ipv4/tcp_illinois.c
+++ net-2.6.22/net/ipv4/tcp_illinois.c
@@ -23,74 +23,106 @@
 #define ALPHA_MIN	((3*ALPHA_SCALE)/10)	/* ~0.3 */
 #define ALPHA_MAX	(10*ALPHA_SCALE)	/* 10.0 */
 #define ALPHA_BASE	ALPHA_SCALE		/* 1.0 */
+#define U32_MAX		((u32)~0U)
+#define RTT_MAX		(U32_MAX / ALPHA_MAX)	/* 3.3 secs */
 
 #define BETA_SHIFT	6
 #define BETA_SCALE	(1u<<BETA_SHIFT)
-#define BETA_MIN	(BETA_SCALE/8)		/* 0.8 */
-#define BETA_MAX	(BETA_SCALE/2)
-#define BETA_BASE	BETA_MAX		/* 0.5 */
-
-#define THETA		5
+#define BETA_MIN	(BETA_SCALE/8)		/* 0.125 */
+#define BETA_MAX	(BETA_SCALE/2)		/* 0.5 */
+#define BETA_BASE	BETA_MAX
 
 static int win_thresh __read_mostly = 15;
-module_param(win_thresh, int, 0644);
+module_param(win_thresh, int, 0);
 MODULE_PARM_DESC(win_thresh, "Window threshold for starting adaptive sizing");
 
-#define MAX_RTT		0x7fffffff
+static int theta __read_mostly = 5;
+module_param(theta, int, 0);
+MODULE_PARM_DESC(theta, "# of fast RTT's before full growth");
 
 /* TCP Illinois Parameters */
-struct tcp_illinois {
-	u32	last_alpha;
-	u32	min_rtt;
-	u32	max_rtt;
-	u32	rtt_low;
-	u32	rtt_cnt;
-	u64	sum_rtt;
+struct illinois {
+	u64	sum_rtt;	/* sum of rtt's measured within last rtt */
+	u16	cnt_rtt;	/* # of rtts measured within last rtt */
+	u32	base_rtt;	/* min of all rtt in usec */
+	u32	max_rtt;	/* max of all rtt in usec */
+	u32	end_seq;	/* right edge of current RTT */
+	u32	alpha;		/* Additive increase */
+	u32	beta;		/* Muliplicative decrease */
+	u16	acked;		/* # packets acked by current ACK */
+	u8	rtt_above;	/* average rtt has gone above threshold */
+	u8	rtt_low;	/* # of rtts measurements below threshold */
 };
 
+static void rtt_reset(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct illinois *ca = inet_csk_ca(sk);
+
+	ca->end_seq = tp->snd_nxt;
+	ca->cnt_rtt = 0;
+	ca->sum_rtt = 0;
+
+	/* TODO: age max_rtt? */
+}
+
 static void tcp_illinois_init(struct sock *sk)
 {
-	struct tcp_illinois *ca = inet_csk_ca(sk);
+	struct illinois *ca = inet_csk_ca(sk);
+
+	ca->alpha = ALPHA_MAX;
+	ca->beta = BETA_BASE;
+	ca->base_rtt = 0x7fffffff;
+	ca->max_rtt = 0;
 
-	ca->last_alpha = ALPHA_BASE;
-	ca->min_rtt = 0x7fffffff;
+	ca->acked = 0;
+	ca->rtt_low = 0;
+	ca->rtt_above = 0;
+
+	rtt_reset(sk);
 }
 
-/*
- * Keep track of min, max and average RTT
- */
-static void tcp_illinois_rtt_calc(struct sock *sk, u32 rtt)
+/* Measure RTT for each ack. */
+static void tcp_illinois_rtt_sample(struct sock *sk, u32 rtt)
 {
-	struct tcp_illinois *ca = inet_csk_ca(sk);
+	struct illinois *ca = inet_csk_ca(sk);
+
+	/* ignore bogus values, this prevents wraparound in alpha math */
+	if (rtt > RTT_MAX)
+		rtt = RTT_MAX;
+
+	/* keep track of minimum RTT seen so far */
+	if (ca->base_rtt > rtt)
+		ca->base_rtt = rtt;
 
-	if (rtt < ca->min_rtt)
-		ca->min_rtt = rtt;
-	if (rtt > ca->max_rtt)
+	/* and max */
+	if (ca->max_rtt < rtt)
 		ca->max_rtt = rtt;
 
-	if (++ca->rtt_cnt == 1)
-		ca->sum_rtt = rtt;
-	else
-		ca->sum_rtt += rtt;
+	++ca->cnt_rtt;
+	ca->sum_rtt += rtt;
 }
 
-/* max queuing delay */
-static inline u32 max_delay(const struct tcp_illinois *ca)
+/* Capture count of packets covered by ack, to adjust for delayed acks */
+static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked)
 {
-	return ca->max_rtt - ca->min_rtt;
+	struct illinois *ca = inet_csk_ca(sk);
+	ca->acked = pkts_acked;
 }
 
-/* average queueing delay */
-static u32 avg_delay(struct tcp_illinois *ca)
+/* Maximum queuing delay */
+static inline u32 max_delay(const struct illinois *ca)
 {
-	u64 avg_rtt = ca->sum_rtt;
-
-	do_div(avg_rtt, ca->rtt_cnt);
+	return ca->max_rtt - ca->base_rtt;
+}
 
-	ca->sum_rtt = 0;
-	ca->rtt_cnt = 0;
+/* Average queuing delay */
+static inline u32 avg_delay(const struct illinois *ca)
+{
+	u64 t = ca->sum_rtt;
 
-	return avg_rtt - ca->min_rtt;
+	do_div(t, ca->cnt_rtt);
+	return t - ca->base_rtt;
 }
 
 /*
@@ -101,32 +133,31 @@ static u32 avg_delay(struct tcp_illinois
  * A. If average delay is at minimum (we are uncongested),
  *    then use large alpha (10.0) to increase faster.
  * B. If average delay is at maximum (getting congested)
- *    then use small alpha (1.0)
+ *    then use small alpha (0.3)
  *
  * The result is a convex window growth curve.
  */
-static u32 alpha(const struct sock *sk)
+static u32 alpha(struct illinois *ca, u32 da, u32 dm)
 {
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct tcp_illinois *ca = inet_csk_ca(sk);
-	u32 dm = max_delay(ca);
-	u32 da = avg_delay(ca);
-	u32 d1, a;
-
-	if (tp->snd_cwnd < win_thresh)
-		return ALPHA_BASE;	/* same as Reno (1.0) */
+	u32 d1 = dm / 100;	/* Low threshold */
 
-	d1 = dm / 100;
 	if (da <= d1) {
-		/* Don't let noise force agressive response */
-		if (ca->rtt_low < THETA) {
-			++ca->rtt_low;
-			return ca->last_alpha;
-		} else
+		/* If never got out of low delay zone, then use max */
+		if (!ca->rtt_above)
 			return ALPHA_MAX;
+
+		/* Wait for 5 good RTT's before allowing alpha to go alpha max.
+		 * This prevents one good RTT from causing sudden window increase.
+		 */
+		if (++ca->rtt_low < theta)
+			return ca->alpha;
+
+		ca->rtt_low = 0;
+		ca->rtt_above = 0;
+		return ALPHA_MAX;
 	}
 
-	ca->rtt_low = 0;
+	ca->rtt_above = 1;
 
 	/*
 	 * Based on:
@@ -146,37 +177,8 @@ static u32 alpha(const struct sock *sk)
 
 	dm -= d1;
 	da -= d1;
-
-	a = (dm * ALPHA_MAX) / (dm - (da  * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN);
-	ca->last_alpha = a;
-	return a;
-}
-
-/*
- * Increase window in response to successful acknowledgment.
- */
-static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt,
-				    u32 in_flight, int flag)
-{
-	struct tcp_sock *tp = tcp_sk(sk);
-
-	/* RFC2861 only increase cwnd if fully utilized */
-	if (!tcp_is_cwnd_limited(sk, in_flight))
-		return;
-
-	/* In slow start */
-	if (tp->snd_cwnd <= tp->snd_ssthresh)
-		tcp_slow_start(tp);
-
-	else {
-		/* additive increase  cwnd += alpha / cwnd */
-		if ((tp->snd_cwnd_cnt * alpha(sk)) >> ALPHA_SHIFT >= tp->snd_cwnd) {
-			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
-				tp->snd_cwnd++;
-			tp->snd_cwnd_cnt = 0;
-		} else
-			tp->snd_cwnd_cnt++;
-	}
+	return (dm * ALPHA_MAX) /
+		(dm + (da  * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN);
 }
 
 /*
@@ -187,20 +189,14 @@ static void tcp_illinois_cong_avoid(stru
  * If delay is up to 80% of max then beta = 1/2
  * In between is a linear function
  */
-static inline u32 beta(struct sock *sk)
+static u32 beta(u32 da, u32 dm)
 {
-	struct tcp_sock *tp = tcp_sk(sk);
-	struct tcp_illinois *ca = inet_csk_ca(sk);
-	u32 dm = max_delay(ca);
-	u32 da = avg_delay(ca);
 	u32 d2, d3;
 
-	if (tp->snd_cwnd < win_thresh)
-		return BETA_BASE;
-
 	d2 = dm / 10;
 	if (da <= d2)
 		return BETA_MIN;
+
 	d3 = (8 * dm) / 10;
 	if (da >= d3 || d3 <= d2)
 		return BETA_MAX;
@@ -222,31 +218,107 @@ static inline u32 beta(struct sock *sk)
 		/ (d3 - d2);
 }
 
+/* Update alpha and beta values once per RTT */
+static void update_params(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct illinois *ca = inet_csk_ca(sk);
+
+	if (tp->snd_cwnd < win_thresh) {
+		ca->alpha = ALPHA_BASE;
+		ca->beta = BETA_BASE;
+	} else if (ca->cnt_rtt > 0) {
+		u32 dm = max_delay(ca);
+		u32 da = avg_delay(ca);
+
+		ca->alpha = alpha(ca, da, dm);
+		ca->beta = beta(da, dm);
+	}
+
+	rtt_reset(sk);
+}
+
+/*
+ * In case of loss, reset to default values
+ */
+static void tcp_illinois_state(struct sock *sk, u8 new_state)
+{
+	struct illinois *ca = inet_csk_ca(sk);
+
+	if (new_state == TCP_CA_Loss) {
+		ca->alpha = ALPHA_BASE;
+		ca->beta = BETA_BASE;
+		ca->rtt_low = 0;
+		ca->rtt_above = 0;
+		rtt_reset(sk);
+	}
+}
+
+/*
+ * Increase window in response to successful acknowledgment.
+ */
+static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt,
+				    u32 in_flight, int flag)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct illinois *ca = inet_csk_ca(sk);
+
+	if (after(ack, ca->end_seq))
+		update_params(sk);
+
+	/* RFC2861 only increase cwnd if fully utilized */
+	if (!tcp_is_cwnd_limited(sk, in_flight))
+		return;
+
+	/* In slow start */
+	if (tp->snd_cwnd <= tp->snd_ssthresh)
+		tcp_slow_start(tp);
+
+	else {
+		u32 delta;
+
+		/* snd_cwnd_cnt is # of packets since last cwnd increment */
+		tp->snd_cwnd_cnt += ca->acked;
+		ca->acked = 1;
+
+		/* This is close approximation of:
+		 * tp->snd_cwnd += alpha/tp->snd_cwnd
+		*/
+		delta = (tp->snd_cwnd_cnt * ca->alpha) >> ALPHA_SHIFT;
+		if (delta >= tp->snd_cwnd) {
+			tp->snd_cwnd = min(tp->snd_cwnd + delta / tp->snd_cwnd,
+				     	   (u32) tp->snd_cwnd_clamp);
+			tp->snd_cwnd_cnt = 0;
+		}
+	}
+}
+
 static u32 tcp_illinois_ssthresh(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
+	struct illinois *ca = inet_csk_ca(sk);
 
 	/* Multiplicative decrease */
-	return max((tp->snd_cwnd * beta(sk)) >> BETA_SHIFT, 2U);
+	return max((tp->snd_cwnd * ca->beta) >> BETA_SHIFT, 2U);
 }
 
-/* Extract info for TCP socket info provided via netlink.
- * We aren't really doing Vegas, but we can provide RTT info
- */
-static void tcp_illinois_get_info(struct sock *sk, u32 ext,
-			       struct sk_buff *skb)
+
+/* Extract info for Tcp socket info provided via netlink. */
+static void tcp_illinois_info(struct sock *sk, u32 ext,
+			      struct sk_buff *skb)
 {
-	const struct tcp_illinois *ca = inet_csk_ca(sk);
+	const struct illinois *ca = inet_csk_ca(sk);
 
 	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
 		struct tcpvegas_info info = {
 			.tcpv_enabled = 1,
-			.tcpv_rttcnt = ca->rtt_cnt,
-			.tcpv_minrtt = ca->min_rtt,
+			.tcpv_rttcnt = ca->cnt_rtt,
+			.tcpv_minrtt = ca->base_rtt,
 		};
-		u64 avg_rtt = ca->sum_rtt;
-		do_div(avg_rtt, ca->rtt_cnt);
-		info.tcpv_rtt = avg_rtt;
+		u64 t = ca->sum_rtt;
+
+		do_div(t, ca->cnt_rtt);
+		info.tcpv_rtt = t;
 
 		nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
 	}
@@ -257,8 +329,10 @@ static struct tcp_congestion_ops tcp_ill
 	.ssthresh	= tcp_illinois_ssthresh,
 	.min_cwnd	= tcp_reno_min_cwnd,
 	.cong_avoid	= tcp_illinois_cong_avoid,
-	.rtt_sample	= tcp_illinois_rtt_calc,
-	.get_info	= tcp_illinois_get_info,
+	.set_state	= tcp_illinois_state,
+	.rtt_sample	= tcp_illinois_rtt_sample,
+	.get_info	= tcp_illinois_info,
+	.pkts_acked	= tcp_illinois_acked,
 
 	.owner		= THIS_MODULE,
 	.name		= "illinois",
@@ -266,7 +340,7 @@ static struct tcp_congestion_ops tcp_ill
 
 static int __init tcp_illinois_register(void)
 {
-	BUILD_BUG_ON(sizeof(struct tcp_illinois) > ICSK_CA_PRIV_SIZE);
+	BUILD_BUG_ON(sizeof(struct illinois) > ICSK_CA_PRIV_SIZE);
 	return tcp_register_congestion_control(&tcp_illinois);
 }
 
@@ -281,4 +355,4 @@ module_exit(tcp_illinois_unregister);
 MODULE_AUTHOR("Stephen Hemminger, Shao Liu");
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("TCP Illinois");
-MODULE_VERSION("0.3");
+MODULE_VERSION("1.0");

-- 

[PATCH 3/4] tcp: congestion control API update

[Stephen Hemminger]

[-- Attachment #1: tcp-cong-api-update.patch --]
[-- Type: text/plain, Size: 15152 bytes --]

Do some simple changes to make congestion control API faster/cleaner.
* use ktime_t rather than timeval
* merge rtt sampling into existing ack callback
  this means one indirect call versus two per ack.
* use flags bits to store options/settings

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

---
 include/linux/skbuff.h  |    5 +++++
 include/net/tcp.h       |    9 +++++----
 net/ipv4/tcp_bic.c      |    2 +-
 net/ipv4/tcp_cong.c     |   14 +++++++-------
 net/ipv4/tcp_cubic.c    |    2 +-
 net/ipv4/tcp_htcp.c     |    2 +-
 net/ipv4/tcp_illinois.c |   16 +++++++---------
 net/ipv4/tcp_input.c    |   25 ++++++++-----------------
 net/ipv4/tcp_lp.c       |    8 +++++---
 net/ipv4/tcp_output.c   |    2 +-
 net/ipv4/tcp_vegas.c    |   10 +++++++---
 net/ipv4/tcp_veno.c     |   10 +++++++---
 net/ipv4/tcp_westwood.c |    2 +-
 net/ipv4/tcp_yeah.c     |    6 ++++--
 net/ipv4/tcp_yeah.h     |    7 +++++--
 15 files changed, 65 insertions(+), 55 deletions(-)

--- net-2.6.22.orig/include/linux/skbuff.h
+++ net-2.6.22/include/linux/skbuff.h
@@ -1569,6 +1569,11 @@ static inline void __net_timestamp(struc
 	skb->tstamp = ktime_get_real();
 }
 
+static inline ktime_t net_timedelta(ktime_t t)
+{
+	return ktime_sub(ktime_get_real(), t);
+}
+
 
 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
--- net-2.6.22.orig/include/net/tcp.h
+++ net-2.6.22/include/net/tcp.h
@@ -629,9 +629,12 @@ enum tcp_ca_event {
 #define TCP_CA_MAX	128
 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
 
+#define TCP_CONG_NON_RESTRICTED 0x1
+#define TCP_CONG_RTT_STAMP	0x2
+
 struct tcp_congestion_ops {
 	struct list_head	list;
-	int	non_restricted;
+	unsigned long flags;
 
 	/* initialize private data (optional) */
 	void (*init)(struct sock *sk);
@@ -645,8 +648,6 @@ struct tcp_congestion_ops {
 	/* do new cwnd calculation (required) */
 	void (*cong_avoid)(struct sock *sk, u32 ack,
 			   u32 rtt, u32 in_flight, int good_ack);
-	/* round trip time sample per acked packet (optional) */
-	void (*rtt_sample)(struct sock *sk, u32 usrtt);
 	/* call before changing ca_state (optional) */
 	void (*set_state)(struct sock *sk, u8 new_state);
 	/* call when cwnd event occurs (optional) */
@@ -654,7 +655,7 @@ struct tcp_congestion_ops {
 	/* new value of cwnd after loss (optional) */
 	u32  (*undo_cwnd)(struct sock *sk);
 	/* hook for packet ack accounting (optional) */
-	void (*pkts_acked)(struct sock *sk, u32 num_acked);
+	void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last);
 	/* get info for inet_diag (optional) */
 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
 
--- net-2.6.22.orig/net/ipv4/tcp_bic.c
+++ net-2.6.22/net/ipv4/tcp_bic.c
@@ -206,7 +206,7 @@ static void bictcp_state(struct sock *sk
 /* Track delayed acknowledgment ratio using sliding window
  * ratio = (15*ratio + sample) / 16
  */
-static void bictcp_acked(struct sock *sk, u32 cnt)
+static void bictcp_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 
--- net-2.6.22.orig/net/ipv4/tcp_cong.c
+++ net-2.6.22/net/ipv4/tcp_cong.c
@@ -126,7 +126,7 @@ int tcp_set_default_congestion_control(c
 #endif
 
 	if (ca) {
-		ca->non_restricted = 1;	/* default is always allowed */
+		ca->flags |= TCP_CONG_NON_RESTRICTED;	/* default is always allowed */
 		list_move(&ca->list, &tcp_cong_list);
 		ret = 0;
 	}
@@ -181,7 +181,7 @@ void tcp_get_allowed_congestion_control(
 	*buf = '\0';
 	rcu_read_lock();
 	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
-		if (!ca->non_restricted)
+		if (!(ca->flags & TCP_CONG_NON_RESTRICTED))
 			continue;
 		offs += snprintf(buf + offs, maxlen - offs,
 				 "%s%s",
@@ -212,16 +212,16 @@ int tcp_set_allowed_congestion_control(c
 		}
 	}
 
-	/* pass 2 clear */
+	/* pass 2 clear old values */
 	list_for_each_entry_rcu(ca, &tcp_cong_list, list)
-		ca->non_restricted = 0;
+		ca->flags &= ~TCP_CONG_NON_RESTRICTED;
 
 	/* pass 3 mark as allowed */
 	while ((name = strsep(&val, " ")) && *name) {
 		ca = tcp_ca_find(name);
 		WARN_ON(!ca);
 		if (ca)
-			ca->non_restricted = 1;
+			ca->flags |= TCP_CONG_NON_RESTRICTED;
 	}
 out:
 	spin_unlock(&tcp_cong_list_lock);
@@ -256,7 +256,7 @@ int tcp_set_congestion_control(struct so
 	if (!ca)
 		err = -ENOENT;
 
-	else if (!(ca->non_restricted || capable(CAP_NET_ADMIN)))
+	else if (!((ca->flags & TCP_CONG_NON_RESTRICTED) || capable(CAP_NET_ADMIN)))
 		err = -EPERM;
 
 	else if (!try_module_get(ca->owner))
@@ -371,8 +371,8 @@ u32 tcp_reno_min_cwnd(const struct sock 
 EXPORT_SYMBOL_GPL(tcp_reno_min_cwnd);
 
 struct tcp_congestion_ops tcp_reno = {
+	.flags		= TCP_CONG_NON_RESTRICTED,
 	.name		= "reno",
-	.non_restricted = 1,
 	.owner		= THIS_MODULE,
 	.ssthresh	= tcp_reno_ssthresh,
 	.cong_avoid	= tcp_reno_cong_avoid,
--- net-2.6.22.orig/net/ipv4/tcp_cubic.c
+++ net-2.6.22/net/ipv4/tcp_cubic.c
@@ -334,7 +334,7 @@ static void bictcp_state(struct sock *sk
 /* Track delayed acknowledgment ratio using sliding window
  * ratio = (15*ratio + sample) / 16
  */
-static void bictcp_acked(struct sock *sk, u32 cnt)
+static void bictcp_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 
--- net-2.6.22.orig/net/ipv4/tcp_htcp.c
+++ net-2.6.22/net/ipv4/tcp_htcp.c
@@ -98,7 +98,7 @@ static inline void measure_rtt(struct so
 	}
 }
 
-static void measure_achieved_throughput(struct sock *sk, u32 pkts_acked)
+static void measure_achieved_throughput(struct sock *sk, u32 pkts_acked, ktime_t last)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	const struct tcp_sock *tp = tcp_sk(sk);
--- net-2.6.22.orig/net/ipv4/tcp_illinois.c
+++ net-2.6.22/net/ipv4/tcp_illinois.c
@@ -83,9 +83,14 @@ static void tcp_illinois_init(struct soc
 }
 
 /* Measure RTT for each ack. */
-static void tcp_illinois_rtt_sample(struct sock *sk, u32 rtt)
+static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked, ktime_t last)
 {
 	struct illinois *ca = inet_csk_ca(sk);
+	u32 rtt;
+
+	ca->acked = pkts_acked;
+
+	rtt = ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC;
 
 	/* ignore bogus values, this prevents wraparound in alpha math */
 	if (rtt > RTT_MAX)
@@ -103,13 +108,6 @@ static void tcp_illinois_rtt_sample(stru
 	ca->sum_rtt += rtt;
 }
 
-/* Capture count of packets covered by ack, to adjust for delayed acks */
-static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked)
-{
-	struct illinois *ca = inet_csk_ca(sk);
-	ca->acked = pkts_acked;
-}
-
 /* Maximum queuing delay */
 static inline u32 max_delay(const struct illinois *ca)
 {
@@ -325,12 +323,12 @@ static void tcp_illinois_info(struct soc
 }
 
 static struct tcp_congestion_ops tcp_illinois = {
+	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_illinois_init,
 	.ssthresh	= tcp_illinois_ssthresh,
 	.min_cwnd	= tcp_reno_min_cwnd,
 	.cong_avoid	= tcp_illinois_cong_avoid,
 	.set_state	= tcp_illinois_state,
-	.rtt_sample	= tcp_illinois_rtt_sample,
 	.get_info	= tcp_illinois_info,
 	.pkts_acked	= tcp_illinois_acked,
 
--- net-2.6.22.orig/net/ipv4/tcp_input.c
+++ net-2.6.22/net/ipv4/tcp_input.c
@@ -2402,14 +2402,6 @@ static int tcp_tso_acked(struct sock *sk
 	return acked;
 }
 
-static u32 tcp_usrtt(struct timeval *tv)
-{
-	struct timeval now;
-
-	do_gettimeofday(&now);
-	return (now.tv_sec - tv->tv_sec) * 1000000 + (now.tv_usec - tv->tv_usec);
-}
-
 /* Remove acknowledged frames from the retransmission queue. */
 static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p)
 {
@@ -2420,9 +2412,7 @@ static int tcp_clean_rtx_queue(struct so
 	int acked = 0;
 	__s32 seq_rtt = -1;
 	u32 pkts_acked = 0;
-	void (*rtt_sample)(struct sock *sk, u32 usrtt)
-		= icsk->icsk_ca_ops->rtt_sample;
-	struct timeval tv = { .tv_sec = 0, .tv_usec = 0 };
+	ktime_t last_ackt = ktime_set(0,0);
 
 	while ((skb = tcp_write_queue_head(sk)) &&
 	       skb != tcp_send_head(sk)) {
@@ -2471,7 +2461,7 @@ static int tcp_clean_rtx_queue(struct so
 				seq_rtt = -1;
 			} else if (seq_rtt < 0) {
 				seq_rtt = now - scb->when;
-				skb_get_timestamp(skb, &tv);
+				last_ackt = skb->tstamp;
 			}
 			if (sacked & TCPCB_SACKED_ACKED)
 				tp->sacked_out -= tcp_skb_pcount(skb);
@@ -2484,7 +2474,7 @@ static int tcp_clean_rtx_queue(struct so
 			}
 		} else if (seq_rtt < 0) {
 			seq_rtt = now - scb->when;
-			skb_get_timestamp(skb, &tv);
+			last_ackt = skb->tstamp;
 		}
 		tcp_dec_pcount_approx(&tp->fackets_out, skb);
 		tcp_packets_out_dec(tp, skb);
@@ -2494,13 +2484,14 @@ static int tcp_clean_rtx_queue(struct so
 	}
 
 	if (acked&FLAG_ACKED) {
+		const struct tcp_congestion_ops *ca_ops
+			= inet_csk(sk)->icsk_ca_ops;
+
 		tcp_ack_update_rtt(sk, acked, seq_rtt);
 		tcp_ack_packets_out(sk);
-		if (rtt_sample && !(acked & FLAG_RETRANS_DATA_ACKED))
-			(*rtt_sample)(sk, tcp_usrtt(&tv));
 
-		if (icsk->icsk_ca_ops->pkts_acked)
-			icsk->icsk_ca_ops->pkts_acked(sk, pkts_acked);
+		if (ca_ops->pkts_acked)
+			ca_ops->pkts_acked(sk, pkts_acked, last_ackt);
 	}
 
 #if FASTRETRANS_DEBUG > 0
--- net-2.6.22.orig/net/ipv4/tcp_lp.c
+++ net-2.6.22/net/ipv4/tcp_lp.c
@@ -218,7 +218,7 @@ static u32 tcp_lp_owd_calculator(struct 
  *   3. calc smoothed OWD (SOWD).
  * Most ideas come from the original TCP-LP implementation.
  */
-static void tcp_lp_rtt_sample(struct sock *sk, u32 usrtt)
+static void tcp_lp_rtt_sample(struct sock *sk, u32 rtt)
 {
 	struct lp *lp = inet_csk_ca(sk);
 	s64 mowd = tcp_lp_owd_calculator(sk);
@@ -261,11 +261,13 @@ static void tcp_lp_rtt_sample(struct soc
  * newReno in increase case.
  * We work it out by following the idea from TCP-LP's paper directly
  */
-static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked)
+static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked, ktime_t last)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct lp *lp = inet_csk_ca(sk);
 
+	tcp_lp_rtt_sample(sk,  ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC);
+
 	/* calc inference */
 	if (tcp_time_stamp > tp->rx_opt.rcv_tsecr)
 		lp->inference = 3 * (tcp_time_stamp - tp->rx_opt.rcv_tsecr);
@@ -312,11 +314,11 @@ static void tcp_lp_pkts_acked(struct soc
 }
 
 static struct tcp_congestion_ops tcp_lp = {
+	.flags = TCP_CONG_RTT_STAMP,
 	.init = tcp_lp_init,
 	.ssthresh = tcp_reno_ssthresh,
 	.cong_avoid = tcp_lp_cong_avoid,
 	.min_cwnd = tcp_reno_min_cwnd,
-	.rtt_sample = tcp_lp_rtt_sample,
 	.pkts_acked = tcp_lp_pkts_acked,
 
 	.owner = THIS_MODULE,
--- net-2.6.22.orig/net/ipv4/tcp_output.c
+++ net-2.6.22/net/ipv4/tcp_output.c
@@ -409,7 +409,7 @@ static int tcp_transmit_skb(struct sock 
 	/* If congestion control is doing timestamping, we must
 	 * take such a timestamp before we potentially clone/copy.
 	 */
-	if (icsk->icsk_ca_ops->rtt_sample)
+	if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
 		__net_timestamp(skb);
 
 	if (likely(clone_it)) {
--- net-2.6.22.orig/net/ipv4/tcp_vegas.c
+++ net-2.6.22/net/ipv4/tcp_vegas.c
@@ -120,10 +120,13 @@ static void tcp_vegas_init(struct sock *
  *   o min-filter RTT samples from a much longer window (forever for now)
  *     to find the propagation delay (baseRTT)
  */
-static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt)
+static void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	struct vegas *vegas = inet_csk_ca(sk);
-	u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
+	u32 vrtt;
+
+	/* Never allow zero rtt or baseRTT */
+	vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1;
 
 	/* Filter to find propagation delay: */
 	if (vrtt < vegas->baseRTT)
@@ -353,11 +356,12 @@ static void tcp_vegas_get_info(struct so
 }
 
 static struct tcp_congestion_ops tcp_vegas = {
+	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_vegas_init,
 	.ssthresh	= tcp_reno_ssthresh,
 	.cong_avoid	= tcp_vegas_cong_avoid,
 	.min_cwnd	= tcp_reno_min_cwnd,
-	.rtt_sample	= tcp_vegas_rtt_calc,
+	.pkts_acked	= tcp_vegas_pkts_acked,
 	.set_state	= tcp_vegas_state,
 	.cwnd_event	= tcp_vegas_cwnd_event,
 	.get_info	= tcp_vegas_get_info,
--- net-2.6.22.orig/net/ipv4/tcp_veno.c
+++ net-2.6.22/net/ipv4/tcp_veno.c
@@ -69,10 +69,13 @@ static void tcp_veno_init(struct sock *s
 }
 
 /* Do rtt sampling needed for Veno. */
-static void tcp_veno_rtt_calc(struct sock *sk, u32 usrtt)
+static void tcp_veno_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	struct veno *veno = inet_csk_ca(sk);
-	u32 vrtt = usrtt + 1;	/* Never allow zero rtt or basertt */
+	u32 vrtt;
+
+	/* Never allow zero rtt or baseRTT */
+	vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1;
 
 	/* Filter to find propagation delay: */
 	if (vrtt < veno->basertt)
@@ -199,10 +202,11 @@ static u32 tcp_veno_ssthresh(struct sock
 }
 
 static struct tcp_congestion_ops tcp_veno = {
+	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_veno_init,
 	.ssthresh	= tcp_veno_ssthresh,
 	.cong_avoid	= tcp_veno_cong_avoid,
-	.rtt_sample	= tcp_veno_rtt_calc,
+	.pkts_acked	= tcp_veno_pkts_acked,
 	.set_state	= tcp_veno_state,
 	.cwnd_event	= tcp_veno_cwnd_event,
 
--- net-2.6.22.orig/net/ipv4/tcp_westwood.c
+++ net-2.6.22/net/ipv4/tcp_westwood.c
@@ -100,7 +100,7 @@ static void westwood_filter(struct westw
  * Called after processing group of packets.
  * but all westwood needs is the last sample of srtt.
  */
-static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt)
+static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	struct westwood *w = inet_csk_ca(sk);
 	if (cnt > 0)
--- net-2.6.22.orig/net/ipv4/tcp_yeah.c
+++ net-2.6.22/net/ipv4/tcp_yeah.c
@@ -64,13 +64,15 @@ static void tcp_yeah_init(struct sock *s
 }
 
 
-static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked)
+static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked, ktime_t last)
 {
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	struct yeah *yeah = inet_csk_ca(sk);
 
 	if (icsk->icsk_ca_state == TCP_CA_Open)
 		yeah->pkts_acked = pkts_acked;
+
+	tcp_vegas_pkts_acked(sk, pkts_acked, last);
 }
 
 static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack,
@@ -237,11 +239,11 @@ static u32 tcp_yeah_ssthresh(struct sock
 }
 
 static struct tcp_congestion_ops tcp_yeah = {
+	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_yeah_init,
 	.ssthresh	= tcp_yeah_ssthresh,
 	.cong_avoid	= tcp_yeah_cong_avoid,
 	.min_cwnd	= tcp_reno_min_cwnd,
-	.rtt_sample	= tcp_vegas_rtt_calc,
 	.set_state	= tcp_vegas_state,
 	.cwnd_event	= tcp_vegas_cwnd_event,
 	.get_info	= tcp_vegas_get_info,
--- net-2.6.22.orig/net/ipv4/tcp_yeah.h
+++ net-2.6.22/net/ipv4/tcp_yeah.h
@@ -81,10 +81,13 @@ static void tcp_vegas_state(struct sock 
  *   o min-filter RTT samples from a much longer window (forever for now)
  *     to find the propagation delay (baseRTT)
  */
-static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt)
+static void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	struct vegas *vegas = inet_csk_ca(sk);
-	u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
+	u32 vrtt;
+
+	/* Never allow zero rtt or baseRTT */
+	vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1;
 
 	/* Filter to find propagation delay: */
 	if (vrtt < vegas->baseRTT)

-- 

[PATCH 4/4] TCP YEAH: use vegas dont copy it

[Stephen Hemminger]

[-- Attachment #1: tcp-yeah-clean.patch --]
[-- Type: text/plain, Size: 13729 bytes --]

Rather than using a copy of vegas code, the YEAH code should just
have it exported so there is common code.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>


---
 net/ipv4/tcp_vegas.c |   31 +++++-------
 net/ipv4/tcp_vegas.h |   24 +++++++++
 net/ipv4/tcp_yeah.c  |   53 +++++++++-----------
 net/ipv4/tcp_yeah.h  |  131 ---------------------------------------------------
 4 files changed, 61 insertions(+), 178 deletions(-)

--- net-2.6.22.orig/net/ipv4/tcp_vegas.c
+++ net-2.6.22/net/ipv4/tcp_vegas.c
@@ -38,6 +38,8 @@
 
 #include <net/tcp.h>
 
+#include "tcp_vegas.h"
+
 /* Default values of the Vegas variables, in fixed-point representation
  * with V_PARAM_SHIFT bits to the right of the binary point.
  */
@@ -54,17 +56,6 @@ module_param(gamma, int, 0644);
 MODULE_PARM_DESC(gamma, "limit on increase (scale by 2)");
 
 
-/* Vegas variables */
-struct vegas {
-	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	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) */
-};
-
 /* There are several situations when we must "re-start" Vegas:
  *
  *  o when a connection is established
@@ -81,7 +72,7 @@ struct vegas {
  * Instead we must wait until the completion of an RTT during
  * which we actually receive ACKs.
  */
-static inline void vegas_enable(struct sock *sk)
+static void vegas_enable(struct sock *sk)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
 	struct vegas *vegas = inet_csk_ca(sk);
@@ -104,13 +95,14 @@ static inline void vegas_disable(struct 
 	vegas->doing_vegas_now = 0;
 }
 
-static void tcp_vegas_init(struct sock *sk)
+void tcp_vegas_init(struct sock *sk)
 {
 	struct vegas *vegas = inet_csk_ca(sk);
 
 	vegas->baseRTT = 0x7fffffff;
 	vegas_enable(sk);
 }
+EXPORT_SYMBOL_GPL(tcp_vegas_init);
 
 /* Do RTT sampling needed for Vegas.
  * Basically we:
@@ -120,7 +112,7 @@ static void tcp_vegas_init(struct sock *
  *   o min-filter RTT samples from a much longer window (forever for now)
  *     to find the propagation delay (baseRTT)
  */
-static void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
+void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
 {
 	struct vegas *vegas = inet_csk_ca(sk);
 	u32 vrtt;
@@ -138,8 +130,9 @@ static void tcp_vegas_pkts_acked(struct 
 	vegas->minRTT = min(vegas->minRTT, vrtt);
 	vegas->cntRTT++;
 }
+EXPORT_SYMBOL_GPL(tcp_vegas_pkts_acked);
 
-static void tcp_vegas_state(struct sock *sk, u8 ca_state)
+void tcp_vegas_state(struct sock *sk, u8 ca_state)
 {
 
 	if (ca_state == TCP_CA_Open)
@@ -147,6 +140,7 @@ static void tcp_vegas_state(struct sock 
 	else
 		vegas_disable(sk);
 }
+EXPORT_SYMBOL_GPL(tcp_vegas_state);
 
 /*
  * If the connection is idle and we are restarting,
@@ -157,12 +151,13 @@ static void tcp_vegas_state(struct sock 
  * packets, _then_ we can make Vegas calculations
  * again.
  */
-static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
+void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
 {
 	if (event == CA_EVENT_CWND_RESTART ||
 	    event == CA_EVENT_TX_START)
 		tcp_vegas_init(sk);
 }
+EXPORT_SYMBOL_GPL(tcp_vegas_cwnd_event);
 
 static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack,
 				 u32 seq_rtt, u32 in_flight, int flag)
@@ -339,8 +334,7 @@ static void tcp_vegas_cong_avoid(struct 
 }
 
 /* Extract info for Tcp socket info provided via netlink. */
-static void tcp_vegas_get_info(struct sock *sk, u32 ext,
-			       struct sk_buff *skb)
+void tcp_vegas_get_info(struct sock *sk, u32 ext, struct sk_buff *skb)
 {
 	const struct vegas *ca = inet_csk_ca(sk);
 	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
@@ -354,6 +348,7 @@ static void tcp_vegas_get_info(struct so
 		nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
 	}
 }
+EXPORT_SYMBOL_GPL(tcp_vegas_get_info);
 
 static struct tcp_congestion_ops tcp_vegas = {
 	.flags		= TCP_CONG_RTT_STAMP,
--- /dev/null
+++ net-2.6.22/net/ipv4/tcp_vegas.h
@@ -0,0 +1,24 @@
+/*
+ * TCP Vegas congestion control interface
+ */
+#ifndef __TCP_VEGAS_H
+#define __TCP_VEGAS_H 1
+
+/* Vegas variables */
+struct vegas {
+	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	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) */
+};
+
+extern void tcp_vegas_init(struct sock *sk);
+extern void tcp_vegas_state(struct sock *sk, u8 ca_state);
+extern void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last);
+extern void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event);
+extern void tcp_vegas_get_info(struct sock *sk, u32 ext, struct sk_buff *skb);
+
+#endif	/* __TCP_VEGAS_H */
--- net-2.6.22.orig/net/ipv4/tcp_yeah.c
+++ net-2.6.22/net/ipv4/tcp_yeah.c
@@ -6,13 +6,14 @@
  *    http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
  *
  */
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/inet_diag.h>
 
-#include "tcp_yeah.h"
+#include <net/tcp.h>
 
-/* 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
+#include "tcp_vegas.h"
 
 #define TCP_YEAH_ALPHA       80 //lin number of packets queued at the bottleneck
 #define TCP_YEAH_GAMMA        1 //lin fraction of queue to be removed per rtt
@@ -26,14 +27,7 @@
 
 /* YeAH variables */
 struct yeah {
-	/* Vegas */
-	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	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) */
+	struct vegas vegas;	/* must be first */
 
 	/* YeAH */
 	u32 lastQ;
@@ -84,9 +78,10 @@ static void tcp_yeah_cong_avoid(struct s
 	if (!tcp_is_cwnd_limited(sk, in_flight))
 		return;
 
-	if (tp->snd_cwnd <= tp->snd_ssthresh) {
+	if (tp->snd_cwnd <= tp->snd_ssthresh)
 		tcp_slow_start(tp);
-	} else if (!yeah->doing_reno_now) {
+
+	else if (!yeah->doing_reno_now) {
 		/* Scalable */
 
 		tp->snd_cwnd_cnt+=yeah->pkts_acked;
@@ -110,19 +105,19 @@ static void tcp_yeah_cong_avoid(struct s
 		}
 	}
 
-	/* The key players are v_beg_snd_una and v_beg_snd_nxt.
+	/* The key players are v_vegas.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
+	 * we will calculate that (v_beg_snd_nxt - v_vegas.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)
+	 *     (v_beg_snd_nxt - v_vegas.beg_snd_una) / (rtt duration)
 	 *
-	 * Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
+	 * Unfortunately, v_vegas.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 yeahly with the boundaries of RTTs.
 	 *
@@ -132,7 +127,7 @@ static void tcp_yeah_cong_avoid(struct s
 	 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
 	 */
 
-	if (after(ack, yeah->beg_snd_nxt)) {
+	if (after(ack, yeah->vegas.beg_snd_nxt)) {
 
 		/* We do the Vegas calculations only if we got enough RTT
 		 * samples that we can be reasonably sure that we got
@@ -143,7 +138,7 @@ static void tcp_yeah_cong_avoid(struct s
 		 * If  we have 3 samples, we should be OK.
 		 */
 
-		if (yeah->cntRTT > 2) {
+		if (yeah->vegas.cntRTT > 2) {
 			u32 rtt, queue;
 			u64 bw;
 
@@ -158,18 +153,18 @@ static void tcp_yeah_cong_avoid(struct s
 			 * of delayed ACKs, at the cost of noticing congestion
 			 * a bit later.
 			 */
-			rtt = yeah->minRTT;
+			rtt = yeah->vegas.minRTT;
 
 			/* Compute excess number of packets above bandwidth
 			 * Avoid doing full 64 bit divide.
 			 */
 			bw = tp->snd_cwnd;
-			bw *= rtt - yeah->baseRTT;
+			bw *= rtt - yeah->vegas.baseRTT;
 			do_div(bw, rtt);
 			queue = bw;
 
 			if (queue > TCP_YEAH_ALPHA ||
-			    rtt - yeah->baseRTT > (yeah->baseRTT / TCP_YEAH_PHY)) {
+			    rtt - yeah->vegas.baseRTT > (yeah->vegas.baseRTT / TCP_YEAH_PHY)) {
 				if (queue > TCP_YEAH_ALPHA
 				    && tp->snd_cwnd > yeah->reno_count) {
 					u32 reduction = min(queue / TCP_YEAH_GAMMA ,
@@ -208,13 +203,13 @@ static void tcp_yeah_cong_avoid(struct s
 		/* Save the extent of the current window so we can use this
 		 * at the end of the next RTT.
 		 */
-		yeah->beg_snd_una  = yeah->beg_snd_nxt;
-		yeah->beg_snd_nxt  = tp->snd_nxt;
-		yeah->beg_snd_cwnd = tp->snd_cwnd;
+		yeah->vegas.beg_snd_una  = yeah->vegas.beg_snd_nxt;
+		yeah->vegas.beg_snd_nxt  = tp->snd_nxt;
+		yeah->vegas.beg_snd_cwnd = tp->snd_cwnd;
 
 		/* Wipe the slate clean for the next RTT. */
-		yeah->cntRTT = 0;
-		yeah->minRTT = 0x7fffffff;
+		yeah->vegas.cntRTT = 0;
+		yeah->vegas.minRTT = 0x7fffffff;
 	}
 }
 
--- net-2.6.22.orig/net/ipv4/tcp_yeah.h
+++ net-2.6.22/net/ipv4/tcp_yeah.h
@@ -5,134 +5,3 @@
 #include <asm/div64.h>
 
 #include <net/tcp.h>
-
-/* Vegas variables */
-struct vegas {
-	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	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) */
-};
-
-/* 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.
- */
-static inline void vegas_enable(struct sock *sk)
-{
-	const struct tcp_sock *tp = tcp_sk(sk);
-	struct vegas *vegas = inet_csk_ca(sk);
-
-	/* Begin taking Vegas samples next time we send something. */
-	vegas->doing_vegas_now = 1;
-
-	/* Set the beginning of the next send window. */
-	vegas->beg_snd_nxt = tp->snd_nxt;
-
-	vegas->cntRTT = 0;
-	vegas->minRTT = 0x7fffffff;
-}
-
-/* Stop taking Vegas samples for now. */
-static inline void vegas_disable(struct sock *sk)
-{
-	struct vegas *vegas = inet_csk_ca(sk);
-
-	vegas->doing_vegas_now = 0;
-}
-
-static void tcp_vegas_init(struct sock *sk)
-{
-	struct vegas *vegas = inet_csk_ca(sk);
-
-	vegas->baseRTT = 0x7fffffff;
-	vegas_enable(sk);
-}
-
-static void tcp_vegas_state(struct sock *sk, u8 ca_state)
-{
-
-	if (ca_state == TCP_CA_Open)
-		vegas_enable(sk);
-	else
-		vegas_disable(sk);
-}
-
-/* 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 void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
-{
-	struct vegas *vegas = inet_csk_ca(sk);
-	u32 vrtt;
-
-	/* Never allow zero rtt or baseRTT */
-	vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1;
-
-	/* Filter to find propagation delay: */
-	if (vrtt < vegas->baseRTT)
-		vegas->baseRTT = vrtt;
-
-	/* Find the min RTT during the last RTT to find
-	 * the current prop. delay + queuing delay:
-	 */
-	vegas->minRTT = min(vegas->minRTT, vrtt);
-	vegas->cntRTT++;
-}
-
-/*
- * 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.
- */
-static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
-{
-	if (event == CA_EVENT_CWND_RESTART ||
-	    event == CA_EVENT_TX_START)
-		tcp_vegas_init(sk);
-}
-
-/* Extract info for Tcp socket info provided via netlink. */
-static void tcp_vegas_get_info(struct sock *sk, u32 ext,
-			       struct sk_buff *skb)
-{
-	const struct vegas *ca = inet_csk_ca(sk);
-	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
-		struct tcpvegas_info *info;
-
-		info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO,
-					  sizeof(*info)));
-
-		info->tcpv_enabled = ca->doing_vegas_now;
-		info->tcpv_rttcnt = ca->cntRTT;
-		info->tcpv_rtt = ca->baseRTT;
-		info->tcpv_minrtt = ca->minRTT;
-	rtattr_failure:	;
-	}
-}
-
-

-- 

Re: [PATCH 1/4] tcp: congestion control initialization

[David Miller]

From: Stephen Hemminger <shemminger@linux-founsation.org>
Date: Mon, 23 Apr 2007 20:31:18 -0700

> Change to defer congestion control initialization.
> 
> If setsockopt() was used to change TCP_CONGESTION before
> connection is established, then protocols that use sequence numbers
> to keep track of one RTT interval (vegas, illinois, ...) get confused.
> 
> Change the init hook to be called after handshake.
> 
> Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

Applied.

I'll look this over for backporting, thanks Stephen.

Re: [PATCH 2/4] TCP Illinois update

[David Miller]

From: Stephen Hemminger <shemminger@linux-founsation.org>
Date: Mon, 23 Apr 2007 20:31:19 -0700

> This version more closely matches the paper, and fixes several
> math errors. The biggest difference is that it updates alpha/beta
> once per RTT
> 
> Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

Applied, thanks.

Re: [PATCH 3/4] tcp: congestion control API update

[David Miller]

From: Stephen Hemminger <shemminger@linux-founsation.org>
Date: Mon, 23 Apr 2007 20:31:20 -0700

> Do some simple changes to make congestion control API faster/cleaner.
> * use ktime_t rather than timeval
> * merge rtt sampling into existing ack callback
>   this means one indirect call versus two per ack.
> * use flags bits to store options/settings
> 
> Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

Nice work Stephen.

Applied, thanks a lot.

Re: [PATCH 4/4] TCP YEAH: use vegas dont copy it

[David Miller]

From: Stephen Hemminger <shemminger@linux-founsation.org>
Date: Mon, 23 Apr 2007 20:31:21 -0700

> Rather than using a copy of vegas code, the YEAH code should just
> have it exported so there is common code.
> 
> Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

Excellent cleanup.

Applied, thanks Stephen.