[net-next-2.6.git] / net / ipv4 / tcp_vegas.c

/*
 * TCP Vegas congestion control
 *
 * 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.
 */

#include <linux/config.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/inet_diag.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
static int alpha = 1<<V_PARAM_SHIFT;
static int beta  = 3<<V_PARAM_SHIFT;
static int gamma = 1<<V_PARAM_SHIFT;

module_param(alpha, int, 0644);
MODULE_PARM_DESC(alpha, "lower bound of packets in network (scale by 2)");
module_param(beta, int, 0644);
MODULE_PARM_DESC(beta, "upper bound of packets in network (scale by 2)");
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
 *  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);
}

/* 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_rtt_calc(struct sock *sk, u32 usrtt)
{
	struct vegas *vegas = inet_csk_ca(sk);
	u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */

	/* 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++;
}

static void tcp_vegas_state(struct sock *sk, u8 ca_state)
{

	if (ca_state == TCP_CA_Open)
		vegas_enable(sk);
	else
		vegas_disable(sk);
}

/*
 * 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);
}

static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack,
				 u32 seq_rtt, u32 in_flight, int flag)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct vegas *vegas = inet_csk_ca(sk);

	if (!vegas->doing_vegas_now)
		return tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);

	/* 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, 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 = (vegas->beg_snd_nxt - vegas->beg_snd_una) /
			tp->mss_cache;
		old_snd_cwnd = vegas->beg_snd_cwnd;

		/* Save the extent of the current window so we can use this
		 * at the end of the next RTT.
		 */
		vegas->beg_snd_una  = vegas->beg_snd_nxt;
		vegas->beg_snd_nxt  = tp->snd_nxt;
		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.
		 */
		tcp_vegas_rtt_calc(sk, seq_rtt * 1000);

		/* 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 (vegas->cntRTT <= 2) {
			/* We don't have enough RTT samples to do the Vegas
			 * calculation, so we'll behave like Reno.
			 */
			tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
		} 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 = 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 * 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 > 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);

				}
				tcp_slow_start(tp);
			} else {
				/* Congestion avoidance. */
				u32 next_snd_cwnd;

				/* Figure out where we would like cwnd
				 * to be.
				 */
				if (diff > beta) {
					/* The old window was too fast, so
					 * we slow down.
					 */
					next_snd_cwnd = old_snd_cwnd - 1;
				} else if (diff < 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--;
			}

			if (tp->snd_cwnd < 2)
				tp->snd_cwnd = 2;
			else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
				tp->snd_cwnd = tp->snd_cwnd_clamp;
		}
	}

	/* Wipe the slate clean for the next RTT. */
	vegas->cntRTT = 0;
	vegas->minRTT = 0x7fffffff;
}

/* 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:	;
	}
}

static struct tcp_congestion_ops tcp_vegas = {
	.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,
	.set_state	= tcp_vegas_state,
	.cwnd_event	= tcp_vegas_cwnd_event,
	.get_info	= tcp_vegas_get_info,

	.owner		= THIS_MODULE,
	.name		= "vegas",
};

static int __init tcp_vegas_register(void)
{
	BUG_ON(sizeof(struct vegas) > ICSK_CA_PRIV_SIZE);
	tcp_register_congestion_control(&tcp_vegas);
	return 0;
}

static void __exit tcp_vegas_unregister(void)
{
	tcp_unregister_congestion_control(&tcp_vegas);
}

module_init(tcp_vegas_register);
module_exit(tcp_vegas_unregister);

MODULE_AUTHOR("Stephen Hemminger");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("TCP Vegas");
Commit	Line	Data
b87d8561 SH	1	/*
	2	* TCP Vegas congestion control
	3	*
	4	* This is based on the congestion detection/avoidance scheme described in
	5	* Lawrence S. Brakmo and Larry L. Peterson.
	6	* "TCP Vegas: End to end congestion avoidance on a global internet."
	7	* IEEE Journal on Selected Areas in Communication, 13(8):1465--1480,
	8	* October 1995. Available from:
	9	* ftp://ftp.cs.arizona.edu/xkernel/Papers/jsac.ps
	10	*
	11	* See http://www.cs.arizona.edu/xkernel/ for their implementation.
	12	* The main aspects that distinguish this implementation from the
	13	* Arizona Vegas implementation are:
	14	* o We do not change the loss detection or recovery mechanisms of
	15	* Linux in any way. Linux already recovers from losses quite well,
	16	* using fine-grained timers, NewReno, and FACK.
	17	* o To avoid the performance penalty imposed by increasing cwnd
	18	* only every-other RTT during slow start, we increase during
	19	* every RTT during slow start, just like Reno.
	20	* o Largely to allow continuous cwnd growth during slow start,
	21	* we use the rate at which ACKs come back as the "actual"
	22	* rate, rather than the rate at which data is sent.
	23	* o To speed convergence to the right rate, we set the cwnd
	24	* to achieve the right ("actual") rate when we exit slow start.
	25	* o To filter out the noise caused by delayed ACKs, we use the
	26	* minimum RTT sample observed during the last RTT to calculate
	27	* the actual rate.
	28	* o When the sender re-starts from idle, it waits until it has
	29	* received ACKs for an entire flight of new data before making
	30	* a cwnd adjustment decision. The original Vegas implementation
	31	* assumed senders never went idle.
	32	*/
	33
	34	#include <linux/config.h>
	35	#include <linux/mm.h>
	36	#include <linux/module.h>
	37	#include <linux/skbuff.h>
a8c2190e	38	#include <linux/inet_diag.h>
b87d8561 SH	39
	40	#include <net/tcp.h>
	41
	42	/* Default values of the Vegas variables, in fixed-point representation
	43	* with V_PARAM_SHIFT bits to the right of the binary point.
	44	*/
	45	#define V_PARAM_SHIFT 1
	46	static int alpha = 1<<V_PARAM_SHIFT;
	47	static int beta = 3<<V_PARAM_SHIFT;
	48	static int gamma = 1<<V_PARAM_SHIFT;
	49
	50	module_param(alpha, int, 0644);
	51	MODULE_PARM_DESC(alpha, "lower bound of packets in network (scale by 2)");
	52	module_param(beta, int, 0644);
	53	MODULE_PARM_DESC(beta, "upper bound of packets in network (scale by 2)");
	54	module_param(gamma, int, 0644);
	55	MODULE_PARM_DESC(gamma, "limit on increase (scale by 2)");
	56
	57
	58	/* Vegas variables */
	59	struct vegas {
	60	u32 beg_snd_nxt; /* right edge during last RTT */
	61	u32 beg_snd_una; /* left edge during last RTT */
	62	u32 beg_snd_cwnd; /* saves the size of the cwnd */
	63	u8 doing_vegas_now;/* if true, do vegas for this RTT */
	64	u16 cntRTT; /* # of RTTs measured within last RTT */
	65	u32 minRTT; /* min of RTTs measured within last RTT (in usec) */
	66	u32 baseRTT; /* the min of all Vegas RTT measurements seen (in usec) */
	67	};
	68
	69	/* There are several situations when we must "re-start" Vegas:
	70	*
	71	* o when a connection is established
	72	* o after an RTO
	73	* o after fast recovery
	74	* o when we send a packet and there is no outstanding
	75	* unacknowledged data (restarting an idle connection)
	76	*
	77	* In these circumstances we cannot do a Vegas calculation at the
	78	* end of the first RTT, because any calculation we do is using
	79	* stale info -- both the saved cwnd and congestion feedback are
	80	* stale.
	81	*
	82	* Instead we must wait until the completion of an RTT during
	83	* which we actually receive ACKs.
	84	*/
6687e988	85	static inline void vegas_enable(struct sock *sk)
b87d8561	86	{
6687e988 ACM	87	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	88	struct vegas *vegas = inet_csk_ca(sk);
b87d8561 SH	89
	90	/* Begin taking Vegas samples next time we send something. */
	91	vegas->doing_vegas_now = 1;
	92
	93	/* Set the beginning of the next send window. */
	94	vegas->beg_snd_nxt = tp->snd_nxt;
	95
	96	vegas->cntRTT = 0;
	97	vegas->minRTT = 0x7fffffff;
	98	}
	99
	100	/* Stop taking Vegas samples for now. */
6687e988	101	static inline void vegas_disable(struct sock *sk)
b87d8561	102	{
6687e988	103	struct vegas *vegas = inet_csk_ca(sk);
b87d8561 SH	104
	105	vegas->doing_vegas_now = 0;
	106	}
	107
6687e988	108	static void tcp_vegas_init(struct sock *sk)
b87d8561	109	{
6687e988	110	struct vegas *vegas = inet_csk_ca(sk);
b87d8561 SH	111
b87d8561 SH	112	vegas->baseRTT = 0x7fffffff;
6687e988	113	vegas_enable(sk);
b87d8561 SH	114	}
	115
	116	/* Do RTT sampling needed for Vegas.
	117	* Basically we:
	118	* o min-filter RTT samples from within an RTT to get the current
	119	* propagation delay + queuing delay (we are min-filtering to try to
	120	* avoid the effects of delayed ACKs)
	121	* o min-filter RTT samples from a much longer window (forever for now)
	122	* to find the propagation delay (baseRTT)
	123	*/
6687e988	124	static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt)
b87d8561	125	{
6687e988	126	struct vegas *vegas = inet_csk_ca(sk);
b87d8561 SH	127	u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
	128
	129	/* Filter to find propagation delay: */
	130	if (vrtt < vegas->baseRTT)
	131	vegas->baseRTT = vrtt;
	132
	133	/* Find the min RTT during the last RTT to find
	134	* the current prop. delay + queuing delay:
	135	*/
	136	vegas->minRTT = min(vegas->minRTT, vrtt);
	137	vegas->cntRTT++;
	138	}
	139
6687e988	140	static void tcp_vegas_state(struct sock *sk, u8 ca_state)
b87d8561 SH	141	{
	142
	143	if (ca_state == TCP_CA_Open)
6687e988	144	vegas_enable(sk);
b87d8561	145	else
6687e988	146	vegas_disable(sk);
b87d8561 SH	147	}
	148
	149	/*
	150	* If the connection is idle and we are restarting,
	151	* then we don't want to do any Vegas calculations
	152	* until we get fresh RTT samples. So when we
	153	* restart, we reset our Vegas state to a clean
	154	* slate. After we get acks for this flight of
	155	* packets, _then_ we can make Vegas calculations
	156	* again.
	157	*/
6687e988	158	static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
b87d8561 SH	159	{
	160	if (event == CA_EVENT_CWND_RESTART \|\|
	161	event == CA_EVENT_TX_START)
6687e988	162	tcp_vegas_init(sk);
b87d8561 SH	163	}
b87d8561 SH	164
6687e988	165	static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack,
b87d8561 SH	166	u32 seq_rtt, u32 in_flight, int flag)
b87d8561 SH	167	{
6687e988 ACM	168	struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	169	struct vegas *vegas = inet_csk_ca(sk);
b87d8561 SH	170
b87d8561 SH	171	if (!vegas->doing_vegas_now)
6687e988	172	return tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
b87d8561 SH	173
	174	/* The key players are v_beg_snd_una and v_beg_snd_nxt.
	175	*
	176	* These are so named because they represent the approximate values
	177	* of snd_una and snd_nxt at the beginning of the current RTT. More
	178	* precisely, they represent the amount of data sent during the RTT.
	179	* At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
	180	* we will calculate that (v_beg_snd_nxt - v_beg_snd_una) outstanding
	181	* bytes of data have been ACKed during the course of the RTT, giving
	182	* an "actual" rate of:
	183	*
	184	* (v_beg_snd_nxt - v_beg_snd_una) / (rtt duration)
	185	*
	186	* Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
	187	* because delayed ACKs can cover more than one segment, so they
	188	* don't line up nicely with the boundaries of RTTs.
	189	*
	190	* Another unfortunate fact of life is that delayed ACKs delay the
	191	* advance of the left edge of our send window, so that the number
	192	* of bytes we send in an RTT is often less than our cwnd will allow.
	193	* So we keep track of our cwnd separately, in v_beg_snd_cwnd.
	194	*/
	195
	196	if (after(ack, vegas->beg_snd_nxt)) {
	197	/* Do the Vegas once-per-RTT cwnd adjustment. */
	198	u32 old_wnd, old_snd_cwnd;
	199
	200
	201	/* Here old_wnd is essentially the window of data that was
	202	* sent during the previous RTT, and has all
	203	* been acknowledged in the course of the RTT that ended
	204	* with the ACK we just received. Likewise, old_snd_cwnd
	205	* is the cwnd during the previous RTT.
	206	*/
	207	old_wnd = (vegas->beg_snd_nxt - vegas->beg_snd_una) /
	208	tp->mss_cache;
	209	old_snd_cwnd = vegas->beg_snd_cwnd;
	210
	211	/* Save the extent of the current window so we can use this
	212	* at the end of the next RTT.
	213	*/
	214	vegas->beg_snd_una = vegas->beg_snd_nxt;
	215	vegas->beg_snd_nxt = tp->snd_nxt;
	216	vegas->beg_snd_cwnd = tp->snd_cwnd;
	217
	218	/* Take into account the current RTT sample too, to
	219	* decrease the impact of delayed acks. This double counts
	220	* this sample since we count it for the next window as well,
	221	* but that's not too awful, since we're taking the min,
	222	* rather than averaging.
	223	*/
6687e988	224	tcp_vegas_rtt_calc(sk, seq_rtt * 1000);
b87d8561 SH	225
	226	/* We do the Vegas calculations only if we got enough RTT
	227	* samples that we can be reasonably sure that we got
	228	* at least one RTT sample that wasn't from a delayed ACK.
	229	* If we only had 2 samples total,
	230	* then that means we're getting only 1 ACK per RTT, which
	231	* means they're almost certainly delayed ACKs.
	232	* If we have 3 samples, we should be OK.
	233	*/
	234
	235	if (vegas->cntRTT <= 2) {
	236	/* We don't have enough RTT samples to do the Vegas
	237	* calculation, so we'll behave like Reno.
	238	*/
c050970a	239	tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
b87d8561 SH	240	} else {
	241	u32 rtt, target_cwnd, diff;
	242
	243	/* We have enough RTT samples, so, using the Vegas
	244	* algorithm, we determine if we should increase or
	245	* decrease cwnd, and by how much.
	246	*/
	247
	248	/* Pluck out the RTT we are using for the Vegas
	249	* calculations. This is the min RTT seen during the
	250	* last RTT. Taking the min filters out the effects
	251	* of delayed ACKs, at the cost of noticing congestion
	252	* a bit later.
	253	*/
	254	rtt = vegas->minRTT;
	255
	256	/* Calculate the cwnd we should have, if we weren't
	257	* going too fast.
	258	*
	259	* This is:
	260	* (actual rate in segments) * baseRTT
	261	* We keep it as a fixed point number with
	262	* V_PARAM_SHIFT bits to the right of the binary point.
	263	*/
	264	target_cwnd = ((old_wnd * vegas->baseRTT)
	265	<< V_PARAM_SHIFT) / rtt;
	266
	267	/* Calculate the difference between the window we had,
	268	* and the window we would like to have. This quantity
	269	* is the "Diff" from the Arizona Vegas papers.
	270	*
	271	* Again, this is a fixed point number with
	272	* V_PARAM_SHIFT bits to the right of the binary
	273	* point.
	274	*/
	275	diff = (old_wnd << V_PARAM_SHIFT) - target_cwnd;
	276
7faffa1c	277	if (tp->snd_cwnd <= tp->snd_ssthresh) {
b87d8561 SH	278	/* Slow start. */
	279	if (diff > gamma) {
	280	/* Going too fast. Time to slow down
	281	* and switch to congestion avoidance.
	282	*/
	283	tp->snd_ssthresh = 2;
	284
	285	/* Set cwnd to match the actual rate
	286	* exactly:
	287	* cwnd = (actual rate) * baseRTT
	288	* Then we add 1 because the integer
	289	* truncation robs us of full link
	290	* utilization.
	291	*/
	292	tp->snd_cwnd = min(tp->snd_cwnd,
	293	(target_cwnd >>
	294	V_PARAM_SHIFT)+1);
	295
	296	}
7faffa1c	297	tcp_slow_start(tp);
b87d8561 SH	298	} else {
	299	/* Congestion avoidance. */
	300	u32 next_snd_cwnd;
	301
	302	/* Figure out where we would like cwnd
	303	* to be.
	304	*/
	305	if (diff > beta) {
	306	/* The old window was too fast, so
	307	* we slow down.
	308	*/
	309	next_snd_cwnd = old_snd_cwnd - 1;
	310	} else if (diff < alpha) {
	311	/* We don't have enough extra packets
	312	* in the network, so speed up.
	313	*/
	314	next_snd_cwnd = old_snd_cwnd + 1;
	315	} else {
	316	/* Sending just as fast as we
	317	* should be.
	318	*/
	319	next_snd_cwnd = old_snd_cwnd;
	320	}
	321
	322	/* Adjust cwnd upward or downward, toward the
	323	* desired value.
	324	*/
	325	if (next_snd_cwnd > tp->snd_cwnd)
	326	tp->snd_cwnd++;
	327	else if (next_snd_cwnd < tp->snd_cwnd)
	328	tp->snd_cwnd--;
	329	}
b87d8561	330
7faffa1c SH	331	if (tp->snd_cwnd < 2)
	332	tp->snd_cwnd = 2;
	333	else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
	334	tp->snd_cwnd = tp->snd_cwnd_clamp;
	335	}
b87d8561 SH	336	}
b87d8561 SH	337
7faffa1c SH	338	/* Wipe the slate clean for the next RTT. */
	339	vegas->cntRTT = 0;
	340	vegas->minRTT = 0x7fffffff;
b87d8561 SH	341	}
	342
	343	/* Extract info for Tcp socket info provided via netlink. */
6687e988	344	static void tcp_vegas_get_info(struct sock *sk, u32 ext,
b87d8561 SH	345	struct sk_buff *skb)
b87d8561 SH	346	{
6687e988	347	const struct vegas *ca = inet_csk_ca(sk);
73c1f4a0	348	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
b87d8561 SH	349	struct tcpvegas_info *info;
b87d8561 SH	350
73c1f4a0	351	info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO,
b87d8561 SH	352	sizeof(*info)));
	353
	354	info->tcpv_enabled = ca->doing_vegas_now;
	355	info->tcpv_rttcnt = ca->cntRTT;
	356	info->tcpv_rtt = ca->baseRTT;
	357	info->tcpv_minrtt = ca->minRTT;
	358	rtattr_failure: ;
	359	}
	360	}
	361
	362	static struct tcp_congestion_ops tcp_vegas = {
	363	.init = tcp_vegas_init,
	364	.ssthresh = tcp_reno_ssthresh,
	365	.cong_avoid = tcp_vegas_cong_avoid,
	366	.min_cwnd = tcp_reno_min_cwnd,
	367	.rtt_sample = tcp_vegas_rtt_calc,
	368	.set_state = tcp_vegas_state,
	369	.cwnd_event = tcp_vegas_cwnd_event,
	370	.get_info = tcp_vegas_get_info,
	371
	372	.owner = THIS_MODULE,
	373	.name = "vegas",
	374	};
	375
	376	static int __init tcp_vegas_register(void)
	377	{
6687e988	378	BUG_ON(sizeof(struct vegas) > ICSK_CA_PRIV_SIZE);
b87d8561 SH	379	tcp_register_congestion_control(&tcp_vegas);
	380	return 0;
	381	}
	382
	383	static void __exit tcp_vegas_unregister(void)
	384	{
	385	tcp_unregister_congestion_control(&tcp_vegas);
	386	}
	387
	388	module_init(tcp_vegas_register);
	389	module_exit(tcp_vegas_unregister);
	390
	391	MODULE_AUTHOR("Stephen Hemminger");
	392	MODULE_LICENSE("GPL");
	393	MODULE_DESCRIPTION("TCP Vegas");