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

/*
 * TCP Westwood+: end-to-end bandwidth estimation for TCP
 *
 *      Angelo Dell'Aera: author of the first version of TCP Westwood+ in Linux 2.4
 *
 * Support at http://c3lab.poliba.it/index.php/Westwood
 * Main references in literature:
 *
 * - Mascolo S, Casetti, M. Gerla et al.
 *   "TCP Westwood: bandwidth estimation for TCP" Proc. ACM Mobicom 2001
 *
 * - A. Grieco, s. Mascolo
 *   "Performance evaluation of New Reno, Vegas, Westwood+ TCP" ACM Computer
 *     Comm. Review, 2004
 *
 * - A. Dell'Aera, L. Grieco, S. Mascolo.
 *   "Linux 2.4 Implementation of Westwood+ TCP with Rate-Halving :
 *    A Performance Evaluation Over the Internet" (ICC 2004), Paris, June 2004
 *
 * Westwood+ employs end-to-end bandwidth measurement to set cwnd and
 * ssthresh after packet loss. The probing phase is as the original Reno.
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/inet_diag.h>
#include <net/tcp.h>

/* TCP Westwood structure */
struct westwood {
	u32    bw_ns_est;        /* first bandwidth estimation..not too smoothed 8) */
	u32    bw_est;           /* bandwidth estimate */
	u32    rtt_win_sx;       /* here starts a new evaluation... */
	u32    bk;
	u32    snd_una;          /* used for evaluating the number of acked bytes */
	u32    cumul_ack;
	u32    accounted;
	u32    rtt;
	u32    rtt_min;          /* minimum observed RTT */
	u8     first_ack;        /* flag which infers that this is the first ack */
	u8     reset_rtt_min;    /* Reset RTT min to next RTT sample*/
};


/* TCP Westwood functions and constants */
#define TCP_WESTWOOD_RTT_MIN   (HZ/20)	/* 50ms */
#define TCP_WESTWOOD_INIT_RTT  (20*HZ)	/* maybe too conservative?! */

/*
 * @tcp_westwood_create
 * This function initializes fields used in TCP Westwood+,
 * it is called after the initial SYN, so the sequence numbers
 * are correct but new passive connections we have no
 * information about RTTmin at this time so we simply set it to
 * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
 * since in this way we're sure it will be updated in a consistent
 * way as soon as possible. It will reasonably happen within the first
 * RTT period of the connection lifetime.
 */
static void tcp_westwood_init(struct sock *sk)
{
	struct westwood *w = inet_csk_ca(sk);

	w->bk = 0;
	w->bw_ns_est = 0;
	w->bw_est = 0;
	w->accounted = 0;
	w->cumul_ack = 0;
	w->reset_rtt_min = 1;
	w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
	w->rtt_win_sx = tcp_time_stamp;
	w->snd_una = tcp_sk(sk)->snd_una;
	w->first_ack = 1;
}

/*
 * @westwood_do_filter
 * Low-pass filter. Implemented using constant coefficients.
 */
static inline u32 westwood_do_filter(u32 a, u32 b)
{
	return (((7 * a) + b) >> 3);
}

static void westwood_filter(struct westwood *w, u32 delta)
{
	/* If the filter is empty fill it with the first sample of bandwidth  */
	if (w->bw_ns_est == 0 && w->bw_est == 0) {
		w->bw_ns_est = w->bk / delta;
		w->bw_est = w->bw_ns_est;
	} else {
		w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
		w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
	}
}

/*
 * @westwood_pkts_acked
 * 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)
{
	struct westwood *w = inet_csk_ca(sk);
	if (cnt > 0)
		w->rtt = tcp_sk(sk)->srtt >> 3;
}

/*
 * @westwood_update_window
 * It updates RTT evaluation window if it is the right moment to do
 * it. If so it calls filter for evaluating bandwidth.
 */
static void westwood_update_window(struct sock *sk)
{
	struct westwood *w = inet_csk_ca(sk);
	s32 delta = tcp_time_stamp - w->rtt_win_sx;

	/* Initialize w->snd_una with the first acked sequence number in order
	 * to fix mismatch between tp->snd_una and w->snd_una for the first
	 * bandwidth sample
	 */
	if (w->first_ack) {
		w->snd_una = tcp_sk(sk)->snd_una;
		w->first_ack = 0;
	}

	/*
	 * See if a RTT-window has passed.
	 * Be careful since if RTT is less than
	 * 50ms we don't filter but we continue 'building the sample'.
	 * This minimum limit was chosen since an estimation on small
	 * time intervals is better to avoid...
	 * Obviously on a LAN we reasonably will always have
	 * right_bound = left_bound + WESTWOOD_RTT_MIN
	 */
	if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
		westwood_filter(w, delta);

		w->bk = 0;
		w->rtt_win_sx = tcp_time_stamp;
	}
}

static inline void update_rtt_min(struct westwood *w)
{
	if (w->reset_rtt_min) {
		w->rtt_min = w->rtt;
		w->reset_rtt_min = 0;
	} else
		w->rtt_min = min(w->rtt, w->rtt_min);
}


/*
 * @westwood_fast_bw
 * It is called when we are in fast path. In particular it is called when
 * header prediction is successful. In such case in fact update is
 * straight forward and doesn't need any particular care.
 */
static inline void westwood_fast_bw(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);

	westwood_update_window(sk);

	w->bk += tp->snd_una - w->snd_una;
	w->snd_una = tp->snd_una;
	update_rtt_min(w);
}

/*
 * @westwood_acked_count
 * This function evaluates cumul_ack for evaluating bk in case of
 * delayed or partial acks.
 */
static inline u32 westwood_acked_count(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);

	w->cumul_ack = tp->snd_una - w->snd_una;

	/* If cumul_ack is 0 this is a dupack since it's not moving
	 * tp->snd_una.
	 */
	if (!w->cumul_ack) {
		w->accounted += tp->mss_cache;
		w->cumul_ack = tp->mss_cache;
	}

	if (w->cumul_ack > tp->mss_cache) {
		/* Partial or delayed ack */
		if (w->accounted >= w->cumul_ack) {
			w->accounted -= w->cumul_ack;
			w->cumul_ack = tp->mss_cache;
		} else {
			w->cumul_ack -= w->accounted;
			w->accounted = 0;
		}
	}

	w->snd_una = tp->snd_una;

	return w->cumul_ack;
}


/*
 * TCP Westwood
 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
 * so avoids ever returning 0.
 */
static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct westwood *w = inet_csk_ca(sk);
	return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
}

static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);

	switch (event) {
	case CA_EVENT_FAST_ACK:
		westwood_fast_bw(sk);
		break;

	case CA_EVENT_COMPLETE_CWR:
		tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
		break;

	case CA_EVENT_FRTO:
		tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
		/* Update RTT_min when next ack arrives */
		w->reset_rtt_min = 1;
		break;

	case CA_EVENT_SLOW_ACK:
		westwood_update_window(sk);
		w->bk += westwood_acked_count(sk);
		update_rtt_min(w);
		break;

	default:
		/* don't care */
		break;
	}
}


/* Extract info for Tcp socket info provided via netlink. */
static void tcp_westwood_info(struct sock *sk, u32 ext,
			      struct sk_buff *skb)
{
	const struct westwood *ca = inet_csk_ca(sk);
	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
		struct rtattr *rta;
		struct tcpvegas_info *info;

		rta = __RTA_PUT(skb, INET_DIAG_VEGASINFO, sizeof(*info));
		info = RTA_DATA(rta);
		info->tcpv_enabled = 1;
		info->tcpv_rttcnt = 0;
		info->tcpv_rtt = jiffies_to_usecs(ca->rtt);
		info->tcpv_minrtt = jiffies_to_usecs(ca->rtt_min);
	rtattr_failure:	;
	}
}


static struct tcp_congestion_ops tcp_westwood = {
	.init		= tcp_westwood_init,
	.ssthresh	= tcp_reno_ssthresh,
	.cong_avoid	= tcp_reno_cong_avoid,
	.min_cwnd	= tcp_westwood_bw_rttmin,
	.cwnd_event	= tcp_westwood_event,
	.get_info	= tcp_westwood_info,
	.pkts_acked	= tcp_westwood_pkts_acked,

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

static int __init tcp_westwood_register(void)
{
	BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
	return tcp_register_congestion_control(&tcp_westwood);
}

static void __exit tcp_westwood_unregister(void)
{
	tcp_unregister_congestion_control(&tcp_westwood);
}

module_init(tcp_westwood_register);
module_exit(tcp_westwood_unregister);

MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("TCP Westwood+");
Commit	Line	Data
87270762	1	/*
b7d7a9e3	2	* TCP Westwood+: end-to-end bandwidth estimation for TCP
87270762	3	*
b7d7a9e3 LDC	4	* Angelo Dell'Aera: author of the first version of TCP Westwood+ in Linux 2.4
	5	*
	6	* Support at http://c3lab.poliba.it/index.php/Westwood
	7	* Main references in literature:
	8	*
	9	* - Mascolo S, Casetti, M. Gerla et al.
	10	* "TCP Westwood: bandwidth estimation for TCP" Proc. ACM Mobicom 2001
	11	*
	12	* - A. Grieco, s. Mascolo
	13	* "Performance evaluation of New Reno, Vegas, Westwood+ TCP" ACM Computer
	14	* Comm. Review, 2004
	15	*
	16	* - A. Dell'Aera, L. Grieco, S. Mascolo.
	17	* "Linux 2.4 Implementation of Westwood+ TCP with Rate-Halving :
	18	* A Performance Evaluation Over the Internet" (ICC 2004), Paris, June 2004
	19	*
	20	* Westwood+ employs end-to-end bandwidth measurement to set cwnd and
	21	* ssthresh after packet loss. The probing phase is as the original Reno.
87270762 SH	22	*/
87270762 SH	23
87270762 SH	24	#include <linux/mm.h>
	25	#include <linux/module.h>
	26	#include <linux/skbuff.h>
a8c2190e	27	#include <linux/inet_diag.h>
87270762 SH	28	#include <net/tcp.h>
	29
	30	/* TCP Westwood structure */
	31	struct westwood {
	32	u32 bw_ns_est; /* first bandwidth estimation..not too smoothed 8) */
	33	u32 bw_est; /* bandwidth estimate */
	34	u32 rtt_win_sx; /* here starts a new evaluation... */
	35	u32 bk;
	36	u32 snd_una; /* used for evaluating the number of acked bytes */
	37	u32 cumul_ack;
	38	u32 accounted;
	39	u32 rtt;
	40	u32 rtt_min; /* minimum observed RTT */
f61e2901	41	u8 first_ack; /* flag which infers that this is the first ack */
bc726a71	42	u8 reset_rtt_min; /* Reset RTT min to next RTT sample*/
87270762 SH	43	};
	44
	45
	46	/* TCP Westwood functions and constants */
	47	#define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
	48	#define TCP_WESTWOOD_INIT_RTT (20HZ) / maybe too conservative?! */
	49
	50	/*
	51	* @tcp_westwood_create
	52	* This function initializes fields used in TCP Westwood+,
	53	* it is called after the initial SYN, so the sequence numbers
	54	* are correct but new passive connections we have no
	55	* information about RTTmin at this time so we simply set it to
	56	* TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
	57	* since in this way we're sure it will be updated in a consistent
	58	* way as soon as possible. It will reasonably happen within the first
	59	* RTT period of the connection lifetime.
	60	*/
6687e988	61	static void tcp_westwood_init(struct sock *sk)
87270762	62	{
6687e988	63	struct westwood *w = inet_csk_ca(sk);
87270762 SH	64
87270762 SH	65	w->bk = 0;
e905a9ed YH	66	w->bw_ns_est = 0;
	67	w->bw_est = 0;
	68	w->accounted = 0;
	69	w->cumul_ack = 0;
bc726a71	70	w->reset_rtt_min = 1;
87270762 SH	71	w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
87270762 SH	72	w->rtt_win_sx = tcp_time_stamp;
6687e988	73	w->snd_una = tcp_sk(sk)->snd_una;
f61e2901	74	w->first_ack = 1;
87270762 SH	75	}
	76
	77	/*
	78	* @westwood_do_filter
	79	* Low-pass filter. Implemented using constant coefficients.
	80	*/
	81	static inline u32 westwood_do_filter(u32 a, u32 b)
	82	{
	83	return (((7 * a) + b) >> 3);
	84	}
	85
b3a92eab	86	static void westwood_filter(struct westwood *w, u32 delta)
87270762	87	{
b3a92eab LDC	88	/* If the filter is empty fill it with the first sample of bandwidth */
	89	if (w->bw_ns_est == 0 && w->bw_est == 0) {
	90	w->bw_ns_est = w->bk / delta;
	91	w->bw_est = w->bw_ns_est;
	92	} else {
	93	w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
	94	w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
	95	}
87270762 SH	96	}
	97
	98	/*
	99	* @westwood_pkts_acked
	100	* Called after processing group of packets.
	101	* but all westwood needs is the last sample of srtt.
	102	*/
6687e988	103	static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt)
87270762	104	{
6687e988	105	struct westwood *w = inet_csk_ca(sk);
87270762	106	if (cnt > 0)
6687e988	107	w->rtt = tcp_sk(sk)->srtt >> 3;
87270762 SH	108	}
	109
	110	/*
	111	* @westwood_update_window
	112	* It updates RTT evaluation window if it is the right moment to do
	113	* it. If so it calls filter for evaluating bandwidth.
	114	*/
6687e988	115	static void westwood_update_window(struct sock *sk)
87270762	116	{
6687e988	117	struct westwood *w = inet_csk_ca(sk);
87270762 SH	118	s32 delta = tcp_time_stamp - w->rtt_win_sx;
87270762 SH	119
b7d7a9e3	120	/* Initialize w->snd_una with the first acked sequence number in order
f61e2901 SH	121	* to fix mismatch between tp->snd_una and w->snd_una for the first
	122	* bandwidth sample
	123	*/
e905a9ed	124	if (w->first_ack) {
f61e2901 SH	125	w->snd_una = tcp_sk(sk)->snd_una;
	126	w->first_ack = 0;
	127	}
	128
87270762 SH	129	/*
	130	* See if a RTT-window has passed.
	131	* Be careful since if RTT is less than
	132	* 50ms we don't filter but we continue 'building the sample'.
	133	* This minimum limit was chosen since an estimation on small
	134	* time intervals is better to avoid...
	135	* Obviously on a LAN we reasonably will always have
	136	* right_bound = left_bound + WESTWOOD_RTT_MIN
	137	*/
	138	if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
	139	westwood_filter(w, delta);
	140
	141	w->bk = 0;
	142	w->rtt_win_sx = tcp_time_stamp;
	143	}
	144	}
	145
bc726a71 LDC	146	static inline void update_rtt_min(struct westwood *w)
	147	{
	148	if (w->reset_rtt_min) {
	149	w->rtt_min = w->rtt;
e905a9ed	150	w->reset_rtt_min = 0;
bc726a71 LDC	151	} else
	152	w->rtt_min = min(w->rtt, w->rtt_min);
	153	}
	154
	155
87270762 SH	156	/*
	157	* @westwood_fast_bw
	158	* It is called when we are in fast path. In particular it is called when
	159	* header prediction is successful. In such case in fact update is
	160	* straight forward and doesn't need any particular care.
	161	*/
6687e988	162	static inline void westwood_fast_bw(struct sock *sk)
87270762	163	{
6687e988 ACM	164	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	165	struct westwood *w = inet_csk_ca(sk);
87270762	166
6687e988	167	westwood_update_window(sk);
87270762 SH	168
	169	w->bk += tp->snd_una - w->snd_una;
	170	w->snd_una = tp->snd_una;
bc726a71	171	update_rtt_min(w);
87270762 SH	172	}
	173
	174	/*
	175	* @westwood_acked_count
	176	* This function evaluates cumul_ack for evaluating bk in case of
	177	* delayed or partial acks.
	178	*/
6687e988	179	static inline u32 westwood_acked_count(struct sock *sk)
87270762	180	{
6687e988 ACM	181	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	182	struct westwood *w = inet_csk_ca(sk);
87270762 SH	183
	184	w->cumul_ack = tp->snd_una - w->snd_una;
	185
e905a9ed YH	186	/* If cumul_ack is 0 this is a dupack since it's not moving
	187	* tp->snd_una.
	188	*/
	189	if (!w->cumul_ack) {
87270762 SH	190	w->accounted += tp->mss_cache;
	191	w->cumul_ack = tp->mss_cache;
	192	}
	193
e905a9ed	194	if (w->cumul_ack > tp->mss_cache) {
87270762 SH	195	/* Partial or delayed ack */
	196	if (w->accounted >= w->cumul_ack) {
	197	w->accounted -= w->cumul_ack;
	198	w->cumul_ack = tp->mss_cache;
	199	} else {
	200	w->cumul_ack -= w->accounted;
	201	w->accounted = 0;
	202	}
	203	}
	204
	205	w->snd_una = tp->snd_una;
	206
	207	return w->cumul_ack;
	208	}
	209
87270762 SH	210
	211	/*
	212	* TCP Westwood
	213	* Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
	214	* in packets we use mss_cache). Rttmin is guaranteed to be >= 2
	215	* so avoids ever returning 0.
	216	*/
72dc5b92	217	static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
87270762	218	{
72dc5b92 SH	219	const struct tcp_sock *tp = tcp_sk(sk);
	220	const struct westwood *w = inet_csk_ca(sk);
	221	return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
87270762 SH	222	}
87270762 SH	223
6687e988	224	static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
87270762	225	{
6687e988 ACM	226	struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	227	struct westwood *w = inet_csk_ca(sk);
b7d7a9e3	228
2de979bd	229	switch (event) {
87270762	230	case CA_EVENT_FAST_ACK:
6687e988	231	westwood_fast_bw(sk);
87270762 SH	232	break;
	233
	234	case CA_EVENT_COMPLETE_CWR:
72dc5b92	235	tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
87270762 SH	236	break;
	237
	238	case CA_EVENT_FRTO:
72dc5b92	239	tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
e905a9ed	240	/* Update RTT_min when next ack arrives */
bc726a71	241	w->reset_rtt_min = 1;
87270762 SH	242	break;
	243
	244	case CA_EVENT_SLOW_ACK:
6687e988 ACM	245	westwood_update_window(sk);
6687e988 ACM	246	w->bk += westwood_acked_count(sk);
bc726a71	247	update_rtt_min(w);
87270762 SH	248	break;
	249
	250	default:
	251	/* don't care */
	252	break;
	253	}
	254	}
	255
	256
	257	/* Extract info for Tcp socket info provided via netlink. */
6687e988	258	static void tcp_westwood_info(struct sock *sk, u32 ext,
87270762 SH	259	struct sk_buff *skb)
87270762 SH	260	{
6687e988	261	const struct westwood *ca = inet_csk_ca(sk);
73c1f4a0	262	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
87270762 SH	263	struct rtattr *rta;
	264	struct tcpvegas_info *info;
	265
73c1f4a0	266	rta = __RTA_PUT(skb, INET_DIAG_VEGASINFO, sizeof(*info));
87270762 SH	267	info = RTA_DATA(rta);
	268	info->tcpv_enabled = 1;
	269	info->tcpv_rttcnt = 0;
	270	info->tcpv_rtt = jiffies_to_usecs(ca->rtt);
	271	info->tcpv_minrtt = jiffies_to_usecs(ca->rtt_min);
	272	rtattr_failure: ;
	273	}
	274	}
	275
	276
	277	static struct tcp_congestion_ops tcp_westwood = {
	278	.init = tcp_westwood_init,
	279	.ssthresh = tcp_reno_ssthresh,
	280	.cong_avoid = tcp_reno_cong_avoid,
72dc5b92	281	.min_cwnd = tcp_westwood_bw_rttmin,
87270762 SH	282	.cwnd_event = tcp_westwood_event,
	283	.get_info = tcp_westwood_info,
	284	.pkts_acked = tcp_westwood_pkts_acked,
	285
	286	.owner = THIS_MODULE,
	287	.name = "westwood"
	288	};
	289
	290	static int __init tcp_westwood_register(void)
	291	{
74975d40	292	BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
87270762 SH	293	return tcp_register_congestion_control(&tcp_westwood);
	294	}
	295
	296	static void __exit tcp_westwood_unregister(void)
	297	{
	298	tcp_unregister_congestion_control(&tcp_westwood);
	299	}
	300
	301	module_init(tcp_westwood_register);
	302	module_exit(tcp_westwood_unregister);
	303
	304	MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
	305	MODULE_LICENSE("GPL");
	306	MODULE_DESCRIPTION("TCP Westwood+");