[linux.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_jiffies32;
	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,
				    const struct ack_sample *sample)
{
	struct westwood *w = inet_csk_ca(sk);

	if (sample->rtt_us > 0)
		w->rtt = usecs_to_jiffies(sample->rtt_us);
}

/*
 * @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_jiffies32 - 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_jiffies32;
	}
}

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_ack(struct sock *sk, u32 ack_flags)
{
	if (ack_flags & CA_ACK_SLOWPATH) {
		struct westwood *w = inet_csk_ca(sk);

		westwood_update_window(sk);
		w->bk += westwood_acked_count(sk);

		update_rtt_min(w);
		return;
	}

	westwood_fast_bw(sk);
}

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_COMPLETE_CWR:
		tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
		break;
	case CA_EVENT_LOSS:
		tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
		/* Update RTT_min when next ack arrives */
		w->reset_rtt_min = 1;
		break;
	default:
		/* don't care */
		break;
	}
}

/* Extract info for Tcp socket info provided via netlink. */
static size_t tcp_westwood_info(struct sock *sk, u32 ext, int *attr,
				union tcp_cc_info *info)
{
	const struct westwood *ca = inet_csk_ca(sk);

	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
		info->vegas.tcpv_enabled = 1;
		info->vegas.tcpv_rttcnt	= 0;
		info->vegas.tcpv_rtt	= jiffies_to_usecs(ca->rtt);
		info->vegas.tcpv_minrtt	= jiffies_to_usecs(ca->rtt_min);

		*attr = INET_DIAG_VEGASINFO;
		return sizeof(struct tcpvegas_info);
	}
	return 0;
}

static struct tcp_congestion_ops tcp_westwood __read_mostly = {
	.init		= tcp_westwood_init,
	.ssthresh	= tcp_reno_ssthresh,
	.cong_avoid	= tcp_reno_cong_avoid,
	.undo_cwnd      = tcp_reno_undo_cwnd,
	.cwnd_event	= tcp_westwood_event,
	.in_ack_event	= tcp_westwood_ack,
	.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	};
87270762 SH	44
87270762 SH	45	/* TCP Westwood functions and constants */
	46	#define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
	47	#define TCP_WESTWOOD_INIT_RTT (20HZ) / maybe too conservative?! */
	48
	49	/*
	50	* @tcp_westwood_create
	51	* This function initializes fields used in TCP Westwood+,
	52	* it is called after the initial SYN, so the sequence numbers
	53	* are correct but new passive connections we have no
	54	* information about RTTmin at this time so we simply set it to
	55	* TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
	56	* since in this way we're sure it will be updated in a consistent
	57	* way as soon as possible. It will reasonably happen within the first
	58	* RTT period of the connection lifetime.
	59	*/
6687e988	60	static void tcp_westwood_init(struct sock *sk)
87270762	61	{
6687e988	62	struct westwood *w = inet_csk_ca(sk);
87270762 SH	63
87270762 SH	64	w->bk = 0;
e905a9ed YH	65	w->bw_ns_est = 0;
	66	w->bw_est = 0;
	67	w->accounted = 0;
	68	w->cumul_ack = 0;
bc726a71	69	w->reset_rtt_min = 1;
87270762	70	w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
ad5ad69e	71	w->rtt_win_sx = tcp_jiffies32;
6687e988	72	w->snd_una = tcp_sk(sk)->snd_una;
f61e2901	73	w->first_ack = 1;
87270762 SH	74	}
	75
	76	/*
	77	* @westwood_do_filter
	78	* Low-pass filter. Implemented using constant coefficients.
	79	*/
	80	static inline u32 westwood_do_filter(u32 a, u32 b)
	81	{
a02cec21	82	return ((7 * a) + b) >> 3;
87270762 SH	83	}
87270762 SH	84
b3a92eab	85	static void westwood_filter(struct westwood *w, u32 delta)
87270762	86	{
b3a92eab LDC	87	/* If the filter is empty fill it with the first sample of bandwidth */
	88	if (w->bw_ns_est == 0 && w->bw_est == 0) {
	89	w->bw_ns_est = w->bk / delta;
	90	w->bw_est = w->bw_ns_est;
	91	} else {
	92	w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
	93	w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
	94	}
87270762 SH	95	}
	96
	97	/*
	98	* @westwood_pkts_acked
	99	* Called after processing group of packets.
	100	* but all westwood needs is the last sample of srtt.
	101	*/
756ee172 LB	102	static void tcp_westwood_pkts_acked(struct sock *sk,
756ee172 LB	103	const struct ack_sample *sample)
87270762	104	{
6687e988	105	struct westwood *w = inet_csk_ca(sk);
30cfd0ba	106
756ee172 LB	107	if (sample->rtt_us > 0)
756ee172 LB	108	w->rtt = usecs_to_jiffies(sample->rtt_us);
87270762 SH	109	}
	110
	111	/*
	112	* @westwood_update_window
	113	* It updates RTT evaluation window if it is the right moment to do
	114	* it. If so it calls filter for evaluating bandwidth.
	115	*/
6687e988	116	static void westwood_update_window(struct sock *sk)
87270762	117	{
6687e988	118	struct westwood *w = inet_csk_ca(sk);
ad5ad69e	119	s32 delta = tcp_jiffies32 - w->rtt_win_sx;
87270762	120
b7d7a9e3	121	/* Initialize w->snd_una with the first acked sequence number in order
f61e2901 SH	122	* to fix mismatch between tp->snd_una and w->snd_una for the first
	123	* bandwidth sample
	124	*/
e905a9ed	125	if (w->first_ack) {
f61e2901 SH	126	w->snd_una = tcp_sk(sk)->snd_una;
	127	w->first_ack = 0;
	128	}
	129
87270762 SH	130	/*
	131	* See if a RTT-window has passed.
	132	* Be careful since if RTT is less than
	133	* 50ms we don't filter but we continue 'building the sample'.
	134	* This minimum limit was chosen since an estimation on small
	135	* time intervals is better to avoid...
	136	* Obviously on a LAN we reasonably will always have
	137	* right_bound = left_bound + WESTWOOD_RTT_MIN
	138	*/
	139	if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
	140	westwood_filter(w, delta);
	141
	142	w->bk = 0;
ad5ad69e	143	w->rtt_win_sx = tcp_jiffies32;
87270762 SH	144	}
	145	}
	146
bc726a71 LDC	147	static inline void update_rtt_min(struct westwood *w)
	148	{
	149	if (w->reset_rtt_min) {
	150	w->rtt_min = w->rtt;
e905a9ed	151	w->reset_rtt_min = 0;
bc726a71 LDC	152	} else
	153	w->rtt_min = min(w->rtt, w->rtt_min);
	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	* TCP Westwood
	212	* Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
	213	* in packets we use mss_cache). Rttmin is guaranteed to be >= 2
	214	* so avoids ever returning 0.
	215	*/
72dc5b92	216	static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
87270762	217	{
72dc5b92 SH	218	const struct tcp_sock *tp = tcp_sk(sk);
72dc5b92 SH	219	const struct westwood *w = inet_csk_ca(sk);
688d1945	220
72dc5b92	221	return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
87270762 SH	222	}
87270762 SH	223
7354c8c3 FW	224	static void tcp_westwood_ack(struct sock *sk, u32 ack_flags)
	225	{
	226	if (ack_flags & CA_ACK_SLOWPATH) {
	227	struct westwood *w = inet_csk_ca(sk);
	228
	229	westwood_update_window(sk);
	230	w->bk += westwood_acked_count(sk);
	231
	232	update_rtt_min(w);
	233	return;
	234	}
	235
	236	westwood_fast_bw(sk);
	237	}
	238
6687e988	239	static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
87270762	240	{
6687e988 ACM	241	struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	242	struct westwood *w = inet_csk_ca(sk);
b7d7a9e3	243
2de979bd	244	switch (event) {
87270762	245	case CA_EVENT_COMPLETE_CWR:
72dc5b92	246	tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
87270762	247	break;
9b44190d	248	case CA_EVENT_LOSS:
72dc5b92	249	tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
e905a9ed	250	/* Update RTT_min when next ack arrives */
bc726a71	251	w->reset_rtt_min = 1;
87270762	252	break;
87270762 SH	253	default:
	254	/* don't care */
	255	break;
	256	}
	257	}
	258
87270762	259	/* Extract info for Tcp socket info provided via netlink. */
31ccd0e6 ED	260	static size_t tcp_westwood_info(struct sock sk, u32 ext, int attr,
31ccd0e6 ED	261	union tcp_cc_info *info)
87270762	262	{
6687e988	263	const struct westwood *ca = inet_csk_ca(sk);
688d1945	264
73c1f4a0	265	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
31ccd0e6 ED	266	info->vegas.tcpv_enabled = 1;
31ccd0e6 ED	267	info->vegas.tcpv_rttcnt = 0;
be7164cd L	268	info->vegas.tcpv_rtt = jiffies_to_usecs(ca->rtt);
be7164cd L	269	info->vegas.tcpv_minrtt = jiffies_to_usecs(ca->rtt_min);
26728105	270
31ccd0e6 ED	271	*attr = INET_DIAG_VEGASINFO;
31ccd0e6 ED	272	return sizeof(struct tcpvegas_info);
87270762	273	}
521f1cf1	274	return 0;
87270762 SH	275	}
87270762 SH	276
a252bebe	277	static struct tcp_congestion_ops tcp_westwood __read_mostly = {
87270762 SH	278	.init = tcp_westwood_init,
	279	.ssthresh = tcp_reno_ssthresh,
	280	.cong_avoid = tcp_reno_cong_avoid,
e9799183	281	.undo_cwnd = tcp_reno_undo_cwnd,
87270762	282	.cwnd_event = tcp_westwood_event,
7354c8c3	283	.in_ack_event = tcp_westwood_ack,
87270762 SH	284	.get_info = tcp_westwood_info,
	285	.pkts_acked = tcp_westwood_pkts_acked,
	286
	287	.owner = THIS_MODULE,
	288	.name = "westwood"
	289	};
	290
	291	static int __init tcp_westwood_register(void)
	292	{
74975d40	293	BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
87270762 SH	294	return tcp_register_congestion_control(&tcp_westwood);
	295	}
	296
	297	static void __exit tcp_westwood_unregister(void)
	298	{
	299	tcp_unregister_congestion_control(&tcp_westwood);
	300	}
	301
	302	module_init(tcp_westwood_register);
	303	module_exit(tcp_westwood_unregister);
	304
	305	MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
	306	MODULE_LICENSE("GPL");
	307	MODULE_DESCRIPTION("TCP Westwood+");