[mirror_ubuntu-artful-kernel.git] / net / ipv4 / tcp_cong.c

/*
 * Plugable TCP congestion control support and newReno
 * congestion control.
 * Based on ideas from I/O scheduler suport and Web100.
 *
 * Copyright (C) 2005 Stephen Hemminger <shemminger@osdl.org>
 */

#include <linux/module.h>
#include <linux/mm.h>
#include <linux/types.h>
#include <linux/list.h>
#include <net/tcp.h>

static DEFINE_SPINLOCK(tcp_cong_list_lock);
static LIST_HEAD(tcp_cong_list);

/* Simple linear search, don't expect many entries! */
static struct tcp_congestion_ops *tcp_ca_find(const char *name)
{
	struct tcp_congestion_ops *e;

	list_for_each_entry_rcu(e, &tcp_cong_list, list) {
		if (strcmp(e->name, name) == 0)
			return e;
	}

	return NULL;
}

/*
 * Attach new congestion control algorthim to the list
 * of available options.
 */
int tcp_register_congestion_control(struct tcp_congestion_ops *ca)
{
	int ret = 0;

	/* all algorithms must implement ssthresh and cong_avoid ops */
	if (!ca->ssthresh || !ca->cong_avoid) {
		printk(KERN_ERR "TCP %s does not implement required ops\n",
		       ca->name);
		return -EINVAL;
	}

	spin_lock(&tcp_cong_list_lock);
	if (tcp_ca_find(ca->name)) {
		printk(KERN_NOTICE "TCP %s already registered\n", ca->name);
		ret = -EEXIST;
	} else {
		list_add_tail_rcu(&ca->list, &tcp_cong_list);
		printk(KERN_INFO "TCP %s registered\n", ca->name);
	}
	spin_unlock(&tcp_cong_list_lock);

	return ret;
}
EXPORT_SYMBOL_GPL(tcp_register_congestion_control);

/*
 * Remove congestion control algorithm, called from
 * the module's remove function.  Module ref counts are used
 * to ensure that this can't be done till all sockets using
 * that method are closed.
 */
void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca)
{
	spin_lock(&tcp_cong_list_lock);
	list_del_rcu(&ca->list);
	spin_unlock(&tcp_cong_list_lock);
}
EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control);

/* Assign choice of congestion control. */
void tcp_init_congestion_control(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_congestion_ops *ca;

	if (icsk->icsk_ca_ops != &tcp_init_congestion_ops)
		return;

	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_unlock();

	if (icsk->icsk_ca_ops->init)
		icsk->icsk_ca_ops->init(sk);
}

/* Manage refcounts on socket close. */
void tcp_cleanup_congestion_control(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);

	if (icsk->icsk_ca_ops->release)
		icsk->icsk_ca_ops->release(sk);
	module_put(icsk->icsk_ca_ops->owner);
}

/* Used by sysctl to change default congestion control */
int tcp_set_default_congestion_control(const char *name)
{
	struct tcp_congestion_ops *ca;
	int ret = -ENOENT;

	spin_lock(&tcp_cong_list_lock);
	ca = tcp_ca_find(name);
#ifdef CONFIG_KMOD
	if (!ca) {
		spin_unlock(&tcp_cong_list_lock);

		request_module("tcp_%s", name);
		spin_lock(&tcp_cong_list_lock);
		ca = tcp_ca_find(name);
	}
#endif

	if (ca) {
		list_move(&ca->list, &tcp_cong_list);
		ret = 0;
	}
	spin_unlock(&tcp_cong_list_lock);

	return ret;
}

/* Set default value from kernel configuration at bootup */
static int __init tcp_congestion_default(void)
{
	return tcp_set_default_congestion_control(CONFIG_DEFAULT_TCP_CONG);
}
late_initcall(tcp_congestion_default);


/* Get current default congestion control */
void tcp_get_default_congestion_control(char *name)
{
	struct tcp_congestion_ops *ca;
	/* We will always have reno... */
	BUG_ON(list_empty(&tcp_cong_list));

	rcu_read_lock();
	ca = list_entry(tcp_cong_list.next, struct tcp_congestion_ops, list);
	strncpy(name, ca->name, TCP_CA_NAME_MAX);
	rcu_read_unlock();
}

/* Change congestion control for socket */
int tcp_set_congestion_control(struct sock *sk, const char *name)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_congestion_ops *ca;
	int err = 0;

	rcu_read_lock();
	ca = tcp_ca_find(name);
	if (ca == icsk->icsk_ca_ops)
		goto out;

	if (!ca)
		err = -ENOENT;

	else if (!try_module_get(ca->owner))
		err = -EBUSY;

	else {
		tcp_cleanup_congestion_control(sk);
		icsk->icsk_ca_ops = ca;
		if (icsk->icsk_ca_ops->init)
			icsk->icsk_ca_ops->init(sk);
	}
 out:
	rcu_read_unlock();
	return err;
}


/*
 * Linear increase during slow start
 */
void tcp_slow_start(struct tcp_sock *tp)
{
	if (sysctl_tcp_abc) {
		/* RFC3465: Slow Start
		 * TCP sender SHOULD increase cwnd by the number of
		 * previously unacknowledged bytes ACKed by each incoming
		 * acknowledgment, provided the increase is not more than L
		 */
		if (tp->bytes_acked < tp->mss_cache)
			return;

		/* We MAY increase by 2 if discovered delayed ack */
		if (sysctl_tcp_abc > 1 && tp->bytes_acked >= 2*tp->mss_cache) {
			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
				tp->snd_cwnd++;
		}
	}
	tp->bytes_acked = 0;

	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
		tp->snd_cwnd++;
}
EXPORT_SYMBOL_GPL(tcp_slow_start);

/*
 * TCP Reno congestion control
 * This is special case used for fallback as well.
 */
/* This is Jacobson's slow start and congestion avoidance.
 * SIGCOMM '88, p. 328.
 */
void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 rtt, u32 in_flight,
			 int flag)
{
	struct tcp_sock *tp = tcp_sk(sk);

	if (!tcp_is_cwnd_limited(sk, in_flight))
		return;

	/* In "safe" area, increase. */
        if (tp->snd_cwnd <= tp->snd_ssthresh)
		tcp_slow_start(tp);

 	/* In dangerous area, increase slowly. */
	else if (sysctl_tcp_abc) {
 		/* RFC3465: Appropriate Byte Count
 		 * increase once for each full cwnd acked
 		 */
 		if (tp->bytes_acked >= tp->snd_cwnd*tp->mss_cache) {
 			tp->bytes_acked -= tp->snd_cwnd*tp->mss_cache;
 			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
 				tp->snd_cwnd++;
 		}
 	} else {
 		/* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd */
 		if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
 			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
 				tp->snd_cwnd++;
 			tp->snd_cwnd_cnt = 0;
 		} else
 			tp->snd_cwnd_cnt++;
 	}
}
EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);

/* Slow start threshold is half the congestion window (min 2) */
u32 tcp_reno_ssthresh(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	return max(tp->snd_cwnd >> 1U, 2U);
}
EXPORT_SYMBOL_GPL(tcp_reno_ssthresh);

/* Lower bound on congestion window with halving. */
u32 tcp_reno_min_cwnd(const struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	return tp->snd_ssthresh/2;
}
EXPORT_SYMBOL_GPL(tcp_reno_min_cwnd);

struct tcp_congestion_ops tcp_reno = {
	.name		= "reno",
	.owner		= THIS_MODULE,
	.ssthresh	= tcp_reno_ssthresh,
	.cong_avoid	= tcp_reno_cong_avoid,
	.min_cwnd	= tcp_reno_min_cwnd,
};

/* Initial congestion control used (until SYN)
 * really reno under another name so we can tell difference
 * during tcp_set_default_congestion_control
 */
struct tcp_congestion_ops tcp_init_congestion_ops  = {
	.name		= "",
	.owner		= THIS_MODULE,
	.ssthresh	= tcp_reno_ssthresh,
	.cong_avoid	= tcp_reno_cong_avoid,
	.min_cwnd	= tcp_reno_min_cwnd,
};
EXPORT_SYMBOL_GPL(tcp_init_congestion_ops);
Commit	Line	Data
317a76f9 SH	1	/*
	2	* Plugable TCP congestion control support and newReno
	3	* congestion control.
	4	* Based on ideas from I/O scheduler suport and Web100.
	5	*
	6	* Copyright (C) 2005 Stephen Hemminger <shemminger@osdl.org>
	7	*/
	8
317a76f9 SH	9	#include <linux/module.h>
	10	#include <linux/mm.h>
	11	#include <linux/types.h>
	12	#include <linux/list.h>
	13	#include <net/tcp.h>
	14
	15	static DEFINE_SPINLOCK(tcp_cong_list_lock);
	16	static LIST_HEAD(tcp_cong_list);
	17
	18	/* Simple linear search, don't expect many entries! */
	19	static struct tcp_congestion_ops tcp_ca_find(const char name)
	20	{
	21	struct tcp_congestion_ops *e;
	22
5f8ef48d	23	list_for_each_entry_rcu(e, &tcp_cong_list, list) {
317a76f9 SH	24	if (strcmp(e->name, name) == 0)
	25	return e;
	26	}
	27
	28	return NULL;
	29	}
	30
	31	/*
	32	* Attach new congestion control algorthim to the list
	33	* of available options.
	34	*/
	35	int tcp_register_congestion_control(struct tcp_congestion_ops *ca)
	36	{
	37	int ret = 0;
	38
	39	/* all algorithms must implement ssthresh and cong_avoid ops */
72dc5b92	40	if (!ca->ssthresh \|\| !ca->cong_avoid) {
317a76f9 SH	41	printk(KERN_ERR "TCP %s does not implement required ops\n",
	42	ca->name);
	43	return -EINVAL;
	44	}
	45
	46	spin_lock(&tcp_cong_list_lock);
	47	if (tcp_ca_find(ca->name)) {
	48	printk(KERN_NOTICE "TCP %s already registered\n", ca->name);
	49	ret = -EEXIST;
	50	} else {
3d2573f7	51	list_add_tail_rcu(&ca->list, &tcp_cong_list);
317a76f9 SH	52	printk(KERN_INFO "TCP %s registered\n", ca->name);
	53	}
	54	spin_unlock(&tcp_cong_list_lock);
	55
	56	return ret;
	57	}
	58	EXPORT_SYMBOL_GPL(tcp_register_congestion_control);
	59
	60	/*
	61	* Remove congestion control algorithm, called from
	62	* the module's remove function. Module ref counts are used
	63	* to ensure that this can't be done till all sockets using
	64	* that method are closed.
	65	*/
	66	void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca)
	67	{
	68	spin_lock(&tcp_cong_list_lock);
	69	list_del_rcu(&ca->list);
	70	spin_unlock(&tcp_cong_list_lock);
	71	}
	72	EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control);
	73
	74	/* Assign choice of congestion control. */
6687e988	75	void tcp_init_congestion_control(struct sock *sk)
317a76f9	76	{
6687e988	77	struct inet_connection_sock *icsk = inet_csk(sk);
317a76f9 SH	78	struct tcp_congestion_ops *ca;
317a76f9 SH	79
6687e988	80	if (icsk->icsk_ca_ops != &tcp_init_congestion_ops)
5f8ef48d SH	81	return;
5f8ef48d SH	82
317a76f9 SH	83	rcu_read_lock();
	84	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
	85	if (try_module_get(ca->owner)) {
6687e988	86	icsk->icsk_ca_ops = ca;
317a76f9 SH	87	break;
	88	}
	89
	90	}
	91	rcu_read_unlock();
	92
6687e988 ACM	93	if (icsk->icsk_ca_ops->init)
6687e988 ACM	94	icsk->icsk_ca_ops->init(sk);
317a76f9 SH	95	}
	96
	97	/* Manage refcounts on socket close. */
6687e988	98	void tcp_cleanup_congestion_control(struct sock *sk)
317a76f9	99	{
6687e988 ACM	100	struct inet_connection_sock *icsk = inet_csk(sk);
	101
	102	if (icsk->icsk_ca_ops->release)
	103	icsk->icsk_ca_ops->release(sk);
	104	module_put(icsk->icsk_ca_ops->owner);
317a76f9 SH	105	}
	106
	107	/* Used by sysctl to change default congestion control */
	108	int tcp_set_default_congestion_control(const char *name)
	109	{
	110	struct tcp_congestion_ops *ca;
	111	int ret = -ENOENT;
	112
	113	spin_lock(&tcp_cong_list_lock);
	114	ca = tcp_ca_find(name);
	115	#ifdef CONFIG_KMOD
	116	if (!ca) {
	117	spin_unlock(&tcp_cong_list_lock);
	118
	119	request_module("tcp_%s", name);
	120	spin_lock(&tcp_cong_list_lock);
	121	ca = tcp_ca_find(name);
	122	}
	123	#endif
	124
	125	if (ca) {
	126	list_move(&ca->list, &tcp_cong_list);
	127	ret = 0;
	128	}
	129	spin_unlock(&tcp_cong_list_lock);
	130
	131	return ret;
	132	}
	133
b1736a71 SH	134	/* Set default value from kernel configuration at bootup */
	135	static int __init tcp_congestion_default(void)
	136	{
	137	return tcp_set_default_congestion_control(CONFIG_DEFAULT_TCP_CONG);
	138	}
	139	late_initcall(tcp_congestion_default);
	140
	141
317a76f9 SH	142	/* Get current default congestion control */
	143	void tcp_get_default_congestion_control(char *name)
	144	{
	145	struct tcp_congestion_ops *ca;
	146	/* We will always have reno... */
	147	BUG_ON(list_empty(&tcp_cong_list));
	148
	149	rcu_read_lock();
	150	ca = list_entry(tcp_cong_list.next, struct tcp_congestion_ops, list);
	151	strncpy(name, ca->name, TCP_CA_NAME_MAX);
	152	rcu_read_unlock();
	153	}
	154
5f8ef48d	155	/* Change congestion control for socket */
6687e988	156	int tcp_set_congestion_control(struct sock sk, const char name)
5f8ef48d	157	{
6687e988	158	struct inet_connection_sock *icsk = inet_csk(sk);
5f8ef48d SH	159	struct tcp_congestion_ops *ca;
	160	int err = 0;
	161
	162	rcu_read_lock();
	163	ca = tcp_ca_find(name);
6687e988	164	if (ca == icsk->icsk_ca_ops)
5f8ef48d SH	165	goto out;
	166
	167	if (!ca)
	168	err = -ENOENT;
	169
	170	else if (!try_module_get(ca->owner))
	171	err = -EBUSY;
	172
	173	else {
6687e988 ACM	174	tcp_cleanup_congestion_control(sk);
	175	icsk->icsk_ca_ops = ca;
	176	if (icsk->icsk_ca_ops->init)
	177	icsk->icsk_ca_ops->init(sk);
5f8ef48d SH	178	}
	179	out:
	180	rcu_read_unlock();
	181	return err;
	182	}
	183
40efc6fa SH	184
	185	/*
	186	* Linear increase during slow start
	187	*/
	188	void tcp_slow_start(struct tcp_sock *tp)
	189	{
	190	if (sysctl_tcp_abc) {
	191	/* RFC3465: Slow Start
	192	* TCP sender SHOULD increase cwnd by the number of
	193	* previously unacknowledged bytes ACKed by each incoming
	194	* acknowledgment, provided the increase is not more than L
	195	*/
	196	if (tp->bytes_acked < tp->mss_cache)
	197	return;
	198
	199	/* We MAY increase by 2 if discovered delayed ack */
3fdf3f0c	200	if (sysctl_tcp_abc > 1 && tp->bytes_acked >= 2*tp->mss_cache) {
40efc6fa SH	201	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	202	tp->snd_cwnd++;
	203	}
	204	}
	205	tp->bytes_acked = 0;
	206
	207	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	208	tp->snd_cwnd++;
	209	}
	210	EXPORT_SYMBOL_GPL(tcp_slow_start);
	211
317a76f9 SH	212	/*
	213	* TCP Reno congestion control
	214	* This is special case used for fallback as well.
	215	*/
	216	/* This is Jacobson's slow start and congestion avoidance.
	217	* SIGCOMM '88, p. 328.
	218	*/
6687e988	219	void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 rtt, u32 in_flight,
317a76f9 SH	220	int flag)
317a76f9 SH	221	{
6687e988 ACM	222	struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	223
f4805ede	224	if (!tcp_is_cwnd_limited(sk, in_flight))
317a76f9 SH	225	return;
317a76f9 SH	226
7faffa1c SH	227	/* In "safe" area, increase. */
	228	if (tp->snd_cwnd <= tp->snd_ssthresh)
	229	tcp_slow_start(tp);
9772efb9 SH	230
	231	/* In dangerous area, increase slowly. */
	232	else if (sysctl_tcp_abc) {
c3e5d877	233	/* RFC3465: Appropriate Byte Count
9772efb9 SH	234	* increase once for each full cwnd acked
	235	*/
	236	if (tp->bytes_acked >= tp->snd_cwnd*tp->mss_cache) {
	237	tp->bytes_acked -= tp->snd_cwnd*tp->mss_cache;
	238	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	239	tp->snd_cwnd++;
	240	}
	241	} else {
	242	/* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd */
	243	if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
	244	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	245	tp->snd_cwnd++;
	246	tp->snd_cwnd_cnt = 0;
	247	} else
	248	tp->snd_cwnd_cnt++;
	249	}
317a76f9 SH	250	}
	251	EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);
	252
	253	/* Slow start threshold is half the congestion window (min 2) */
6687e988	254	u32 tcp_reno_ssthresh(struct sock *sk)
317a76f9	255	{
6687e988	256	const struct tcp_sock *tp = tcp_sk(sk);
317a76f9 SH	257	return max(tp->snd_cwnd >> 1U, 2U);
	258	}
	259	EXPORT_SYMBOL_GPL(tcp_reno_ssthresh);
	260
72dc5b92 SH	261	/* Lower bound on congestion window with halving. */
72dc5b92 SH	262	u32 tcp_reno_min_cwnd(const struct sock *sk)
317a76f9	263	{
6687e988	264	const struct tcp_sock *tp = tcp_sk(sk);
317a76f9 SH	265	return tp->snd_ssthresh/2;
	266	}
	267	EXPORT_SYMBOL_GPL(tcp_reno_min_cwnd);
	268
	269	struct tcp_congestion_ops tcp_reno = {
	270	.name = "reno",
	271	.owner = THIS_MODULE,
	272	.ssthresh = tcp_reno_ssthresh,
	273	.cong_avoid = tcp_reno_cong_avoid,
	274	.min_cwnd = tcp_reno_min_cwnd,
	275	};
	276
5f8ef48d SH	277	/* Initial congestion control used (until SYN)
	278	* really reno under another name so we can tell difference
	279	* during tcp_set_default_congestion_control
	280	*/
	281	struct tcp_congestion_ops tcp_init_congestion_ops = {
	282	.name = "",
	283	.owner = THIS_MODULE,
	284	.ssthresh = tcp_reno_ssthresh,
	285	.cong_avoid = tcp_reno_cong_avoid,
	286	.min_cwnd = tcp_reno_min_cwnd,
	287	};
	288	EXPORT_SYMBOL_GPL(tcp_init_congestion_ops);