[linux.git] / net / ipv4 / tcp_cubic.c

/*
 * TCP CUBIC: Binary Increase Congestion control for TCP v2.0
 *
 * This is from the implementation of CUBIC TCP in
 * Injong Rhee, Lisong Xu.
 *  "CUBIC: A New TCP-Friendly High-Speed TCP Variant
 *  in PFLDnet 2005
 * Available from:
 *  http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/cubic-paper.pdf
 *
 * Unless CUBIC is enabled and congestion window is large
 * this behaves the same as the original Reno.
 */

#include <linux/config.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <net/tcp.h>
#include <asm/div64.h>

#define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
					 * max_cwnd = snd_cwnd * beta
					 */
#define BICTCP_B		4	 /*
					  * In binary search,
					  * go to point (max+min)/N
					  */
#define	BICTCP_HZ		10	/* BIC HZ 2^10 = 1024 */

static int fast_convergence = 1;
static int max_increment = 16;
static int beta = 819;		/* = 819/1024 (BICTCP_BETA_SCALE) */
static int initial_ssthresh = 100;
static int bic_scale = 41;
static int tcp_friendliness = 1;

static u32 cube_rtt_scale;
static u32 beta_scale;
static u64 cube_factor;

/* Note parameters that are used for precomputing scale factors are read-only */
module_param(fast_convergence, int, 0644);
MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
module_param(max_increment, int, 0644);
MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
module_param(beta, int, 0444);
MODULE_PARM_DESC(beta, "beta for multiplicative increase");
module_param(initial_ssthresh, int, 0644);
MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
module_param(bic_scale, int, 0444);
MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
module_param(tcp_friendliness, int, 0644);
MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");

#include <asm/div64.h>

/* BIC TCP Parameters */
struct bictcp {
	u32	cnt;		/* increase cwnd by 1 after ACKs */
	u32 	last_max_cwnd;	/* last maximum snd_cwnd */
	u32	loss_cwnd;	/* congestion window at last loss */
	u32	last_cwnd;	/* the last snd_cwnd */
	u32	last_time;	/* time when updated last_cwnd */
	u32	bic_origin_point;/* origin point of bic function */
	u32	bic_K;		/* time to origin point from the beginning of the current epoch */
	u32	delay_min;	/* min delay */
	u32	epoch_start;	/* beginning of an epoch */
	u32	ack_cnt;	/* number of acks */
	u32	tcp_cwnd;	/* estimated tcp cwnd */
#define ACK_RATIO_SHIFT	4
	u32	delayed_ack;	/* estimate the ratio of Packets/ACKs << 4 */
};

static inline void bictcp_reset(struct bictcp *ca)
{
	ca->cnt = 0;
	ca->last_max_cwnd = 0;
	ca->loss_cwnd = 0;
	ca->last_cwnd = 0;
	ca->last_time = 0;
	ca->bic_origin_point = 0;
	ca->bic_K = 0;
	ca->delay_min = 0;
	ca->epoch_start = 0;
	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	ca->ack_cnt = 0;
	ca->tcp_cwnd = 0;
}

static void bictcp_init(struct sock *sk)
{
	bictcp_reset(inet_csk_ca(sk));
	if (initial_ssthresh)
		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}

/* 64bit divisor, dividend and result. dynamic precision */
static inline u_int64_t div64_64(u_int64_t dividend, u_int64_t divisor)
{
	u_int32_t d = divisor;

	if (divisor > 0xffffffffULL) {
		unsigned int shift = fls(divisor >> 32);

		d = divisor >> shift;
		dividend >>= shift;
	}

	/* avoid 64 bit division if possible */
	if (dividend >> 32)
		do_div(dividend, d);
	else
		dividend = (uint32_t) dividend / d;

	return dividend;
}

/*
 * calculate the cubic root of x using Newton-Raphson
 */
static u32 cubic_root(u64 a)
{
	u32 x, x1;

	/* Initial estimate is based on:
	 * cbrt(x) = exp(log(x) / 3)
	 */
	x = 1u << (fls64(a)/3);

	/*
	 * Iteration based on:
	 *                         2
	 * x    = ( 2 * x  +  a / x  ) / 3
	 *  k+1          k         k
	 */
	do {
		x1 = x;
		x = (2 * x + (uint32_t) div64_64(a, x*x)) / 3;
	} while (abs(x1 - x) > 1);

	return x;
}

/*
 * Compute congestion window to use.
 */
static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
{
	u64 offs;
	u32 delta, t, bic_target, min_cnt, max_cnt;

	ca->ack_cnt++;	/* count the number of ACKs */

	if (ca->last_cwnd == cwnd &&
	    (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
		return;

	ca->last_cwnd = cwnd;
	ca->last_time = tcp_time_stamp;

	if (ca->epoch_start == 0) {
		ca->epoch_start = tcp_time_stamp;	/* record the beginning of an epoch */
		ca->ack_cnt = 1;			/* start counting */
		ca->tcp_cwnd = cwnd;			/* syn with cubic */

		if (ca->last_max_cwnd <= cwnd) {
			ca->bic_K = 0;
			ca->bic_origin_point = cwnd;
		} else {
			/* Compute new K based on
			 * (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
			 */
			ca->bic_K = cubic_root(cube_factor
					       * (ca->last_max_cwnd - cwnd));
			ca->bic_origin_point = ca->last_max_cwnd;
		}
	}

        /* cubic function - calc*/
        /* calculate c * time^3 / rtt,
         *  while considering overflow in calculation of time^3
	 * (so time^3 is done by using 64 bit)
	 * and without the support of division of 64bit numbers
	 * (so all divisions are done by using 32 bit)
         *  also NOTE the unit of those veriables
         *	  time  = (t - K) / 2^bictcp_HZ
         *	  c = bic_scale >> 10
	 * rtt  = (srtt >> 3) / HZ
	 * !!! The following code does not have overflow problems,
	 * if the cwnd < 1 million packets !!!
         */

	/* change the unit from HZ to bictcp_HZ */
        t = ((tcp_time_stamp + ca->delay_min - ca->epoch_start)
	     << BICTCP_HZ) / HZ;

        if (t < ca->bic_K)		/* t - K */
		offs = ca->bic_K - t;
        else
                offs = t - ca->bic_K;

	/* c/rtt * (t-K)^3 */
	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
        if (t < ca->bic_K)                                	/* below origin*/
                bic_target = ca->bic_origin_point - delta;
        else                                                	/* above origin*/
                bic_target = ca->bic_origin_point + delta;

        /* cubic function - calc bictcp_cnt*/
        if (bic_target > cwnd) {
		ca->cnt = cwnd / (bic_target - cwnd);
        } else {
                ca->cnt = 100 * cwnd;              /* very small increment*/
        }

	if (ca->delay_min > 0) {
		/* max increment = Smax * rtt / 0.1  */
		min_cnt = (cwnd * HZ * 8)/(10 * max_increment * ca->delay_min);
		if (ca->cnt < min_cnt)
			ca->cnt = min_cnt;
	}

        /* slow start and low utilization  */
	if (ca->loss_cwnd == 0)		/* could be aggressive in slow start */
		ca->cnt = 50;

	/* TCP Friendly */
	if (tcp_friendliness) {
		u32 scale = beta_scale;
		delta = (cwnd * scale) >> 3;
	        while (ca->ack_cnt > delta) {		/* update tcp cwnd */
	                ca->ack_cnt -= delta;
        	        ca->tcp_cwnd++;
		}

		if (ca->tcp_cwnd > cwnd){	/* if bic is slower than tcp */
			delta = ca->tcp_cwnd - cwnd;
			max_cnt = cwnd / delta;
			if (ca->cnt > max_cnt)
				ca->cnt = max_cnt;
		}
        }

	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	if (ca->cnt == 0)			/* cannot be zero */
		ca->cnt = 1;
}


/* Keep track of minimum rtt */
static inline void measure_delay(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);
	u32 delay;

	/* No time stamp */
	if (!(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) ||
	     /* Discard delay samples right after fast recovery */
	    (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
		return;

	delay = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
	if (delay == 0)
		delay = 1;

	/* first time call or link delay decreases */
	if (ca->delay_min == 0 || ca->delay_min > delay)
		ca->delay_min = delay;
}

static void bictcp_cong_avoid(struct sock *sk, u32 ack,
			      u32 seq_rtt, u32 in_flight, int data_acked)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	if (data_acked)
		measure_delay(sk);

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

	if (tp->snd_cwnd <= tp->snd_ssthresh)
		tcp_slow_start(tp);
	else {
		bictcp_update(ca, tp->snd_cwnd);

		/* In dangerous area, increase slowly.
		 * In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
		 */
		if (tp->snd_cwnd_cnt >= ca->cnt) {
			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
				tp->snd_cwnd++;
			tp->snd_cwnd_cnt = 0;
		} else
			tp->snd_cwnd_cnt++;
	}

}

static u32 bictcp_recalc_ssthresh(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->epoch_start = 0;	/* end of epoch */

	/* Wmax and fast convergence */
	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
		ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
			/ (2 * BICTCP_BETA_SCALE);
	else
		ca->last_max_cwnd = tp->snd_cwnd;

	ca->loss_cwnd = tp->snd_cwnd;

	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
}

static u32 bictcp_undo_cwnd(struct sock *sk)
{
	struct bictcp *ca = inet_csk_ca(sk);

	return max(tcp_sk(sk)->snd_cwnd, ca->last_max_cwnd);
}

static u32 bictcp_min_cwnd(struct sock *sk)
{
	return tcp_sk(sk)->snd_ssthresh;
}

static void bictcp_state(struct sock *sk, u8 new_state)
{
	if (new_state == TCP_CA_Loss)
		bictcp_reset(inet_csk_ca(sk));
}

/* Track delayed acknowledgment ratio using sliding window
 * ratio = (15*ratio + sample) / 16
 */
static void bictcp_acked(struct sock *sk, u32 cnt)
{
	const struct inet_connection_sock *icsk = inet_csk(sk);

	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
		struct bictcp *ca = inet_csk_ca(sk);
		cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
		ca->delayed_ack += cnt;
	}
}


static struct tcp_congestion_ops cubictcp = {
	.init		= bictcp_init,
	.ssthresh	= bictcp_recalc_ssthresh,
	.cong_avoid	= bictcp_cong_avoid,
	.set_state	= bictcp_state,
	.undo_cwnd	= bictcp_undo_cwnd,
	.min_cwnd	= bictcp_min_cwnd,
	.pkts_acked     = bictcp_acked,
	.owner		= THIS_MODULE,
	.name		= "cubic",
};

static int __init cubictcp_register(void)
{
	BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);

	/* Precompute a bunch of the scaling factors that are used per-packet
	 * based on SRTT of 100ms
	 */

	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);

	cube_rtt_scale = (bic_scale << 3) / 10;	/* 1024*c/rtt */

	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	 *  so K = cubic_root( (wmax-cwnd)*rtt/c )
	 * the unit of K is bictcp_HZ=2^10, not HZ
	 *
	 *  c = bic_scale >> 10
	 *  rtt = 100ms
	 *
	 * the following code has been designed and tested for
	 * cwnd < 1 million packets
	 * RTT < 100 seconds
	 * HZ < 1,000,00  (corresponding to 10 nano-second)
	 */

	/* 1/c * 2^2*bictcp_HZ * srtt */
	cube_factor = 1ull << (10+3*BICTCP_HZ); /* 2^40 */

	/* divide by bic_scale and by constant Srtt (100ms) */
	do_div(cube_factor, bic_scale * 10);

	return tcp_register_congestion_control(&cubictcp);
}

static void __exit cubictcp_unregister(void)
{
	tcp_unregister_congestion_control(&cubictcp);
}

module_init(cubictcp_register);
module_exit(cubictcp_unregister);

MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("CUBIC TCP");
MODULE_VERSION("2.0");
Commit	Line	Data
df3271f3 SH	1	/*
	2	* TCP CUBIC: Binary Increase Congestion control for TCP v2.0
	3	*
	4	* This is from the implementation of CUBIC TCP in
	5	* Injong Rhee, Lisong Xu.
	6	* "CUBIC: A New TCP-Friendly High-Speed TCP Variant
	7	* in PFLDnet 2005
	8	* Available from:
	9	* http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/cubic-paper.pdf
	10	*
	11	* Unless CUBIC is enabled and congestion window is large
	12	* this behaves the same as the original Reno.
	13	*/
	14
	15	#include <linux/config.h>
	16	#include <linux/mm.h>
	17	#include <linux/module.h>
	18	#include <net/tcp.h>
89b3d9aa	19	#include <asm/div64.h>
df3271f3 SH	20
	21	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	22	* max_cwnd = snd_cwnd * beta
	23	*/
	24	#define BICTCP_B 4 /*
	25	* In binary search,
	26	* go to point (max+min)/N
	27	*/
	28	#define BICTCP_HZ 10 /* BIC HZ 2^10 = 1024 */
	29
	30	static int fast_convergence = 1;
	31	static int max_increment = 16;
	32	static int beta = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
	33	static int initial_ssthresh = 100;
	34	static int bic_scale = 41;
	35	static int tcp_friendliness = 1;
	36
89b3d9aa SH	37	static u32 cube_rtt_scale;
	38	static u32 beta_scale;
	39	static u64 cube_factor;
	40
	41	/* Note parameters that are used for precomputing scale factors are read-only */
df3271f3 SH	42	module_param(fast_convergence, int, 0644);
	43	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
	44	module_param(max_increment, int, 0644);
	45	MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
89b3d9aa	46	module_param(beta, int, 0444);
df3271f3 SH	47	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
	48	module_param(initial_ssthresh, int, 0644);
	49	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
89b3d9aa	50	module_param(bic_scale, int, 0444);
df3271f3 SH	51	MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
	52	module_param(tcp_friendliness, int, 0644);
	53	MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
	54
9eb2d627	55	#include <asm/div64.h>
df3271f3 SH	56
	57	/* BIC TCP Parameters */
	58	struct bictcp {
	59	u32 cnt; /* increase cwnd by 1 after ACKs */
	60	u32 last_max_cwnd; /* last maximum snd_cwnd */
	61	u32 loss_cwnd; /* congestion window at last loss */
	62	u32 last_cwnd; /* the last snd_cwnd */
	63	u32 last_time; /* time when updated last_cwnd */
	64	u32 bic_origin_point;/* origin point of bic function */
	65	u32 bic_K; /* time to origin point from the beginning of the current epoch */
	66	u32 delay_min; /* min delay */
	67	u32 epoch_start; /* beginning of an epoch */
	68	u32 ack_cnt; /* number of acks */
	69	u32 tcp_cwnd; /* estimated tcp cwnd */
	70	#define ACK_RATIO_SHIFT 4
	71	u32 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
	72	};
	73
	74	static inline void bictcp_reset(struct bictcp *ca)
	75	{
	76	ca->cnt = 0;
	77	ca->last_max_cwnd = 0;
	78	ca->loss_cwnd = 0;
	79	ca->last_cwnd = 0;
	80	ca->last_time = 0;
	81	ca->bic_origin_point = 0;
	82	ca->bic_K = 0;
	83	ca->delay_min = 0;
	84	ca->epoch_start = 0;
	85	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	86	ca->ack_cnt = 0;
	87	ca->tcp_cwnd = 0;
	88	}
	89
	90	static void bictcp_init(struct sock *sk)
	91	{
	92	bictcp_reset(inet_csk_ca(sk));
	93	if (initial_ssthresh)
	94	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
	95	}
	96
9eb2d627 SH	97	/* 64bit divisor, dividend and result. dynamic precision */
	98	static inline u_int64_t div64_64(u_int64_t dividend, u_int64_t divisor)
	99	{
	100	u_int32_t d = divisor;
	101
	102	if (divisor > 0xffffffffULL) {
	103	unsigned int shift = fls(divisor >> 32);
	104
	105	d = divisor >> shift;
	106	dividend >>= shift;
	107	}
	108
	109	/* avoid 64 bit division if possible */
	110	if (dividend >> 32)
	111	do_div(dividend, d);
	112	else
	113	dividend = (uint32_t) dividend / d;
	114
	115	return dividend;
	116	}
df3271f3 SH	117
df3271f3 SH	118	/*
9eb2d627	119	* calculate the cubic root of x using Newton-Raphson
df3271f3	120	*/
9eb2d627	121	static u32 cubic_root(u64 a)
df3271f3	122	{
9eb2d627 SH	123	u32 x, x1;
	124
	125	/* Initial estimate is based on:
	126	* cbrt(x) = exp(log(x) / 3)
	127	*/
	128	x = 1u << (fls64(a)/3);
	129
	130	/*
	131	* Iteration based on:
	132	* 2
	133	* x = ( 2 * x + a / x ) / 3
	134	* k+1 k k
	135	*/
	136	do {
	137	x1 = x;
	138	x = (2 * x + (uint32_t) div64_64(a, x*x)) / 3;
	139	} while (abs(x1 - x) > 1);
	140
	141	return x;
df3271f3 SH	142	}
df3271f3 SH	143
df3271f3 SH	144	/*
	145	* Compute congestion window to use.
	146	*/
	147	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	148	{
89b3d9aa SH	149	u64 offs;
89b3d9aa SH	150	u32 delta, t, bic_target, min_cnt, max_cnt;
df3271f3 SH	151
	152	ca->ack_cnt++; /* count the number of ACKs */
	153
	154	if (ca->last_cwnd == cwnd &&
	155	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	156	return;
	157
	158	ca->last_cwnd = cwnd;
	159	ca->last_time = tcp_time_stamp;
	160
df3271f3 SH	161	if (ca->epoch_start == 0) {
	162	ca->epoch_start = tcp_time_stamp; /* record the beginning of an epoch */
	163	ca->ack_cnt = 1; /* start counting */
	164	ca->tcp_cwnd = cwnd; /* syn with cubic */
	165
	166	if (ca->last_max_cwnd <= cwnd) {
	167	ca->bic_K = 0;
	168	ca->bic_origin_point = cwnd;
	169	} else {
89b3d9aa SH	170	/* Compute new K based on
	171	* (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
	172	*/
	173	ca->bic_K = cubic_root(cube_factor
	174	* (ca->last_max_cwnd - cwnd));
df3271f3 SH	175	ca->bic_origin_point = ca->last_max_cwnd;
	176	}
	177	}
	178
	179	/* cubic function - calc*/
	180	/* calculate c * time^3 / rtt,
	181	* while considering overflow in calculation of time^3
89b3d9aa	182	* (so time^3 is done by using 64 bit)
df3271f3	183	* and without the support of division of 64bit numbers
89b3d9aa	184	* (so all divisions are done by using 32 bit)
df3271f3 SH	185	* also NOTE the unit of those veriables
	186	* time = (t - K) / 2^bictcp_HZ
	187	* c = bic_scale >> 10
	188	* rtt = (srtt >> 3) / HZ
	189	* !!! The following code does not have overflow problems,
	190	* if the cwnd < 1 million packets !!!
	191	*/
	192
	193	/* change the unit from HZ to bictcp_HZ */
	194	t = ((tcp_time_stamp + ca->delay_min - ca->epoch_start)
	195	<< BICTCP_HZ) / HZ;
	196
	197	if (t < ca->bic_K) /* t - K */
89b3d9aa	198	offs = ca->bic_K - t;
df3271f3	199	else
89b3d9aa	200	offs = t - ca->bic_K;
df3271f3	201
89b3d9aa SH	202	/* c/rtt * (t-K)^3 */
89b3d9aa SH	203	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
df3271f3	204	if (t < ca->bic_K) /* below origin*/
89b3d9aa	205	bic_target = ca->bic_origin_point - delta;
df3271f3	206	else /* above origin*/
89b3d9aa	207	bic_target = ca->bic_origin_point + delta;
df3271f3 SH	208
	209	/* cubic function - calc bictcp_cnt*/
	210	if (bic_target > cwnd) {
	211	ca->cnt = cwnd / (bic_target - cwnd);
	212	} else {
	213	ca->cnt = 100 * cwnd; /* very small increment*/
	214	}
	215
	216	if (ca->delay_min > 0) {
	217	/* max increment = Smax * rtt / 0.1 */
	218	min_cnt = (cwnd * HZ * 8)/(10 * max_increment * ca->delay_min);
	219	if (ca->cnt < min_cnt)
	220	ca->cnt = min_cnt;
	221	}
	222
	223	/* slow start and low utilization */
	224	if (ca->loss_cwnd == 0) /* could be aggressive in slow start */
	225	ca->cnt = 50;
	226
	227	/* TCP Friendly */
	228	if (tcp_friendliness) {
89b3d9aa SH	229	u32 scale = beta_scale;
	230	delta = (cwnd * scale) >> 3;
	231	while (ca->ack_cnt > delta) { /* update tcp cwnd */
	232	ca->ack_cnt -= delta;
df3271f3 SH	233	ca->tcp_cwnd++;
	234	}
	235
	236	if (ca->tcp_cwnd > cwnd){ /* if bic is slower than tcp */
89b3d9aa SH	237	delta = ca->tcp_cwnd - cwnd;
89b3d9aa SH	238	max_cnt = cwnd / delta;
df3271f3 SH	239	if (ca->cnt > max_cnt)
	240	ca->cnt = max_cnt;
	241	}
	242	}
	243
	244	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	245	if (ca->cnt == 0) /* cannot be zero */
	246	ca->cnt = 1;
	247	}
	248
	249
	250	/* Keep track of minimum rtt */
	251	static inline void measure_delay(struct sock *sk)
	252	{
	253	const struct tcp_sock *tp = tcp_sk(sk);
	254	struct bictcp *ca = inet_csk_ca(sk);
	255	u32 delay;
	256
	257	/* No time stamp */
	258	if (!(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) \|\|
	259	/* Discard delay samples right after fast recovery */
	260	(s32)(tcp_time_stamp - ca->epoch_start) < HZ)
	261	return;
	262
	263	delay = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
	264	if (delay == 0)
	265	delay = 1;
	266
	267	/* first time call or link delay decreases */
	268	if (ca->delay_min == 0 \|\| ca->delay_min > delay)
	269	ca->delay_min = delay;
	270	}
	271
	272	static void bictcp_cong_avoid(struct sock *sk, u32 ack,
	273	u32 seq_rtt, u32 in_flight, int data_acked)
	274	{
	275	struct tcp_sock *tp = tcp_sk(sk);
	276	struct bictcp *ca = inet_csk_ca(sk);
	277
	278	if (data_acked)
	279	measure_delay(sk);
	280
	281	if (!tcp_is_cwnd_limited(sk, in_flight))
	282	return;
	283
	284	if (tp->snd_cwnd <= tp->snd_ssthresh)
	285	tcp_slow_start(tp);
	286	else {
	287	bictcp_update(ca, tp->snd_cwnd);
	288
	289	/* In dangerous area, increase slowly.
	290	* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
	291	*/
	292	if (tp->snd_cwnd_cnt >= ca->cnt) {
	293	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	294	tp->snd_cwnd++;
	295	tp->snd_cwnd_cnt = 0;
	296	} else
	297	tp->snd_cwnd_cnt++;
	298	}
	299
	300	}
	301
	302	static u32 bictcp_recalc_ssthresh(struct sock *sk)
303	{
304	const struct tcp_sock *tp = tcp_sk(sk);
305	struct bictcp *ca = inet_csk_ca(sk);
306
307	ca->epoch_start = 0; /* end of epoch */
308
309	/* Wmax and fast convergence */
310	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
311	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
312	/ (2 * BICTCP_BETA_SCALE);
313	else
314	ca->last_max_cwnd = tp->snd_cwnd;
315
316	ca->loss_cwnd = tp->snd_cwnd;
317
318	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
319	}
320
321	static u32 bictcp_undo_cwnd(struct sock *sk)
322	{
323	struct bictcp *ca = inet_csk_ca(sk);
324
325	return max(tcp_sk(sk)->snd_cwnd, ca->last_max_cwnd);
326	}
327
328	static u32 bictcp_min_cwnd(struct sock *sk)
329	{
330	return tcp_sk(sk)->snd_ssthresh;
331	}
332
333	static void bictcp_state(struct sock *sk, u8 new_state)
334	{
335	if (new_state == TCP_CA_Loss)
336	bictcp_reset(inet_csk_ca(sk));
337	}
338
339	/* Track delayed acknowledgment ratio using sliding window
340	* ratio = (15*ratio + sample) / 16
341	*/
342	static void bictcp_acked(struct sock *sk, u32 cnt)
343	{
344	const struct inet_connection_sock *icsk = inet_csk(sk);
345
346	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
347	struct bictcp *ca = inet_csk_ca(sk);
348	cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
349	ca->delayed_ack += cnt;
350	}
351	}
352
353
354	static struct tcp_congestion_ops cubictcp = {
355	.init = bictcp_init,
356	.ssthresh = bictcp_recalc_ssthresh,
357	.cong_avoid = bictcp_cong_avoid,
358	.set_state = bictcp_state,
359	.undo_cwnd = bictcp_undo_cwnd,
360	.min_cwnd = bictcp_min_cwnd,
361	.pkts_acked = bictcp_acked,
362	.owner = THIS_MODULE,
363	.name = "cubic",
364	};
365
366	static int __init cubictcp_register(void)
367	{
368	BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
89b3d9aa SH	369
	370	/* Precompute a bunch of the scaling factors that are used per-packet
	371	* based on SRTT of 100ms
	372	*/
	373
	374	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);
	375
	376	cube_rtt_scale = (bic_scale << 3) / 10; /* 1024c/rtt /
	377
	378	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	379	* so K = cubic_root( (wmax-cwnd)*rtt/c )
	380	* the unit of K is bictcp_HZ=2^10, not HZ
	381	*
	382	* c = bic_scale >> 10
	383	* rtt = 100ms
	384	*
	385	* the following code has been designed and tested for
	386	* cwnd < 1 million packets
	387	* RTT < 100 seconds
	388	* HZ < 1,000,00 (corresponding to 10 nano-second)
	389	*/
	390
	391	/* 1/c * 2^2bictcp_HZ srtt */
	392	cube_factor = 1ull << (10+3BICTCP_HZ); / 2^40 */
	393
	394	/* divide by bic_scale and by constant Srtt (100ms) */
	395	do_div(cube_factor, bic_scale * 10);
	396
df3271f3 SH	397	return tcp_register_congestion_control(&cubictcp);
	398	}
	399
	400	static void __exit cubictcp_unregister(void)
	401	{
	402	tcp_unregister_congestion_control(&cubictcp);
	403	}
	404
	405	module_init(cubictcp_register);
	406	module_exit(cubictcp_unregister);
	407
	408	MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
	409	MODULE_LICENSE("GPL");
	410	MODULE_DESCRIPTION("CUBIC TCP");
	411	MODULE_VERSION("2.0");