xref: /freebsd/sys/netinet/cc/cc_cubic.c (revision 74ca7bf1d4c7173d5575ba168bc4b5f6d181ff5a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008-2010 Lawrence Stewart <lstewart@freebsd.org>
5  * Copyright (c) 2010 The FreeBSD Foundation
6  * All rights reserved.
7  *
8  * This software was developed by Lawrence Stewart while studying at the Centre
9  * for Advanced Internet Architectures, Swinburne University of Technology, made
10  * possible in part by a grant from the Cisco University Research Program Fund
11  * at Community Foundation Silicon Valley.
12  *
13  * Portions of this software were developed at the Centre for Advanced
14  * Internet Architectures, Swinburne University of Technology, Melbourne,
15  * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 /*
40  * An implementation of the CUBIC congestion control algorithm for FreeBSD,
41  * based on the Internet Draft "draft-rhee-tcpm-cubic-02" by Rhee, Xu and Ha.
42  * Originally released as part of the NewTCP research project at Swinburne
43  * University of Technology's Centre for Advanced Internet Architectures,
44  * Melbourne, Australia, which was made possible in part by a grant from the
45  * Cisco University Research Program Fund at Community Foundation Silicon
46  * Valley. More details are available at:
47  *   http://caia.swin.edu.au/urp/newtcp/
48  */
49 
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52 
53 #include <sys/param.h>
54 #include <sys/kernel.h>
55 #include <sys/malloc.h>
56 #include <sys/module.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 
62 #include <net/vnet.h>
63 
64 #include <netinet/tcp.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/tcp_timer.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet/cc/cc.h>
69 #include <netinet/cc/cc_cubic.h>
70 #include <netinet/cc/cc_module.h>
71 
72 static void	cubic_ack_received(struct cc_var *ccv, uint16_t type);
73 static void	cubic_cb_destroy(struct cc_var *ccv);
74 static int	cubic_cb_init(struct cc_var *ccv);
75 static void	cubic_cong_signal(struct cc_var *ccv, uint32_t type);
76 static void	cubic_conn_init(struct cc_var *ccv);
77 static int	cubic_mod_init(void);
78 static void	cubic_post_recovery(struct cc_var *ccv);
79 static void	cubic_record_rtt(struct cc_var *ccv);
80 static void	cubic_ssthresh_update(struct cc_var *ccv);
81 
82 struct cubic {
83 	/* Cubic K in fixed point form with CUBIC_SHIFT worth of precision. */
84 	int64_t		K;
85 	/* Sum of RTT samples across an epoch in ticks. */
86 	int64_t		sum_rtt_ticks;
87 	/* cwnd at the most recent congestion event. */
88 	unsigned long	max_cwnd;
89 	/* cwnd at the previous congestion event. */
90 	unsigned long	prev_max_cwnd;
91 	/* Cached value for t_maxseg when K was computed */
92 	uint32_t        k_maxseg;
93 	/* Number of congestion events. */
94 	uint32_t	num_cong_events;
95 	/* Minimum observed rtt in ticks. */
96 	int		min_rtt_ticks;
97 	/* Mean observed rtt between congestion epochs. */
98 	int		mean_rtt_ticks;
99 	/* ACKs since last congestion event. */
100 	int		epoch_ack_count;
101 	/* Time of last congestion event in ticks. */
102 	int		t_last_cong;
103 };
104 
105 static MALLOC_DEFINE(M_CUBIC, "cubic data",
106     "Per connection data required for the CUBIC congestion control algorithm");
107 
108 struct cc_algo cubic_cc_algo = {
109 	.name = "cubic",
110 	.ack_received = cubic_ack_received,
111 	.cb_destroy = cubic_cb_destroy,
112 	.cb_init = cubic_cb_init,
113 	.cong_signal = cubic_cong_signal,
114 	.conn_init = cubic_conn_init,
115 	.mod_init = cubic_mod_init,
116 	.post_recovery = cubic_post_recovery,
117 };
118 
119 static void
120 cubic_ack_received(struct cc_var *ccv, uint16_t type)
121 {
122 	struct cubic *cubic_data;
123 	unsigned long w_tf, w_cubic_next;
124 	int ticks_since_cong;
125 
126 	cubic_data = ccv->cc_data;
127 	cubic_record_rtt(ccv);
128 
129 	if (ccv->flags & CCF_MAX_CWND)
130 		return;
131 
132 	/*
133 	 * Regular ACK and we're not in cong/fast recovery and we're cwnd
134 	 * limited and we're either not doing ABC or are slow starting or are
135 	 * doing ABC and we've sent a cwnd's worth of bytes.
136 	 */
137 	if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) &&
138 	    (ccv->flags & CCF_CWND_LIMITED) && (!V_tcp_do_rfc3465 ||
139 	    CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) ||
140 	    (V_tcp_do_rfc3465 && ccv->flags & CCF_ABC_SENTAWND))) {
141 		 /* Use the logic in NewReno ack_received() for slow start. */
142 		if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) ||
143 		    cubic_data->min_rtt_ticks == TCPTV_SRTTBASE)
144 			newreno_cc_algo.ack_received(ccv, type);
145 		else {
146 			ticks_since_cong = ticks - cubic_data->t_last_cong;
147 
148 			/*
149 			 * The mean RTT is used to best reflect the equations in
150 			 * the I-D. Using min_rtt in the tf_cwnd calculation
151 			 * causes w_tf to grow much faster than it should if the
152 			 * RTT is dominated by network buffering rather than
153 			 * propagation delay.
154 			 */
155 			w_tf = tf_cwnd(ticks_since_cong,
156 			    cubic_data->mean_rtt_ticks, cubic_data->max_cwnd,
157 			    CCV(ccv, t_maxseg));
158 
159 			if (ccv->flags & CCF_CHG_MAX_CWND || cubic_data->k_maxseg != CCV(ccv, t_maxseg)) {
160 				cubic_data->K = cubic_k(cubic_data->max_cwnd / CCV(ccv, t_maxseg));
161 				cubic_data->k_maxseg = CCV(ccv, t_maxseg);
162 				ccv->flags &= ~(CCF_MAX_CWND|CCF_CHG_MAX_CWND);
163 			}
164 
165 			w_cubic_next = cubic_cwnd(ticks_since_cong +
166 			    cubic_data->mean_rtt_ticks, cubic_data->max_cwnd,
167 			    CCV(ccv, t_maxseg), cubic_data->K);
168 
169 			ccv->flags &= ~CCF_ABC_SENTAWND;
170 
171 			if (w_cubic_next < w_tf)
172 				/*
173 				 * TCP-friendly region, follow tf
174 				 * cwnd growth.
175 				 */
176 				CCV(ccv, snd_cwnd) = ulmin(w_tf, TCP_MAXWIN << CCV(ccv, snd_scale));
177 
178 			else if (CCV(ccv, snd_cwnd) < w_cubic_next) {
179 				/*
180 				 * Concave or convex region, follow CUBIC
181 				 * cwnd growth.
182 				 */
183 				if (w_cubic_next >= TCP_MAXWIN << CCV(ccv, snd_scale)) {
184 					w_cubic_next = TCP_MAXWIN << CCV(ccv, snd_scale);
185 					ccv->flags |= CCF_MAX_CWND;
186 				}
187 				w_cubic_next = ulmin(w_cubic_next, TCP_MAXWIN << CCV(ccv, snd_scale));
188 				if (V_tcp_do_rfc3465)
189 					CCV(ccv, snd_cwnd) = w_cubic_next;
190 				else
191 					CCV(ccv, snd_cwnd) += ((w_cubic_next -
192 					    CCV(ccv, snd_cwnd)) *
193 					    CCV(ccv, t_maxseg)) /
194 					    CCV(ccv, snd_cwnd);
195 			}
196 
197 			/*
198 			 * If we're not in slow start and we're probing for a
199 			 * new cwnd limit at the start of a connection
200 			 * (happens when hostcache has a relevant entry),
201 			 * keep updating our current estimate of the
202 			 * max_cwnd.
203 			 */
204 			if (cubic_data->num_cong_events == 0 &&
205 			    cubic_data->max_cwnd < CCV(ccv, snd_cwnd)) {
206 				cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
207 				ccv->flags |= CCF_CHG_MAX_CWND;
208 			}
209 		}
210 	}
211 }
212 
213 static void
214 cubic_cb_destroy(struct cc_var *ccv)
215 {
216 
217 	if (ccv->cc_data != NULL)
218 		free(ccv->cc_data, M_CUBIC);
219 }
220 
221 static int
222 cubic_cb_init(struct cc_var *ccv)
223 {
224 	struct cubic *cubic_data;
225 
226 	cubic_data = malloc(sizeof(struct cubic), M_CUBIC, M_NOWAIT|M_ZERO);
227 
228 	if (cubic_data == NULL)
229 		return (ENOMEM);
230 
231 	/* Init some key variables with sensible defaults. */
232 	cubic_data->t_last_cong = ticks;
233 	cubic_data->min_rtt_ticks = TCPTV_SRTTBASE;
234 	cubic_data->mean_rtt_ticks = 1;
235 
236 	ccv->cc_data = cubic_data;
237 
238 	return (0);
239 }
240 
241 /*
242  * Perform any necessary tasks before we enter congestion recovery.
243  */
244 static void
245 cubic_cong_signal(struct cc_var *ccv, uint32_t type)
246 {
247 	struct cubic *cubic_data;
248 
249 	cubic_data = ccv->cc_data;
250 
251 	switch (type) {
252 	case CC_NDUPACK:
253 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
254 			if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
255 				cubic_ssthresh_update(ccv);
256 				cubic_data->num_cong_events++;
257 				cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
258 				cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
259 				ccv->flags |= CCF_CHG_MAX_CWND;
260 			}
261 			ENTER_RECOVERY(CCV(ccv, t_flags));
262 		}
263 		break;
264 
265 	case CC_ECN:
266 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
267 			cubic_ssthresh_update(ccv);
268 			cubic_data->num_cong_events++;
269 			cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
270 			cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
271 			cubic_data->t_last_cong = ticks;
272 			ccv->flags |= CCF_CHG_MAX_CWND;
273 			ccv->flags &= ~CCF_MAX_CWND;
274 			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
275 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
276 		}
277 		break;
278 
279 	case CC_RTO:
280 		/*
281 		 * Grab the current time and record it so we know when the
282 		 * most recent congestion event was. Only record it when the
283 		 * timeout has fired more than once, as there is a reasonable
284 		 * chance the first one is a false alarm and may not indicate
285 		 * congestion.
286 		 */
287 		if (CCV(ccv, t_rxtshift) >= 2) {
288 			cubic_data->num_cong_events++;
289 			cubic_data->t_last_cong = ticks;
290 			ccv->flags &= ~CCF_MAX_CWND;
291 		}
292 		break;
293 	}
294 }
295 
296 static void
297 cubic_conn_init(struct cc_var *ccv)
298 {
299 	struct cubic *cubic_data;
300 
301 	cubic_data = ccv->cc_data;
302 
303 	/*
304 	 * Ensure we have a sane initial value for max_cwnd recorded. Without
305 	 * this here bad things happen when entries from the TCP hostcache
306 	 * get used.
307 	 */
308 	cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
309 	ccv->flags |= CCF_CHG_MAX_CWND;
310 }
311 
312 static int
313 cubic_mod_init(void)
314 {
315 
316 	cubic_cc_algo.after_idle = newreno_cc_algo.after_idle;
317 
318 	return (0);
319 }
320 
321 /*
322  * Perform any necessary tasks before we exit congestion recovery.
323  */
324 static void
325 cubic_post_recovery(struct cc_var *ccv)
326 {
327 	struct cubic *cubic_data;
328 	int pipe;
329 
330 	cubic_data = ccv->cc_data;
331 	pipe = 0;
332 
333 	/* Fast convergence heuristic. */
334 	if (cubic_data->max_cwnd < cubic_data->prev_max_cwnd) {
335 		cubic_data->max_cwnd = (cubic_data->max_cwnd * CUBIC_FC_FACTOR)
336 		    >> CUBIC_SHIFT;
337 		ccv->flags |= CCF_CHG_MAX_CWND;
338 	}
339 
340 	if (IN_FASTRECOVERY(CCV(ccv, t_flags))) {
341 		/*
342 		 * If inflight data is less than ssthresh, set cwnd
343 		 * conservatively to avoid a burst of data, as suggested in
344 		 * the NewReno RFC. Otherwise, use the CUBIC method.
345 		 *
346 		 * XXXLAS: Find a way to do this without needing curack
347 		 */
348 		if (V_tcp_do_rfc6675_pipe)
349 			pipe = tcp_compute_pipe(ccv->ccvc.tcp);
350 		else
351 			pipe = CCV(ccv, snd_max) - ccv->curack;
352 
353 		if (pipe < CCV(ccv, snd_ssthresh))
354 			CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg);
355 		else
356 			/* Update cwnd based on beta and adjusted max_cwnd. */
357 			CCV(ccv, snd_cwnd) = max(1, ((CUBIC_BETA *
358 			    cubic_data->max_cwnd) >> CUBIC_SHIFT));
359 	}
360 	cubic_data->t_last_cong = ticks;
361 	ccv->flags &= ~CCF_MAX_CWND;
362 
363 	/* Calculate the average RTT between congestion epochs. */
364 	if (cubic_data->epoch_ack_count > 0 &&
365 	    cubic_data->sum_rtt_ticks >= cubic_data->epoch_ack_count) {
366 		cubic_data->mean_rtt_ticks = (int)(cubic_data->sum_rtt_ticks /
367 		    cubic_data->epoch_ack_count);
368 	}
369 
370 	cubic_data->epoch_ack_count = 0;
371 	cubic_data->sum_rtt_ticks = 0;
372 }
373 
374 /*
375  * Record the min RTT and sum samples for the epoch average RTT calculation.
376  */
377 static void
378 cubic_record_rtt(struct cc_var *ccv)
379 {
380 	struct cubic *cubic_data;
381 	int t_srtt_ticks;
382 
383 	/* Ignore srtt until a min number of samples have been taken. */
384 	if (CCV(ccv, t_rttupdated) >= CUBIC_MIN_RTT_SAMPLES) {
385 		cubic_data = ccv->cc_data;
386 		t_srtt_ticks = CCV(ccv, t_srtt) / TCP_RTT_SCALE;
387 
388 		/*
389 		 * Record the current SRTT as our minrtt if it's the smallest
390 		 * we've seen or minrtt is currently equal to its initialised
391 		 * value.
392 		 *
393 		 * XXXLAS: Should there be some hysteresis for minrtt?
394 		 */
395 		if ((t_srtt_ticks < cubic_data->min_rtt_ticks ||
396 		    cubic_data->min_rtt_ticks == TCPTV_SRTTBASE)) {
397 			cubic_data->min_rtt_ticks = max(1, t_srtt_ticks);
398 
399 			/*
400 			 * If the connection is within its first congestion
401 			 * epoch, ensure we prime mean_rtt_ticks with a
402 			 * reasonable value until the epoch average RTT is
403 			 * calculated in cubic_post_recovery().
404 			 */
405 			if (cubic_data->min_rtt_ticks >
406 			    cubic_data->mean_rtt_ticks)
407 				cubic_data->mean_rtt_ticks =
408 				    cubic_data->min_rtt_ticks;
409 		}
410 
411 		/* Sum samples for epoch average RTT calculation. */
412 		cubic_data->sum_rtt_ticks += t_srtt_ticks;
413 		cubic_data->epoch_ack_count++;
414 	}
415 }
416 
417 /*
418  * Update the ssthresh in the event of congestion.
419  */
420 static void
421 cubic_ssthresh_update(struct cc_var *ccv)
422 {
423 	struct cubic *cubic_data;
424 
425 	cubic_data = ccv->cc_data;
426 
427 	/*
428 	 * On the first congestion event, set ssthresh to cwnd * 0.5, on
429 	 * subsequent congestion events, set it to cwnd * beta.
430 	 */
431 	if (cubic_data->num_cong_events == 0)
432 		CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd) >> 1;
433 	else
434 		CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) *
435 		    CUBIC_BETA) >> CUBIC_SHIFT;
436 }
437 
438 
439 DECLARE_CC_MODULE(cubic, &cubic_cc_algo);
440