xref: /freebsd/sys/netinet/cc/cc_cubic.c (revision 3a9fd8242b35884921dfc4e886f284a75870a536)
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/limits.h>
56 #include <sys/malloc.h>
57 #include <sys/module.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
62 
63 #include <net/vnet.h>
64 
65 #include <net/route.h>
66 #include <net/route/nhop.h>
67 
68 #include <netinet/in_pcb.h>
69 #include <netinet/tcp.h>
70 #include <netinet/tcp_seq.h>
71 #include <netinet/tcp_timer.h>
72 #include <netinet/tcp_var.h>
73 #include <netinet/cc/cc.h>
74 #include <netinet/cc/cc_cubic.h>
75 #include <netinet/cc/cc_module.h>
76 
77 static void	cubic_ack_received(struct cc_var *ccv, uint16_t type);
78 static void	cubic_cb_destroy(struct cc_var *ccv);
79 static int	cubic_cb_init(struct cc_var *ccv, void *ptr);
80 static void	cubic_cong_signal(struct cc_var *ccv, uint32_t type);
81 static void	cubic_conn_init(struct cc_var *ccv);
82 static int	cubic_mod_init(void);
83 static void	cubic_post_recovery(struct cc_var *ccv);
84 static void	cubic_record_rtt(struct cc_var *ccv);
85 static void	cubic_ssthresh_update(struct cc_var *ccv, uint32_t maxseg);
86 static void	cubic_after_idle(struct cc_var *ccv);
87 static size_t	cubic_data_sz(void);
88 
89 struct cubic {
90 	/* Cubic K in fixed point form with CUBIC_SHIFT worth of precision. */
91 	int64_t		K;
92 	/* Sum of RTT samples across an epoch in ticks. */
93 	int64_t		sum_rtt_ticks;
94 	/* cwnd at the most recent congestion event. */
95 	unsigned long	max_cwnd;
96 	/* cwnd at the previous congestion event. */
97 	unsigned long	prev_max_cwnd;
98 	/* A copy of prev_max_cwnd. Used for CC_RTO_ERR */
99 	unsigned long	prev_max_cwnd_cp;
100 	/* various flags */
101 	uint32_t	flags;
102 #define CUBICFLAG_CONG_EVENT	0x00000001	/* congestion experienced */
103 #define CUBICFLAG_IN_SLOWSTART	0x00000002	/* in slow start */
104 #define CUBICFLAG_IN_APPLIMIT	0x00000004	/* application limited */
105 #define CUBICFLAG_RTO_EVENT	0x00000008	/* RTO experienced */
106 	/* Minimum observed rtt in ticks. */
107 	int		min_rtt_ticks;
108 	/* Mean observed rtt between congestion epochs. */
109 	int		mean_rtt_ticks;
110 	/* ACKs since last congestion event. */
111 	int		epoch_ack_count;
112 	/* Timestamp (in ticks) of arriving in congestion avoidance from last
113 	 * congestion event.
114 	 */
115 	int		t_last_cong;
116 	/* Timestamp (in ticks) of a previous congestion event. Used for
117 	 * CC_RTO_ERR.
118 	 */
119 	int		t_last_cong_prev;
120 };
121 
122 struct cc_algo cubic_cc_algo = {
123 	.name = "cubic",
124 	.ack_received = cubic_ack_received,
125 	.cb_destroy = cubic_cb_destroy,
126 	.cb_init = cubic_cb_init,
127 	.cong_signal = cubic_cong_signal,
128 	.conn_init = cubic_conn_init,
129 	.mod_init = cubic_mod_init,
130 	.post_recovery = cubic_post_recovery,
131 	.after_idle = cubic_after_idle,
132 	.cc_data_sz = cubic_data_sz
133 };
134 
135 static void
136 cubic_ack_received(struct cc_var *ccv, uint16_t type)
137 {
138 	struct cubic *cubic_data;
139 	unsigned long w_tf, w_cubic_next;
140 	int ticks_since_cong;
141 
142 	cubic_data = ccv->cc_data;
143 	cubic_record_rtt(ccv);
144 
145 	/*
146 	 * For a regular ACK and we're not in cong/fast recovery and
147 	 * we're cwnd limited, always recalculate cwnd.
148 	 */
149 	if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) &&
150 	    (ccv->flags & CCF_CWND_LIMITED)) {
151 		 /* Use the logic in NewReno ack_received() for slow start. */
152 		if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) ||
153 		    cubic_data->min_rtt_ticks == TCPTV_SRTTBASE) {
154 			cubic_data->flags |= CUBICFLAG_IN_SLOWSTART;
155 			newreno_cc_ack_received(ccv, type);
156 		} else {
157 			if ((cubic_data->flags & CUBICFLAG_RTO_EVENT) &&
158 			    (cubic_data->flags & CUBICFLAG_IN_SLOWSTART)) {
159 				/* RFC8312 Section 4.7 */
160 				cubic_data->flags &= ~(CUBICFLAG_RTO_EVENT |
161 						       CUBICFLAG_IN_SLOWSTART);
162 				cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
163 				cubic_data->K = 0;
164 			} else if (cubic_data->flags & (CUBICFLAG_IN_SLOWSTART |
165 						 CUBICFLAG_IN_APPLIMIT)) {
166 				cubic_data->flags &= ~(CUBICFLAG_IN_SLOWSTART |
167 						       CUBICFLAG_IN_APPLIMIT);
168 				cubic_data->t_last_cong = ticks;
169 				cubic_data->K = cubic_k(cubic_data->max_cwnd /
170 							CCV(ccv, t_maxseg));
171 			}
172 			if ((ticks_since_cong =
173 			    ticks - cubic_data->t_last_cong) < 0) {
174 				/*
175 				 * dragging t_last_cong along
176 				 */
177 				ticks_since_cong = INT_MAX;
178 				cubic_data->t_last_cong = ticks - INT_MAX;
179 			}
180 			/*
181 			 * The mean RTT is used to best reflect the equations in
182 			 * the I-D. Using min_rtt in the tf_cwnd calculation
183 			 * causes w_tf to grow much faster than it should if the
184 			 * RTT is dominated by network buffering rather than
185 			 * propagation delay.
186 			 */
187 			w_tf = tf_cwnd(ticks_since_cong,
188 			    cubic_data->mean_rtt_ticks, cubic_data->max_cwnd,
189 			    CCV(ccv, t_maxseg));
190 
191 			w_cubic_next = cubic_cwnd(ticks_since_cong +
192 			    cubic_data->mean_rtt_ticks, cubic_data->max_cwnd,
193 			    CCV(ccv, t_maxseg), cubic_data->K);
194 
195 			ccv->flags &= ~CCF_ABC_SENTAWND;
196 
197 			if (w_cubic_next < w_tf) {
198 				/*
199 				 * TCP-friendly region, follow tf
200 				 * cwnd growth.
201 				 */
202 				if (CCV(ccv, snd_cwnd) < w_tf)
203 					CCV(ccv, snd_cwnd) = ulmin(w_tf, INT_MAX);
204 			} else if (CCV(ccv, snd_cwnd) < w_cubic_next) {
205 				/*
206 				 * Concave or convex region, follow CUBIC
207 				 * cwnd growth.
208 				 * Only update snd_cwnd, if it doesn't shrink.
209 				 */
210 				CCV(ccv, snd_cwnd) = ulmin(w_cubic_next,
211 				    INT_MAX);
212 			}
213 
214 			/*
215 			 * If we're not in slow start and we're probing for a
216 			 * new cwnd limit at the start of a connection
217 			 * (happens when hostcache has a relevant entry),
218 			 * keep updating our current estimate of the
219 			 * max_cwnd.
220 			 */
221 			if (((cubic_data->flags & CUBICFLAG_CONG_EVENT) == 0) &&
222 			    cubic_data->max_cwnd < CCV(ccv, snd_cwnd)) {
223 				cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
224 				cubic_data->K = cubic_k(cubic_data->max_cwnd /
225 				    CCV(ccv, t_maxseg));
226 			}
227 		}
228 	} else if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) &&
229 	    !(ccv->flags & CCF_CWND_LIMITED)) {
230 		cubic_data->flags |= CUBICFLAG_IN_APPLIMIT;
231 	}
232 }
233 
234 /*
235  * This is a Cubic specific implementation of after_idle.
236  *   - Reset cwnd by calling New Reno implementation of after_idle.
237  *   - Reset t_last_cong.
238  */
239 static void
240 cubic_after_idle(struct cc_var *ccv)
241 {
242 	struct cubic *cubic_data;
243 
244 	cubic_data = ccv->cc_data;
245 
246 	cubic_data->max_cwnd = ulmax(cubic_data->max_cwnd, CCV(ccv, snd_cwnd));
247 	cubic_data->K = cubic_k(cubic_data->max_cwnd / CCV(ccv, t_maxseg));
248 
249 	newreno_cc_after_idle(ccv);
250 	cubic_data->t_last_cong = ticks;
251 }
252 
253 static void
254 cubic_cb_destroy(struct cc_var *ccv)
255 {
256 	free(ccv->cc_data, M_CC_MEM);
257 }
258 
259 static size_t
260 cubic_data_sz(void)
261 {
262 	return (sizeof(struct cubic));
263 }
264 
265 static int
266 cubic_cb_init(struct cc_var *ccv, void *ptr)
267 {
268 	struct cubic *cubic_data;
269 
270 	INP_WLOCK_ASSERT(ccv->ccvc.tcp->t_inpcb);
271 	if (ptr == NULL) {
272 		cubic_data = malloc(sizeof(struct cubic), M_CC_MEM, M_NOWAIT|M_ZERO);
273 		if (cubic_data == NULL)
274 			return (ENOMEM);
275 	} else
276 		cubic_data = ptr;
277 
278 	/* Init some key variables with sensible defaults. */
279 	cubic_data->t_last_cong = ticks;
280 	cubic_data->min_rtt_ticks = TCPTV_SRTTBASE;
281 	cubic_data->mean_rtt_ticks = 1;
282 
283 	ccv->cc_data = cubic_data;
284 
285 	return (0);
286 }
287 
288 /*
289  * Perform any necessary tasks before we enter congestion recovery.
290  */
291 static void
292 cubic_cong_signal(struct cc_var *ccv, uint32_t type)
293 {
294 	struct cubic *cubic_data;
295 	u_int mss;
296 
297 	cubic_data = ccv->cc_data;
298 	mss = tcp_maxseg(ccv->ccvc.tcp);
299 
300 	switch (type) {
301 	case CC_NDUPACK:
302 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
303 			if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
304 				cubic_ssthresh_update(ccv, mss);
305 				cubic_data->flags |= CUBICFLAG_CONG_EVENT;
306 				cubic_data->t_last_cong = ticks;
307 				cubic_data->K = cubic_k(cubic_data->max_cwnd / mss);
308 			}
309 			ENTER_RECOVERY(CCV(ccv, t_flags));
310 		}
311 		break;
312 
313 	case CC_ECN:
314 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
315 			cubic_ssthresh_update(ccv, mss);
316 			cubic_data->flags |= CUBICFLAG_CONG_EVENT;
317 			cubic_data->t_last_cong = ticks;
318 			cubic_data->K = cubic_k(cubic_data->max_cwnd / mss);
319 			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
320 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
321 		}
322 		break;
323 
324 	case CC_RTO:
325 		/* RFC8312 Section 4.7 */
326 		if (CCV(ccv, t_rxtshift) == 1) {
327 			cubic_data->t_last_cong_prev = cubic_data->t_last_cong;
328 			cubic_data->prev_max_cwnd_cp = cubic_data->prev_max_cwnd;
329 		}
330 		cubic_data->flags |= CUBICFLAG_CONG_EVENT | CUBICFLAG_RTO_EVENT;
331 		cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
332 		CCV(ccv, snd_ssthresh) = ((uint64_t)CCV(ccv, snd_cwnd) *
333 					  CUBIC_BETA) >> CUBIC_SHIFT;
334 		CCV(ccv, snd_cwnd) = mss;
335 		break;
336 
337 	case CC_RTO_ERR:
338 		cubic_data->flags &= ~(CUBICFLAG_CONG_EVENT | CUBICFLAG_RTO_EVENT);
339 		cubic_data->max_cwnd = cubic_data->prev_max_cwnd;
340 		cubic_data->prev_max_cwnd = cubic_data->prev_max_cwnd_cp;
341 		cubic_data->t_last_cong = cubic_data->t_last_cong_prev;
342 		cubic_data->K = cubic_k(cubic_data->max_cwnd / mss);
343 		break;
344 	}
345 }
346 
347 static void
348 cubic_conn_init(struct cc_var *ccv)
349 {
350 	struct cubic *cubic_data;
351 
352 	cubic_data = ccv->cc_data;
353 
354 	/*
355 	 * Ensure we have a sane initial value for max_cwnd recorded. Without
356 	 * this here bad things happen when entries from the TCP hostcache
357 	 * get used.
358 	 */
359 	cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
360 }
361 
362 static int
363 cubic_mod_init(void)
364 {
365 	return (0);
366 }
367 
368 /*
369  * Perform any necessary tasks before we exit congestion recovery.
370  */
371 static void
372 cubic_post_recovery(struct cc_var *ccv)
373 {
374 	struct cubic *cubic_data;
375 	int pipe;
376 
377 	cubic_data = ccv->cc_data;
378 	pipe = 0;
379 
380 	if (IN_FASTRECOVERY(CCV(ccv, t_flags))) {
381 		/*
382 		 * If inflight data is less than ssthresh, set cwnd
383 		 * conservatively to avoid a burst of data, as suggested in
384 		 * the NewReno RFC. Otherwise, use the CUBIC method.
385 		 *
386 		 * XXXLAS: Find a way to do this without needing curack
387 		 */
388 		if (V_tcp_do_newsack)
389 			pipe = tcp_compute_pipe(ccv->ccvc.tcp);
390 		else
391 			pipe = CCV(ccv, snd_max) - ccv->curack;
392 
393 		if (pipe < CCV(ccv, snd_ssthresh))
394 			/*
395 			 * Ensure that cwnd does not collapse to 1 MSS under
396 			 * adverse conditions. Implements RFC6582
397 			 */
398 			CCV(ccv, snd_cwnd) = max(pipe, CCV(ccv, t_maxseg)) +
399 			    CCV(ccv, t_maxseg);
400 		else
401 			/* Update cwnd based on beta and adjusted max_cwnd. */
402 			CCV(ccv, snd_cwnd) = max(((uint64_t)cubic_data->max_cwnd *
403 			    CUBIC_BETA) >> CUBIC_SHIFT,
404 			    2 * CCV(ccv, t_maxseg));
405 	}
406 
407 	/* Calculate the average RTT between congestion epochs. */
408 	if (cubic_data->epoch_ack_count > 0 &&
409 	    cubic_data->sum_rtt_ticks >= cubic_data->epoch_ack_count) {
410 		cubic_data->mean_rtt_ticks = (int)(cubic_data->sum_rtt_ticks /
411 		    cubic_data->epoch_ack_count);
412 	}
413 
414 	cubic_data->epoch_ack_count = 0;
415 	cubic_data->sum_rtt_ticks = 0;
416 }
417 
418 /*
419  * Record the min RTT and sum samples for the epoch average RTT calculation.
420  */
421 static void
422 cubic_record_rtt(struct cc_var *ccv)
423 {
424 	struct cubic *cubic_data;
425 	int t_srtt_ticks;
426 
427 	/* Ignore srtt until a min number of samples have been taken. */
428 	if (CCV(ccv, t_rttupdated) >= CUBIC_MIN_RTT_SAMPLES) {
429 		cubic_data = ccv->cc_data;
430 		t_srtt_ticks = CCV(ccv, t_srtt) / TCP_RTT_SCALE;
431 
432 		/*
433 		 * Record the current SRTT as our minrtt if it's the smallest
434 		 * we've seen or minrtt is currently equal to its initialised
435 		 * value.
436 		 *
437 		 * XXXLAS: Should there be some hysteresis for minrtt?
438 		 */
439 		if ((t_srtt_ticks < cubic_data->min_rtt_ticks ||
440 		    cubic_data->min_rtt_ticks == TCPTV_SRTTBASE)) {
441 			cubic_data->min_rtt_ticks = max(1, t_srtt_ticks);
442 
443 			/*
444 			 * If the connection is within its first congestion
445 			 * epoch, ensure we prime mean_rtt_ticks with a
446 			 * reasonable value until the epoch average RTT is
447 			 * calculated in cubic_post_recovery().
448 			 */
449 			if (cubic_data->min_rtt_ticks >
450 			    cubic_data->mean_rtt_ticks)
451 				cubic_data->mean_rtt_ticks =
452 				    cubic_data->min_rtt_ticks;
453 		}
454 
455 		/* Sum samples for epoch average RTT calculation. */
456 		cubic_data->sum_rtt_ticks += t_srtt_ticks;
457 		cubic_data->epoch_ack_count++;
458 	}
459 }
460 
461 /*
462  * Update the ssthresh in the event of congestion.
463  */
464 static void
465 cubic_ssthresh_update(struct cc_var *ccv, uint32_t maxseg)
466 {
467 	struct cubic *cubic_data;
468 	uint32_t ssthresh;
469 	uint32_t cwnd;
470 
471 	cubic_data = ccv->cc_data;
472 	cwnd = CCV(ccv, snd_cwnd);
473 
474 	/* Fast convergence heuristic. */
475 	if (cwnd < cubic_data->max_cwnd) {
476 		cwnd = ((uint64_t)cwnd * CUBIC_FC_FACTOR) >> CUBIC_SHIFT;
477 	}
478 	cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
479 	cubic_data->max_cwnd = cwnd;
480 
481 	/*
482 	 * On the first congestion event, set ssthresh to cwnd * 0.5
483 	 * and reduce max_cwnd to cwnd * beta. This aligns the cubic concave
484 	 * region appropriately. On subsequent congestion events, set
485 	 * ssthresh to cwnd * beta.
486 	 */
487 	if ((cubic_data->flags & CUBICFLAG_CONG_EVENT) == 0) {
488 		ssthresh = cwnd >> 1;
489 		cubic_data->max_cwnd = ((uint64_t)cwnd *
490 		    CUBIC_BETA) >> CUBIC_SHIFT;
491 	} else {
492 		ssthresh = ((uint64_t)cwnd *
493 		    CUBIC_BETA) >> CUBIC_SHIFT;
494 	}
495 	CCV(ccv, snd_ssthresh) = max(ssthresh, 2 * maxseg);
496 }
497 
498 DECLARE_CC_MODULE(cubic, &cubic_cc_algo);
499 MODULE_VERSION(cubic, 2);
500