xref: /netbsd-src/sys/netinet/tcp_congctl.c (revision fc4f42693f9b1c31f39f9cf50af1bf2010325808)
1 /*	$NetBSD: tcp_congctl.c,v 1.24 2018/03/29 07:46:43 maxv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
9  * Facility, NASA Ames Research Center.
10  * This code is derived from software contributed to The NetBSD Foundation
11  * by Charles M. Hannum.
12  * This code is derived from software contributed to The NetBSD Foundation
13  * by Rui Paulo.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
26  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34  * POSSIBILITY OF SUCH DAMAGE.
35  */
36 
37 /*
38  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
39  * All rights reserved.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the project nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  */
65 
66 /*
67  *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
68  *
69  * NRL grants permission for redistribution and use in source and binary
70  * forms, with or without modification, of the software and documentation
71  * created at NRL provided that the following conditions are met:
72  *
73  * 1. Redistributions of source code must retain the above copyright
74  *    notice, this list of conditions and the following disclaimer.
75  * 2. Redistributions in binary form must reproduce the above copyright
76  *    notice, this list of conditions and the following disclaimer in the
77  *    documentation and/or other materials provided with the distribution.
78  * 3. All advertising materials mentioning features or use of this software
79  *    must display the following acknowledgements:
80  *      This product includes software developed by the University of
81  *      California, Berkeley and its contributors.
82  *      This product includes software developed at the Information
83  *      Technology Division, US Naval Research Laboratory.
84  * 4. Neither the name of the NRL nor the names of its contributors
85  *    may be used to endorse or promote products derived from this software
86  *    without specific prior written permission.
87  *
88  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
89  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
90  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
91  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
92  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
93  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
94  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
95  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
96  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
97  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
98  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
99  *
100  * The views and conclusions contained in the software and documentation
101  * are those of the authors and should not be interpreted as representing
102  * official policies, either expressed or implied, of the US Naval
103  * Research Laboratory (NRL).
104  */
105 
106 /*
107  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
108  *	The Regents of the University of California.  All rights reserved.
109  *
110  * Redistribution and use in source and binary forms, with or without
111  * modification, are permitted provided that the following conditions
112  * are met:
113  * 1. Redistributions of source code must retain the above copyright
114  *    notice, this list of conditions and the following disclaimer.
115  * 2. Redistributions in binary form must reproduce the above copyright
116  *    notice, this list of conditions and the following disclaimer in the
117  *    documentation and/or other materials provided with the distribution.
118  * 3. Neither the name of the University nor the names of its contributors
119  *    may be used to endorse or promote products derived from this software
120  *    without specific prior written permission.
121  *
122  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
123  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
124  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
125  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
126  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
127  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
128  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
129  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
130  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
131  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
132  * SUCH DAMAGE.
133  *
134  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
135  */
136 
137 #include <sys/cdefs.h>
138 __KERNEL_RCSID(0, "$NetBSD: tcp_congctl.c,v 1.24 2018/03/29 07:46:43 maxv Exp $");
139 
140 #ifdef _KERNEL_OPT
141 #include "opt_inet.h"
142 #include "opt_tcp_debug.h"
143 #include "opt_tcp_congctl.h"
144 #endif
145 
146 #include <sys/param.h>
147 #include <sys/systm.h>
148 #include <sys/malloc.h>
149 #include <sys/mbuf.h>
150 #include <sys/protosw.h>
151 #include <sys/socket.h>
152 #include <sys/socketvar.h>
153 #include <sys/errno.h>
154 #include <sys/syslog.h>
155 #include <sys/pool.h>
156 #include <sys/domain.h>
157 #include <sys/kernel.h>
158 #include <sys/mutex.h>
159 
160 #include <net/if.h>
161 
162 #include <netinet/in.h>
163 #include <netinet/in_systm.h>
164 #include <netinet/ip.h>
165 #include <netinet/in_pcb.h>
166 #include <netinet/in_var.h>
167 #include <netinet/ip_var.h>
168 
169 #ifdef INET6
170 #include <netinet/ip6.h>
171 #include <netinet6/ip6_var.h>
172 #include <netinet6/in6_pcb.h>
173 #include <netinet6/ip6_var.h>
174 #include <netinet6/in6_var.h>
175 #include <netinet/icmp6.h>
176 #endif
177 
178 #include <netinet/tcp.h>
179 #include <netinet/tcp_fsm.h>
180 #include <netinet/tcp_seq.h>
181 #include <netinet/tcp_timer.h>
182 #include <netinet/tcp_var.h>
183 #include <netinet/tcpip.h>
184 #include <netinet/tcp_congctl.h>
185 #ifdef TCP_DEBUG
186 #include <netinet/tcp_debug.h>
187 #endif
188 
189 /*
190  * TODO:
191  *   consider separating the actual implementations in another file.
192  */
193 
194 static void tcp_common_congestion_exp(struct tcpcb *, int, int);
195 
196 static int  tcp_reno_do_fast_retransmit(struct tcpcb *, const struct tcphdr *);
197 static int  tcp_reno_fast_retransmit(struct tcpcb *, const struct tcphdr *);
198 static void tcp_reno_slow_retransmit(struct tcpcb *);
199 static void tcp_reno_fast_retransmit_newack(struct tcpcb *,
200     const struct tcphdr *);
201 static void tcp_reno_newack(struct tcpcb *, const struct tcphdr *);
202 static void tcp_reno_congestion_exp(struct tcpcb *tp);
203 
204 static int  tcp_newreno_fast_retransmit(struct tcpcb *, const struct tcphdr *);
205 static void tcp_newreno_fast_retransmit_newack(struct tcpcb *,
206 	const struct tcphdr *);
207 static void tcp_newreno_newack(struct tcpcb *, const struct tcphdr *);
208 
209 static int tcp_cubic_fast_retransmit(struct tcpcb *, const struct tcphdr *);
210 static void tcp_cubic_slow_retransmit(struct tcpcb *tp);
211 static void tcp_cubic_newack(struct tcpcb *, const struct tcphdr *);
212 static void tcp_cubic_congestion_exp(struct tcpcb *);
213 
214 static void tcp_congctl_fillnames(void);
215 
216 extern int tcprexmtthresh;
217 
218 MALLOC_DEFINE(M_TCPCONGCTL, "tcpcongctl", "TCP congestion control structures");
219 
220 /* currently selected global congestion control */
221 char tcp_congctl_global_name[TCPCC_MAXLEN];
222 
223 /* available global congestion control algorithms */
224 char tcp_congctl_avail[10 * TCPCC_MAXLEN];
225 
226 /*
227  * Used to list the available congestion control algorithms.
228  */
229 TAILQ_HEAD(, tcp_congctlent) tcp_congctlhd =
230     TAILQ_HEAD_INITIALIZER(tcp_congctlhd);
231 
232 static struct tcp_congctlent * tcp_congctl_global;
233 
234 static kmutex_t tcp_congctl_mtx;
235 
236 void
237 tcp_congctl_init(void)
238 {
239 	int r __diagused;
240 
241 	mutex_init(&tcp_congctl_mtx, MUTEX_DEFAULT, IPL_NONE);
242 
243 	/* Base algorithms. */
244 	r = tcp_congctl_register("reno", &tcp_reno_ctl);
245 	KASSERT(r == 0);
246 	r = tcp_congctl_register("newreno", &tcp_newreno_ctl);
247 	KASSERT(r == 0);
248 	r = tcp_congctl_register("cubic", &tcp_cubic_ctl);
249 	KASSERT(r == 0);
250 
251 	/* NewReno is the default. */
252 #ifndef TCP_CONGCTL_DEFAULT
253 #define TCP_CONGCTL_DEFAULT "newreno"
254 #endif
255 
256 	r = tcp_congctl_select(NULL, TCP_CONGCTL_DEFAULT);
257 	KASSERT(r == 0);
258 }
259 
260 /*
261  * Register a congestion algorithm and select it if we have none.
262  */
263 int
264 tcp_congctl_register(const char *name, const struct tcp_congctl *tcc)
265 {
266 	struct tcp_congctlent *ntcc, *tccp;
267 
268 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent)
269 		if (!strcmp(name, tccp->congctl_name)) {
270 			/* name already registered */
271 			return EEXIST;
272 		}
273 
274 	ntcc = malloc(sizeof(*ntcc), M_TCPCONGCTL, M_WAITOK|M_ZERO);
275 
276 	strlcpy(ntcc->congctl_name, name, sizeof(ntcc->congctl_name) - 1);
277 	ntcc->congctl_ctl = tcc;
278 
279 	TAILQ_INSERT_TAIL(&tcp_congctlhd, ntcc, congctl_ent);
280 	tcp_congctl_fillnames();
281 
282 	if (TAILQ_FIRST(&tcp_congctlhd) == ntcc)
283 		tcp_congctl_select(NULL, name);
284 
285 	return 0;
286 }
287 
288 int
289 tcp_congctl_unregister(const char *name)
290 {
291 	struct tcp_congctlent *tccp, *rtccp;
292 	unsigned int size;
293 
294 	rtccp = NULL;
295 	size = 0;
296 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
297 		if (!strcmp(name, tccp->congctl_name))
298 			rtccp = tccp;
299 		size++;
300 	}
301 
302 	if (!rtccp)
303 		return ENOENT;
304 
305 	if (size <= 1 || tcp_congctl_global == rtccp || rtccp->congctl_refcnt)
306 		return EBUSY;
307 
308 	TAILQ_REMOVE(&tcp_congctlhd, rtccp, congctl_ent);
309 	free(rtccp, M_TCPCONGCTL);
310 	tcp_congctl_fillnames();
311 
312 	return 0;
313 }
314 
315 /*
316  * Select a congestion algorithm by name.
317  */
318 int
319 tcp_congctl_select(struct tcpcb *tp, const char *name)
320 {
321 	struct tcp_congctlent *tccp, *old_tccp, *new_tccp;
322 	bool old_found, new_found;
323 
324 	KASSERT(name);
325 
326 	old_found = (tp == NULL || tp->t_congctl == NULL);
327 	old_tccp = NULL;
328 	new_found = false;
329 	new_tccp = NULL;
330 
331 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
332 		if (!old_found && tccp->congctl_ctl == tp->t_congctl) {
333 			old_tccp = tccp;
334 			old_found = true;
335 		}
336 
337 		if (!new_found && !strcmp(name, tccp->congctl_name)) {
338 			new_tccp = tccp;
339 			new_found = true;
340 		}
341 
342 		if (new_found && old_found) {
343 			if (tp) {
344 				mutex_enter(&tcp_congctl_mtx);
345 				if (old_tccp)
346 					old_tccp->congctl_refcnt--;
347 				tp->t_congctl = new_tccp->congctl_ctl;
348 				new_tccp->congctl_refcnt++;
349 				mutex_exit(&tcp_congctl_mtx);
350 			} else {
351 				tcp_congctl_global = new_tccp;
352 				strlcpy(tcp_congctl_global_name,
353 				    new_tccp->congctl_name,
354 				    sizeof(tcp_congctl_global_name) - 1);
355 			}
356 			return 0;
357 		}
358 	}
359 
360 	return EINVAL;
361 }
362 
363 void
364 tcp_congctl_release(struct tcpcb *tp)
365 {
366 	struct tcp_congctlent *tccp;
367 
368 	KASSERT(tp->t_congctl);
369 
370 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
371 		if (tccp->congctl_ctl == tp->t_congctl) {
372 			tccp->congctl_refcnt--;
373 			return;
374 		}
375 	}
376 }
377 
378 /*
379  * Returns the name of a congestion algorithm.
380  */
381 const char *
382 tcp_congctl_bystruct(const struct tcp_congctl *tcc)
383 {
384 	struct tcp_congctlent *tccp;
385 
386 	KASSERT(tcc);
387 
388 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent)
389 		if (tccp->congctl_ctl == tcc)
390 			return tccp->congctl_name;
391 
392 	return NULL;
393 }
394 
395 static void
396 tcp_congctl_fillnames(void)
397 {
398 	struct tcp_congctlent *tccp;
399 	const char *delim = " ";
400 
401 	tcp_congctl_avail[0] = '\0';
402 	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
403 		strlcat(tcp_congctl_avail, tccp->congctl_name,
404 		    sizeof(tcp_congctl_avail) - 1);
405 		if (TAILQ_NEXT(tccp, congctl_ent))
406 			strlcat(tcp_congctl_avail, delim,
407 			    sizeof(tcp_congctl_avail) - 1);
408 	}
409 
410 }
411 
412 /* ------------------------------------------------------------------------ */
413 
414 /*
415  * Common stuff
416  */
417 
418 /* Window reduction (1-beta) for [New]Reno: 0.5 */
419 #define RENO_BETAA 1
420 #define RENO_BETAB 2
421 /* Window reduction (1-beta) for Cubic: 0.8 */
422 #define CUBIC_BETAA 4
423 #define CUBIC_BETAB 5
424 /* Draft Rhee Section 4.1 */
425 #define CUBIC_CA 4
426 #define CUBIC_CB 10
427 
428 static void
429 tcp_common_congestion_exp(struct tcpcb *tp, int betaa, int betab)
430 {
431 	u_int win;
432 
433 	/*
434 	 * Reduce the congestion window and the slow start threshold.
435 	 */
436 	win = min(tp->snd_wnd, tp->snd_cwnd) * betaa / betab / tp->t_segsz;
437 	if (win < 2)
438 		win = 2;
439 
440 	tp->snd_ssthresh = win * tp->t_segsz;
441 	tp->snd_recover = tp->snd_max;
442 	tp->snd_cwnd = tp->snd_ssthresh;
443 
444 	/*
445 	 * When using TCP ECN, notify the peer that
446 	 * we reduced the cwnd.
447 	 */
448 	if (TCP_ECN_ALLOWED(tp))
449 		tp->t_flags |= TF_ECN_SND_CWR;
450 }
451 
452 
453 /* ------------------------------------------------------------------------ */
454 
455 /*
456  * TCP/Reno congestion control.
457  */
458 static void
459 tcp_reno_congestion_exp(struct tcpcb *tp)
460 {
461 
462 	tcp_common_congestion_exp(tp, RENO_BETAA, RENO_BETAB);
463 }
464 
465 static int
466 tcp_reno_do_fast_retransmit(struct tcpcb *tp, const struct tcphdr *th)
467 {
468 	/*
469 	 * Dup acks mean that packets have left the
470 	 * network (they're now cached at the receiver)
471 	 * so bump cwnd by the amount in the receiver
472 	 * to keep a constant cwnd packets in the
473 	 * network.
474 	 *
475 	 * If we are using TCP/SACK, then enter
476 	 * Fast Recovery if the receiver SACKs
477 	 * data that is tcprexmtthresh * MSS
478 	 * bytes past the last ACKed segment,
479 	 * irrespective of the number of DupAcks.
480 	 */
481 
482 	tcp_seq onxt = tp->snd_nxt;
483 
484 	tp->t_partialacks = 0;
485 	TCP_TIMER_DISARM(tp, TCPT_REXMT);
486 	tp->t_rtttime = 0;
487 	if (TCP_SACK_ENABLED(tp)) {
488 		tp->t_dupacks = tcprexmtthresh;
489 		tp->sack_newdata = tp->snd_nxt;
490 		tp->snd_cwnd = tp->t_segsz;
491 		(void) tcp_output(tp);
492 		return 0;
493 	}
494 	tp->snd_nxt = th->th_ack;
495 	tp->snd_cwnd = tp->t_segsz;
496 	(void) tcp_output(tp);
497 	tp->snd_cwnd = tp->snd_ssthresh + tp->t_segsz * tp->t_dupacks;
498 	if (SEQ_GT(onxt, tp->snd_nxt))
499 		tp->snd_nxt = onxt;
500 
501 	return 0;
502 }
503 
504 static int
505 tcp_reno_fast_retransmit(struct tcpcb *tp, const struct tcphdr *th)
506 {
507 
508 	/*
509 	 * We know we're losing at the current
510 	 * window size so do congestion avoidance
511 	 * (set ssthresh to half the current window
512 	 * and pull our congestion window back to
513 	 * the new ssthresh).
514 	 */
515 
516 	tcp_reno_congestion_exp(tp);
517 	return tcp_reno_do_fast_retransmit(tp, th);
518 }
519 
520 static void
521 tcp_reno_slow_retransmit(struct tcpcb *tp)
522 {
523 	u_int win;
524 
525 	/*
526 	 * Close the congestion window down to one segment
527 	 * (we'll open it by one segment for each ack we get).
528 	 * Since we probably have a window's worth of unacked
529 	 * data accumulated, this "slow start" keeps us from
530 	 * dumping all that data as back-to-back packets (which
531 	 * might overwhelm an intermediate gateway).
532 	 *
533 	 * There are two phases to the opening: Initially we
534 	 * open by one mss on each ack.  This makes the window
535 	 * size increase exponentially with time.  If the
536 	 * window is larger than the path can handle, this
537 	 * exponential growth results in dropped packet(s)
538 	 * almost immediately.  To get more time between
539 	 * drops but still "push" the network to take advantage
540 	 * of improving conditions, we switch from exponential
541 	 * to linear window opening at some threshhold size.
542 	 * For a threshhold, we use half the current window
543 	 * size, truncated to a multiple of the mss.
544 	 *
545 	 * (the minimum cwnd that will give us exponential
546 	 * growth is 2 mss.  We don't allow the threshhold
547 	 * to go below this.)
548 	 */
549 
550 	win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
551 	if (win < 2)
552 		win = 2;
553 	/* Loss Window MUST be one segment. */
554 	tp->snd_cwnd = tp->t_segsz;
555 	tp->snd_ssthresh = win * tp->t_segsz;
556 	tp->t_partialacks = -1;
557 	tp->t_dupacks = 0;
558 	tp->t_bytes_acked = 0;
559 
560 	if (TCP_ECN_ALLOWED(tp))
561 		tp->t_flags |= TF_ECN_SND_CWR;
562 }
563 
564 static void
565 tcp_reno_fast_retransmit_newack(struct tcpcb *tp,
566     const struct tcphdr *th)
567 {
568 	if (tp->t_partialacks < 0) {
569 		/*
570 		 * We were not in fast recovery.  Reset the duplicate ack
571 		 * counter.
572 		 */
573 		tp->t_dupacks = 0;
574 	} else {
575 		/*
576 		 * Clamp the congestion window to the crossover point and
577 		 * exit fast recovery.
578 		 */
579 		if (tp->snd_cwnd > tp->snd_ssthresh)
580 			tp->snd_cwnd = tp->snd_ssthresh;
581 		tp->t_partialacks = -1;
582 		tp->t_dupacks = 0;
583 		tp->t_bytes_acked = 0;
584 		if (TCP_SACK_ENABLED(tp) && SEQ_GT(th->th_ack, tp->snd_fack))
585 			tp->snd_fack = th->th_ack;
586 	}
587 }
588 
589 static void
590 tcp_reno_newack(struct tcpcb *tp, const struct tcphdr *th)
591 {
592 	/*
593 	 * When new data is acked, open the congestion window.
594 	 */
595 
596 	u_int cw = tp->snd_cwnd;
597 	u_int incr = tp->t_segsz;
598 
599 	if (tcp_do_abc) {
600 
601 		/*
602 		 * RFC 3465 Appropriate Byte Counting (ABC)
603 		 */
604 
605 		int acked = th->th_ack - tp->snd_una;
606 
607 		if (cw >= tp->snd_ssthresh) {
608 			tp->t_bytes_acked += acked;
609 			if (tp->t_bytes_acked >= cw) {
610 				/* Time to increase the window. */
611 				tp->t_bytes_acked -= cw;
612 			} else {
613 				/* No need to increase yet. */
614 				incr = 0;
615 			}
616 		} else {
617 			/*
618 			 * use 2*SMSS or 1*SMSS for the "L" param,
619 			 * depending on sysctl setting.
620 			 *
621 			 * (See RFC 3465 2.3 Choosing the Limit)
622 			 */
623 			u_int abc_lim;
624 
625 			abc_lim = (tcp_abc_aggressive == 0 ||
626 			    tp->snd_nxt != tp->snd_max) ? incr : incr * 2;
627 			incr = min(acked, abc_lim);
628 		}
629 	} else {
630 
631 		/*
632 		 * If the window gives us less than ssthresh packets
633 		 * in flight, open exponentially (segsz per packet).
634 		 * Otherwise open linearly: segsz per window
635 		 * (segsz^2 / cwnd per packet).
636 		 */
637 
638 		if (cw >= tp->snd_ssthresh) {
639 			incr = incr * incr / cw;
640 		}
641 	}
642 
643 	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
644 }
645 
646 const struct tcp_congctl tcp_reno_ctl = {
647 	.fast_retransmit = tcp_reno_fast_retransmit,
648 	.slow_retransmit = tcp_reno_slow_retransmit,
649 	.fast_retransmit_newack = tcp_reno_fast_retransmit_newack,
650 	.newack = tcp_reno_newack,
651 	.cong_exp = tcp_reno_congestion_exp,
652 };
653 
654 /*
655  * TCP/NewReno Congestion control.
656  */
657 static int
658 tcp_newreno_fast_retransmit(struct tcpcb *tp, const struct tcphdr *th)
659 {
660 
661 	if (SEQ_LT(th->th_ack, tp->snd_high)) {
662 		/*
663 		 * False fast retransmit after timeout.
664 		 * Do not enter fast recovery
665 		 */
666 		tp->t_dupacks = 0;
667 		return 1;
668 	}
669 	/*
670 	 * Fast retransmit is same as reno.
671 	 */
672 	return tcp_reno_fast_retransmit(tp, th);
673 }
674 
675 /*
676  * Implement the NewReno response to a new ack, checking for partial acks in
677  * fast recovery.
678  */
679 static void
680 tcp_newreno_fast_retransmit_newack(struct tcpcb *tp, const struct tcphdr *th)
681 {
682 	if (tp->t_partialacks < 0) {
683 		/*
684 		 * We were not in fast recovery.  Reset the duplicate ack
685 		 * counter.
686 		 */
687 		tp->t_dupacks = 0;
688 	} else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
689 		/*
690 		 * This is a partial ack.  Retransmit the first unacknowledged
691 		 * segment and deflate the congestion window by the amount of
692 		 * acknowledged data.  Do not exit fast recovery.
693 		 */
694 		tcp_seq onxt = tp->snd_nxt;
695 		u_long ocwnd = tp->snd_cwnd;
696 		int sack_num_segs = 1, sack_bytes_rxmt = 0;
697 
698 		/*
699 		 * snd_una has not yet been updated and the socket's send
700 		 * buffer has not yet drained off the ACK'd data, so we
701 		 * have to leave snd_una as it was to get the correct data
702 		 * offset in tcp_output().
703 		 */
704 		tp->t_partialacks++;
705 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
706 		tp->t_rtttime = 0;
707 
708 		if (TCP_SACK_ENABLED(tp)) {
709 			/*
710 			 * Partial ack handling within a sack recovery episode.
711 			 * Keeping this very simple for now. When a partial ack
712 			 * is received, force snd_cwnd to a value that will
713 			 * allow the sender to transmit no more than 2 segments.
714 			 * If necessary, a fancier scheme can be adopted at a
715 			 * later point, but for now, the goal is to prevent the
716 			 * sender from bursting a large amount of data in the
717 			 * midst of sack recovery.
718 		 	 */
719 
720 			/*
721 			 * send one or 2 segments based on how much
722 			 * new data was acked
723 			 */
724 			if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2)
725 				sack_num_segs = 2;
726 			(void)tcp_sack_output(tp, &sack_bytes_rxmt);
727 			tp->snd_cwnd = sack_bytes_rxmt +
728 			    (tp->snd_nxt - tp->sack_newdata) +
729 			    sack_num_segs * tp->t_segsz;
730 			tp->t_flags |= TF_ACKNOW;
731 			(void) tcp_output(tp);
732 		} else {
733 			tp->snd_nxt = th->th_ack;
734 			/*
735 			 * Set snd_cwnd to one segment beyond ACK'd offset
736 			 * snd_una is not yet updated when we're called
737 			 */
738 			tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
739 			(void) tcp_output(tp);
740 			tp->snd_cwnd = ocwnd;
741 			if (SEQ_GT(onxt, tp->snd_nxt))
742 				tp->snd_nxt = onxt;
743 			/*
744 			 * Partial window deflation.  Relies on fact that
745 			 * tp->snd_una not updated yet.
746 		 	 */
747 			tp->snd_cwnd -= (th->th_ack - tp->snd_una -
748 			    tp->t_segsz);
749 		}
750 	} else {
751 		/*
752 		 * Complete ack.  Inflate the congestion window to ssthresh
753 		 * and exit fast recovery.
754 		 *
755 		 * Window inflation should have left us with approx.
756 		 * snd_ssthresh outstanding data.  But in case we
757 		 * would be inclined to send a burst, better to do
758 		 * it via the slow start mechanism.
759 		 */
760 		if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
761 			tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
762 			    + tp->t_segsz;
763 		else
764 			tp->snd_cwnd = tp->snd_ssthresh;
765 		tp->t_partialacks = -1;
766 		tp->t_dupacks = 0;
767 		tp->t_bytes_acked = 0;
768 		if (TCP_SACK_ENABLED(tp) && SEQ_GT(th->th_ack, tp->snd_fack))
769 			tp->snd_fack = th->th_ack;
770 	}
771 }
772 
773 static void
774 tcp_newreno_newack(struct tcpcb *tp, const struct tcphdr *th)
775 {
776 	/*
777 	 * If we are still in fast recovery (meaning we are using
778 	 * NewReno and we have only received partial acks), do not
779 	 * inflate the window yet.
780 	 */
781 	if (tp->t_partialacks < 0)
782 		tcp_reno_newack(tp, th);
783 }
784 
785 
786 const struct tcp_congctl tcp_newreno_ctl = {
787 	.fast_retransmit = tcp_newreno_fast_retransmit,
788 	.slow_retransmit = tcp_reno_slow_retransmit,
789 	.fast_retransmit_newack = tcp_newreno_fast_retransmit_newack,
790 	.newack = tcp_newreno_newack,
791 	.cong_exp = tcp_reno_congestion_exp,
792 };
793 
794 /*
795  * CUBIC - http://tools.ietf.org/html/draft-rhee-tcpm-cubic-02
796  */
797 
798 /* Cubic prototypes */
799 static void	tcp_cubic_update_ctime(struct tcpcb *tp);
800 static uint32_t	tcp_cubic_diff_ctime(struct tcpcb *);
801 static uint32_t	tcp_cubic_cbrt(uint32_t);
802 static ulong	tcp_cubic_getW(struct tcpcb *, uint32_t, uint32_t);
803 
804 /* Cubic TIME functions - XXX I don't like using timevals and microuptime */
805 /*
806  * Set congestion timer to now
807  */
808 static void
809 tcp_cubic_update_ctime(struct tcpcb *tp)
810 {
811 	struct timeval now_timeval;
812 
813 	getmicrouptime(&now_timeval);
814 	tp->snd_cubic_ctime = now_timeval.tv_sec * 1000 +
815 	    now_timeval.tv_usec / 1000;
816 }
817 
818 /*
819  * miliseconds from last congestion
820  */
821 static uint32_t
822 tcp_cubic_diff_ctime(struct tcpcb *tp)
823 {
824 	struct timeval now_timeval;
825 
826 	getmicrouptime(&now_timeval);
827 	return now_timeval.tv_sec * 1000 + now_timeval.tv_usec / 1000 -
828 	    tp->snd_cubic_ctime;
829 }
830 
831 /*
832  * Approximate cubic root
833  */
834 #define CBRT_ROUNDS 30
835 static uint32_t
836 tcp_cubic_cbrt(uint32_t v)
837 {
838 	int i, rounds = CBRT_ROUNDS;
839 	uint64_t x = v / 3;
840 
841 	/* We fail to calculate correct for small numbers */
842 	if (v == 0)
843 		return 0;
844 	else if (v < 4)
845 		return 1;
846 
847 	/*
848 	 * largest x that 2*x^3+3*x fits 64bit
849 	 * Avoid overflow for a time cost
850 	 */
851 	if (x > 2097151)
852 		rounds += 10;
853 
854 	for (i = 0; i < rounds; i++)
855 		if (rounds == CBRT_ROUNDS)
856 			x = (v + 2 * x * x * x) / (3 * x * x);
857 		else
858 			/* Avoid overflow */
859 			x = v / (3 * x * x) + 2 * x / 3;
860 
861 	return (uint32_t)x;
862 }
863 
864 /* Draft Rhee Section 3.1 - get W(t+rtt) - Eq. 1 */
865 static ulong
866 tcp_cubic_getW(struct tcpcb *tp, uint32_t ms_elapsed, uint32_t rtt)
867 {
868 	uint32_t K;
869 	long tK3;
870 
871 	/* Section 3.1 Eq. 2 */
872 	K = tcp_cubic_cbrt(tp->snd_cubic_wmax / CUBIC_BETAB *
873 	    CUBIC_CB / CUBIC_CA);
874 	/*  (t-K)^3 - not clear why is the measure unit mattering */
875 	tK3 = (long)(ms_elapsed + rtt) - (long)K;
876 	tK3 = tK3 * tK3 * tK3;
877 
878 	return CUBIC_CA * tK3 / CUBIC_CB + tp->snd_cubic_wmax;
879 }
880 
881 static void
882 tcp_cubic_congestion_exp(struct tcpcb *tp)
883 {
884 
885 	/*
886 	 * Congestion - Set WMax and shrink cwnd
887 	 */
888 	tcp_cubic_update_ctime(tp);
889 
890 	/* Section 3.6 - Fast Convergence */
891 	if (tp->snd_cubic_wmax < tp->snd_cubic_wmax_last) {
892 		tp->snd_cubic_wmax_last = tp->snd_cubic_wmax;
893 		tp->snd_cubic_wmax = tp->snd_cubic_wmax / 2 +
894 		    tp->snd_cubic_wmax * CUBIC_BETAA / CUBIC_BETAB / 2;
895 	} else {
896 		tp->snd_cubic_wmax_last = tp->snd_cubic_wmax;
897 		tp->snd_cubic_wmax = tp->snd_cwnd;
898 	}
899 
900 	tp->snd_cubic_wmax = max(tp->t_segsz, tp->snd_cubic_wmax);
901 
902 	/* Shrink CWND */
903 	tcp_common_congestion_exp(tp, CUBIC_BETAA, CUBIC_BETAB);
904 }
905 
906 static int
907 tcp_cubic_fast_retransmit(struct tcpcb *tp, const struct tcphdr *th)
908 {
909 
910 	if (SEQ_LT(th->th_ack, tp->snd_high)) {
911 		/* See newreno */
912 		tp->t_dupacks = 0;
913 		return 1;
914 	}
915 
916 	/*
917 	 * mark WMax
918 	 */
919 	tcp_cubic_congestion_exp(tp);
920 
921 	/* Do fast retransmit */
922 	return tcp_reno_do_fast_retransmit(tp, th);
923 }
924 
925 static void
926 tcp_cubic_newack(struct tcpcb *tp, const struct tcphdr *th)
927 {
928 	uint32_t ms_elapsed, rtt;
929 	u_long w_tcp;
930 
931 	/* Congestion avoidance and not in fast recovery and usable rtt */
932 	if (tp->snd_cwnd > tp->snd_ssthresh && tp->t_partialacks < 0 &&
933 	    /*
934 	     * t_srtt is 1/32 units of slow ticks
935 	     * converting it in ms would be equal to
936 	     * (t_srtt >> 5) * 1000 / PR_SLOWHZ ~= (t_srtt << 5) / PR_SLOWHZ
937 	     */
938 	    (rtt = (tp->t_srtt << 5) / PR_SLOWHZ) > 0) {
939 		ms_elapsed = tcp_cubic_diff_ctime(tp);
940 
941 		/* Compute W_tcp(t) */
942 		w_tcp = tp->snd_cubic_wmax * CUBIC_BETAA / CUBIC_BETAB +
943 		    ms_elapsed / rtt / 3;
944 
945 		if (tp->snd_cwnd > w_tcp) {
946 			/* Not in TCP friendly mode */
947 			tp->snd_cwnd += (tcp_cubic_getW(tp, ms_elapsed, rtt) -
948 			    tp->snd_cwnd) / tp->snd_cwnd;
949 		} else {
950 			/* friendly TCP mode */
951 			tp->snd_cwnd = w_tcp;
952 		}
953 
954 		/* Make sure we are within limits */
955 		tp->snd_cwnd = max(tp->snd_cwnd, tp->t_segsz);
956 		tp->snd_cwnd = min(tp->snd_cwnd, TCP_MAXWIN << tp->snd_scale);
957 	} else {
958 		/* Use New Reno */
959 		tcp_newreno_newack(tp, th);
960 	}
961 }
962 
963 static void
964 tcp_cubic_slow_retransmit(struct tcpcb *tp)
965 {
966 
967 	/* Timeout - Mark new congestion */
968 	tcp_cubic_congestion_exp(tp);
969 
970 	/* Loss Window MUST be one segment. */
971 	tp->snd_cwnd = tp->t_segsz;
972 	tp->t_partialacks = -1;
973 	tp->t_dupacks = 0;
974 	tp->t_bytes_acked = 0;
975 
976 	if (TCP_ECN_ALLOWED(tp))
977 		tp->t_flags |= TF_ECN_SND_CWR;
978 }
979 
980 const struct tcp_congctl tcp_cubic_ctl = {
981 	.fast_retransmit = tcp_cubic_fast_retransmit,
982 	.slow_retransmit = tcp_cubic_slow_retransmit,
983 	.fast_retransmit_newack = tcp_newreno_fast_retransmit_newack,
984 	.newack = tcp_cubic_newack,
985 	.cong_exp = tcp_cubic_congestion_exp,
986 };
987