tcp_congctl.c revision 1.4
1/*	$NetBSD: tcp_congctl.c,v 1.4 2006/10/10 11:12:39 rpaulo 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 * 3. All advertising materials mentioning features or use of this software
24 *    must display the following acknowledgement:
25 *	This product includes software developed by the NetBSD
26 *	Foundation, Inc. and its contributors.
27 * 4. Neither the name of The NetBSD Foundation nor the names of its
28 *    contributors may be used to endorse or promote products derived
29 *    from this software without specific prior written permission.
30 *
31 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
32 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
33 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
34 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
35 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
37 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
38 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
39 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
40 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
41 * POSSIBILITY OF SUCH DAMAGE.
42 */
43
44/*
45 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
46 * All rights reserved.
47 *
48 * Redistribution and use in source and binary forms, with or without
49 * modification, are permitted provided that the following conditions
50 * are met:
51 * 1. Redistributions of source code must retain the above copyright
52 *    notice, this list of conditions and the following disclaimer.
53 * 2. Redistributions in binary form must reproduce the above copyright
54 *    notice, this list of conditions and the following disclaimer in the
55 *    documentation and/or other materials provided with the distribution.
56 * 3. Neither the name of the project nor the names of its contributors
57 *    may be used to endorse or promote products derived from this software
58 *    without specific prior written permission.
59 *
60 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
63 * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70 * SUCH DAMAGE.
71 */
72
73/*
74 *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
75 *
76 * NRL grants permission for redistribution and use in source and binary
77 * forms, with or without modification, of the software and documentation
78 * created at NRL provided that the following conditions are met:
79 *
80 * 1. Redistributions of source code must retain the above copyright
81 *    notice, this list of conditions and the following disclaimer.
82 * 2. Redistributions in binary form must reproduce the above copyright
83 *    notice, this list of conditions and the following disclaimer in the
84 *    documentation and/or other materials provided with the distribution.
85 * 3. All advertising materials mentioning features or use of this software
86 *    must display the following acknowledgements:
87 *      This product includes software developed by the University of
88 *      California, Berkeley and its contributors.
89 *      This product includes software developed at the Information
90 *      Technology Division, US Naval Research Laboratory.
91 * 4. Neither the name of the NRL nor the names of its contributors
92 *    may be used to endorse or promote products derived from this software
93 *    without specific prior written permission.
94 *
95 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
96 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
97 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
98 * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
99 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
100 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
101 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
102 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
103 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
104 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
105 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
106 *
107 * The views and conclusions contained in the software and documentation
108 * are those of the authors and should not be interpreted as representing
109 * official policies, either expressed or implied, of the US Naval
110 * Research Laboratory (NRL).
111 */
112
113/*
114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
115 *	The Regents of the University of California.  All rights reserved.
116 *
117 * Redistribution and use in source and binary forms, with or without
118 * modification, are permitted provided that the following conditions
119 * are met:
120 * 1. Redistributions of source code must retain the above copyright
121 *    notice, this list of conditions and the following disclaimer.
122 * 2. Redistributions in binary form must reproduce the above copyright
123 *    notice, this list of conditions and the following disclaimer in the
124 *    documentation and/or other materials provided with the distribution.
125 * 3. Neither the name of the University nor the names of its contributors
126 *    may be used to endorse or promote products derived from this software
127 *    without specific prior written permission.
128 *
129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
132 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
139 * SUCH DAMAGE.
140 *
141 *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
142 */
143
144#include <sys/cdefs.h>
145__KERNEL_RCSID(0, "$NetBSD: tcp_congctl.c,v 1.4 2006/10/10 11:12:39 rpaulo Exp $");
146
147#include "opt_inet.h"
148#include "opt_tcp_debug.h"
149#include "opt_tcp_congctl.h"
150
151#include <sys/param.h>
152#include <sys/systm.h>
153#include <sys/malloc.h>
154#include <sys/mbuf.h>
155#include <sys/protosw.h>
156#include <sys/socket.h>
157#include <sys/socketvar.h>
158#include <sys/errno.h>
159#include <sys/syslog.h>
160#include <sys/pool.h>
161#include <sys/domain.h>
162#include <sys/kernel.h>
163#include <sys/lock.h>
164
165#include <net/if.h>
166#include <net/route.h>
167
168#include <netinet/in.h>
169#include <netinet/in_systm.h>
170#include <netinet/ip.h>
171#include <netinet/in_pcb.h>
172#include <netinet/in_var.h>
173#include <netinet/ip_var.h>
174
175#ifdef INET6
176#ifndef INET
177#include <netinet/in.h>
178#endif
179#include <netinet/ip6.h>
180#include <netinet6/ip6_var.h>
181#include <netinet6/in6_pcb.h>
182#include <netinet6/ip6_var.h>
183#include <netinet6/in6_var.h>
184#include <netinet/icmp6.h>
185#include <netinet6/nd6.h>
186#endif
187
188#include <netinet/tcp.h>
189#include <netinet/tcp_fsm.h>
190#include <netinet/tcp_seq.h>
191#include <netinet/tcp_timer.h>
192#include <netinet/tcp_var.h>
193#include <netinet/tcpip.h>
194#include <netinet/tcp_congctl.h>
195#ifdef TCP_DEBUG
196#include <netinet/tcp_debug.h>
197#endif
198
199/*
200 * TODO:
201 *   consider separating the actual implementations in another file.
202 */
203
204static int  tcp_reno_fast_retransmit(struct tcpcb *, struct tcphdr *);
205static void tcp_reno_slow_retransmit(struct tcpcb *);
206static void tcp_reno_fast_retransmit_newack(struct tcpcb *, struct tcphdr *);
207static void tcp_reno_newack(struct tcpcb *, struct tcphdr *);
208
209static int  tcp_newreno_fast_retransmit(struct tcpcb *, struct tcphdr *);
210static void tcp_newreno_fast_retransmit_newack(struct tcpcb *,
211	struct tcphdr *);
212static void tcp_newreno_newack(struct tcpcb *, struct tcphdr *);
213
214
215static void tcp_congctl_fillnames(void);
216
217extern int tcprexmtthresh;
218
219MALLOC_DEFINE(M_TCPCONGCTL, "tcpcongctl", "TCP congestion control structures");
220
221/*
222 * Used to list the available congestion control algorithms.
223 */
224struct tcp_congctlent {
225	TAILQ_ENTRY(tcp_congctlent) congctl_ent;
226	char               congctl_name[TCPCC_MAXLEN];
227	struct tcp_congctl *congctl_ctl;
228};
229TAILQ_HEAD(, tcp_congctlent) tcp_congctlhd;
230
231struct simplelock tcp_congctl_slock;
232
233void
234tcp_congctl_init(void)
235{
236	int r;
237
238	TAILQ_INIT(&tcp_congctlhd);
239	simple_lock_init(&tcp_congctl_slock);
240
241	/* Base algorithms. */
242	r = tcp_congctl_register("reno", &tcp_reno_ctl);
243	KASSERT(r == 0);
244	r = tcp_congctl_register("newreno", &tcp_newreno_ctl);
245	KASSERT(r == 0);
246
247	/* NewReno is the default. */
248#ifndef TCP_CONGCTL_DEFAULT
249#define TCP_CONGCTL_DEFAULT "newreno"
250#endif
251
252	r = tcp_congctl_select(NULL, TCP_CONGCTL_DEFAULT);
253	KASSERT(r == 0);
254}
255
256/*
257 * Register a congestion algorithm and select it if we have none.
258 */
259int
260tcp_congctl_register(const char *name, struct tcp_congctl *tcc)
261{
262	struct tcp_congctlent *ntcc, *tccp;
263
264	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent)
265		if (!strcmp(name, tccp->congctl_name)) {
266			/* name already registered */
267			return EEXIST;
268		}
269
270	ntcc = malloc(sizeof(*ntcc), M_TCPCONGCTL, M_WAITOK);
271
272	strlcpy(ntcc->congctl_name, name, sizeof(ntcc->congctl_name) - 1);
273	ntcc->congctl_ctl = tcc;
274
275	TAILQ_INSERT_TAIL(&tcp_congctlhd, ntcc, congctl_ent);
276	tcp_congctl_fillnames();
277
278	if (TAILQ_FIRST(&tcp_congctlhd) == ntcc)
279		tcp_congctl_select(NULL, name);
280
281	return 0;
282}
283
284int
285tcp_congctl_unregister(const char *name)
286{
287	struct tcp_congctlent *tccp, *rtccp;
288	unsigned int size;
289
290	rtccp = NULL;
291	size = 0;
292	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
293		if (!strcmp(name, tccp->congctl_name))
294			rtccp = tccp;
295		size++;
296	}
297
298	if (!rtccp)
299		return ENOENT;
300
301	if (size <= 1 || tcp_congctl_global == rtccp->congctl_ctl ||
302	    rtccp->congctl_ctl->refcnt)
303		return EBUSY;
304
305	TAILQ_REMOVE(&tcp_congctlhd, rtccp, congctl_ent);
306	free(rtccp, M_TCPCONGCTL);
307	tcp_congctl_fillnames();
308
309	return 0;
310}
311
312/*
313 * Select a congestion algorithm by name.
314 */
315int
316tcp_congctl_select(struct tcpcb *tp, const char *name)
317{
318	struct tcp_congctlent *tccp;
319
320	KASSERT(name);
321
322	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent)
323		if (!strcmp(name, tccp->congctl_name)) {
324			if (tp) {
325				simple_lock(&tcp_congctl_slock);
326				tp->t_congctl->refcnt--;
327				tp->t_congctl = tccp->congctl_ctl;
328				tp->t_congctl->refcnt++;
329				simple_unlock(&tcp_congctl_slock);
330			} else {
331				tcp_congctl_global = tccp->congctl_ctl;
332				strlcpy(tcp_congctl_global_name,
333				    tccp->congctl_name,
334				    sizeof(tcp_congctl_global_name) - 1);
335			}
336			return 0;
337		}
338
339	return EINVAL;
340}
341
342/*
343 * Returns the name of a congestion algorithm.
344 */
345const char *
346tcp_congctl_bystruct(const struct tcp_congctl *tcc)
347{
348	struct tcp_congctlent *tccp;
349
350	KASSERT(tcc);
351
352	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent)
353		if (tccp->congctl_ctl == tcc)
354			return tccp->congctl_name;
355
356	return NULL;
357}
358
359static void
360tcp_congctl_fillnames(void)
361{
362	struct tcp_congctlent *tccp;
363	const char *delim = " ";
364
365	tcp_congctl_avail[0] = '\0';
366	TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) {
367		strlcat(tcp_congctl_avail, tccp->congctl_name,
368		    sizeof(tcp_congctl_avail) - 1);
369		if (TAILQ_NEXT(tccp, congctl_ent))
370			strlcat(tcp_congctl_avail, delim,
371			    sizeof(tcp_congctl_avail) - 1);
372	}
373
374}
375
376/* ------------------------------------------------------------------------ */
377
378inline void
379tcp_reno_congestion_exp(struct tcpcb *tp)
380{
381	u_int win;
382
383	/*
384	 * Halve the congestion window and reduce the
385	 * slow start threshold.
386	 */
387	win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
388	if (win < 2)
389		win = 2;
390
391	tp->snd_ssthresh = win * tp->t_segsz;
392	tp->snd_recover = tp->snd_max;
393	tp->snd_cwnd = tp->snd_ssthresh;
394
395	if (TCP_ECN_ALLOWED(tp))
396		tp->t_flags |= TF_ECN_SND_CWR;
397}
398
399
400/*
401 * TCP/Reno congestion control.
402 */
403static int
404tcp_reno_fast_retransmit(struct tcpcb *tp, struct tcphdr *th)
405{
406	tcp_seq onxt;
407
408	onxt = tp->snd_nxt;
409	tcp_reno_congestion_exp(tp);
410	tp->t_partialacks = 0;
411	TCP_TIMER_DISARM(tp, TCPT_REXMT);
412	tp->t_rtttime = 0;
413	if (TCP_SACK_ENABLED(tp)) {
414		tp->t_dupacks = tcprexmtthresh;
415		tp->sack_newdata = tp->snd_nxt;
416		tp->snd_cwnd = tp->t_segsz;
417		(void) tcp_output(tp);
418		return 0;
419	}
420	tp->snd_nxt = th->th_ack;
421	tp->snd_cwnd = tp->t_segsz;
422	(void) tcp_output(tp);
423	tp->snd_cwnd = tp->snd_ssthresh + tp->t_segsz * tp->t_dupacks;
424	if (SEQ_GT(onxt, tp->snd_nxt))
425		tp->snd_nxt = onxt;
426
427	return 0;
428}
429
430static void
431tcp_reno_slow_retransmit(struct tcpcb *tp)
432{
433	u_int win;
434
435	/*
436	 * Close the congestion window down to one segment
437	 * (we'll open it by one segment for each ack we get).
438	 * Since we probably have a window's worth of unacked
439	 * data accumulated, this "slow start" keeps us from
440	 * dumping all that data as back-to-back packets (which
441	 * might overwhelm an intermediate gateway).
442	 *
443	 * There are two phases to the opening: Initially we
444	 * open by one mss on each ack.  This makes the window
445	 * size increase exponentially with time.  If the
446	 * window is larger than the path can handle, this
447	 * exponential growth results in dropped packet(s)
448	 * almost immediately.  To get more time between
449	 * drops but still "push" the network to take advantage
450	 * of improving conditions, we switch from exponential
451	 * to linear window opening at some threshhold size.
452	 * For a threshhold, we use half the current window
453	 * size, truncated to a multiple of the mss.
454	 *
455	 * (the minimum cwnd that will give us exponential
456	 * growth is 2 mss.  We don't allow the threshhold
457	 * to go below this.)
458	 */
459
460	win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
461	if (win < 2)
462		win = 2;
463	/* Loss Window MUST be one segment. */
464	tp->snd_cwnd = tp->t_segsz;
465	tp->snd_ssthresh = win * tp->t_segsz;
466	tp->t_partialacks = -1;
467	tp->t_dupacks = 0;
468}
469
470static void
471tcp_reno_fast_retransmit_newack(struct tcpcb *tp, struct tcphdr *th)
472{
473	if (tp->t_partialacks < 0) {
474		/*
475		 * We were not in fast recovery.  Reset the duplicate ack
476		 * counter.
477		 */
478		tp->t_dupacks = 0;
479	} else {
480		/*
481		 * Clamp the congestion window to the crossover point and
482		 * exit fast recovery.
483		 */
484		if (tp->snd_cwnd > tp->snd_ssthresh)
485			tp->snd_cwnd = tp->snd_ssthresh;
486		tp->t_partialacks = -1;
487		tp->t_dupacks = 0;
488	}
489}
490
491static void
492tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th)
493{
494	/*
495	 * When new data is acked, open the congestion window.
496	 * If the window gives us less than ssthresh packets
497	 * in flight, open exponentially (segsz per packet).
498	 * Otherwise open linearly: segsz per window
499	 * (segsz^2 / cwnd per packet).
500	 */
501
502	u_int cw = tp->snd_cwnd;
503	u_int incr = tp->t_segsz;
504
505	if (cw >= tp->snd_ssthresh)
506		incr = incr * incr / cw;
507
508	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
509}
510
511struct tcp_congctl tcp_reno_ctl = {
512	.fast_retransmit = tcp_reno_fast_retransmit,
513	.slow_retransmit = tcp_reno_slow_retransmit,
514	.fast_retransmit_newack = tcp_reno_fast_retransmit_newack,
515	.newack = tcp_reno_newack,
516};
517
518/*
519 * TCP/NewReno Congestion control.
520 */
521static int
522tcp_newreno_fast_retransmit(struct tcpcb *tp, struct tcphdr *th)
523{
524	if (SEQ_LT(th->th_ack, tp->snd_high)) {
525		/*
526		 * False fast retransmit after timeout.
527		 * Do not enter fast recovery
528		 */
529		tp->t_dupacks = 0;
530		return 1;
531	} else {
532		/*
533		 * Fast retransmit is same as reno.
534		 */
535		return tcp_reno_fast_retransmit(tp, th);
536	}
537
538	return 0;
539}
540
541/*
542 * Implement the NewReno response to a new ack, checking for partial acks in
543 * fast recovery.
544 */
545static void
546tcp_newreno_fast_retransmit_newack(struct tcpcb *tp, struct tcphdr *th)
547{
548	if (tp->t_partialacks < 0) {
549		/*
550		 * We were not in fast recovery.  Reset the duplicate ack
551		 * counter.
552		 */
553		tp->t_dupacks = 0;
554	} else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
555		/*
556		 * This is a partial ack.  Retransmit the first unacknowledged
557		 * segment and deflate the congestion window by the amount of
558		 * acknowledged data.  Do not exit fast recovery.
559		 */
560		tcp_seq onxt = tp->snd_nxt;
561		u_long ocwnd = tp->snd_cwnd;
562
563		/*
564		 * snd_una has not yet been updated and the socket's send
565		 * buffer has not yet drained off the ACK'd data, so we
566		 * have to leave snd_una as it was to get the correct data
567		 * offset in tcp_output().
568		 */
569		if (++tp->t_partialacks == 1)
570			TCP_TIMER_DISARM(tp, TCPT_REXMT);
571		tp->t_rtttime = 0;
572		tp->snd_nxt = th->th_ack;
573		/*
574		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
575		 * is not yet updated when we're called.
576		 */
577		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
578		(void) tcp_output(tp);
579		tp->snd_cwnd = ocwnd;
580		if (SEQ_GT(onxt, tp->snd_nxt))
581			tp->snd_nxt = onxt;
582		/*
583		 * Partial window deflation.  Relies on fact that tp->snd_una
584		 * not updated yet.
585		 */
586		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
587	} else {
588		/*
589		 * Complete ack.  Inflate the congestion window to ssthresh
590		 * and exit fast recovery.
591		 *
592		 * Window inflation should have left us with approx.
593		 * snd_ssthresh outstanding data.  But in case we
594		 * would be inclined to send a burst, better to do
595		 * it via the slow start mechanism.
596		 */
597		if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
598			tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
599			    + tp->t_segsz;
600		else
601			tp->snd_cwnd = tp->snd_ssthresh;
602		tp->t_partialacks = -1;
603		tp->t_dupacks = 0;
604	}
605}
606
607static void
608tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th)
609{
610	/*
611	 * If we are still in fast recovery (meaning we are using
612	 * NewReno and we have only received partial acks), do not
613	 * inflate the window yet.
614	 */
615	if (tp->t_partialacks < 0)
616		tcp_reno_newack(tp, th);
617}
618
619
620struct tcp_congctl tcp_newreno_ctl = {
621	.fast_retransmit = tcp_newreno_fast_retransmit,
622	.slow_retransmit = tcp_reno_slow_retransmit,
623	.fast_retransmit_newack = tcp_newreno_fast_retransmit_newack,
624	.newack = tcp_newreno_newack,
625};
626
627
628