xref: /freebsd/sys/netinet/tcp_timer.c (revision 271c3a9060f2ee55607ebe146523f888e1db2654)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet6.h"
36 #include "opt_tcpdebug.h"
37 
38 #include <sys/param.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mbuf.h>
42 #include <sys/mutex.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/vimage.h>
49 
50 #include <net/route.h>
51 
52 #include <netinet/in.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/in_systm.h>
55 #ifdef INET6
56 #include <netinet6/in6_pcb.h>
57 #endif
58 #include <netinet/ip_var.h>
59 #include <netinet/tcp.h>
60 #include <netinet/tcp_fsm.h>
61 #include <netinet/tcp_timer.h>
62 #include <netinet/tcp_var.h>
63 #include <netinet/tcpip.h>
64 #ifdef TCPDEBUG
65 #include <netinet/tcp_debug.h>
66 #endif
67 
68 int	tcp_keepinit;
69 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
70     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
71 
72 int	tcp_keepidle;
73 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
74     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
75 
76 int	tcp_keepintvl;
77 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
78     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
79 
80 int	tcp_delacktime;
81 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
82     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
83     "Time before a delayed ACK is sent");
84 
85 int	tcp_msl;
86 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
87     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
88 
89 int	tcp_rexmit_min;
90 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
91     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
92     "Minimum Retransmission Timeout");
93 
94 int	tcp_rexmit_slop;
95 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
96     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
97     "Retransmission Timer Slop");
98 
99 static int	always_keepalive = 1;
100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
101     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
102 
103 int    tcp_fast_finwait2_recycle = 0;
104 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
105     &tcp_fast_finwait2_recycle, 0,
106     "Recycle closed FIN_WAIT_2 connections faster");
107 
108 int    tcp_finwait2_timeout;
109 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
110     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
111 
112 
113 static int	tcp_keepcnt = TCPTV_KEEPCNT;
114 	/* max idle probes */
115 int	tcp_maxpersistidle;
116 	/* max idle time in persist */
117 int	tcp_maxidle;
118 
119 /*
120  * Tcp protocol timeout routine called every 500 ms.
121  * Updates timestamps used for TCP
122  * causes finite state machine actions if timers expire.
123  */
124 void
125 tcp_slowtimo(void)
126 {
127 
128 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
129 	INP_INFO_WLOCK(&V_tcbinfo);
130 	(void) tcp_tw_2msl_scan(0);
131 	INP_INFO_WUNLOCK(&V_tcbinfo);
132 }
133 
134 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
135     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
136 
137 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
138     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
139 
140 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
141 
142 static int tcp_timer_race;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
144     0, "Count of t_inpcb races on tcp_discardcb");
145 
146 /*
147  * TCP timer processing.
148  */
149 
150 void
151 tcp_timer_delack(void *xtp)
152 {
153 	struct tcpcb *tp = xtp;
154 	struct inpcb *inp;
155 
156 	INP_INFO_RLOCK(&V_tcbinfo);
157 	inp = tp->t_inpcb;
158 	/*
159 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
160 	 * tear-down mean we need it as a work-around for races between
161 	 * timers and tcp_discardcb().
162 	 *
163 	 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
164 	 */
165 	if (inp == NULL) {
166 		tcp_timer_race++;
167 		INP_INFO_RUNLOCK(&V_tcbinfo);
168 		return;
169 	}
170 	INP_WLOCK(inp);
171 	INP_INFO_RUNLOCK(&V_tcbinfo);
172 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack)
173 	    || !callout_active(&tp->t_timers->tt_delack)) {
174 		INP_WUNLOCK(inp);
175 		return;
176 	}
177 	callout_deactivate(&tp->t_timers->tt_delack);
178 
179 	tp->t_flags |= TF_ACKNOW;
180 	V_tcpstat.tcps_delack++;
181 	(void) tcp_output(tp);
182 	INP_WUNLOCK(inp);
183 }
184 
185 void
186 tcp_timer_2msl(void *xtp)
187 {
188 	struct tcpcb *tp = xtp;
189 	struct inpcb *inp;
190 #ifdef TCPDEBUG
191 	int ostate;
192 
193 	ostate = tp->t_state;
194 #endif
195 	/*
196 	 * XXXRW: Does this actually happen?
197 	 */
198 	INP_INFO_WLOCK(&V_tcbinfo);
199 	inp = tp->t_inpcb;
200 	/*
201 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
202 	 * tear-down mean we need it as a work-around for races between
203 	 * timers and tcp_discardcb().
204 	 *
205 	 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
206 	 */
207 	if (inp == NULL) {
208 		tcp_timer_race++;
209 		INP_INFO_WUNLOCK(&V_tcbinfo);
210 		return;
211 	}
212 	INP_WLOCK(inp);
213 	tcp_free_sackholes(tp);
214 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) ||
215 	    !callout_active(&tp->t_timers->tt_2msl)) {
216 		INP_WUNLOCK(tp->t_inpcb);
217 		INP_INFO_WUNLOCK(&V_tcbinfo);
218 		return;
219 	}
220 	callout_deactivate(&tp->t_timers->tt_2msl);
221 	/*
222 	 * 2 MSL timeout in shutdown went off.  If we're closed but
223 	 * still waiting for peer to close and connection has been idle
224 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
225 	 * control block.  Otherwise, check again in a bit.
226 	 *
227 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
228 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
229 	 * Ignore fact that there were recent incoming segments.
230 	 */
231 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
232 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
233 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
234 		V_tcpstat.tcps_finwait2_drops++;
235 		tp = tcp_close(tp);
236 	} else {
237 		if (tp->t_state != TCPS_TIME_WAIT &&
238 		   (ticks - tp->t_rcvtime) <= tcp_maxidle)
239 		       callout_reset(&tp->t_timers->tt_2msl, tcp_keepintvl,
240 				     tcp_timer_2msl, tp);
241 	       else
242 		       tp = tcp_close(tp);
243        }
244 
245 #ifdef TCPDEBUG
246 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
247 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
248 			  PRU_SLOWTIMO);
249 #endif
250 	if (tp != NULL)
251 		INP_WUNLOCK(inp);
252 	INP_INFO_WUNLOCK(&V_tcbinfo);
253 }
254 
255 void
256 tcp_timer_keep(void *xtp)
257 {
258 	struct tcpcb *tp = xtp;
259 	struct tcptemp *t_template;
260 	struct inpcb *inp;
261 #ifdef TCPDEBUG
262 	int ostate;
263 
264 	ostate = tp->t_state;
265 #endif
266 	INP_INFO_WLOCK(&V_tcbinfo);
267 	inp = tp->t_inpcb;
268 	/*
269 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
270 	 * tear-down mean we need it as a work-around for races between
271 	 * timers and tcp_discardcb().
272 	 *
273 	 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
274 	 */
275 	if (inp == NULL) {
276 		tcp_timer_race++;
277 		INP_INFO_WUNLOCK(&V_tcbinfo);
278 		return;
279 	}
280 	INP_WLOCK(inp);
281 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep)
282 	    || !callout_active(&tp->t_timers->tt_keep)) {
283 		INP_WUNLOCK(inp);
284 		INP_INFO_WUNLOCK(&V_tcbinfo);
285 		return;
286 	}
287 	callout_deactivate(&tp->t_timers->tt_keep);
288 	/*
289 	 * Keep-alive timer went off; send something
290 	 * or drop connection if idle for too long.
291 	 */
292 	V_tcpstat.tcps_keeptimeo++;
293 	if (tp->t_state < TCPS_ESTABLISHED)
294 		goto dropit;
295 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
296 	    tp->t_state <= TCPS_CLOSING) {
297 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
298 			goto dropit;
299 		/*
300 		 * Send a packet designed to force a response
301 		 * if the peer is up and reachable:
302 		 * either an ACK if the connection is still alive,
303 		 * or an RST if the peer has closed the connection
304 		 * due to timeout or reboot.
305 		 * Using sequence number tp->snd_una-1
306 		 * causes the transmitted zero-length segment
307 		 * to lie outside the receive window;
308 		 * by the protocol spec, this requires the
309 		 * correspondent TCP to respond.
310 		 */
311 		V_tcpstat.tcps_keepprobe++;
312 		t_template = tcpip_maketemplate(inp);
313 		if (t_template) {
314 			tcp_respond(tp, t_template->tt_ipgen,
315 				    &t_template->tt_t, (struct mbuf *)NULL,
316 				    tp->rcv_nxt, tp->snd_una - 1, 0);
317 			free(t_template, M_TEMP);
318 		}
319 		callout_reset(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
320 	} else
321 		callout_reset(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
322 
323 #ifdef TCPDEBUG
324 	if (inp->inp_socket->so_options & SO_DEBUG)
325 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
326 			  PRU_SLOWTIMO);
327 #endif
328 	INP_WUNLOCK(inp);
329 	INP_INFO_WUNLOCK(&V_tcbinfo);
330 	return;
331 
332 dropit:
333 	V_tcpstat.tcps_keepdrops++;
334 	tp = tcp_drop(tp, ETIMEDOUT);
335 
336 #ifdef TCPDEBUG
337 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
338 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
339 			  PRU_SLOWTIMO);
340 #endif
341 	if (tp != NULL)
342 		INP_WUNLOCK(tp->t_inpcb);
343 	INP_INFO_WUNLOCK(&V_tcbinfo);
344 }
345 
346 void
347 tcp_timer_persist(void *xtp)
348 {
349 	struct tcpcb *tp = xtp;
350 	struct inpcb *inp;
351 #ifdef TCPDEBUG
352 	int ostate;
353 
354 	ostate = tp->t_state;
355 #endif
356 	INP_INFO_WLOCK(&V_tcbinfo);
357 	inp = tp->t_inpcb;
358 	/*
359 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
360 	 * tear-down mean we need it as a work-around for races between
361 	 * timers and tcp_discardcb().
362 	 *
363 	 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
364 	 */
365 	if (inp == NULL) {
366 		tcp_timer_race++;
367 		INP_INFO_WUNLOCK(&V_tcbinfo);
368 		return;
369 	}
370 	INP_WLOCK(inp);
371 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
372 	    || !callout_active(&tp->t_timers->tt_persist)) {
373 		INP_WUNLOCK(inp);
374 		INP_INFO_WUNLOCK(&V_tcbinfo);
375 		return;
376 	}
377 	callout_deactivate(&tp->t_timers->tt_persist);
378 	/*
379 	 * Persistance timer into zero window.
380 	 * Force a byte to be output, if possible.
381 	 */
382 	V_tcpstat.tcps_persisttimeo++;
383 	/*
384 	 * Hack: if the peer is dead/unreachable, we do not
385 	 * time out if the window is closed.  After a full
386 	 * backoff, drop the connection if the idle time
387 	 * (no responses to probes) reaches the maximum
388 	 * backoff that we would use if retransmitting.
389 	 */
390 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
391 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
392 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
393 		V_tcpstat.tcps_persistdrop++;
394 		tp = tcp_drop(tp, ETIMEDOUT);
395 		goto out;
396 	}
397 	tcp_setpersist(tp);
398 	tp->t_flags |= TF_FORCEDATA;
399 	(void) tcp_output(tp);
400 	tp->t_flags &= ~TF_FORCEDATA;
401 
402 out:
403 #ifdef TCPDEBUG
404 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
405 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
406 #endif
407 	if (tp != NULL)
408 		INP_WUNLOCK(inp);
409 	INP_INFO_WUNLOCK(&V_tcbinfo);
410 }
411 
412 void
413 tcp_timer_rexmt(void * xtp)
414 {
415 	struct tcpcb *tp = xtp;
416 	int rexmt;
417 	int headlocked;
418 	struct inpcb *inp;
419 #ifdef TCPDEBUG
420 	int ostate;
421 
422 	ostate = tp->t_state;
423 #endif
424 	INP_INFO_WLOCK(&V_tcbinfo);
425 	headlocked = 1;
426 	inp = tp->t_inpcb;
427 	/*
428 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
429 	 * tear-down mean we need it as a work-around for races between
430 	 * timers and tcp_discardcb().
431 	 *
432 	 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
433 	 */
434 	if (inp == NULL) {
435 		tcp_timer_race++;
436 		INP_INFO_WUNLOCK(&V_tcbinfo);
437 		return;
438 	}
439 	INP_WLOCK(inp);
440 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt)
441 	    || !callout_active(&tp->t_timers->tt_rexmt)) {
442 		INP_WUNLOCK(inp);
443 		INP_INFO_WUNLOCK(&V_tcbinfo);
444 		return;
445 	}
446 	callout_deactivate(&tp->t_timers->tt_rexmt);
447 	tcp_free_sackholes(tp);
448 	/*
449 	 * Retransmission timer went off.  Message has not
450 	 * been acked within retransmit interval.  Back off
451 	 * to a longer retransmit interval and retransmit one segment.
452 	 */
453 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
454 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
455 		V_tcpstat.tcps_timeoutdrop++;
456 		tp = tcp_drop(tp, tp->t_softerror ?
457 			      tp->t_softerror : ETIMEDOUT);
458 		goto out;
459 	}
460 	INP_INFO_WUNLOCK(&V_tcbinfo);
461 	headlocked = 0;
462 	if (tp->t_rxtshift == 1) {
463 		/*
464 		 * first retransmit; record ssthresh and cwnd so they can
465 		 * be recovered if this turns out to be a "bad" retransmit.
466 		 * A retransmit is considered "bad" if an ACK for this
467 		 * segment is received within RTT/2 interval; the assumption
468 		 * here is that the ACK was already in flight.  See
469 		 * "On Estimating End-to-End Network Path Properties" by
470 		 * Allman and Paxson for more details.
471 		 */
472 		tp->snd_cwnd_prev = tp->snd_cwnd;
473 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
474 		tp->snd_recover_prev = tp->snd_recover;
475 		if (IN_FASTRECOVERY(tp))
476 		  tp->t_flags |= TF_WASFRECOVERY;
477 		else
478 		  tp->t_flags &= ~TF_WASFRECOVERY;
479 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
480 	}
481 	V_tcpstat.tcps_rexmttimeo++;
482 	if (tp->t_state == TCPS_SYN_SENT)
483 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
484 	else
485 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
486 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
487 		      tp->t_rttmin, TCPTV_REXMTMAX);
488 	/*
489 	 * Disable rfc1323 if we havn't got any response to
490 	 * our third SYN to work-around some broken terminal servers
491 	 * (most of which have hopefully been retired) that have bad VJ
492 	 * header compression code which trashes TCP segments containing
493 	 * unknown-to-them TCP options.
494 	 */
495 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
496 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
497 	/*
498 	 * If we backed off this far, our srtt estimate is probably bogus.
499 	 * Clobber it so we'll take the next rtt measurement as our srtt;
500 	 * move the current srtt into rttvar to keep the current
501 	 * retransmit times until then.
502 	 */
503 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
504 #ifdef INET6
505 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
506 			in6_losing(tp->t_inpcb);
507 		else
508 #endif
509 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
510 		tp->t_srtt = 0;
511 	}
512 	tp->snd_nxt = tp->snd_una;
513 	tp->snd_recover = tp->snd_max;
514 	/*
515 	 * Force a segment to be sent.
516 	 */
517 	tp->t_flags |= TF_ACKNOW;
518 	/*
519 	 * If timing a segment in this window, stop the timer.
520 	 */
521 	tp->t_rtttime = 0;
522 	/*
523 	 * Close the congestion window down to one segment
524 	 * (we'll open it by one segment for each ack we get).
525 	 * Since we probably have a window's worth of unacked
526 	 * data accumulated, this "slow start" keeps us from
527 	 * dumping all that data as back-to-back packets (which
528 	 * might overwhelm an intermediate gateway).
529 	 *
530 	 * There are two phases to the opening: Initially we
531 	 * open by one mss on each ack.  This makes the window
532 	 * size increase exponentially with time.  If the
533 	 * window is larger than the path can handle, this
534 	 * exponential growth results in dropped packet(s)
535 	 * almost immediately.  To get more time between
536 	 * drops but still "push" the network to take advantage
537 	 * of improving conditions, we switch from exponential
538 	 * to linear window opening at some threshhold size.
539 	 * For a threshhold, we use half the current window
540 	 * size, truncated to a multiple of the mss.
541 	 *
542 	 * (the minimum cwnd that will give us exponential
543 	 * growth is 2 mss.  We don't allow the threshhold
544 	 * to go below this.)
545 	 */
546 	{
547 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
548 		if (win < 2)
549 			win = 2;
550 		tp->snd_cwnd = tp->t_maxseg;
551 		tp->snd_ssthresh = win * tp->t_maxseg;
552 		tp->t_dupacks = 0;
553 	}
554 	EXIT_FASTRECOVERY(tp);
555 	(void) tcp_output(tp);
556 
557 out:
558 #ifdef TCPDEBUG
559 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
560 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
561 			  PRU_SLOWTIMO);
562 #endif
563 	if (tp != NULL)
564 		INP_WUNLOCK(inp);
565 	if (headlocked)
566 		INP_INFO_WUNLOCK(&V_tcbinfo);
567 }
568 
569 void
570 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
571 {
572 	struct callout *t_callout;
573 	void *f_callout;
574 
575 	switch (timer_type) {
576 		case TT_DELACK:
577 			t_callout = &tp->t_timers->tt_delack;
578 			f_callout = tcp_timer_delack;
579 			break;
580 		case TT_REXMT:
581 			t_callout = &tp->t_timers->tt_rexmt;
582 			f_callout = tcp_timer_rexmt;
583 			break;
584 		case TT_PERSIST:
585 			t_callout = &tp->t_timers->tt_persist;
586 			f_callout = tcp_timer_persist;
587 			break;
588 		case TT_KEEP:
589 			t_callout = &tp->t_timers->tt_keep;
590 			f_callout = tcp_timer_keep;
591 			break;
592 		case TT_2MSL:
593 			t_callout = &tp->t_timers->tt_2msl;
594 			f_callout = tcp_timer_2msl;
595 			break;
596 		default:
597 			panic("bad timer_type");
598 		}
599 	if (delta == 0) {
600 		callout_stop(t_callout);
601 	} else {
602 		callout_reset(t_callout, delta, f_callout, tp);
603 	}
604 }
605 
606 int
607 tcp_timer_active(struct tcpcb *tp, int timer_type)
608 {
609 	struct callout *t_callout;
610 
611 	switch (timer_type) {
612 		case TT_DELACK:
613 			t_callout = &tp->t_timers->tt_delack;
614 			break;
615 		case TT_REXMT:
616 			t_callout = &tp->t_timers->tt_rexmt;
617 			break;
618 		case TT_PERSIST:
619 			t_callout = &tp->t_timers->tt_persist;
620 			break;
621 		case TT_KEEP:
622 			t_callout = &tp->t_timers->tt_keep;
623 			break;
624 		case TT_2MSL:
625 			t_callout = &tp->t_timers->tt_2msl;
626 			break;
627 		default:
628 			panic("bad timer_type");
629 		}
630 	return callout_active(t_callout);
631 }
632