xref: /netbsd-src/sys/netinet/tcp_timer.c (revision ed75d7a867996c84cfa88e3b8906816277e957f7)
1 /*	$NetBSD: tcp_timer.c,v 1.95 2018/05/03 07:13:48 maxv Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998, 2001, 2005 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38  * Facility, NASA Ames Research Center.
39  * This code is derived from software contributed to The NetBSD Foundation
40  * by Charles M. Hannum.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
52  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
53  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
54  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
55  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
56  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
57  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
58  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
59  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
60  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
61  * POSSIBILITY OF SUCH DAMAGE.
62  */
63 
64 /*
65  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
66  *	The Regents of the University of California.  All rights reserved.
67  *
68  * Redistribution and use in source and binary forms, with or without
69  * modification, are permitted provided that the following conditions
70  * are met:
71  * 1. Redistributions of source code must retain the above copyright
72  *    notice, this list of conditions and the following disclaimer.
73  * 2. Redistributions in binary form must reproduce the above copyright
74  *    notice, this list of conditions and the following disclaimer in the
75  *    documentation and/or other materials provided with the distribution.
76  * 3. Neither the name of the University nor the names of its contributors
77  *    may be used to endorse or promote products derived from this software
78  *    without specific prior written permission.
79  *
80  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
81  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
82  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
83  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
84  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
85  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
86  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
87  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
88  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
89  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
90  * SUCH DAMAGE.
91  *
92  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
93  */
94 
95 #include <sys/cdefs.h>
96 __KERNEL_RCSID(0, "$NetBSD: tcp_timer.c,v 1.95 2018/05/03 07:13:48 maxv Exp $");
97 
98 #ifdef _KERNEL_OPT
99 #include "opt_inet.h"
100 #include "opt_tcp_debug.h"
101 #include "opt_net_mpsafe.h"
102 #endif
103 
104 #include <sys/param.h>
105 #include <sys/systm.h>
106 #include <sys/mbuf.h>
107 #include <sys/socket.h>
108 #include <sys/socketvar.h>
109 #include <sys/protosw.h>
110 #include <sys/errno.h>
111 #include <sys/kernel.h>
112 #include <sys/callout.h>
113 #include <sys/workqueue.h>
114 
115 #include <net/if.h>
116 
117 #include <netinet/in.h>
118 #include <netinet/in_systm.h>
119 #include <netinet/ip.h>
120 #include <netinet/in_pcb.h>
121 #include <netinet/ip_var.h>
122 #include <netinet/ip_icmp.h>
123 
124 #ifdef INET6
125 #include <netinet/ip6.h>
126 #include <netinet6/in6_pcb.h>
127 #endif
128 
129 #include <netinet/tcp.h>
130 #include <netinet/tcp_fsm.h>
131 #include <netinet/tcp_seq.h>
132 #include <netinet/tcp_timer.h>
133 #include <netinet/tcp_var.h>
134 #include <netinet/tcp_private.h>
135 #include <netinet/tcp_congctl.h>
136 #ifdef TCP_DEBUG
137 #include <netinet/tcp_debug.h>
138 #endif
139 
140 /*
141  * Various tunable timer parameters.  These are initialized in tcp_init(),
142  * unless they are patched.
143  */
144 u_int	tcp_keepinit = 0;
145 u_int	tcp_keepidle = 0;
146 u_int	tcp_keepintvl = 0;
147 u_int	tcp_keepcnt = 0;		/* max idle probes */
148 
149 int	tcp_maxpersistidle = 0;		/* max idle time in persist */
150 
151 static callout_t	tcp_slowtimo_ch;
152 #ifdef NET_MPSAFE
153 static struct workqueue	*tcp_slowtimo_wq;
154 static struct work	tcp_slowtimo_wk;
155 #endif
156 
157 static void tcp_slowtimo_work(struct work *, void *);
158 static void tcp_slowtimo(void *);
159 
160 /*
161  * Time to delay the ACK.  This is initialized in tcp_init(), unless
162  * its patched.
163  */
164 int	tcp_delack_ticks = 0;
165 
166 void	tcp_timer_rexmt(void *);
167 void	tcp_timer_persist(void *);
168 void	tcp_timer_keep(void *);
169 void	tcp_timer_2msl(void *);
170 
171 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
172 	tcp_timer_rexmt,
173 	tcp_timer_persist,
174 	tcp_timer_keep,
175 	tcp_timer_2msl,
176 };
177 
178 /*
179  * Timer state initialization, called from tcp_init().
180  */
181 void
182 tcp_timer_init(void)
183 {
184 
185 	if (tcp_keepinit == 0)
186 		tcp_keepinit = TCPTV_KEEP_INIT;
187 
188 	if (tcp_keepidle == 0)
189 		tcp_keepidle = TCPTV_KEEP_IDLE;
190 
191 	if (tcp_keepintvl == 0)
192 		tcp_keepintvl = TCPTV_KEEPINTVL;
193 
194 	if (tcp_keepcnt == 0)
195 		tcp_keepcnt = TCPTV_KEEPCNT;
196 
197 	if (tcp_maxpersistidle == 0)
198 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
199 
200 	if (tcp_delack_ticks == 0)
201 		tcp_delack_ticks = TCP_DELACK_TICKS;
202 }
203 
204 void
205 tcp_slowtimo_init(void)
206 {
207 #ifdef NET_MPSAFE
208 	int error;
209 
210 	error = workqueue_create(&tcp_slowtimo_wq, "tcp_slowtimo",
211 	    tcp_slowtimo_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
212 	if (error != 0)
213 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
214 #endif
215 	callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
216 	callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
217 }
218 
219 /*
220  * Callout to process delayed ACKs for a TCPCB.
221  */
222 void
223 tcp_delack(void *arg)
224 {
225 	struct tcpcb *tp = arg;
226 
227 	/*
228 	 * If tcp_output() wasn't able to transmit the ACK
229 	 * for whatever reason, it will restart the delayed
230 	 * ACK callout.
231 	 */
232 
233 	mutex_enter(softnet_lock);
234 	if ((tp->t_flags & (TF_DEAD | TF_DELACK)) != TF_DELACK) {
235 		mutex_exit(softnet_lock);
236 		return;
237 	}
238 	if (!callout_expired(&tp->t_delack_ch)) {
239 		mutex_exit(softnet_lock);
240 		return;
241 	}
242 
243 	tp->t_flags |= TF_ACKNOW;
244 	KERNEL_LOCK(1, NULL);
245 	(void) tcp_output(tp);
246 	KERNEL_UNLOCK_ONE(NULL);
247 	mutex_exit(softnet_lock);
248 }
249 
250 /*
251  * Tcp protocol timeout routine called every 500 ms.
252  * Updates the timers in all active tcb's and
253  * causes finite state machine actions if timers expire.
254  */
255 static void
256 tcp_slowtimo_work(struct work *wk, void *arg)
257 {
258 
259 	mutex_enter(softnet_lock);
260 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
261 	tcp_now++;					/* for timestamps */
262 	mutex_exit(softnet_lock);
263 
264 	callout_schedule(&tcp_slowtimo_ch, hz / PR_SLOWHZ);
265 }
266 
267 static void
268 tcp_slowtimo(void *arg)
269 {
270 
271 #ifdef NET_MPSAFE
272 	workqueue_enqueue(tcp_slowtimo_wq, &tcp_slowtimo_wk, NULL);
273 #else
274 	tcp_slowtimo_work(NULL, NULL);
275 #endif
276 }
277 
278 /*
279  * Cancel all timers for TCP tp.
280  */
281 void
282 tcp_canceltimers(struct tcpcb *tp)
283 {
284 	int i;
285 
286 	for (i = 0; i < TCPT_NTIMERS; i++)
287 		TCP_TIMER_DISARM(tp, i);
288 }
289 
290 const int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
291     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
292 
293 const int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
294 
295 /*
296  * TCP timer processing.
297  */
298 
299 void
300 tcp_timer_rexmt(void *arg)
301 {
302 	struct tcpcb *tp = arg;
303 	uint32_t rto;
304 #ifdef TCP_DEBUG
305 	struct socket *so = NULL;
306 	short ostate;
307 #endif
308 
309 	mutex_enter(softnet_lock);
310 	if ((tp->t_flags & TF_DEAD) != 0) {
311 		mutex_exit(softnet_lock);
312 		return;
313 	}
314 	if (!callout_expired(&tp->t_timer[TCPT_REXMT])) {
315 		mutex_exit(softnet_lock);
316 		return;
317 	}
318 
319 	KERNEL_LOCK(1, NULL);
320 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
321 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
322 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
323 		extern struct sockaddr_in icmpsrc;
324 		struct icmp icmp;
325 
326 		tp->t_flags &= ~TF_PMTUD_PEND;
327 
328 		/* XXX create fake icmp message with relevant entries */
329 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
330 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
331 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
332 		icmpsrc.sin_addr = tp->t_inpcb->inp_faddr;
333 		icmp_mtudisc(&icmp, icmpsrc.sin_addr);
334 
335 		/*
336 		 * Notify all connections to the same peer about
337 		 * new mss and trigger retransmit.
338 		 */
339 		in_pcbnotifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
340 		    tcp_mtudisc);
341 		KERNEL_UNLOCK_ONE(NULL);
342 		mutex_exit(softnet_lock);
343  		return;
344  	}
345 #ifdef TCP_DEBUG
346 	if (tp->t_inpcb)
347 		so = tp->t_inpcb->inp_socket;
348 #ifdef INET6
349 	if (tp->t_in6pcb)
350 		so = tp->t_in6pcb->in6p_socket;
351 #endif
352 	ostate = tp->t_state;
353 #endif /* TCP_DEBUG */
354 
355 	/*
356 	 * Clear the SACK scoreboard, reset FACK estimate.
357 	 */
358 	tcp_free_sackholes(tp);
359 	tp->snd_fack = tp->snd_una;
360 
361 	/*
362 	 * Retransmission timer went off.  Message has not
363 	 * been acked within retransmit interval.  Back off
364 	 * to a longer retransmit interval and retransmit one segment.
365 	 */
366 
367 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
368 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
369 		TCP_STATINC(TCP_STAT_TIMEOUTDROP);
370 		tp = tcp_drop(tp, tp->t_softerror ?
371 		    tp->t_softerror : ETIMEDOUT);
372 		goto out;
373 	}
374 	TCP_STATINC(TCP_STAT_REXMTTIMEO);
375 	rto = TCP_REXMTVAL(tp);
376 	if (rto < tp->t_rttmin)
377 		rto = tp->t_rttmin;
378 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
379 	    tp->t_rttmin, TCPTV_REXMTMAX);
380 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
381 
382 	/*
383 	 * If we are losing and we are trying path MTU discovery,
384 	 * try turning it off.  This will avoid black holes in
385 	 * the network which suppress or fail to send "packet
386 	 * too big" ICMP messages.  We should ideally do
387 	 * lots more sophisticated searching to find the right
388 	 * value here...
389 	 */
390 	if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
391 		TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
392 
393 		/* try turning PMTUD off */
394 		if (tp->t_inpcb)
395 			tp->t_mtudisc = 0;
396 #ifdef INET6
397 		/* try using IPv6 minimum MTU */
398 		if (tp->t_in6pcb)
399 			tp->t_mtudisc = 0;
400 #endif
401 
402 		/* XXX: more sophisticated Black hole recovery code? */
403 	}
404 
405 	/*
406 	 * If losing, let the lower level know and try for
407 	 * a better route.  Also, if we backed off this far,
408 	 * our srtt estimate is probably bogus.  Clobber it
409 	 * so we'll take the next rtt measurement as our srtt;
410 	 * move the current srtt into rttvar to keep the current
411 	 * retransmit times until then.
412 	 */
413 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
414 		if (tp->t_inpcb)
415 			in_losing(tp->t_inpcb);
416 #ifdef INET6
417 		if (tp->t_in6pcb)
418 			in6_losing(tp->t_in6pcb);
419 #endif
420 		/*
421 		 * This operation is not described in RFC2988.  The
422 		 * point is to keep srtt+4*rttvar constant, so we
423 		 * should shift right 2 bits to divide by 4, and then
424 		 * shift right one bit because the storage
425 		 * representation of rttvar is 1/16s vs 1/32s for
426 		 * srtt.
427 		 */
428 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
429 		tp->t_srtt = 0;
430 	}
431 	tp->snd_nxt = tp->snd_una;
432 	tp->snd_high = tp->snd_max;
433 	/*
434 	 * If timing a segment in this window, stop the timer.
435 	 */
436 	tp->t_rtttime = 0;
437 	/*
438 	 * Remember if we are retransmitting a SYN, because if
439 	 * we do, set the initial congestion window must be set
440 	 * to 1 segment.
441 	 */
442 	if (tp->t_state == TCPS_SYN_SENT)
443 		tp->t_flags |= TF_SYN_REXMT;
444 
445 	/*
446 	 * Adjust congestion control parameters.
447 	 */
448 	tp->t_congctl->slow_retransmit(tp);
449 
450 	(void) tcp_output(tp);
451 
452  out:
453 #ifdef TCP_DEBUG
454 	if (tp && so->so_options & SO_DEBUG)
455 		tcp_trace(TA_USER, ostate, tp, NULL,
456 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
457 #endif
458 	KERNEL_UNLOCK_ONE(NULL);
459 	mutex_exit(softnet_lock);
460 }
461 
462 void
463 tcp_timer_persist(void *arg)
464 {
465 	struct tcpcb *tp = arg;
466 	uint32_t rto;
467 #ifdef TCP_DEBUG
468 	struct socket *so = NULL;
469 	short ostate;
470 #endif
471 
472 	mutex_enter(softnet_lock);
473 	if ((tp->t_flags & TF_DEAD) != 0) {
474 		mutex_exit(softnet_lock);
475 		return;
476 	}
477 	if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
478 		mutex_exit(softnet_lock);
479 		return;
480 	}
481 
482 	KERNEL_LOCK(1, NULL);
483 #ifdef TCP_DEBUG
484 	if (tp->t_inpcb)
485 		so = tp->t_inpcb->inp_socket;
486 #ifdef INET6
487 	if (tp->t_in6pcb)
488 		so = tp->t_in6pcb->in6p_socket;
489 #endif
490 
491 	ostate = tp->t_state;
492 #endif /* TCP_DEBUG */
493 
494 	/*
495 	 * Persistance timer into zero window.
496 	 * Force a byte to be output, if possible.
497 	 */
498 
499 	/*
500 	 * Hack: if the peer is dead/unreachable, we do not
501 	 * time out if the window is closed.  After a full
502 	 * backoff, drop the connection if the idle time
503 	 * (no responses to probes) reaches the maximum
504 	 * backoff that we would use if retransmitting.
505 	 */
506 	rto = TCP_REXMTVAL(tp);
507 	if (rto < tp->t_rttmin)
508 		rto = tp->t_rttmin;
509 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
510 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
511 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
512 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
513 		tp = tcp_drop(tp, ETIMEDOUT);
514 		goto out;
515 	}
516 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
517 	tcp_setpersist(tp);
518 	tp->t_force = 1;
519 	(void) tcp_output(tp);
520 	tp->t_force = 0;
521 
522  out:
523 #ifdef TCP_DEBUG
524 	if (tp && so->so_options & SO_DEBUG)
525 		tcp_trace(TA_USER, ostate, tp, NULL,
526 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
527 #endif
528 	KERNEL_UNLOCK_ONE(NULL);
529 	mutex_exit(softnet_lock);
530 }
531 
532 void
533 tcp_timer_keep(void *arg)
534 {
535 	struct tcpcb *tp = arg;
536 	struct socket *so = NULL;	/* Quell compiler warning */
537 #ifdef TCP_DEBUG
538 	short ostate;
539 #endif
540 
541 	mutex_enter(softnet_lock);
542 	if ((tp->t_flags & TF_DEAD) != 0) {
543 		mutex_exit(softnet_lock);
544 		return;
545 	}
546 	if (!callout_expired(&tp->t_timer[TCPT_KEEP])) {
547 		mutex_exit(softnet_lock);
548 		return;
549 	}
550 
551 	KERNEL_LOCK(1, NULL);
552 
553 #ifdef TCP_DEBUG
554 	ostate = tp->t_state;
555 #endif /* TCP_DEBUG */
556 
557 	/*
558 	 * Keep-alive timer went off; send something
559 	 * or drop connection if idle for too long.
560 	 */
561 
562 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
563 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
564 		goto dropit;
565 	if (tp->t_inpcb)
566 		so = tp->t_inpcb->inp_socket;
567 #ifdef INET6
568 	if (tp->t_in6pcb)
569 		so = tp->t_in6pcb->in6p_socket;
570 #endif
571 	KASSERT(so != NULL);
572 	if (so->so_options & SO_KEEPALIVE &&
573 	    tp->t_state <= TCPS_CLOSE_WAIT) {
574 	    	if ((tp->t_maxidle > 0) &&
575 		    ((tcp_now - tp->t_rcvtime) >=
576 		     tp->t_keepidle + tp->t_maxidle))
577 			goto dropit;
578 		/*
579 		 * Send a packet designed to force a response
580 		 * if the peer is up and reachable:
581 		 * either an ACK if the connection is still alive,
582 		 * or an RST if the peer has closed the connection
583 		 * due to timeout or reboot.
584 		 * Using sequence number tp->snd_una-1
585 		 * causes the transmitted zero-length segment
586 		 * to lie outside the receive window;
587 		 * by the protocol spec, this requires the
588 		 * correspondent TCP to respond.
589 		 */
590 		TCP_STATINC(TCP_STAT_KEEPPROBE);
591 
592 		(void)tcp_respond(tp, tp->t_template,
593 		    NULL, NULL, tp->rcv_nxt,
594 		    tp->snd_una - 1, 0);
595 
596 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepintvl);
597 	} else
598 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
599 
600 #ifdef TCP_DEBUG
601 	if (tp && so->so_options & SO_DEBUG)
602 		tcp_trace(TA_USER, ostate, tp, NULL,
603 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
604 #endif
605 	KERNEL_UNLOCK_ONE(NULL);
606 	mutex_exit(softnet_lock);
607 	return;
608 
609  dropit:
610 	TCP_STATINC(TCP_STAT_KEEPDROPS);
611 	(void) tcp_drop(tp, ETIMEDOUT);
612 	KERNEL_UNLOCK_ONE(NULL);
613 	mutex_exit(softnet_lock);
614 }
615 
616 void
617 tcp_timer_2msl(void *arg)
618 {
619 	struct tcpcb *tp = arg;
620 #ifdef TCP_DEBUG
621 	struct socket *so = NULL;
622 	short ostate;
623 #endif
624 
625 	mutex_enter(softnet_lock);
626 	if ((tp->t_flags & TF_DEAD) != 0) {
627 		mutex_exit(softnet_lock);
628 		return;
629 	}
630 	if (!callout_expired(&tp->t_timer[TCPT_2MSL])) {
631 		mutex_exit(softnet_lock);
632 		return;
633 	}
634 
635 	/*
636 	 * 2 MSL timeout went off, clear the SACK scoreboard, reset
637 	 * the FACK estimate.
638 	 */
639 	KERNEL_LOCK(1, NULL);
640 	tcp_free_sackholes(tp);
641 	tp->snd_fack = tp->snd_una;
642 
643 #ifdef TCP_DEBUG
644 	if (tp->t_inpcb)
645 		so = tp->t_inpcb->inp_socket;
646 #ifdef INET6
647 	if (tp->t_in6pcb)
648 		so = tp->t_in6pcb->in6p_socket;
649 #endif
650 
651 	ostate = tp->t_state;
652 #endif /* TCP_DEBUG */
653 
654 	/*
655 	 * 2 MSL timeout in shutdown went off.  If we're closed but
656 	 * still waiting for peer to close and connection has been idle
657 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
658 	 * control block.  Otherwise, check again in a bit.
659 	 */
660 	if (tp->t_state != TCPS_TIME_WAIT &&
661 	    ((tp->t_maxidle == 0) ||
662 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
663 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
664 	else
665 		tp = tcp_close(tp);
666 
667 #ifdef TCP_DEBUG
668 	if (tp && so->so_options & SO_DEBUG)
669 		tcp_trace(TA_USER, ostate, tp, NULL,
670 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
671 #endif
672 	KERNEL_UNLOCK_ONE(NULL);
673 	mutex_exit(softnet_lock);
674 }
675