Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.79
      1 /*	$NetBSD: tcp_timer.c,v 1.79 2008/04/12 05:58:22 thorpej 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  * 3. All advertising materials mentioning features or use of this software
     51  *    must display the following acknowledgement:
     52  *	This product includes software developed by the NetBSD
     53  *	Foundation, Inc. and its contributors.
     54  * 4. Neither the name of The NetBSD Foundation nor the names of its
     55  *    contributors may be used to endorse or promote products derived
     56  *    from this software without specific prior written permission.
     57  *
     58  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     59  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     60  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     61  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     62  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     63  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     64  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     65  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     66  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     67  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     68  * POSSIBILITY OF SUCH DAMAGE.
     69  */
     70 
     71 /*
     72  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
     73  *	The Regents of the University of California.  All rights reserved.
     74  *
     75  * Redistribution and use in source and binary forms, with or without
     76  * modification, are permitted provided that the following conditions
     77  * are met:
     78  * 1. Redistributions of source code must retain the above copyright
     79  *    notice, this list of conditions and the following disclaimer.
     80  * 2. Redistributions in binary form must reproduce the above copyright
     81  *    notice, this list of conditions and the following disclaimer in the
     82  *    documentation and/or other materials provided with the distribution.
     83  * 3. Neither the name of the University nor the names of its contributors
     84  *    may be used to endorse or promote products derived from this software
     85  *    without specific prior written permission.
     86  *
     87  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     88  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     89  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     90  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     91  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     92  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     93  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     94  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     95  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     96  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     97  * SUCH DAMAGE.
     98  *
     99  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
    100  */
    101 
    102 #include <sys/cdefs.h>
    103 __KERNEL_RCSID(0, "$NetBSD: tcp_timer.c,v 1.79 2008/04/12 05:58:22 thorpej Exp $");
    104 
    105 #include "opt_inet.h"
    106 #include "opt_tcp_debug.h"
    107 
    108 #include <sys/param.h>
    109 #include <sys/systm.h>
    110 #include <sys/malloc.h>
    111 #include <sys/mbuf.h>
    112 #include <sys/socket.h>
    113 #include <sys/socketvar.h>
    114 #include <sys/protosw.h>
    115 #include <sys/errno.h>
    116 #include <sys/kernel.h>
    117 
    118 #include <net/if.h>
    119 #include <net/route.h>
    120 
    121 #include <netinet/in.h>
    122 #include <netinet/in_systm.h>
    123 #include <netinet/ip.h>
    124 #include <netinet/in_pcb.h>
    125 #include <netinet/ip_var.h>
    126 #include <netinet/ip_icmp.h>
    127 
    128 #ifdef INET6
    129 #ifndef INET
    130 #include <netinet/in.h>
    131 #endif
    132 #include <netinet/ip6.h>
    133 #include <netinet6/in6_pcb.h>
    134 #endif
    135 
    136 #include <netinet/tcp.h>
    137 #include <netinet/tcp_fsm.h>
    138 #include <netinet/tcp_seq.h>
    139 #include <netinet/tcp_timer.h>
    140 #include <netinet/tcp_var.h>
    141 #include <netinet/tcp_private.h>
    142 #include <netinet/tcp_congctl.h>
    143 #include <netinet/tcpip.h>
    144 #ifdef TCP_DEBUG
    145 #include <netinet/tcp_debug.h>
    146 #endif
    147 
    148 /*
    149  * Various tunable timer parameters.  These are initialized in tcp_init(),
    150  * unless they are patched.
    151  */
    152 u_int	tcp_keepinit = 0;
    153 u_int	tcp_keepidle = 0;
    154 u_int	tcp_keepintvl = 0;
    155 u_int	tcp_keepcnt = 0;		/* max idle probes */
    156 
    157 int	tcp_maxpersistidle = 0;		/* max idle time in persist */
    158 
    159 /*
    160  * Time to delay the ACK.  This is initialized in tcp_init(), unless
    161  * its patched.
    162  */
    163 int	tcp_delack_ticks = 0;
    164 
    165 void	tcp_timer_rexmt(void *);
    166 void	tcp_timer_persist(void *);
    167 void	tcp_timer_keep(void *);
    168 void	tcp_timer_2msl(void *);
    169 
    170 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
    171 	tcp_timer_rexmt,
    172 	tcp_timer_persist,
    173 	tcp_timer_keep,
    174 	tcp_timer_2msl,
    175 };
    176 
    177 /*
    178  * Timer state initialization, called from tcp_init().
    179  */
    180 void
    181 tcp_timer_init(void)
    182 {
    183 
    184 	if (tcp_keepinit == 0)
    185 		tcp_keepinit = TCPTV_KEEP_INIT;
    186 
    187 	if (tcp_keepidle == 0)
    188 		tcp_keepidle = TCPTV_KEEP_IDLE;
    189 
    190 	if (tcp_keepintvl == 0)
    191 		tcp_keepintvl = TCPTV_KEEPINTVL;
    192 
    193 	if (tcp_keepcnt == 0)
    194 		tcp_keepcnt = TCPTV_KEEPCNT;
    195 
    196 	if (tcp_maxpersistidle == 0)
    197 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
    198 
    199 	if (tcp_delack_ticks == 0)
    200 		tcp_delack_ticks = TCP_DELACK_TICKS;
    201 }
    202 
    203 /*
    204  * Return how many timers are currently being invoked.
    205  */
    206 int
    207 tcp_timers_invoking(struct tcpcb *tp)
    208 {
    209 	int i;
    210 	int count = 0;
    211 
    212 	for (i = 0; i < TCPT_NTIMERS; i++)
    213 		if (callout_invoking(&tp->t_timer[i]))
    214 			count++;
    215 	if (callout_invoking(&tp->t_delack_ch))
    216 		count++;
    217 
    218 	return count;
    219 }
    220 
    221 /*
    222  * Callout to process delayed ACKs for a TCPCB.
    223  */
    224 void
    225 tcp_delack(void *arg)
    226 {
    227 	struct tcpcb *tp = arg;
    228 	int s;
    229 
    230 	/*
    231 	 * If tcp_output() wasn't able to transmit the ACK
    232 	 * for whatever reason, it will restart the delayed
    233 	 * ACK callout.
    234 	 */
    235 
    236 	s = splsoftnet();
    237 	callout_ack(&tp->t_delack_ch);
    238 	if (tcp_isdead(tp)) {
    239 		splx(s);
    240 		return;
    241 	}
    242 
    243 	tp->t_flags |= TF_ACKNOW;
    244 	(void) tcp_output(tp);
    245 	splx(s);
    246 }
    247 
    248 /*
    249  * Tcp protocol timeout routine called every 500 ms.
    250  * Updates the timers in all active tcb's and
    251  * causes finite state machine actions if timers expire.
    252  */
    253 void
    254 tcp_slowtimo(void)
    255 {
    256 	int s;
    257 
    258 	s = splsoftnet();
    259 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    260 	tcp_now++;					/* for timestamps */
    261 	splx(s);
    262 }
    263 
    264 /*
    265  * Cancel all timers for TCP tp.
    266  */
    267 void
    268 tcp_canceltimers(struct tcpcb *tp)
    269 {
    270 	int i;
    271 
    272 	for (i = 0; i < TCPT_NTIMERS; i++)
    273 		TCP_TIMER_DISARM(tp, i);
    274 }
    275 
    276 const int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    277     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    278 
    279 const int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    280 
    281 /*
    282  * TCP timer processing.
    283  */
    284 
    285 void
    286 tcp_timer_rexmt(void *arg)
    287 {
    288 	struct tcpcb *tp = arg;
    289 	uint32_t rto;
    290 	int s;
    291 #ifdef TCP_DEBUG
    292 	struct socket *so = NULL;
    293 	short ostate;
    294 #endif
    295 
    296 	s = splsoftnet();
    297 	callout_ack(&tp->t_timer[TCPT_REXMT]);
    298 	if (tcp_isdead(tp)) {
    299 		splx(s);
    300 		return;
    301 	}
    302 
    303 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
    304 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
    305 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
    306 		extern struct sockaddr_in icmpsrc;
    307 		struct icmp icmp;
    308 
    309 		tp->t_flags &= ~TF_PMTUD_PEND;
    310 
    311 		/* XXX create fake icmp message with relevant entries */
    312 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
    313 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
    314 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
    315 		icmpsrc.sin_addr = tp->t_inpcb->inp_faddr;
    316 		icmp_mtudisc(&icmp, icmpsrc.sin_addr);
    317 
    318 		/*
    319 		 * Notify all connections to the same peer about
    320 		 * new mss and trigger retransmit.
    321 		 */
    322 		in_pcbnotifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
    323 		    tcp_mtudisc);
    324  		splx(s);
    325  		return;
    326  	}
    327 #ifdef TCP_DEBUG
    328 #ifdef INET
    329 	if (tp->t_inpcb)
    330 		so = tp->t_inpcb->inp_socket;
    331 #endif
    332 #ifdef INET6
    333 	if (tp->t_in6pcb)
    334 		so = tp->t_in6pcb->in6p_socket;
    335 #endif
    336 	ostate = tp->t_state;
    337 #endif /* TCP_DEBUG */
    338 
    339 	/*
    340 	 * Clear the SACK scoreboard, reset FACK estimate.
    341 	 */
    342 	tcp_free_sackholes(tp);
    343 	tp->snd_fack = tp->snd_una;
    344 
    345 	/*
    346 	 * Retransmission timer went off.  Message has not
    347 	 * been acked within retransmit interval.  Back off
    348 	 * to a longer retransmit interval and retransmit one segment.
    349 	 */
    350 
    351 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    352 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
    353 		TCP_STATINC(TCP_STAT_TIMEOUTDROP);
    354 		tp = tcp_drop(tp, tp->t_softerror ?
    355 		    tp->t_softerror : ETIMEDOUT);
    356 		goto out;
    357 	}
    358 	TCP_STATINC(TCP_STAT_REXMTTIMEO);
    359 	rto = TCP_REXMTVAL(tp);
    360 	if (rto < tp->t_rttmin)
    361 		rto = tp->t_rttmin;
    362 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    363 	    tp->t_rttmin, TCPTV_REXMTMAX);
    364 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    365 
    366 	/*
    367 	 * If we are losing and we are trying path MTU discovery,
    368 	 * try turning it off.  This will avoid black holes in
    369 	 * the network which suppress or fail to send "packet
    370 	 * too big" ICMP messages.  We should ideally do
    371 	 * lots more sophisticated searching to find the right
    372 	 * value here...
    373 	 */
    374 	if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    375 		TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
    376 
    377 #ifdef INET
    378 		/* try turning PMTUD off */
    379 		if (tp->t_inpcb)
    380 			tp->t_mtudisc = 0;
    381 #endif
    382 #ifdef INET6
    383 		/* try using IPv6 minimum MTU */
    384 		if (tp->t_in6pcb)
    385 			tp->t_mtudisc = 0;
    386 #endif
    387 
    388 		/* XXX: more sophisticated Black hole recovery code? */
    389 	}
    390 
    391 	/*
    392 	 * If losing, let the lower level know and try for
    393 	 * a better route.  Also, if we backed off this far,
    394 	 * our srtt estimate is probably bogus.  Clobber it
    395 	 * so we'll take the next rtt measurement as our srtt;
    396 	 * move the current srtt into rttvar to keep the current
    397 	 * retransmit times until then.
    398 	 */
    399 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    400 #ifdef INET
    401 		if (tp->t_inpcb)
    402 			in_losing(tp->t_inpcb);
    403 #endif
    404 #ifdef INET6
    405 		if (tp->t_in6pcb)
    406 			in6_losing(tp->t_in6pcb);
    407 #endif
    408 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    409 		tp->t_srtt = 0;
    410 	}
    411 	tp->snd_nxt = tp->snd_una;
    412 	tp->snd_high = tp->snd_max;
    413 	/*
    414 	 * If timing a segment in this window, stop the timer.
    415 	 */
    416 	tp->t_rtttime = 0;
    417 	/*
    418 	 * Remember if we are retransmitting a SYN, because if
    419 	 * we do, set the initial congestion window must be set
    420 	 * to 1 segment.
    421 	 */
    422 	if (tp->t_state == TCPS_SYN_SENT)
    423 		tp->t_flags |= TF_SYN_REXMT;
    424 
    425 	/*
    426 	 * Adjust congestion control parameters.
    427 	 */
    428 	tp->t_congctl->slow_retransmit(tp);
    429 
    430 	(void) tcp_output(tp);
    431 
    432  out:
    433 #ifdef TCP_DEBUG
    434 	if (tp && so->so_options & SO_DEBUG)
    435 		tcp_trace(TA_USER, ostate, tp, NULL,
    436 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    437 #endif
    438 	splx(s);
    439 }
    440 
    441 void
    442 tcp_timer_persist(void *arg)
    443 {
    444 	struct tcpcb *tp = arg;
    445 	uint32_t rto;
    446 	int s;
    447 #ifdef TCP_DEBUG
    448 	struct socket *so = NULL;
    449 	short ostate;
    450 #endif
    451 
    452 	s = splsoftnet();
    453 	callout_ack(&tp->t_timer[TCPT_PERSIST]);
    454 	if (tcp_isdead(tp)) {
    455 		splx(s);
    456 		return;
    457 	}
    458 
    459 #ifdef TCP_DEBUG
    460 #ifdef INET
    461 	if (tp->t_inpcb)
    462 		so = tp->t_inpcb->inp_socket;
    463 #endif
    464 #ifdef INET6
    465 	if (tp->t_in6pcb)
    466 		so = tp->t_in6pcb->in6p_socket;
    467 #endif
    468 
    469 	ostate = tp->t_state;
    470 #endif /* TCP_DEBUG */
    471 
    472 	/*
    473 	 * Persistance timer into zero window.
    474 	 * Force a byte to be output, if possible.
    475 	 */
    476 
    477 	/*
    478 	 * Hack: if the peer is dead/unreachable, we do not
    479 	 * time out if the window is closed.  After a full
    480 	 * backoff, drop the connection if the idle time
    481 	 * (no responses to probes) reaches the maximum
    482 	 * backoff that we would use if retransmitting.
    483 	 */
    484 	rto = TCP_REXMTVAL(tp);
    485 	if (rto < tp->t_rttmin)
    486 		rto = tp->t_rttmin;
    487 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    488 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    489 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    490 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
    491 		tp = tcp_drop(tp, ETIMEDOUT);
    492 		goto out;
    493 	}
    494 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
    495 	tcp_setpersist(tp);
    496 	tp->t_force = 1;
    497 	(void) tcp_output(tp);
    498 	tp->t_force = 0;
    499 
    500  out:
    501 #ifdef TCP_DEBUG
    502 	if (tp && so->so_options & SO_DEBUG)
    503 		tcp_trace(TA_USER, ostate, tp, NULL,
    504 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    505 #endif
    506 	splx(s);
    507 }
    508 
    509 void
    510 tcp_timer_keep(void *arg)
    511 {
    512 	struct tcpcb *tp = arg;
    513 	struct socket *so = NULL;	/* Quell compiler warning */
    514 	int s;
    515 #ifdef TCP_DEBUG
    516 	short ostate;
    517 #endif
    518 
    519 	s = splsoftnet();
    520 	callout_ack(&tp->t_timer[TCPT_KEEP]);
    521 	if (tcp_isdead(tp)) {
    522 		splx(s);
    523 		return;
    524 	}
    525 
    526 #ifdef TCP_DEBUG
    527 	ostate = tp->t_state;
    528 #endif /* TCP_DEBUG */
    529 
    530 	/*
    531 	 * Keep-alive timer went off; send something
    532 	 * or drop connection if idle for too long.
    533 	 */
    534 
    535 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
    536 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    537 		goto dropit;
    538 #ifdef INET
    539 	if (tp->t_inpcb)
    540 		so = tp->t_inpcb->inp_socket;
    541 #endif
    542 #ifdef INET6
    543 	if (tp->t_in6pcb)
    544 		so = tp->t_in6pcb->in6p_socket;
    545 #endif
    546 	KASSERT(so != NULL);
    547 	if (so->so_options & SO_KEEPALIVE &&
    548 	    tp->t_state <= TCPS_CLOSE_WAIT) {
    549 	    	if ((tp->t_maxidle > 0) &&
    550 		    ((tcp_now - tp->t_rcvtime) >=
    551 		     tp->t_keepidle + tp->t_maxidle))
    552 			goto dropit;
    553 		/*
    554 		 * Send a packet designed to force a response
    555 		 * if the peer is up and reachable:
    556 		 * either an ACK if the connection is still alive,
    557 		 * or an RST if the peer has closed the connection
    558 		 * due to timeout or reboot.
    559 		 * Using sequence number tp->snd_una-1
    560 		 * causes the transmitted zero-length segment
    561 		 * to lie outside the receive window;
    562 		 * by the protocol spec, this requires the
    563 		 * correspondent TCP to respond.
    564 		 */
    565 		TCP_STATINC(TCP_STAT_KEEPPROBE);
    566 		if (tcp_compat_42) {
    567 			/*
    568 			 * The keepalive packet must have nonzero
    569 			 * length to get a 4.2 host to respond.
    570 			 */
    571 			(void)tcp_respond(tp, tp->t_template,
    572 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
    573 			    tp->snd_una - 1, 0);
    574 		} else {
    575 			(void)tcp_respond(tp, tp->t_template,
    576 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
    577 			    tp->snd_una - 1, 0);
    578 		}
    579 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepintvl);
    580 	} else
    581 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
    582 
    583 #ifdef TCP_DEBUG
    584 	if (tp && so->so_options & SO_DEBUG)
    585 		tcp_trace(TA_USER, ostate, tp, NULL,
    586 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
    587 #endif
    588 	splx(s);
    589 	return;
    590 
    591  dropit:
    592 	TCP_STATINC(TCP_STAT_KEEPDROPS);
    593 	(void) tcp_drop(tp, ETIMEDOUT);
    594 	splx(s);
    595 }
    596 
    597 void
    598 tcp_timer_2msl(void *arg)
    599 {
    600 	struct tcpcb *tp = arg;
    601 	int s;
    602 #ifdef TCP_DEBUG
    603 	struct socket *so = NULL;
    604 	short ostate;
    605 #endif
    606 
    607 	s = splsoftnet();
    608 	callout_ack(&tp->t_timer[TCPT_2MSL]);
    609 	if (tcp_isdead(tp)) {
    610 		splx(s);
    611 		return;
    612 	}
    613 
    614 	/*
    615 	 * 2 MSL timeout went off, clear the SACK scoreboard, reset
    616 	 * the FACK estimate.
    617 	 */
    618 	tcp_free_sackholes(tp);
    619 	tp->snd_fack = tp->snd_una;
    620 
    621 #ifdef TCP_DEBUG
    622 #ifdef INET
    623 	if (tp->t_inpcb)
    624 		so = tp->t_inpcb->inp_socket;
    625 #endif
    626 #ifdef INET6
    627 	if (tp->t_in6pcb)
    628 		so = tp->t_in6pcb->in6p_socket;
    629 #endif
    630 
    631 	ostate = tp->t_state;
    632 #endif /* TCP_DEBUG */
    633 
    634 	/*
    635 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    636 	 * still waiting for peer to close and connection has been idle
    637 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    638 	 * control block.  Otherwise, check again in a bit.
    639 	 */
    640 	if (tp->t_state != TCPS_TIME_WAIT &&
    641 	    ((tp->t_maxidle == 0) ||
    642 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
    643 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
    644 	else
    645 		tp = tcp_close(tp);
    646 
    647 #ifdef TCP_DEBUG
    648 	if (tp && so->so_options & SO_DEBUG)
    649 		tcp_trace(TA_USER, ostate, tp, NULL,
    650 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    651 #endif
    652 	splx(s);
    653 }
    654