Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.83
      1 /*	$NetBSD: tcp_timer.c,v 1.83 2008/11/09 17:32:38 bouyer 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 1.83 2008/11/09 17:32:38 bouyer Exp $");
     97 
     98 #include "opt_inet.h"
     99 #include "opt_tcp_debug.h"
    100 
    101 #include <sys/param.h>
    102 #include <sys/systm.h>
    103 #include <sys/malloc.h>
    104 #include <sys/mbuf.h>
    105 #include <sys/socket.h>
    106 #include <sys/socketvar.h>
    107 #include <sys/protosw.h>
    108 #include <sys/errno.h>
    109 #include <sys/kernel.h>
    110 
    111 #include <net/if.h>
    112 #include <net/route.h>
    113 
    114 #include <netinet/in.h>
    115 #include <netinet/in_systm.h>
    116 #include <netinet/ip.h>
    117 #include <netinet/in_pcb.h>
    118 #include <netinet/ip_var.h>
    119 #include <netinet/ip_icmp.h>
    120 
    121 #ifdef INET6
    122 #ifndef INET
    123 #include <netinet/in.h>
    124 #endif
    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 #include <netinet/tcpip.h>
    137 #ifdef TCP_DEBUG
    138 #include <netinet/tcp_debug.h>
    139 #endif
    140 
    141 /*
    142  * Various tunable timer parameters.  These are initialized in tcp_init(),
    143  * unless they are patched.
    144  */
    145 u_int	tcp_keepinit = 0;
    146 u_int	tcp_keepidle = 0;
    147 u_int	tcp_keepintvl = 0;
    148 u_int	tcp_keepcnt = 0;		/* max idle probes */
    149 
    150 int	tcp_maxpersistidle = 0;		/* max idle time in persist */
    151 
    152 /*
    153  * Time to delay the ACK.  This is initialized in tcp_init(), unless
    154  * its patched.
    155  */
    156 int	tcp_delack_ticks = 0;
    157 
    158 void	tcp_timer_rexmt(void *);
    159 void	tcp_timer_persist(void *);
    160 void	tcp_timer_keep(void *);
    161 void	tcp_timer_2msl(void *);
    162 
    163 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
    164 	tcp_timer_rexmt,
    165 	tcp_timer_persist,
    166 	tcp_timer_keep,
    167 	tcp_timer_2msl,
    168 };
    169 
    170 /*
    171  * Timer state initialization, called from tcp_init().
    172  */
    173 void
    174 tcp_timer_init(void)
    175 {
    176 
    177 	if (tcp_keepinit == 0)
    178 		tcp_keepinit = TCPTV_KEEP_INIT;
    179 
    180 	if (tcp_keepidle == 0)
    181 		tcp_keepidle = TCPTV_KEEP_IDLE;
    182 
    183 	if (tcp_keepintvl == 0)
    184 		tcp_keepintvl = TCPTV_KEEPINTVL;
    185 
    186 	if (tcp_keepcnt == 0)
    187 		tcp_keepcnt = TCPTV_KEEPCNT;
    188 
    189 	if (tcp_maxpersistidle == 0)
    190 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
    191 
    192 	if (tcp_delack_ticks == 0)
    193 		tcp_delack_ticks = TCP_DELACK_TICKS;
    194 }
    195 
    196 /*
    197  * Callout to process delayed ACKs for a TCPCB.
    198  */
    199 void
    200 tcp_delack(void *arg)
    201 {
    202 	struct tcpcb *tp = arg;
    203 
    204 	/*
    205 	 * If tcp_output() wasn't able to transmit the ACK
    206 	 * for whatever reason, it will restart the delayed
    207 	 * ACK callout.
    208 	 */
    209 
    210 	mutex_enter(softnet_lock);
    211 	if ((tp->t_flags & (TF_DEAD | TF_DELACK)) != TF_DELACK) {
    212 		mutex_exit(softnet_lock);
    213 		return;
    214 	}
    215 	if (!callout_expired(&tp->t_delack_ch)) {
    216 		mutex_exit(softnet_lock);
    217 		return;
    218 	}
    219 
    220 	tp->t_flags |= TF_ACKNOW;
    221 	KERNEL_LOCK(1, NULL);
    222 	(void) tcp_output(tp);
    223 	KERNEL_UNLOCK_ONE(NULL);
    224 	mutex_exit(softnet_lock);
    225 }
    226 
    227 /*
    228  * Tcp protocol timeout routine called every 500 ms.
    229  * Updates the timers in all active tcb's and
    230  * causes finite state machine actions if timers expire.
    231  */
    232 void
    233 tcp_slowtimo(void)
    234 {
    235 
    236 	mutex_enter(softnet_lock);
    237 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    238 	tcp_now++;					/* for timestamps */
    239 	mutex_exit(softnet_lock);
    240 }
    241 
    242 /*
    243  * Cancel all timers for TCP tp.
    244  */
    245 void
    246 tcp_canceltimers(struct tcpcb *tp)
    247 {
    248 	int i;
    249 
    250 	for (i = 0; i < TCPT_NTIMERS; i++)
    251 		TCP_TIMER_DISARM(tp, i);
    252 }
    253 
    254 const int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    255     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    256 
    257 const int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    258 
    259 /*
    260  * TCP timer processing.
    261  */
    262 
    263 void
    264 tcp_timer_rexmt(void *arg)
    265 {
    266 	struct tcpcb *tp = arg;
    267 	uint32_t rto;
    268 #ifdef TCP_DEBUG
    269 	struct socket *so = NULL;
    270 	short ostate;
    271 #endif
    272 
    273 	mutex_enter(softnet_lock);
    274 	if ((tp->t_flags & TF_DEAD) != 0) {
    275 		mutex_exit(softnet_lock);
    276 		return;
    277 	}
    278 	if (!callout_expired(&tp->t_timer[TCPT_REXMT])) {
    279 		mutex_exit(softnet_lock);
    280 		return;
    281 	}
    282 
    283 	KERNEL_LOCK(1, NULL);
    284 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
    285 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
    286 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
    287 		extern struct sockaddr_in icmpsrc;
    288 		struct icmp icmp;
    289 
    290 		tp->t_flags &= ~TF_PMTUD_PEND;
    291 
    292 		/* XXX create fake icmp message with relevant entries */
    293 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
    294 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
    295 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
    296 		icmpsrc.sin_addr = tp->t_inpcb->inp_faddr;
    297 		icmp_mtudisc(&icmp, icmpsrc.sin_addr);
    298 
    299 		/*
    300 		 * Notify all connections to the same peer about
    301 		 * new mss and trigger retransmit.
    302 		 */
    303 		in_pcbnotifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
    304 		    tcp_mtudisc);
    305 		KERNEL_UNLOCK_ONE(NULL);
    306 		mutex_exit(softnet_lock);
    307  		return;
    308  	}
    309 #ifdef TCP_DEBUG
    310 #ifdef INET
    311 	if (tp->t_inpcb)
    312 		so = tp->t_inpcb->inp_socket;
    313 #endif
    314 #ifdef INET6
    315 	if (tp->t_in6pcb)
    316 		so = tp->t_in6pcb->in6p_socket;
    317 #endif
    318 	ostate = tp->t_state;
    319 #endif /* TCP_DEBUG */
    320 
    321 	/*
    322 	 * Clear the SACK scoreboard, reset FACK estimate.
    323 	 */
    324 	tcp_free_sackholes(tp);
    325 	tp->snd_fack = tp->snd_una;
    326 
    327 	/*
    328 	 * Retransmission timer went off.  Message has not
    329 	 * been acked within retransmit interval.  Back off
    330 	 * to a longer retransmit interval and retransmit one segment.
    331 	 */
    332 
    333 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    334 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
    335 		TCP_STATINC(TCP_STAT_TIMEOUTDROP);
    336 		tp = tcp_drop(tp, tp->t_softerror ?
    337 		    tp->t_softerror : ETIMEDOUT);
    338 		goto out;
    339 	}
    340 	TCP_STATINC(TCP_STAT_REXMTTIMEO);
    341 	rto = TCP_REXMTVAL(tp);
    342 	if (rto < tp->t_rttmin)
    343 		rto = tp->t_rttmin;
    344 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    345 	    tp->t_rttmin, TCPTV_REXMTMAX);
    346 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    347 
    348 	/*
    349 	 * If we are losing and we are trying path MTU discovery,
    350 	 * try turning it off.  This will avoid black holes in
    351 	 * the network which suppress or fail to send "packet
    352 	 * too big" ICMP messages.  We should ideally do
    353 	 * lots more sophisticated searching to find the right
    354 	 * value here...
    355 	 */
    356 	if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    357 		TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
    358 
    359 #ifdef INET
    360 		/* try turning PMTUD off */
    361 		if (tp->t_inpcb)
    362 			tp->t_mtudisc = 0;
    363 #endif
    364 #ifdef INET6
    365 		/* try using IPv6 minimum MTU */
    366 		if (tp->t_in6pcb)
    367 			tp->t_mtudisc = 0;
    368 #endif
    369 
    370 		/* XXX: more sophisticated Black hole recovery code? */
    371 	}
    372 
    373 	/*
    374 	 * If losing, let the lower level know and try for
    375 	 * a better route.  Also, if we backed off this far,
    376 	 * our srtt estimate is probably bogus.  Clobber it
    377 	 * so we'll take the next rtt measurement as our srtt;
    378 	 * move the current srtt into rttvar to keep the current
    379 	 * retransmit times until then.
    380 	 */
    381 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    382 #ifdef INET
    383 		if (tp->t_inpcb)
    384 			in_losing(tp->t_inpcb);
    385 #endif
    386 #ifdef INET6
    387 		if (tp->t_in6pcb)
    388 			in6_losing(tp->t_in6pcb);
    389 #endif
    390 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    391 		tp->t_srtt = 0;
    392 	}
    393 	tp->snd_nxt = tp->snd_una;
    394 	tp->snd_high = tp->snd_max;
    395 	/*
    396 	 * If timing a segment in this window, stop the timer.
    397 	 */
    398 	tp->t_rtttime = 0;
    399 	/*
    400 	 * Remember if we are retransmitting a SYN, because if
    401 	 * we do, set the initial congestion window must be set
    402 	 * to 1 segment.
    403 	 */
    404 	if (tp->t_state == TCPS_SYN_SENT)
    405 		tp->t_flags |= TF_SYN_REXMT;
    406 
    407 	/*
    408 	 * Adjust congestion control parameters.
    409 	 */
    410 	tp->t_congctl->slow_retransmit(tp);
    411 
    412 	(void) tcp_output(tp);
    413 
    414  out:
    415 #ifdef TCP_DEBUG
    416 	if (tp && so->so_options & SO_DEBUG)
    417 		tcp_trace(TA_USER, ostate, tp, NULL,
    418 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    419 #endif
    420 	KERNEL_UNLOCK_ONE(NULL);
    421 	mutex_exit(softnet_lock);
    422 }
    423 
    424 void
    425 tcp_timer_persist(void *arg)
    426 {
    427 	struct tcpcb *tp = arg;
    428 	uint32_t rto;
    429 #ifdef TCP_DEBUG
    430 	struct socket *so = NULL;
    431 	short ostate;
    432 #endif
    433 
    434 	mutex_enter(softnet_lock);
    435 	if ((tp->t_flags & TF_DEAD) != 0) {
    436 		mutex_exit(softnet_lock);
    437 		return;
    438 	}
    439 	if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
    440 		mutex_exit(softnet_lock);
    441 		return;
    442 	}
    443 
    444 	KERNEL_LOCK(1, NULL);
    445 #ifdef TCP_DEBUG
    446 #ifdef INET
    447 	if (tp->t_inpcb)
    448 		so = tp->t_inpcb->inp_socket;
    449 #endif
    450 #ifdef INET6
    451 	if (tp->t_in6pcb)
    452 		so = tp->t_in6pcb->in6p_socket;
    453 #endif
    454 
    455 	ostate = tp->t_state;
    456 #endif /* TCP_DEBUG */
    457 
    458 	/*
    459 	 * Persistance timer into zero window.
    460 	 * Force a byte to be output, if possible.
    461 	 */
    462 
    463 	/*
    464 	 * Hack: if the peer is dead/unreachable, we do not
    465 	 * time out if the window is closed.  After a full
    466 	 * backoff, drop the connection if the idle time
    467 	 * (no responses to probes) reaches the maximum
    468 	 * backoff that we would use if retransmitting.
    469 	 */
    470 	rto = TCP_REXMTVAL(tp);
    471 	if (rto < tp->t_rttmin)
    472 		rto = tp->t_rttmin;
    473 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    474 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    475 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    476 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
    477 		tp = tcp_drop(tp, ETIMEDOUT);
    478 		goto out;
    479 	}
    480 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
    481 	tcp_setpersist(tp);
    482 	tp->t_force = 1;
    483 	(void) tcp_output(tp);
    484 	tp->t_force = 0;
    485 
    486  out:
    487 #ifdef TCP_DEBUG
    488 	if (tp && so->so_options & SO_DEBUG)
    489 		tcp_trace(TA_USER, ostate, tp, NULL,
    490 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    491 #endif
    492 	KERNEL_UNLOCK_ONE(NULL);
    493 	mutex_exit(softnet_lock);
    494 }
    495 
    496 void
    497 tcp_timer_keep(void *arg)
    498 {
    499 	struct tcpcb *tp = arg;
    500 	struct socket *so = NULL;	/* Quell compiler warning */
    501 #ifdef TCP_DEBUG
    502 	short ostate;
    503 #endif
    504 
    505 	mutex_enter(softnet_lock);
    506 	if ((tp->t_flags & TF_DEAD) != 0) {
    507 		mutex_exit(softnet_lock);
    508 		return;
    509 	}
    510 	if (!callout_expired(&tp->t_timer[TCPT_KEEP])) {
    511 		mutex_exit(softnet_lock);
    512 		return;
    513 	}
    514 
    515 	KERNEL_LOCK(1, NULL);
    516 
    517 #ifdef TCP_DEBUG
    518 	ostate = tp->t_state;
    519 #endif /* TCP_DEBUG */
    520 
    521 	/*
    522 	 * Keep-alive timer went off; send something
    523 	 * or drop connection if idle for too long.
    524 	 */
    525 
    526 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
    527 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    528 		goto dropit;
    529 #ifdef INET
    530 	if (tp->t_inpcb)
    531 		so = tp->t_inpcb->inp_socket;
    532 #endif
    533 #ifdef INET6
    534 	if (tp->t_in6pcb)
    535 		so = tp->t_in6pcb->in6p_socket;
    536 #endif
    537 	KASSERT(so != NULL);
    538 	if (so->so_options & SO_KEEPALIVE &&
    539 	    tp->t_state <= TCPS_CLOSE_WAIT) {
    540 	    	if ((tp->t_maxidle > 0) &&
    541 		    ((tcp_now - tp->t_rcvtime) >=
    542 		     tp->t_keepidle + tp->t_maxidle))
    543 			goto dropit;
    544 		/*
    545 		 * Send a packet designed to force a response
    546 		 * if the peer is up and reachable:
    547 		 * either an ACK if the connection is still alive,
    548 		 * or an RST if the peer has closed the connection
    549 		 * due to timeout or reboot.
    550 		 * Using sequence number tp->snd_una-1
    551 		 * causes the transmitted zero-length segment
    552 		 * to lie outside the receive window;
    553 		 * by the protocol spec, this requires the
    554 		 * correspondent TCP to respond.
    555 		 */
    556 		TCP_STATINC(TCP_STAT_KEEPPROBE);
    557 		if (tcp_compat_42) {
    558 			/*
    559 			 * The keepalive packet must have nonzero
    560 			 * length to get a 4.2 host to respond.
    561 			 */
    562 			(void)tcp_respond(tp, tp->t_template,
    563 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
    564 			    tp->snd_una - 1, 0);
    565 		} else {
    566 			(void)tcp_respond(tp, tp->t_template,
    567 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
    568 			    tp->snd_una - 1, 0);
    569 		}
    570 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepintvl);
    571 	} else
    572 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
    573 
    574 #ifdef TCP_DEBUG
    575 	if (tp && so->so_options & SO_DEBUG)
    576 		tcp_trace(TA_USER, ostate, tp, NULL,
    577 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
    578 #endif
    579 	KERNEL_UNLOCK_ONE(NULL);
    580 	mutex_exit(softnet_lock);
    581 	return;
    582 
    583  dropit:
    584 	TCP_STATINC(TCP_STAT_KEEPDROPS);
    585 	(void) tcp_drop(tp, ETIMEDOUT);
    586 	KERNEL_UNLOCK_ONE(NULL);
    587 	mutex_exit(softnet_lock);
    588 }
    589 
    590 void
    591 tcp_timer_2msl(void *arg)
    592 {
    593 	struct tcpcb *tp = arg;
    594 #ifdef TCP_DEBUG
    595 	struct socket *so = NULL;
    596 	short ostate;
    597 #endif
    598 
    599 	mutex_enter(softnet_lock);
    600 	if ((tp->t_flags & TF_DEAD) != 0) {
    601 		mutex_exit(softnet_lock);
    602 		return;
    603 	}
    604 	if (!callout_expired(&tp->t_timer[TCPT_2MSL])) {
    605 		mutex_exit(softnet_lock);
    606 		return;
    607 	}
    608 
    609 	/*
    610 	 * 2 MSL timeout went off, clear the SACK scoreboard, reset
    611 	 * the FACK estimate.
    612 	 */
    613 	KERNEL_LOCK(1, NULL);
    614 	tcp_free_sackholes(tp);
    615 	tp->snd_fack = tp->snd_una;
    616 
    617 #ifdef TCP_DEBUG
    618 #ifdef INET
    619 	if (tp->t_inpcb)
    620 		so = tp->t_inpcb->inp_socket;
    621 #endif
    622 #ifdef INET6
    623 	if (tp->t_in6pcb)
    624 		so = tp->t_in6pcb->in6p_socket;
    625 #endif
    626 
    627 	ostate = tp->t_state;
    628 #endif /* TCP_DEBUG */
    629 
    630 	/*
    631 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    632 	 * still waiting for peer to close and connection has been idle
    633 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    634 	 * control block.  Otherwise, check again in a bit.
    635 	 */
    636 	if (tp->t_state != TCPS_TIME_WAIT &&
    637 	    ((tp->t_maxidle == 0) ||
    638 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
    639 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
    640 	else
    641 		tp = tcp_close(tp);
    642 
    643 #ifdef TCP_DEBUG
    644 	if (tp && so->so_options & SO_DEBUG)
    645 		tcp_trace(TA_USER, ostate, tp, NULL,
    646 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    647 #endif
    648 	mutex_exit(softnet_lock);
    649 	KERNEL_UNLOCK_ONE(NULL);
    650 }
    651