Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.51
      1 /*	$NetBSD: tcp_timer.c,v 1.51 2001/09/10 20:15:14 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 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  *
     40  * Redistribution and use in source and binary forms, with or without
     41  * modification, are permitted provided that the following conditions
     42  * are met:
     43  * 1. Redistributions of source code must retain the above copyright
     44  *    notice, this list of conditions and the following disclaimer.
     45  * 2. Redistributions in binary form must reproduce the above copyright
     46  *    notice, this list of conditions and the following disclaimer in the
     47  *    documentation and/or other materials provided with the distribution.
     48  * 3. All advertising materials mentioning features or use of this software
     49  *    must display the following acknowledgement:
     50  *	This product includes software developed by the NetBSD
     51  *	Foundation, Inc. and its contributors.
     52  * 4. Neither the name of The NetBSD Foundation nor the names of its
     53  *    contributors may be used to endorse or promote products derived
     54  *    from this software without specific prior written permission.
     55  *
     56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     66  * POSSIBILITY OF SUCH DAMAGE.
     67  */
     68 
     69 /*
     70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
     71  *	The Regents of the University of California.  All rights reserved.
     72  *
     73  * Redistribution and use in source and binary forms, with or without
     74  * modification, are permitted provided that the following conditions
     75  * are met:
     76  * 1. Redistributions of source code must retain the above copyright
     77  *    notice, this list of conditions and the following disclaimer.
     78  * 2. Redistributions in binary form must reproduce the above copyright
     79  *    notice, this list of conditions and the following disclaimer in the
     80  *    documentation and/or other materials provided with the distribution.
     81  * 3. All advertising materials mentioning features or use of this software
     82  *    must display the following acknowledgement:
     83  *	This product includes software developed by the University of
     84  *	California, Berkeley and its contributors.
     85  * 4. Neither the name of the University nor the names of its contributors
     86  *    may be used to endorse or promote products derived from this software
     87  *    without specific prior written permission.
     88  *
     89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     99  * SUCH DAMAGE.
    100  *
    101  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
    102  */
    103 
    104 #include "opt_inet.h"
    105 #include "opt_tcp_debug.h"
    106 
    107 #include <sys/param.h>
    108 #include <sys/systm.h>
    109 #include <sys/malloc.h>
    110 #include <sys/mbuf.h>
    111 #include <sys/socket.h>
    112 #include <sys/socketvar.h>
    113 #include <sys/protosw.h>
    114 #include <sys/errno.h>
    115 #include <sys/kernel.h>
    116 
    117 #include <net/if.h>
    118 #include <net/route.h>
    119 
    120 #include <netinet/in.h>
    121 #include <netinet/in_systm.h>
    122 #include <netinet/ip.h>
    123 #include <netinet/in_pcb.h>
    124 #include <netinet/ip_var.h>
    125 
    126 #ifdef INET6
    127 #ifndef INET
    128 #include <netinet/in.h>
    129 #endif
    130 #include <netinet/ip6.h>
    131 #include <netinet6/in6_pcb.h>
    132 #endif
    133 
    134 #include <netinet/tcp.h>
    135 #include <netinet/tcp_fsm.h>
    136 #include <netinet/tcp_seq.h>
    137 #include <netinet/tcp_timer.h>
    138 #include <netinet/tcp_var.h>
    139 #include <netinet/tcpip.h>
    140 #ifdef TCP_DEBUG
    141 #include <netinet/tcp_debug.h>
    142 #endif
    143 
    144 int	tcp_keepidle = TCPTV_KEEP_IDLE;
    145 int	tcp_keepintvl = TCPTV_KEEPINTVL;
    146 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
    147 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
    148 int	tcp_maxidle;
    149 
    150 /*
    151  * Time to delay the ACK.  This is initialized in tcp_init(), unless
    152  * its patched.
    153  */
    154 int	tcp_delack_ticks = 0;
    155 
    156 void	tcp_timer_rexmt(void *);
    157 void	tcp_timer_persist(void *);
    158 void	tcp_timer_keep(void *);
    159 void	tcp_timer_2msl(void *);
    160 
    161 tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
    162 	tcp_timer_rexmt,
    163 	tcp_timer_persist,
    164 	tcp_timer_keep,
    165 	tcp_timer_2msl,
    166 };
    167 
    168 /*
    169  * Callout to process delayed ACKs for a TCPCB.
    170  */
    171 void
    172 tcp_delack(void *arg)
    173 {
    174 	struct tcpcb *tp = arg;
    175 	int s;
    176 
    177 	/*
    178 	 * If tcp_output() wasn't able to transmit the ACK
    179 	 * for whatever reason, it will restart the delayed
    180 	 * ACK callout.
    181 	 */
    182 
    183 	s = splsoftnet();
    184 	tp->t_flags |= TF_ACKNOW;
    185 	(void) tcp_output(tp);
    186 	splx(s);
    187 }
    188 
    189 /*
    190  * Tcp protocol timeout routine called every 500 ms.
    191  * Updates the timers in all active tcb's and
    192  * causes finite state machine actions if timers expire.
    193  */
    194 void
    195 tcp_slowtimo()
    196 {
    197 	struct inpcb *inp, *ninp;
    198 	struct tcpcb *tp;
    199 #ifdef INET6
    200 	struct in6pcb *in6p, *nin6p;
    201 #endif
    202 	int s;
    203 	long i;
    204 	static int syn_cache_last = 0;
    205 	int skip, mask;
    206 
    207 	skip = mask = 0;
    208 
    209 	s = splsoftnet();
    210 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
    211 	/*
    212 	 * Search through tcb's and update active timers.
    213 	 */
    214 	mask |= 1;
    215 	inp = tcbtable.inpt_queue.cqh_first;
    216 	if (inp == (struct inpcb *)0) {				/* XXX */
    217 		skip |= 1;
    218 		goto dotcb6;
    219 	}
    220 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) {
    221 		ninp = inp->inp_queue.cqe_next;
    222 		tp = intotcpcb(inp);
    223 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
    224 			continue;
    225 		for (i = 0; i < TCPT_NTIMERS; i++) {
    226 			if (TCP_TIMER_ISEXPIRED(tp, i)) {
    227 				TCP_TIMER_DISARM(tp, i);
    228 				(*(tcp_timer_funcs[i]))(tp);
    229 				/* XXX NOT MP SAFE */
    230 				if ((ninp == (void *)&tcbtable.inpt_queue &&
    231 				    tcbtable.inpt_queue.cqh_last != inp) ||
    232 				    ninp->inp_queue.cqe_prev != inp)
    233 					goto tpgone;
    234 			}
    235 		}
    236 tpgone:
    237 		;
    238 	}
    239 dotcb6:
    240 #ifdef INET6
    241 	mask |= 2;
    242 	in6p = tcb6.in6p_next;
    243 	if (in6p == (struct in6pcb *)0) {			/* XXX */
    244 		skip |= 2;
    245 		goto doiss;
    246 	}
    247 	for (; in6p != (struct in6pcb *)&tcb6; in6p = nin6p) {
    248 		nin6p = in6p->in6p_next;
    249 		tp = in6totcpcb(in6p);
    250 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
    251 			continue;
    252 		for (i = 0; i < TCPT_NTIMERS; i++) {
    253 			if (TCP_TIMER_ISEXPIRED(tp, i)) {
    254 				TCP_TIMER_DISARM(tp, i);
    255 				(*(tcp_timer_funcs[i]))(tp);
    256 				/* XXX NOT MP SAFE */
    257 				if ((nin6p == (void *)&tcb6 &&
    258 				    tcb6.in6p_prev != in6p) ||
    259 				    nin6p->in6p_prev != in6p)
    260 					goto tp6gone;
    261 			}
    262 		}
    263 tp6gone:
    264 		;
    265 	}
    266 
    267 doiss:
    268 #endif
    269 	if (mask == skip)
    270 		goto done;
    271 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    272 	tcp_now++;					/* for timestamps */
    273 	if (++syn_cache_last >= tcp_syn_cache_interval) {
    274 		syn_cache_timer();
    275 		syn_cache_last = 0;
    276 	}
    277 done:
    278 	splx(s);
    279 }
    280 
    281 /*
    282  * Cancel all timers for TCP tp.
    283  */
    284 void
    285 tcp_canceltimers(tp)
    286 	struct tcpcb *tp;
    287 {
    288 	int i;
    289 
    290 	for (i = 0; i < TCPT_NTIMERS; i++)
    291 		TCP_TIMER_DISARM(tp, i);
    292 }
    293 
    294 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    295     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    296 
    297 int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    298 
    299 /*
    300  * TCP timer processing.
    301  */
    302 
    303 void
    304 tcp_timer_rexmt(void *arg)
    305 {
    306 	struct tcpcb *tp = arg;
    307 	uint32_t rto;
    308 	int s;
    309 #ifdef TCP_DEBUG
    310 	struct socket *so;
    311 	short ostate;
    312 #endif
    313 
    314 	s = splsoftnet();
    315 
    316 #ifdef TCP_DEBUG
    317 #ifdef INET
    318 	if (tp->t_inpcb)
    319 		so = tp->t_inpcb->inp_socket;
    320 #endif
    321 #ifdef INET6
    322 	if (tp->t_in6pcb)
    323 		so = tp->t_in6pcb->in6p_socket;
    324 #endif
    325 	ostate = tp->t_state;
    326 #endif /* TCP_DEBUG */
    327 
    328 	/*
    329 	 * Retransmission timer went off.  Message has not
    330 	 * been acked within retransmit interval.  Back off
    331 	 * to a longer retransmit interval and retransmit one segment.
    332 	 */
    333 
    334 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    335 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
    336 		tcpstat.tcps_timeoutdrop++;
    337 		tp = tcp_drop(tp, tp->t_softerror ?
    338 		    tp->t_softerror : ETIMEDOUT);
    339 		goto out;
    340 	}
    341 	tcpstat.tcps_rexmttimeo++;
    342 	rto = TCP_REXMTVAL(tp);
    343 	if (rto < tp->t_rttmin)
    344 		rto = tp->t_rttmin;
    345 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    346 	    tp->t_rttmin, TCPTV_REXMTMAX);
    347 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    348 #if 0
    349 	/*
    350 	 * If we are losing and we are trying path MTU discovery,
    351 	 * try turning it off.  This will avoid black holes in
    352 	 * the network which suppress or fail to send "packet
    353 	 * too big" ICMP messages.  We should ideally do
    354 	 * lots more sophisticated searching to find the right
    355 	 * value here...
    356 	 */
    357 	if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    358 		struct rtentry *rt = NULL;
    359 
    360 #ifdef INET
    361 		if (tp->t_inpcb)
    362 			rt = in_pcbrtentry(tp->t_inpcb);
    363 #endif
    364 #ifdef INET6
    365 		if (tp->t_in6pcb)
    366 			rt = in6_pcbrtentry(tp->t_in6pcb);
    367 #endif
    368 
    369 		/* XXX:  Black hole recovery code goes here */
    370 	}
    371 #endif /* 0 */
    372 	/*
    373 	 * If losing, let the lower level know and try for
    374 	 * a better route.  Also, if we backed off this far,
    375 	 * our srtt estimate is probably bogus.  Clobber it
    376 	 * so we'll take the next rtt measurement as our srtt;
    377 	 * move the current srtt into rttvar to keep the current
    378 	 * retransmit times until then.
    379 	 */
    380 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    381 #ifdef INET
    382 		if (tp->t_inpcb)
    383 			in_losing(tp->t_inpcb);
    384 #endif
    385 #ifdef INET6
    386 		if (tp->t_in6pcb)
    387 			in6_losing(tp->t_in6pcb);
    388 #endif
    389 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    390 		tp->t_srtt = 0;
    391 	}
    392 	tp->snd_nxt = tp->snd_una;
    393 	/*
    394 	 * If timing a segment in this window, stop the timer.
    395 	 */
    396 	tp->t_rtttime = 0;
    397 	/*
    398 	 * Remember if we are retransmitting a SYN, because if
    399 	 * we do, set the initial congestion window must be set
    400 	 * to 1 segment.
    401 	 */
    402 	if (tp->t_state == TCPS_SYN_SENT)
    403 		tp->t_flags |= TF_SYN_REXMT;
    404 	/*
    405 	 * Close the congestion window down to one segment
    406 	 * (we'll open it by one segment for each ack we get).
    407 	 * Since we probably have a window's worth of unacked
    408 	 * data accumulated, this "slow start" keeps us from
    409 	 * dumping all that data as back-to-back packets (which
    410 	 * might overwhelm an intermediate gateway).
    411 	 *
    412 	 * There are two phases to the opening: Initially we
    413 	 * open by one mss on each ack.  This makes the window
    414 	 * size increase exponentially with time.  If the
    415 	 * window is larger than the path can handle, this
    416 	 * exponential growth results in dropped packet(s)
    417 	 * almost immediately.  To get more time between
    418 	 * drops but still "push" the network to take advantage
    419 	 * of improving conditions, we switch from exponential
    420 	 * to linear window opening at some threshhold size.
    421 	 * For a threshhold, we use half the current window
    422 	 * size, truncated to a multiple of the mss.
    423 	 *
    424 	 * (the minimum cwnd that will give us exponential
    425 	 * growth is 2 mss.  We don't allow the threshhold
    426 	 * to go below this.)
    427 	 */
    428 	{
    429 	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
    430 	if (win < 2)
    431 		win = 2;
    432 	/* Loss Window MUST be one segment. */
    433 	tp->snd_cwnd = tp->t_segsz;
    434 	tp->snd_ssthresh = win * tp->t_segsz;
    435 	tp->t_dupacks = 0;
    436 	}
    437 	(void) tcp_output(tp);
    438 
    439  out:
    440 #ifdef TCP_DEBUG
    441 	if (tp && so->so_options & SO_DEBUG)
    442 		tcp_trace(TA_USER, ostate, tp, NULL,
    443 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    444 #endif
    445 	splx(s);
    446 }
    447 
    448 void
    449 tcp_timer_persist(void *arg)
    450 {
    451 	struct tcpcb *tp = arg;
    452 	struct socket *so;
    453 	uint32_t rto;
    454 	int s;
    455 #ifdef TCP_DEBUG
    456 	short ostate;
    457 #endif
    458 
    459 	s = splsoftnet();
    460 
    461 #ifdef INET
    462 	if (tp->t_inpcb)
    463 		so = tp->t_inpcb->inp_socket;
    464 #endif
    465 #ifdef INET6
    466 	if (tp->t_in6pcb)
    467 		so = tp->t_in6pcb->in6p_socket;
    468 #endif
    469 
    470 #ifdef TCP_DEBUG
    471 	ostate = tp->t_state;
    472 #endif
    473 
    474 	/*
    475 	 * Persistance timer into zero window.
    476 	 * Force a byte to be output, if possible.
    477 	 */
    478 
    479 	/*
    480 	 * Hack: if the peer is dead/unreachable, we do not
    481 	 * time out if the window is closed.  After a full
    482 	 * backoff, drop the connection if the idle time
    483 	 * (no responses to probes) reaches the maximum
    484 	 * backoff that we would use if retransmitting.
    485 	 */
    486 	rto = TCP_REXMTVAL(tp);
    487 	if (rto < tp->t_rttmin)
    488 		rto = tp->t_rttmin;
    489 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    490 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    491 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    492 		tcpstat.tcps_persistdrops++;
    493 		tp = tcp_drop(tp, ETIMEDOUT);
    494 		goto out;
    495 	}
    496 	tcpstat.tcps_persisttimeo++;
    497 	tcp_setpersist(tp);
    498 	tp->t_force = 1;
    499 	(void) tcp_output(tp);
    500 	tp->t_force = 0;
    501 
    502  out:
    503 #ifdef TCP_DEBUG
    504 	if (tp && so->so_options & SO_DEBUG)
    505 		tcp_trace(TA_USER, ostate, tp, NULL,
    506 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    507 #endif
    508 	splx(s);
    509 }
    510 
    511 void
    512 tcp_timer_keep(void *arg)
    513 {
    514 	struct tcpcb *tp = arg;
    515 	struct socket *so;
    516 	int s;
    517 #ifdef TCP_DEBUG
    518 	short ostate;
    519 #endif
    520 
    521 	s = splsoftnet();
    522 
    523 #ifdef TCP_DEBUG
    524 	ostate = tp->t_state;
    525 #endif /* TCP_DEBUG */
    526 
    527 	/*
    528 	 * Keep-alive timer went off; send something
    529 	 * or drop connection if idle for too long.
    530 	 */
    531 
    532 	tcpstat.tcps_keeptimeo++;
    533 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    534 		goto dropit;
    535 #ifdef INET
    536 	if (tp->t_inpcb)
    537 		so = tp->t_inpcb->inp_socket;
    538 #endif
    539 #ifdef INET6
    540 	if (tp->t_in6pcb)
    541 		so = tp->t_in6pcb->in6p_socket;
    542 #endif
    543 	if (so->so_options & SO_KEEPALIVE &&
    544 	    tp->t_state <= TCPS_CLOSE_WAIT) {
    545 	    	if ((tcp_maxidle > 0) &&
    546 		    ((tcp_now - tp->t_rcvtime) >=
    547 		     tcp_keepidle + tcp_maxidle))
    548 			goto dropit;
    549 		/*
    550 		 * Send a packet designed to force a response
    551 		 * if the peer is up and reachable:
    552 		 * either an ACK if the connection is still alive,
    553 		 * or an RST if the peer has closed the connection
    554 		 * due to timeout or reboot.
    555 		 * Using sequence number tp->snd_una-1
    556 		 * causes the transmitted zero-length segment
    557 		 * to lie outside the receive window;
    558 		 * by the protocol spec, this requires the
    559 		 * correspondent TCP to respond.
    560 		 */
    561 		tcpstat.tcps_keepprobe++;
    562 		if (tcp_compat_42) {
    563 			/*
    564 			 * The keepalive packet must have nonzero
    565 			 * length to get a 4.2 host to respond.
    566 			 */
    567 			(void)tcp_respond(tp, tp->t_template,
    568 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
    569 			    tp->snd_una - 1, 0);
    570 		} else {
    571 			(void)tcp_respond(tp, tp->t_template,
    572 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
    573 			    tp->snd_una - 1, 0);
    574 		}
    575 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
    576 	} else
    577 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
    578 
    579 #ifdef TCP_DEBUG
    580 	if (tp && so->so_options & SO_DEBUG)
    581 		tcp_trace(TA_USER, ostate, tp, NULL,
    582 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
    583 #endif
    584 	splx(s);
    585 	return;
    586 
    587  dropit:
    588 	tcpstat.tcps_keepdrops++;
    589 	(void) tcp_drop(tp, ETIMEDOUT);
    590 	splx(s);
    591 }
    592 
    593 void
    594 tcp_timer_2msl(void *arg)
    595 {
    596 	struct tcpcb *tp = arg;
    597 	struct socket *so;
    598 	int s;
    599 #ifdef TCP_DEBUG
    600 	short ostate;
    601 #endif
    602 
    603 	s = splsoftnet();
    604 
    605 #ifdef INET
    606 	if (tp->t_inpcb)
    607 		so = tp->t_inpcb->inp_socket;
    608 #endif
    609 #ifdef INET6
    610 	if (tp->t_in6pcb)
    611 		so = tp->t_in6pcb->in6p_socket;
    612 #endif
    613 
    614 #ifdef TCP_DEBUG
    615 	ostate = tp->t_state;
    616 #endif
    617 
    618 	/*
    619 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    620 	 * still waiting for peer to close and connection has been idle
    621 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    622 	 * control block.  Otherwise, check again in a bit.
    623 	 */
    624 	if (tp->t_state != TCPS_TIME_WAIT &&
    625 	    ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle)))
    626 		TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
    627 	else
    628 		tp = tcp_close(tp);
    629 
    630 #ifdef TCP_DEBUG
    631 	if (tp && so->so_options & SO_DEBUG)
    632 		tcp_trace(TA_USER, ostate, tp, NULL,
    633 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    634 #endif
    635 	splx(s);
    636 }
    637