Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.89
      1 /*	$NetBSD: tcp_timer.c,v 1.89 2015/08/24 22:21:26 pooka 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 1.89 2015/08/24 22:21:26 pooka Exp $");
     97 
     98 #ifdef _KERNEL_OPT
     99 #include "opt_inet.h"
    100 #include "opt_tcp_debug.h"
    101 #endif
    102 
    103 #include <sys/param.h>
    104 #include <sys/systm.h>
    105 #include <sys/mbuf.h>
    106 #include <sys/socket.h>
    107 #include <sys/socketvar.h>
    108 #include <sys/protosw.h>
    109 #include <sys/errno.h>
    110 #include <sys/kernel.h>
    111 
    112 #include <net/if.h>
    113 #include <net/route.h>
    114 
    115 #include <netinet/in.h>
    116 #include <netinet/in_systm.h>
    117 #include <netinet/ip.h>
    118 #include <netinet/in_pcb.h>
    119 #include <netinet/ip_var.h>
    120 #include <netinet/ip_icmp.h>
    121 
    122 #ifdef INET6
    123 #ifndef INET
    124 #include <netinet/in.h>
    125 #endif
    126 #include <netinet/ip6.h>
    127 #include <netinet6/in6_pcb.h>
    128 #endif
    129 
    130 #include <netinet/tcp.h>
    131 #include <netinet/tcp_fsm.h>
    132 #include <netinet/tcp_seq.h>
    133 #include <netinet/tcp_timer.h>
    134 #include <netinet/tcp_var.h>
    135 #include <netinet/tcp_private.h>
    136 #include <netinet/tcp_congctl.h>
    137 #include <netinet/tcpip.h>
    138 #ifdef TCP_DEBUG
    139 #include <netinet/tcp_debug.h>
    140 #endif
    141 
    142 /*
    143  * Various tunable timer parameters.  These are initialized in tcp_init(),
    144  * unless they are patched.
    145  */
    146 u_int	tcp_keepinit = 0;
    147 u_int	tcp_keepidle = 0;
    148 u_int	tcp_keepintvl = 0;
    149 u_int	tcp_keepcnt = 0;		/* max idle probes */
    150 
    151 int	tcp_maxpersistidle = 0;		/* max idle time in persist */
    152 
    153 /*
    154  * Time to delay the ACK.  This is initialized in tcp_init(), unless
    155  * its patched.
    156  */
    157 int	tcp_delack_ticks = 0;
    158 
    159 void	tcp_timer_rexmt(void *);
    160 void	tcp_timer_persist(void *);
    161 void	tcp_timer_keep(void *);
    162 void	tcp_timer_2msl(void *);
    163 
    164 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
    165 	tcp_timer_rexmt,
    166 	tcp_timer_persist,
    167 	tcp_timer_keep,
    168 	tcp_timer_2msl,
    169 };
    170 
    171 /*
    172  * Timer state initialization, called from tcp_init().
    173  */
    174 void
    175 tcp_timer_init(void)
    176 {
    177 
    178 	if (tcp_keepinit == 0)
    179 		tcp_keepinit = TCPTV_KEEP_INIT;
    180 
    181 	if (tcp_keepidle == 0)
    182 		tcp_keepidle = TCPTV_KEEP_IDLE;
    183 
    184 	if (tcp_keepintvl == 0)
    185 		tcp_keepintvl = TCPTV_KEEPINTVL;
    186 
    187 	if (tcp_keepcnt == 0)
    188 		tcp_keepcnt = TCPTV_KEEPCNT;
    189 
    190 	if (tcp_maxpersistidle == 0)
    191 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
    192 
    193 	if (tcp_delack_ticks == 0)
    194 		tcp_delack_ticks = TCP_DELACK_TICKS;
    195 }
    196 
    197 /*
    198  * Callout to process delayed ACKs for a TCPCB.
    199  */
    200 void
    201 tcp_delack(void *arg)
    202 {
    203 	struct tcpcb *tp = arg;
    204 
    205 	/*
    206 	 * If tcp_output() wasn't able to transmit the ACK
    207 	 * for whatever reason, it will restart the delayed
    208 	 * ACK callout.
    209 	 */
    210 
    211 	mutex_enter(softnet_lock);
    212 	if ((tp->t_flags & (TF_DEAD | TF_DELACK)) != TF_DELACK) {
    213 		mutex_exit(softnet_lock);
    214 		return;
    215 	}
    216 	if (!callout_expired(&tp->t_delack_ch)) {
    217 		mutex_exit(softnet_lock);
    218 		return;
    219 	}
    220 
    221 	tp->t_flags |= TF_ACKNOW;
    222 	KERNEL_LOCK(1, NULL);
    223 	(void) tcp_output(tp);
    224 	KERNEL_UNLOCK_ONE(NULL);
    225 	mutex_exit(softnet_lock);
    226 }
    227 
    228 /*
    229  * Tcp protocol timeout routine called every 500 ms.
    230  * Updates the timers in all active tcb's and
    231  * causes finite state machine actions if timers expire.
    232  */
    233 void
    234 tcp_slowtimo(void *arg)
    235 {
    236 
    237 	mutex_enter(softnet_lock);
    238 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    239 	tcp_now++;					/* for timestamps */
    240 	mutex_exit(softnet_lock);
    241 
    242 	callout_schedule(&tcp_slowtimo_ch, hz / PR_SLOWHZ);
    243 }
    244 
    245 /*
    246  * Cancel all timers for TCP tp.
    247  */
    248 void
    249 tcp_canceltimers(struct tcpcb *tp)
    250 {
    251 	int i;
    252 
    253 	for (i = 0; i < TCPT_NTIMERS; i++)
    254 		TCP_TIMER_DISARM(tp, i);
    255 }
    256 
    257 const int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    258     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    259 
    260 const int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    261 
    262 /*
    263  * TCP timer processing.
    264  */
    265 
    266 void
    267 tcp_timer_rexmt(void *arg)
    268 {
    269 	struct tcpcb *tp = arg;
    270 	uint32_t rto;
    271 #ifdef TCP_DEBUG
    272 	struct socket *so = NULL;
    273 	short ostate;
    274 #endif
    275 
    276 	mutex_enter(softnet_lock);
    277 	if ((tp->t_flags & TF_DEAD) != 0) {
    278 		mutex_exit(softnet_lock);
    279 		return;
    280 	}
    281 	if (!callout_expired(&tp->t_timer[TCPT_REXMT])) {
    282 		mutex_exit(softnet_lock);
    283 		return;
    284 	}
    285 
    286 	KERNEL_LOCK(1, NULL);
    287 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
    288 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
    289 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
    290 		extern struct sockaddr_in icmpsrc;
    291 		struct icmp icmp;
    292 
    293 		tp->t_flags &= ~TF_PMTUD_PEND;
    294 
    295 		/* XXX create fake icmp message with relevant entries */
    296 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
    297 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
    298 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
    299 		icmpsrc.sin_addr = tp->t_inpcb->inp_faddr;
    300 		icmp_mtudisc(&icmp, icmpsrc.sin_addr);
    301 
    302 		/*
    303 		 * Notify all connections to the same peer about
    304 		 * new mss and trigger retransmit.
    305 		 */
    306 		in_pcbnotifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
    307 		    tcp_mtudisc);
    308 		KERNEL_UNLOCK_ONE(NULL);
    309 		mutex_exit(softnet_lock);
    310  		return;
    311  	}
    312 #ifdef TCP_DEBUG
    313 #ifdef INET
    314 	if (tp->t_inpcb)
    315 		so = tp->t_inpcb->inp_socket;
    316 #endif
    317 #ifdef INET6
    318 	if (tp->t_in6pcb)
    319 		so = tp->t_in6pcb->in6p_socket;
    320 #endif
    321 	ostate = tp->t_state;
    322 #endif /* TCP_DEBUG */
    323 
    324 	/*
    325 	 * Clear the SACK scoreboard, reset FACK estimate.
    326 	 */
    327 	tcp_free_sackholes(tp);
    328 	tp->snd_fack = tp->snd_una;
    329 
    330 	/*
    331 	 * Retransmission timer went off.  Message has not
    332 	 * been acked within retransmit interval.  Back off
    333 	 * to a longer retransmit interval and retransmit one segment.
    334 	 */
    335 
    336 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    337 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
    338 		TCP_STATINC(TCP_STAT_TIMEOUTDROP);
    339 		tp = tcp_drop(tp, tp->t_softerror ?
    340 		    tp->t_softerror : ETIMEDOUT);
    341 		goto out;
    342 	}
    343 	TCP_STATINC(TCP_STAT_REXMTTIMEO);
    344 	rto = TCP_REXMTVAL(tp);
    345 	if (rto < tp->t_rttmin)
    346 		rto = tp->t_rttmin;
    347 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    348 	    tp->t_rttmin, TCPTV_REXMTMAX);
    349 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    350 
    351 	/*
    352 	 * If we are losing and we are trying path MTU discovery,
    353 	 * try turning it off.  This will avoid black holes in
    354 	 * the network which suppress or fail to send "packet
    355 	 * too big" ICMP messages.  We should ideally do
    356 	 * lots more sophisticated searching to find the right
    357 	 * value here...
    358 	 */
    359 	if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    360 		TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
    361 
    362 #ifdef INET
    363 		/* try turning PMTUD off */
    364 		if (tp->t_inpcb)
    365 			tp->t_mtudisc = 0;
    366 #endif
    367 #ifdef INET6
    368 		/* try using IPv6 minimum MTU */
    369 		if (tp->t_in6pcb)
    370 			tp->t_mtudisc = 0;
    371 #endif
    372 
    373 		/* XXX: more sophisticated Black hole recovery code? */
    374 	}
    375 
    376 	/*
    377 	 * If losing, let the lower level know and try for
    378 	 * a better route.  Also, if we backed off this far,
    379 	 * our srtt estimate is probably bogus.  Clobber it
    380 	 * so we'll take the next rtt measurement as our srtt;
    381 	 * move the current srtt into rttvar to keep the current
    382 	 * retransmit times until then.
    383 	 */
    384 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    385 #ifdef INET
    386 		if (tp->t_inpcb)
    387 			in_losing(tp->t_inpcb);
    388 #endif
    389 #ifdef INET6
    390 		if (tp->t_in6pcb)
    391 			in6_losing(tp->t_in6pcb);
    392 #endif
    393 		/*
    394 		 * This operation is not described in RFC2988.  The
    395 		 * point is to keep srtt+4*rttvar constant, so we
    396 		 * should shift right 2 bits to divide by 4, and then
    397 		 * shift right one bit because the storage
    398 		 * representation of rttvar is 1/16s vs 1/32s for
    399 		 * srtt.
    400 		 */
    401 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    402 		tp->t_srtt = 0;
    403 	}
    404 	tp->snd_nxt = tp->snd_una;
    405 	tp->snd_high = tp->snd_max;
    406 	/*
    407 	 * If timing a segment in this window, stop the timer.
    408 	 */
    409 	tp->t_rtttime = 0;
    410 	/*
    411 	 * Remember if we are retransmitting a SYN, because if
    412 	 * we do, set the initial congestion window must be set
    413 	 * to 1 segment.
    414 	 */
    415 	if (tp->t_state == TCPS_SYN_SENT)
    416 		tp->t_flags |= TF_SYN_REXMT;
    417 
    418 	/*
    419 	 * Adjust congestion control parameters.
    420 	 */
    421 	tp->t_congctl->slow_retransmit(tp);
    422 
    423 	(void) tcp_output(tp);
    424 
    425  out:
    426 #ifdef TCP_DEBUG
    427 	if (tp && so->so_options & SO_DEBUG)
    428 		tcp_trace(TA_USER, ostate, tp, NULL,
    429 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    430 #endif
    431 	KERNEL_UNLOCK_ONE(NULL);
    432 	mutex_exit(softnet_lock);
    433 }
    434 
    435 void
    436 tcp_timer_persist(void *arg)
    437 {
    438 	struct tcpcb *tp = arg;
    439 	uint32_t rto;
    440 #ifdef TCP_DEBUG
    441 	struct socket *so = NULL;
    442 	short ostate;
    443 #endif
    444 
    445 	mutex_enter(softnet_lock);
    446 	if ((tp->t_flags & TF_DEAD) != 0) {
    447 		mutex_exit(softnet_lock);
    448 		return;
    449 	}
    450 	if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
    451 		mutex_exit(softnet_lock);
    452 		return;
    453 	}
    454 
    455 	KERNEL_LOCK(1, NULL);
    456 #ifdef TCP_DEBUG
    457 #ifdef INET
    458 	if (tp->t_inpcb)
    459 		so = tp->t_inpcb->inp_socket;
    460 #endif
    461 #ifdef INET6
    462 	if (tp->t_in6pcb)
    463 		so = tp->t_in6pcb->in6p_socket;
    464 #endif
    465 
    466 	ostate = tp->t_state;
    467 #endif /* TCP_DEBUG */
    468 
    469 	/*
    470 	 * Persistance timer into zero window.
    471 	 * Force a byte to be output, if possible.
    472 	 */
    473 
    474 	/*
    475 	 * Hack: if the peer is dead/unreachable, we do not
    476 	 * time out if the window is closed.  After a full
    477 	 * backoff, drop the connection if the idle time
    478 	 * (no responses to probes) reaches the maximum
    479 	 * backoff that we would use if retransmitting.
    480 	 */
    481 	rto = TCP_REXMTVAL(tp);
    482 	if (rto < tp->t_rttmin)
    483 		rto = tp->t_rttmin;
    484 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    485 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    486 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    487 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
    488 		tp = tcp_drop(tp, ETIMEDOUT);
    489 		goto out;
    490 	}
    491 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
    492 	tcp_setpersist(tp);
    493 	tp->t_force = 1;
    494 	(void) tcp_output(tp);
    495 	tp->t_force = 0;
    496 
    497  out:
    498 #ifdef TCP_DEBUG
    499 	if (tp && so->so_options & SO_DEBUG)
    500 		tcp_trace(TA_USER, ostate, tp, NULL,
    501 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    502 #endif
    503 	KERNEL_UNLOCK_ONE(NULL);
    504 	mutex_exit(softnet_lock);
    505 }
    506 
    507 void
    508 tcp_timer_keep(void *arg)
    509 {
    510 	struct tcpcb *tp = arg;
    511 	struct socket *so = NULL;	/* Quell compiler warning */
    512 #ifdef TCP_DEBUG
    513 	short ostate;
    514 #endif
    515 
    516 	mutex_enter(softnet_lock);
    517 	if ((tp->t_flags & TF_DEAD) != 0) {
    518 		mutex_exit(softnet_lock);
    519 		return;
    520 	}
    521 	if (!callout_expired(&tp->t_timer[TCPT_KEEP])) {
    522 		mutex_exit(softnet_lock);
    523 		return;
    524 	}
    525 
    526 	KERNEL_LOCK(1, NULL);
    527 
    528 #ifdef TCP_DEBUG
    529 	ostate = tp->t_state;
    530 #endif /* TCP_DEBUG */
    531 
    532 	/*
    533 	 * Keep-alive timer went off; send something
    534 	 * or drop connection if idle for too long.
    535 	 */
    536 
    537 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
    538 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    539 		goto dropit;
    540 #ifdef INET
    541 	if (tp->t_inpcb)
    542 		so = tp->t_inpcb->inp_socket;
    543 #endif
    544 #ifdef INET6
    545 	if (tp->t_in6pcb)
    546 		so = tp->t_in6pcb->in6p_socket;
    547 #endif
    548 	KASSERT(so != NULL);
    549 	if (so->so_options & SO_KEEPALIVE &&
    550 	    tp->t_state <= TCPS_CLOSE_WAIT) {
    551 	    	if ((tp->t_maxidle > 0) &&
    552 		    ((tcp_now - tp->t_rcvtime) >=
    553 		     tp->t_keepidle + tp->t_maxidle))
    554 			goto dropit;
    555 		/*
    556 		 * Send a packet designed to force a response
    557 		 * if the peer is up and reachable:
    558 		 * either an ACK if the connection is still alive,
    559 		 * or an RST if the peer has closed the connection
    560 		 * due to timeout or reboot.
    561 		 * Using sequence number tp->snd_una-1
    562 		 * causes the transmitted zero-length segment
    563 		 * to lie outside the receive window;
    564 		 * by the protocol spec, this requires the
    565 		 * correspondent TCP to respond.
    566 		 */
    567 		TCP_STATINC(TCP_STAT_KEEPPROBE);
    568 		if (tcp_compat_42) {
    569 			/*
    570 			 * The keepalive packet must have nonzero
    571 			 * length to get a 4.2 host to respond.
    572 			 */
    573 			(void)tcp_respond(tp, tp->t_template,
    574 			    NULL, NULL, tp->rcv_nxt - 1,
    575 			    tp->snd_una - 1, 0);
    576 		} else {
    577 			(void)tcp_respond(tp, tp->t_template,
    578 			    NULL, NULL, tp->rcv_nxt,
    579 			    tp->snd_una - 1, 0);
    580 		}
    581 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepintvl);
    582 	} else
    583 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
    584 
    585 #ifdef TCP_DEBUG
    586 	if (tp && so->so_options & SO_DEBUG)
    587 		tcp_trace(TA_USER, ostate, tp, NULL,
    588 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
    589 #endif
    590 	KERNEL_UNLOCK_ONE(NULL);
    591 	mutex_exit(softnet_lock);
    592 	return;
    593 
    594  dropit:
    595 	TCP_STATINC(TCP_STAT_KEEPDROPS);
    596 	(void) tcp_drop(tp, ETIMEDOUT);
    597 	KERNEL_UNLOCK_ONE(NULL);
    598 	mutex_exit(softnet_lock);
    599 }
    600 
    601 void
    602 tcp_timer_2msl(void *arg)
    603 {
    604 	struct tcpcb *tp = arg;
    605 #ifdef TCP_DEBUG
    606 	struct socket *so = NULL;
    607 	short ostate;
    608 #endif
    609 
    610 	mutex_enter(softnet_lock);
    611 	if ((tp->t_flags & TF_DEAD) != 0) {
    612 		mutex_exit(softnet_lock);
    613 		return;
    614 	}
    615 	if (!callout_expired(&tp->t_timer[TCPT_2MSL])) {
    616 		mutex_exit(softnet_lock);
    617 		return;
    618 	}
    619 
    620 	/*
    621 	 * 2 MSL timeout went off, clear the SACK scoreboard, reset
    622 	 * the FACK estimate.
    623 	 */
    624 	KERNEL_LOCK(1, NULL);
    625 	tcp_free_sackholes(tp);
    626 	tp->snd_fack = tp->snd_una;
    627 
    628 #ifdef TCP_DEBUG
    629 #ifdef INET
    630 	if (tp->t_inpcb)
    631 		so = tp->t_inpcb->inp_socket;
    632 #endif
    633 #ifdef INET6
    634 	if (tp->t_in6pcb)
    635 		so = tp->t_in6pcb->in6p_socket;
    636 #endif
    637 
    638 	ostate = tp->t_state;
    639 #endif /* TCP_DEBUG */
    640 
    641 	/*
    642 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    643 	 * still waiting for peer to close and connection has been idle
    644 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    645 	 * control block.  Otherwise, check again in a bit.
    646 	 */
    647 	if (tp->t_state != TCPS_TIME_WAIT &&
    648 	    ((tp->t_maxidle == 0) ||
    649 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
    650 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
    651 	else
    652 		tp = tcp_close(tp);
    653 
    654 #ifdef TCP_DEBUG
    655 	if (tp && so->so_options & SO_DEBUG)
    656 		tcp_trace(TA_USER, ostate, tp, NULL,
    657 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    658 #endif
    659 	mutex_exit(softnet_lock);
    660 	KERNEL_UNLOCK_ONE(NULL);
    661 }
    662