Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.85
      1 /*	$NetBSD: tcp_timer.c,v 1.85 2011/04/20 13:35:52 gdt 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 1.85 2011/04/20 13:35:52 gdt 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 		/*
    391 		 * This operation is not described in RFC2988.  The
    392 		 * point is to keep srtt+4*rttvar constant, so we
    393 		 * should shift right 2 bits to divide by 4, and then
    394 		 * shift right one bit because the storage
    395 		 * representation of rttvar is 1/16s vs 1/32s for
    396 		 * srtt.
    397 		 */
    398 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    399 		tp->t_srtt = 0;
    400 	}
    401 	tp->snd_nxt = tp->snd_una;
    402 	tp->snd_high = tp->snd_max;
    403 	/*
    404 	 * If timing a segment in this window, stop the timer.
    405 	 */
    406 	tp->t_rtttime = 0;
    407 	/*
    408 	 * Remember if we are retransmitting a SYN, because if
    409 	 * we do, set the initial congestion window must be set
    410 	 * to 1 segment.
    411 	 */
    412 	if (tp->t_state == TCPS_SYN_SENT)
    413 		tp->t_flags |= TF_SYN_REXMT;
    414 
    415 	/*
    416 	 * Adjust congestion control parameters.
    417 	 */
    418 	tp->t_congctl->slow_retransmit(tp);
    419 
    420 	(void) tcp_output(tp);
    421 
    422  out:
    423 #ifdef TCP_DEBUG
    424 	if (tp && so->so_options & SO_DEBUG)
    425 		tcp_trace(TA_USER, ostate, tp, NULL,
    426 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    427 #endif
    428 	KERNEL_UNLOCK_ONE(NULL);
    429 	mutex_exit(softnet_lock);
    430 }
    431 
    432 void
    433 tcp_timer_persist(void *arg)
    434 {
    435 	struct tcpcb *tp = arg;
    436 	uint32_t rto;
    437 #ifdef TCP_DEBUG
    438 	struct socket *so = NULL;
    439 	short ostate;
    440 #endif
    441 
    442 	mutex_enter(softnet_lock);
    443 	if ((tp->t_flags & TF_DEAD) != 0) {
    444 		mutex_exit(softnet_lock);
    445 		return;
    446 	}
    447 	if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
    448 		mutex_exit(softnet_lock);
    449 		return;
    450 	}
    451 
    452 	KERNEL_LOCK(1, NULL);
    453 #ifdef TCP_DEBUG
    454 #ifdef INET
    455 	if (tp->t_inpcb)
    456 		so = tp->t_inpcb->inp_socket;
    457 #endif
    458 #ifdef INET6
    459 	if (tp->t_in6pcb)
    460 		so = tp->t_in6pcb->in6p_socket;
    461 #endif
    462 
    463 	ostate = tp->t_state;
    464 #endif /* TCP_DEBUG */
    465 
    466 	/*
    467 	 * Persistance timer into zero window.
    468 	 * Force a byte to be output, if possible.
    469 	 */
    470 
    471 	/*
    472 	 * Hack: if the peer is dead/unreachable, we do not
    473 	 * time out if the window is closed.  After a full
    474 	 * backoff, drop the connection if the idle time
    475 	 * (no responses to probes) reaches the maximum
    476 	 * backoff that we would use if retransmitting.
    477 	 */
    478 	rto = TCP_REXMTVAL(tp);
    479 	if (rto < tp->t_rttmin)
    480 		rto = tp->t_rttmin;
    481 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    482 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    483 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    484 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
    485 		tp = tcp_drop(tp, ETIMEDOUT);
    486 		goto out;
    487 	}
    488 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
    489 	tcp_setpersist(tp);
    490 	tp->t_force = 1;
    491 	(void) tcp_output(tp);
    492 	tp->t_force = 0;
    493 
    494  out:
    495 #ifdef TCP_DEBUG
    496 	if (tp && so->so_options & SO_DEBUG)
    497 		tcp_trace(TA_USER, ostate, tp, NULL,
    498 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    499 #endif
    500 	KERNEL_UNLOCK_ONE(NULL);
    501 	mutex_exit(softnet_lock);
    502 }
    503 
    504 void
    505 tcp_timer_keep(void *arg)
    506 {
    507 	struct tcpcb *tp = arg;
    508 	struct socket *so = NULL;	/* Quell compiler warning */
    509 #ifdef TCP_DEBUG
    510 	short ostate;
    511 #endif
    512 
    513 	mutex_enter(softnet_lock);
    514 	if ((tp->t_flags & TF_DEAD) != 0) {
    515 		mutex_exit(softnet_lock);
    516 		return;
    517 	}
    518 	if (!callout_expired(&tp->t_timer[TCPT_KEEP])) {
    519 		mutex_exit(softnet_lock);
    520 		return;
    521 	}
    522 
    523 	KERNEL_LOCK(1, NULL);
    524 
    525 #ifdef TCP_DEBUG
    526 	ostate = tp->t_state;
    527 #endif /* TCP_DEBUG */
    528 
    529 	/*
    530 	 * Keep-alive timer went off; send something
    531 	 * or drop connection if idle for too long.
    532 	 */
    533 
    534 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
    535 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    536 		goto dropit;
    537 #ifdef INET
    538 	if (tp->t_inpcb)
    539 		so = tp->t_inpcb->inp_socket;
    540 #endif
    541 #ifdef INET6
    542 	if (tp->t_in6pcb)
    543 		so = tp->t_in6pcb->in6p_socket;
    544 #endif
    545 	KASSERT(so != NULL);
    546 	if (so->so_options & SO_KEEPALIVE &&
    547 	    tp->t_state <= TCPS_CLOSE_WAIT) {
    548 	    	if ((tp->t_maxidle > 0) &&
    549 		    ((tcp_now - tp->t_rcvtime) >=
    550 		     tp->t_keepidle + tp->t_maxidle))
    551 			goto dropit;
    552 		/*
    553 		 * Send a packet designed to force a response
    554 		 * if the peer is up and reachable:
    555 		 * either an ACK if the connection is still alive,
    556 		 * or an RST if the peer has closed the connection
    557 		 * due to timeout or reboot.
    558 		 * Using sequence number tp->snd_una-1
    559 		 * causes the transmitted zero-length segment
    560 		 * to lie outside the receive window;
    561 		 * by the protocol spec, this requires the
    562 		 * correspondent TCP to respond.
    563 		 */
    564 		TCP_STATINC(TCP_STAT_KEEPPROBE);
    565 		if (tcp_compat_42) {
    566 			/*
    567 			 * The keepalive packet must have nonzero
    568 			 * length to get a 4.2 host to respond.
    569 			 */
    570 			(void)tcp_respond(tp, tp->t_template,
    571 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
    572 			    tp->snd_una - 1, 0);
    573 		} else {
    574 			(void)tcp_respond(tp, tp->t_template,
    575 			    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
    576 			    tp->snd_una - 1, 0);
    577 		}
    578 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepintvl);
    579 	} else
    580 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
    581 
    582 #ifdef TCP_DEBUG
    583 	if (tp && so->so_options & SO_DEBUG)
    584 		tcp_trace(TA_USER, ostate, tp, NULL,
    585 		    PRU_SLOWTIMO | (TCPT_KEEP << 8));
    586 #endif
    587 	KERNEL_UNLOCK_ONE(NULL);
    588 	mutex_exit(softnet_lock);
    589 	return;
    590 
    591  dropit:
    592 	TCP_STATINC(TCP_STAT_KEEPDROPS);
    593 	(void) tcp_drop(tp, ETIMEDOUT);
    594 	KERNEL_UNLOCK_ONE(NULL);
    595 	mutex_exit(softnet_lock);
    596 }
    597 
    598 void
    599 tcp_timer_2msl(void *arg)
    600 {
    601 	struct tcpcb *tp = arg;
    602 #ifdef TCP_DEBUG
    603 	struct socket *so = NULL;
    604 	short ostate;
    605 #endif
    606 
    607 	mutex_enter(softnet_lock);
    608 	if ((tp->t_flags & TF_DEAD) != 0) {
    609 		mutex_exit(softnet_lock);
    610 		return;
    611 	}
    612 	if (!callout_expired(&tp->t_timer[TCPT_2MSL])) {
    613 		mutex_exit(softnet_lock);
    614 		return;
    615 	}
    616 
    617 	/*
    618 	 * 2 MSL timeout went off, clear the SACK scoreboard, reset
    619 	 * the FACK estimate.
    620 	 */
    621 	KERNEL_LOCK(1, NULL);
    622 	tcp_free_sackholes(tp);
    623 	tp->snd_fack = tp->snd_una;
    624 
    625 #ifdef TCP_DEBUG
    626 #ifdef INET
    627 	if (tp->t_inpcb)
    628 		so = tp->t_inpcb->inp_socket;
    629 #endif
    630 #ifdef INET6
    631 	if (tp->t_in6pcb)
    632 		so = tp->t_in6pcb->in6p_socket;
    633 #endif
    634 
    635 	ostate = tp->t_state;
    636 #endif /* TCP_DEBUG */
    637 
    638 	/*
    639 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    640 	 * still waiting for peer to close and connection has been idle
    641 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    642 	 * control block.  Otherwise, check again in a bit.
    643 	 */
    644 	if (tp->t_state != TCPS_TIME_WAIT &&
    645 	    ((tp->t_maxidle == 0) ||
    646 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
    647 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
    648 	else
    649 		tp = tcp_close(tp);
    650 
    651 #ifdef TCP_DEBUG
    652 	if (tp && so->so_options & SO_DEBUG)
    653 		tcp_trace(TA_USER, ostate, tp, NULL,
    654 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    655 #endif
    656 	mutex_exit(softnet_lock);
    657 	KERNEL_UNLOCK_ONE(NULL);
    658 }
    659