Home | History | Annotate | Line # | Download | only in netinet
      1 /*	$NetBSD: tcp_timer.c,v 1.99 2022/11/04 09:00:58 ozaki-r 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 1.99 2022/11/04 09:00:58 ozaki-r Exp $");
     97 
     98 #ifdef _KERNEL_OPT
     99 #include "opt_inet.h"
    100 #include "opt_tcp_debug.h"
    101 #include "opt_net_mpsafe.h"
    102 #endif
    103 
    104 #include <sys/param.h>
    105 #include <sys/systm.h>
    106 #include <sys/mbuf.h>
    107 #include <sys/socket.h>
    108 #include <sys/socketvar.h>
    109 #include <sys/protosw.h>
    110 #include <sys/errno.h>
    111 #include <sys/kernel.h>
    112 #include <sys/callout.h>
    113 #include <sys/workqueue.h>
    114 #include <sys/cprng.h>
    115 
    116 #include <net/if.h>
    117 
    118 #include <netinet/in.h>
    119 #include <netinet/in_systm.h>
    120 #include <netinet/ip.h>
    121 #include <netinet/in_pcb.h>
    122 #include <netinet/ip_var.h>
    123 #include <netinet/ip_icmp.h>
    124 
    125 #ifdef INET6
    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 #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 static callout_t	tcp_slowtimo_ch;
    153 #ifdef NET_MPSAFE
    154 static struct workqueue	*tcp_slowtimo_wq;
    155 static struct work	tcp_slowtimo_wk;
    156 #endif
    157 
    158 static void tcp_slowtimo_work(struct work *, void *);
    159 static void tcp_slowtimo(void *);
    160 
    161 /*
    162  * Time to delay the ACK.  This is initialized in tcp_init(), unless
    163  * its patched.
    164  */
    165 int	tcp_delack_ticks = 0;
    166 
    167 void	tcp_timer_rexmt(void *);
    168 void	tcp_timer_persist(void *);
    169 void	tcp_timer_keep(void *);
    170 void	tcp_timer_2msl(void *);
    171 
    172 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
    173 	tcp_timer_rexmt,
    174 	tcp_timer_persist,
    175 	tcp_timer_keep,
    176 	tcp_timer_2msl,
    177 };
    178 
    179 /*
    180  * Timer state initialization, called from tcp_init().
    181  */
    182 void
    183 tcp_timer_init(void)
    184 {
    185 
    186 	if (tcp_keepinit == 0)
    187 		tcp_keepinit = TCPTV_KEEP_INIT;
    188 
    189 	if (tcp_keepidle == 0)
    190 		tcp_keepidle = TCPTV_KEEP_IDLE;
    191 
    192 	if (tcp_keepintvl == 0)
    193 		tcp_keepintvl = TCPTV_KEEPINTVL;
    194 
    195 	if (tcp_keepcnt == 0)
    196 		tcp_keepcnt = TCPTV_KEEPCNT;
    197 
    198 	if (tcp_maxpersistidle == 0)
    199 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
    200 
    201 	if (tcp_delack_ticks == 0)
    202 		tcp_delack_ticks = TCP_DELACK_TICKS;
    203 }
    204 
    205 void
    206 tcp_slowtimo_init(void)
    207 {
    208 #ifdef NET_MPSAFE
    209 	int error;
    210 
    211 	error = workqueue_create(&tcp_slowtimo_wq, "tcp_slowtimo",
    212 	    tcp_slowtimo_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
    213 	if (error != 0)
    214 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
    215 #endif
    216 	callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
    217 	callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
    218 }
    219 
    220 /*
    221  * Callout to process delayed ACKs for a TCPCB.
    222  */
    223 void
    224 tcp_delack(void *arg)
    225 {
    226 	struct tcpcb *tp = arg;
    227 
    228 	/*
    229 	 * If tcp_output() wasn't able to transmit the ACK
    230 	 * for whatever reason, it will restart the delayed
    231 	 * ACK callout.
    232 	 */
    233 
    234 	mutex_enter(softnet_lock);
    235 	if ((tp->t_flags & (TF_DEAD | TF_DELACK)) != TF_DELACK) {
    236 		mutex_exit(softnet_lock);
    237 		return;
    238 	}
    239 	if (!callout_expired(&tp->t_delack_ch)) {
    240 		mutex_exit(softnet_lock);
    241 		return;
    242 	}
    243 
    244 	tp->t_flags |= TF_ACKNOW;
    245 	KERNEL_LOCK(1, NULL);
    246 	(void) tcp_output(tp);
    247 	KERNEL_UNLOCK_ONE(NULL);
    248 	mutex_exit(softnet_lock);
    249 }
    250 
    251 /*
    252  * Tcp protocol timeout routine called every 500 ms.
    253  * Updates the timers in all active tcb's and
    254  * causes finite state machine actions if timers expire.
    255  */
    256 static void
    257 tcp_slowtimo_work(struct work *wk, void *arg)
    258 {
    259 
    260 	mutex_enter(softnet_lock);
    261 	tcp_iss_seq += TCP_ISSINCR + (TCP_ISS_RANDOM_MASK & cprng_fast32());
    262 	tcp_now++;					/* for timestamps */
    263 	mutex_exit(softnet_lock);
    264 
    265 	callout_schedule(&tcp_slowtimo_ch, hz / PR_SLOWHZ);
    266 }
    267 
    268 static void
    269 tcp_slowtimo(void *arg)
    270 {
    271 
    272 #ifdef NET_MPSAFE
    273 	workqueue_enqueue(tcp_slowtimo_wq, &tcp_slowtimo_wk, NULL);
    274 #else
    275 	tcp_slowtimo_work(NULL, NULL);
    276 #endif
    277 }
    278 
    279 /*
    280  * Cancel all timers for TCP tp.
    281  */
    282 void
    283 tcp_canceltimers(struct tcpcb *tp)
    284 {
    285 	int i;
    286 
    287 	for (i = 0; i < TCPT_NTIMERS; i++)
    288 		TCP_TIMER_DISARM(tp, i);
    289 }
    290 
    291 const int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    292     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    293 
    294 const int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    295 
    296 /*
    297  * TCP timer processing.
    298  */
    299 
    300 void
    301 tcp_timer_rexmt(void *arg)
    302 {
    303 	struct tcpcb *tp = arg;
    304 	uint32_t rto;
    305 #ifdef TCP_DEBUG
    306 	struct socket *so = NULL;
    307 	short ostate;
    308 #endif
    309 
    310 	mutex_enter(softnet_lock);
    311 	if ((tp->t_flags & TF_DEAD) != 0) {
    312 		mutex_exit(softnet_lock);
    313 		return;
    314 	}
    315 	if (!callout_expired(&tp->t_timer[TCPT_REXMT])) {
    316 		mutex_exit(softnet_lock);
    317 		return;
    318 	}
    319 
    320 	KERNEL_LOCK(1, NULL);
    321 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
    322 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
    323 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
    324 		extern struct sockaddr_in icmpsrc;
    325 		struct icmp icmp;
    326 
    327 		tp->t_flags &= ~TF_PMTUD_PEND;
    328 
    329 		/* XXX create fake icmp message with relevant entries */
    330 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
    331 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
    332 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
    333 		icmpsrc.sin_addr = in4p_faddr(tp->t_inpcb);
    334 		icmp_mtudisc(&icmp, icmpsrc.sin_addr);
    335 
    336 		/*
    337 		 * Notify all connections to the same peer about
    338 		 * new mss and trigger retransmit.
    339 		 */
    340 		inpcb_notifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
    341 		    tcp_mtudisc);
    342 		KERNEL_UNLOCK_ONE(NULL);
    343 		mutex_exit(softnet_lock);
    344  		return;
    345  	}
    346 #ifdef TCP_DEBUG
    347 	so = tp->t_inpcb->inp_socket;
    348 	ostate = tp->t_state;
    349 #endif /* TCP_DEBUG */
    350 
    351 	/*
    352 	 * Clear the SACK scoreboard, reset FACK estimate.
    353 	 */
    354 	tcp_free_sackholes(tp);
    355 	tp->snd_fack = tp->snd_una;
    356 
    357 	/*
    358 	 * Retransmission timer went off.  Message has not
    359 	 * been acked within retransmit interval.  Back off
    360 	 * to a longer retransmit interval and retransmit one segment.
    361 	 */
    362 
    363 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    364 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
    365 		TCP_STATINC(TCP_STAT_TIMEOUTDROP);
    366 		tp = tcp_drop(tp, tp->t_softerror ?
    367 		    tp->t_softerror : ETIMEDOUT);
    368 		goto out;
    369 	}
    370 	TCP_STATINC(TCP_STAT_REXMTTIMEO);
    371 	rto = TCP_REXMTVAL(tp);
    372 	if (rto < tp->t_rttmin)
    373 		rto = tp->t_rttmin;
    374 	TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    375 	    tp->t_rttmin, TCPTV_REXMTMAX);
    376 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    377 
    378 	/*
    379 	 * If we are losing and we are trying path MTU discovery,
    380 	 * try turning it off.  This will avoid black holes in
    381 	 * the network which suppress or fail to send "packet
    382 	 * too big" ICMP messages.  We should ideally do
    383 	 * lots more sophisticated searching to find the right
    384 	 * value here...
    385 	 */
    386 	if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    387 		TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
    388 
    389 		/* try turning PMTUD off or try using IPv6 minimum MTU */
    390 		tp->t_mtudisc = 0;
    391 
    392 		/* XXX: more sophisticated Black hole recovery code? */
    393 	}
    394 
    395 	/*
    396 	 * If losing, let the lower level know and try for
    397 	 * a better route.  Also, if we backed off this far,
    398 	 * our srtt estimate is probably bogus.  Clobber it
    399 	 * so we'll take the next rtt measurement as our srtt;
    400 	 * move the current srtt into rttvar to keep the current
    401 	 * retransmit times until then.
    402 	 */
    403 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    404 		inpcb_losing(tp->t_inpcb);
    405 		/*
    406 		 * This operation is not described in RFC2988.  The
    407 		 * point is to keep srtt+4*rttvar constant, so we
    408 		 * should shift right 2 bits to divide by 4, and then
    409 		 * shift right one bit because the storage
    410 		 * representation of rttvar is 1/16s vs 1/32s for
    411 		 * srtt.
    412 		 */
    413 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    414 		tp->t_srtt = 0;
    415 	}
    416 	tp->snd_nxt = tp->snd_una;
    417 	tp->snd_high = tp->snd_max;
    418 	/*
    419 	 * If timing a segment in this window, stop the timer.
    420 	 */
    421 	tp->t_rtttime = 0;
    422 	/*
    423 	 * Remember if we are retransmitting a SYN, because if
    424 	 * we do, set the initial congestion window must be set
    425 	 * to 1 segment.
    426 	 */
    427 	if (tp->t_state == TCPS_SYN_SENT)
    428 		tp->t_flags |= TF_SYN_REXMT;
    429 
    430 	/*
    431 	 * Adjust congestion control parameters.
    432 	 */
    433 	tp->t_congctl->slow_retransmit(tp);
    434 
    435 	(void) tcp_output(tp);
    436 
    437  out:
    438 #ifdef TCP_DEBUG
    439 	if (tp && so->so_options & SO_DEBUG)
    440 		tcp_trace(TA_USER, ostate, tp, NULL,
    441 		    PRU_SLOWTIMO | (TCPT_REXMT << 8));
    442 #endif
    443 	KERNEL_UNLOCK_ONE(NULL);
    444 	mutex_exit(softnet_lock);
    445 }
    446 
    447 void
    448 tcp_timer_persist(void *arg)
    449 {
    450 	struct tcpcb *tp = arg;
    451 	uint32_t rto;
    452 #ifdef TCP_DEBUG
    453 	struct socket *so = NULL;
    454 	short ostate;
    455 #endif
    456 
    457 	mutex_enter(softnet_lock);
    458 	if ((tp->t_flags & TF_DEAD) != 0) {
    459 		mutex_exit(softnet_lock);
    460 		return;
    461 	}
    462 	if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
    463 		mutex_exit(softnet_lock);
    464 		return;
    465 	}
    466 
    467 	KERNEL_LOCK(1, NULL);
    468 #ifdef TCP_DEBUG
    469 	so = tp->t_inpcb->inp_socket;
    470 	ostate = tp->t_state;
    471 #endif /* TCP_DEBUG */
    472 
    473 	/*
    474 	 * Persistance timer into zero window.
    475 	 * Force a byte to be output, if possible.
    476 	 */
    477 
    478 	/*
    479 	 * Hack: if the peer is dead/unreachable, we do not
    480 	 * time out if the window is closed.  After a full
    481 	 * backoff, drop the connection if the idle time
    482 	 * (no responses to probes) reaches the maximum
    483 	 * backoff that we would use if retransmitting.
    484 	 */
    485 	rto = TCP_REXMTVAL(tp);
    486 	if (rto < tp->t_rttmin)
    487 		rto = tp->t_rttmin;
    488 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    489 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
    490 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
    491 		TCP_STATINC(TCP_STAT_PERSISTDROPS);
    492 		tp = tcp_drop(tp, ETIMEDOUT);
    493 		goto out;
    494 	}
    495 	TCP_STATINC(TCP_STAT_PERSISTTIMEO);
    496 	tcp_setpersist(tp);
    497 	tp->t_force = 1;
    498 	(void) tcp_output(tp);
    499 	tp->t_force = 0;
    500 
    501  out:
    502 #ifdef TCP_DEBUG
    503 	if (tp && so->so_options & SO_DEBUG)
    504 		tcp_trace(TA_USER, ostate, tp, NULL,
    505 		    PRU_SLOWTIMO | (TCPT_PERSIST << 8));
    506 #endif
    507 	KERNEL_UNLOCK_ONE(NULL);
    508 	mutex_exit(softnet_lock);
    509 }
    510 
    511 void
    512 tcp_timer_keep(void *arg)
    513 {
    514 	struct tcpcb *tp = arg;
    515 	struct socket *so = NULL;	/* Quell compiler warning */
    516 #ifdef TCP_DEBUG
    517 	short ostate;
    518 #endif
    519 
    520 	mutex_enter(softnet_lock);
    521 	if ((tp->t_flags & TF_DEAD) != 0) {
    522 		mutex_exit(softnet_lock);
    523 		return;
    524 	}
    525 	if (!callout_expired(&tp->t_timer[TCPT_KEEP])) {
    526 		mutex_exit(softnet_lock);
    527 		return;
    528 	}
    529 
    530 	KERNEL_LOCK(1, NULL);
    531 
    532 #ifdef TCP_DEBUG
    533 	ostate = tp->t_state;
    534 #endif /* TCP_DEBUG */
    535 
    536 	/*
    537 	 * Keep-alive timer went off; send something
    538 	 * or drop connection if idle for too long.
    539 	 */
    540 
    541 	TCP_STATINC(TCP_STAT_KEEPTIMEO);
    542 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    543 		goto dropit;
    544 	so = tp->t_inpcb->inp_socket;
    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 
    566 		(void)tcp_respond(tp, tp->t_template,
    567 		    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 	so = tp->t_inpcb->inp_socket;
    619 	ostate = tp->t_state;
    620 #endif /* TCP_DEBUG */
    621 
    622 	/*
    623 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    624 	 * still waiting for peer to close and connection has been idle
    625 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    626 	 * control block.  Otherwise, check again in a bit.
    627 	 */
    628 	if (tp->t_state != TCPS_TIME_WAIT &&
    629 	    ((tp->t_maxidle == 0) ||
    630 	    ((tcp_now - tp->t_rcvtime) <= tp->t_maxidle)))
    631 	    TCP_TIMER_ARM(tp, TCPT_2MSL, tp->t_keepintvl);
    632 	else
    633 		tp = tcp_close(tp);
    634 
    635 #ifdef TCP_DEBUG
    636 	if (tp && so->so_options & SO_DEBUG)
    637 		tcp_trace(TA_USER, ostate, tp, NULL,
    638 		    PRU_SLOWTIMO | (TCPT_2MSL << 8));
    639 #endif
    640 	KERNEL_UNLOCK_ONE(NULL);
    641 	mutex_exit(softnet_lock);
    642 }
    643