Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.43
      1 /*	$NetBSD: tcp_timer.c,v 1.43 1998/09/10 10:47:00 mouse Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
      9  * Facility, NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
     42  *	The Regents of the University of California.  All rights reserved.
     43  *
     44  * Redistribution and use in source and binary forms, with or without
     45  * modification, are permitted provided that the following conditions
     46  * are met:
     47  * 1. Redistributions of source code must retain the above copyright
     48  *    notice, this list of conditions and the following disclaimer.
     49  * 2. Redistributions in binary form must reproduce the above copyright
     50  *    notice, this list of conditions and the following disclaimer in the
     51  *    documentation and/or other materials provided with the distribution.
     52  * 3. All advertising materials mentioning features or use of this software
     53  *    must display the following acknowledgement:
     54  *	This product includes software developed by the University of
     55  *	California, Berkeley and its contributors.
     56  * 4. Neither the name of the University nor the names of its contributors
     57  *    may be used to endorse or promote products derived from this software
     58  *    without specific prior written permission.
     59  *
     60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     70  * SUCH DAMAGE.
     71  *
     72  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
     73  */
     74 
     75 #include <sys/param.h>
     76 #include <sys/systm.h>
     77 #include <sys/malloc.h>
     78 #include <sys/mbuf.h>
     79 #include <sys/socket.h>
     80 #include <sys/socketvar.h>
     81 #include <sys/protosw.h>
     82 #include <sys/errno.h>
     83 
     84 #include <net/if.h>
     85 #include <net/route.h>
     86 
     87 #include <netinet/in.h>
     88 #include <netinet/in_systm.h>
     89 #include <netinet/ip.h>
     90 #include <netinet/in_pcb.h>
     91 #include <netinet/ip_var.h>
     92 #include <netinet/tcp.h>
     93 #include <netinet/tcp_fsm.h>
     94 #include <netinet/tcp_seq.h>
     95 #include <netinet/tcp_timer.h>
     96 #include <netinet/tcp_var.h>
     97 #include <netinet/tcpip.h>
     98 
     99 int	tcp_keepidle = TCPTV_KEEP_IDLE;
    100 int	tcp_keepintvl = TCPTV_KEEPINTVL;
    101 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
    102 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
    103 int	tcp_maxidle;
    104 
    105 struct tcp_delack_head tcp_delacks;
    106 
    107 /*
    108  * Fast timeout routine for processing delayed acks
    109  */
    110 void
    111 tcp_fasttimo()
    112 {
    113 	register struct tcpcb *tp, *ntp;
    114 	int s;
    115 
    116 	s = splsoftnet();
    117 	for (tp = tcp_delacks.lh_first; tp != NULL; tp = ntp) {
    118 		/*
    119 		 * If tcp_output() can't transmit the ACK for whatever
    120 		 * reason, it will remain on the queue for the next
    121 		 * time the heartbeat ticks.
    122 		 */
    123 		ntp = tp->t_delack.le_next;
    124 		tp->t_flags |= TF_ACKNOW;
    125 		(void) tcp_output(tp);
    126 	}
    127 	splx(s);
    128 }
    129 
    130 /*
    131  * Tcp protocol timeout routine called every 500 ms.
    132  * Updates the timers in all active tcb's and
    133  * causes finite state machine actions if timers expire.
    134  */
    135 void
    136 tcp_slowtimo()
    137 {
    138 	register struct inpcb *inp, *ninp;
    139 	register struct tcpcb *tp;
    140 	int s;
    141 	register long i;
    142 	static int syn_cache_last = 0;
    143 
    144 	s = splsoftnet();
    145 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
    146 	/*
    147 	 * Search through tcb's and update active timers.
    148 	 */
    149 	inp = tcbtable.inpt_queue.cqh_first;
    150 	if (inp == (struct inpcb *)0) {				/* XXX */
    151 		splx(s);
    152 		return;
    153 	}
    154 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) {
    155 		ninp = inp->inp_queue.cqe_next;
    156 		tp = intotcpcb(inp);
    157 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
    158 			continue;
    159 		for (i = 0; i < TCPT_NTIMERS; i++) {
    160 			if (TCP_TIMER_ISEXPIRED(tp, i)) {
    161 				TCP_TIMER_DISARM(tp, i);
    162 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
    163 				    PRU_SLOWTIMO, (struct mbuf *)0,
    164 				    (struct mbuf *)i, (struct mbuf *)0,
    165 				    (struct proc *)0);
    166 				/* XXX NOT MP SAFE */
    167 				if ((ninp == (void *)&tcbtable.inpt_queue &&
    168 				    tcbtable.inpt_queue.cqh_last != inp) ||
    169 				    ninp->inp_queue.cqe_prev != inp)
    170 					goto tpgone;
    171 			}
    172 		}
    173 		tp->t_idle++;
    174 		if (tp->t_rtt)
    175 			tp->t_rtt++;
    176 tpgone:
    177 		;
    178 	}
    179 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    180 	tcp_now++;					/* for timestamps */
    181 	if (++syn_cache_last >= tcp_syn_cache_interval) {
    182 		syn_cache_timer();
    183 		syn_cache_last = 0;
    184 	}
    185 	splx(s);
    186 }
    187 
    188 /*
    189  * Cancel all timers for TCP tp.
    190  */
    191 void
    192 tcp_canceltimers(tp)
    193 	struct tcpcb *tp;
    194 {
    195 	register int i;
    196 
    197 	for (i = 0; i < TCPT_NTIMERS; i++)
    198 		TCP_TIMER_DISARM(tp, i);
    199 }
    200 
    201 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    202     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    203 
    204 int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    205 
    206 /*
    207  * TCP timer processing.
    208  */
    209 struct tcpcb *
    210 tcp_timers(tp, timer)
    211 	register struct tcpcb *tp;
    212 	int timer;
    213 {
    214 	short	rto;
    215 
    216 	switch (timer) {
    217 
    218 	/*
    219 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    220 	 * still waiting for peer to close and connection has been idle
    221 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    222 	 * control block.  Otherwise, check again in a bit.
    223 	 */
    224 	case TCPT_2MSL:
    225 		if (tp->t_state != TCPS_TIME_WAIT &&
    226 		    ((tcp_maxidle == 0) || (tp->t_idle <= tcp_maxidle)))
    227 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
    228 		else
    229 			tp = tcp_close(tp);
    230 		break;
    231 
    232 	/*
    233 	 * Retransmission timer went off.  Message has not
    234 	 * been acked within retransmit interval.  Back off
    235 	 * to a longer retransmit interval and retransmit one segment.
    236 	 */
    237 	case TCPT_REXMT:
    238 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    239 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
    240 			tcpstat.tcps_timeoutdrop++;
    241 			tp = tcp_drop(tp, tp->t_softerror ?
    242 			    tp->t_softerror : ETIMEDOUT);
    243 			break;
    244 		}
    245 		tcpstat.tcps_rexmttimeo++;
    246 		rto = TCP_REXMTVAL(tp);
    247 		if (rto < tp->t_rttmin)
    248 			rto = tp->t_rttmin;
    249 		TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
    250 		    tp->t_rttmin, TCPTV_REXMTMAX);
    251 		TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
    252 #if 0
    253 		/*
    254 		 * If we are losing and we are trying path MTU discovery,
    255 		 * try turning it off.  This will avoid black holes in
    256 		 * the network which suppress or fail to send "packet
    257 		 * too big" ICMP messages.  We should ideally do
    258 		 * lots more sophisticated searching to find the right
    259 		 * value here...
    260 		 */
    261 		if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
    262 			struct inpcb *inp = tp->t_inpcb;
    263 			struct rtentry *rt = in_pcbrtentry(inp);
    264 
    265 			/* XXX:  Black hole recovery code goes here */
    266 		}
    267 #endif
    268 		/*
    269 		 * If losing, let the lower level know and try for
    270 		 * a better route.  Also, if we backed off this far,
    271 		 * our srtt estimate is probably bogus.  Clobber it
    272 		 * so we'll take the next rtt measurement as our srtt;
    273 		 * move the current srtt into rttvar to keep the current
    274 		 * retransmit times until then.
    275 		 */
    276 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    277 			in_losing(tp->t_inpcb);
    278 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    279 			tp->t_srtt = 0;
    280 		}
    281 		tp->snd_nxt = tp->snd_una;
    282 		/*
    283 		 * If timing a segment in this window, stop the timer.
    284 		 */
    285 		tp->t_rtt = 0;
    286 		/*
    287 		 * Remember if we are retransmitting a SYN, because if
    288 		 * we do, set the initial congestion window must be set
    289 		 * to 1 segment.
    290 		 */
    291 		if (tp->t_state == TCPS_SYN_SENT)
    292 			tp->t_flags |= TF_SYN_REXMT;
    293 		/*
    294 		 * Close the congestion window down to one segment
    295 		 * (we'll open it by one segment for each ack we get).
    296 		 * Since we probably have a window's worth of unacked
    297 		 * data accumulated, this "slow start" keeps us from
    298 		 * dumping all that data as back-to-back packets (which
    299 		 * might overwhelm an intermediate gateway).
    300 		 *
    301 		 * There are two phases to the opening: Initially we
    302 		 * open by one mss on each ack.  This makes the window
    303 		 * size increase exponentially with time.  If the
    304 		 * window is larger than the path can handle, this
    305 		 * exponential growth results in dropped packet(s)
    306 		 * almost immediately.  To get more time between
    307 		 * drops but still "push" the network to take advantage
    308 		 * of improving conditions, we switch from exponential
    309 		 * to linear window opening at some threshhold size.
    310 		 * For a threshhold, we use half the current window
    311 		 * size, truncated to a multiple of the mss.
    312 		 *
    313 		 * (the minimum cwnd that will give us exponential
    314 		 * growth is 2 mss.  We don't allow the threshhold
    315 		 * to go below this.)
    316 		 */
    317 		{
    318 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
    319 		if (win < 2)
    320 			win = 2;
    321 		/* Loss Window MUST be one segment. */
    322 		tp->snd_cwnd = tp->t_segsz;
    323 		tp->snd_ssthresh = win * tp->t_segsz;
    324 		tp->t_dupacks = 0;
    325 		}
    326 		(void) tcp_output(tp);
    327 		break;
    328 
    329 	/*
    330 	 * Persistance timer into zero window.
    331 	 * Force a byte to be output, if possible.
    332 	 */
    333 	case TCPT_PERSIST:
    334 		/*
    335 		 * Hack: if the peer is dead/unreachable, we do not
    336 		 * time out if the window is closed.  After a full
    337 		 * backoff, drop the connection if the idle time
    338 		 * (no responses to probes) reaches the maximum
    339 		 * backoff that we would use if retransmitting.
    340 		 */
    341 		rto = TCP_REXMTVAL(tp);
    342 		if (rto < tp->t_rttmin)
    343 			rto = tp->t_rttmin;
    344 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    345 		    (tp->t_idle >= tcp_maxpersistidle ||
    346 		    tp->t_idle >= rto * tcp_totbackoff)) {
    347 			tcpstat.tcps_persistdrops++;
    348 			tp = tcp_drop(tp, ETIMEDOUT);
    349 			break;
    350 		}
    351 		tcpstat.tcps_persisttimeo++;
    352 		tcp_setpersist(tp);
    353 		tp->t_force = 1;
    354 		(void) tcp_output(tp);
    355 		tp->t_force = 0;
    356 		break;
    357 
    358 	/*
    359 	 * Keep-alive timer went off; send something
    360 	 * or drop connection if idle for too long.
    361 	 */
    362 	case TCPT_KEEP:
    363 		tcpstat.tcps_keeptimeo++;
    364 		if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    365 			goto dropit;
    366 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
    367 		    tp->t_state <= TCPS_CLOSE_WAIT) {
    368 		    	if ((tcp_maxidle > 0) &&
    369 			    (tp->t_idle >= tcp_keepidle + tcp_maxidle))
    370 				goto dropit;
    371 			/*
    372 			 * Send a packet designed to force a response
    373 			 * if the peer is up and reachable:
    374 			 * either an ACK if the connection is still alive,
    375 			 * or an RST if the peer has closed the connection
    376 			 * due to timeout or reboot.
    377 			 * Using sequence number tp->snd_una-1
    378 			 * causes the transmitted zero-length segment
    379 			 * to lie outside the receive window;
    380 			 * by the protocol spec, this requires the
    381 			 * correspondent TCP to respond.
    382 			 */
    383 			tcpstat.tcps_keepprobe++;
    384 			if (tcp_compat_42) {
    385 				/*
    386 				 * The keepalive packet must have nonzero
    387 				 * length to get a 4.2 host to respond.
    388 				 */
    389 				(void)tcp_respond(tp, tp->t_template,
    390 				    (struct mbuf *)NULL, tp->rcv_nxt - 1,
    391 				    tp->snd_una - 1, 0);
    392 			} else {
    393 				(void)tcp_respond(tp, tp->t_template,
    394 				    (struct mbuf *)NULL, tp->rcv_nxt,
    395 				    tp->snd_una - 1, 0);
    396 			}
    397 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
    398 		} else
    399 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
    400 		break;
    401 	dropit:
    402 		tcpstat.tcps_keepdrops++;
    403 		tp = tcp_drop(tp, ETIMEDOUT);
    404 		break;
    405 	}
    406 	return (tp);
    407 }
    408