Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.29
      1 /*	$NetBSD: tcp_timer.c,v 1.29 1998/01/12 03:00:43 scottr Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
      5  *	The Regents of the University of California.  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. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
     36  */
     37 
     38 #include "opt_tcp_compat_42.h"
     39 
     40 #ifndef TUBA_INCLUDE
     41 #include <sys/param.h>
     42 #include <sys/systm.h>
     43 #include <sys/malloc.h>
     44 #include <sys/mbuf.h>
     45 #include <sys/socket.h>
     46 #include <sys/socketvar.h>
     47 #include <sys/protosw.h>
     48 #include <sys/errno.h>
     49 
     50 #include <net/if.h>
     51 #include <net/route.h>
     52 
     53 #include <netinet/in.h>
     54 #include <netinet/in_systm.h>
     55 #include <netinet/ip.h>
     56 #include <netinet/in_pcb.h>
     57 #include <netinet/ip_var.h>
     58 #include <netinet/tcp.h>
     59 #include <netinet/tcp_fsm.h>
     60 #include <netinet/tcp_seq.h>
     61 #include <netinet/tcp_timer.h>
     62 #include <netinet/tcp_var.h>
     63 #include <netinet/tcpip.h>
     64 
     65 int	tcp_keepidle = TCPTV_KEEP_IDLE;
     66 int	tcp_keepintvl = TCPTV_KEEPINTVL;
     67 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
     68 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
     69 int	tcp_maxidle;
     70 #else /* TUBA_INCLUDE */
     71 
     72 extern	int tcp_keepcnt;
     73 extern	int tcp_maxpersistidle;
     74 #endif /* TUBA_INCLUDE */
     75 
     76 struct tcp_delack_head tcp_delacks;
     77 
     78 /*
     79  * Fast timeout routine for processing delayed acks
     80  */
     81 void
     82 tcp_fasttimo()
     83 {
     84 	register struct tcpcb *tp, *ntp;
     85 	int s;
     86 
     87 	s = splsoftnet();
     88 	for (tp = tcp_delacks.lh_first; tp != NULL; tp = ntp) {
     89 		/*
     90 		 * If tcp_output() can't transmit the ACK for whatever
     91 		 * reason, it will remain on the queue for the next
     92 		 * time the heartbeat ticks.
     93 		 */
     94 		ntp = tp->t_delack.le_next;
     95 		tp->t_flags |= TF_ACKNOW;
     96 		(void) tcp_output(tp);
     97 	}
     98 	splx(s);
     99 }
    100 
    101 /*
    102  * Tcp protocol timeout routine called every 500 ms.
    103  * Updates the timers in all active tcb's and
    104  * causes finite state machine actions if timers expire.
    105  */
    106 void
    107 tcp_slowtimo()
    108 {
    109 	register struct inpcb *inp, *ninp;
    110 	register struct tcpcb *tp;
    111 	int s;
    112 	register long i;
    113 	static int syn_cache_last = 0;
    114 
    115 	s = splsoftnet();
    116 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
    117 	/*
    118 	 * Search through tcb's and update active timers.
    119 	 */
    120 	inp = tcbtable.inpt_queue.cqh_first;
    121 	if (inp == (struct inpcb *)0) {				/* XXX */
    122 		splx(s);
    123 		return;
    124 	}
    125 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) {
    126 		ninp = inp->inp_queue.cqe_next;
    127 		tp = intotcpcb(inp);
    128 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
    129 			continue;
    130 		for (i = 0; i < TCPT_NTIMERS; i++) {
    131 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
    132 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
    133 				    PRU_SLOWTIMO, (struct mbuf *)0,
    134 				    (struct mbuf *)i, (struct mbuf *)0,
    135 				    (struct proc *)0);
    136 				/* XXX NOT MP SAFE */
    137 				if ((ninp == (void *)&tcbtable.inpt_queue &&
    138 				    tcbtable.inpt_queue.cqh_last != inp) ||
    139 				    ninp->inp_queue.cqe_prev != inp)
    140 					goto tpgone;
    141 			}
    142 		}
    143 		tp->t_idle++;
    144 		if (tp->t_rtt)
    145 			tp->t_rtt++;
    146 tpgone:
    147 		;
    148 	}
    149 #if NRND == 0 /* Do we need to do this when using random() ? */
    150 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
    151 #ifdef TCP_COMPAT_42
    152 	if ((int)tcp_iss_seq < 0)
    153 		tcp_iss_seq = 0;			/* XXX */
    154 #endif
    155 #endif
    156 	tcp_now++;					/* for timestamps */
    157 	if (++syn_cache_last >= tcp_syn_cache_interval) {
    158 		syn_cache_timer(syn_cache_last);
    159 		syn_cache_last = 0;
    160 	}
    161 	splx(s);
    162 }
    163 #ifndef TUBA_INCLUDE
    164 
    165 /*
    166  * Cancel all timers for TCP tp.
    167  */
    168 void
    169 tcp_canceltimers(tp)
    170 	struct tcpcb *tp;
    171 {
    172 	register int i;
    173 
    174 	for (i = 0; i < TCPT_NTIMERS; i++)
    175 		tp->t_timer[i] = 0;
    176 }
    177 
    178 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    179     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    180 
    181 int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
    182 
    183 /*
    184  * TCP timer processing.
    185  */
    186 struct tcpcb *
    187 tcp_timers(tp, timer)
    188 	register struct tcpcb *tp;
    189 	int timer;
    190 {
    191 
    192 	switch (timer) {
    193 
    194 	/*
    195 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    196 	 * still waiting for peer to close and connection has been idle
    197 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    198 	 * control block.  Otherwise, check again in a bit.
    199 	 */
    200 	case TCPT_2MSL:
    201 		if (tp->t_state != TCPS_TIME_WAIT &&
    202 		    tp->t_idle <= tcp_maxidle)
    203 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
    204 		else
    205 			tp = tcp_close(tp);
    206 		break;
    207 
    208 	/*
    209 	 * Retransmission timer went off.  Message has not
    210 	 * been acked within retransmit interval.  Back off
    211 	 * to a longer retransmit interval and retransmit one segment.
    212 	 */
    213 	case TCPT_REXMT:
    214 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    215 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
    216 			tcpstat.tcps_timeoutdrop++;
    217 			tp = tcp_drop(tp, tp->t_softerror ?
    218 			    tp->t_softerror : ETIMEDOUT);
    219 			break;
    220 		}
    221 		tcpstat.tcps_rexmttimeo++;
    222 		TCPT_RANGESET(tp->t_rxtcur,
    223 		    TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift],
    224 		    tp->t_rttmin, TCPTV_REXMTMAX);
    225 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
    226 		/*
    227 		 * If losing, let the lower level know and try for
    228 		 * a better route.  Also, if we backed off this far,
    229 		 * our srtt estimate is probably bogus.  Clobber it
    230 		 * so we'll take the next rtt measurement as our srtt;
    231 		 * move the current srtt into rttvar to keep the current
    232 		 * retransmit times until then.
    233 		 */
    234 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    235 			in_losing(tp->t_inpcb);
    236 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    237 			tp->t_srtt = 0;
    238 		}
    239 		tp->snd_nxt = tp->snd_una;
    240 		/*
    241 		 * If timing a segment in this window, stop the timer.
    242 		 */
    243 		tp->t_rtt = 0;
    244 		/*
    245 		 * Close the congestion window down to the initial window
    246 		 * (we'll open it by one segment for each ack we get).
    247 		 * Since we probably have a window's worth of unacked
    248 		 * data accumulated, this "slow start" keeps us from
    249 		 * dumping all that data as back-to-back packets (which
    250 		 * might overwhelm an intermediate gateway).
    251 		 *
    252 		 * There are two phases to the opening: Initially we
    253 		 * open by one mss on each ack.  This makes the window
    254 		 * size increase exponentially with time.  If the
    255 		 * window is larger than the path can handle, this
    256 		 * exponential growth results in dropped packet(s)
    257 		 * almost immediately.  To get more time between
    258 		 * drops but still "push" the network to take advantage
    259 		 * of improving conditions, we switch from exponential
    260 		 * to linear window opening at some threshhold size.
    261 		 * For a threshhold, we use half the current window
    262 		 * size, truncated to a multiple of the mss.
    263 		 *
    264 		 * (the minimum cwnd that will give us exponential
    265 		 * growth is 2 mss.  We don't allow the threshhold
    266 		 * to go below this.)
    267 		 */
    268 		{
    269 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
    270 		if (win < 2)
    271 			win = 2;
    272 		tp->snd_cwnd = TCP_INITIAL_WINDOW(tp->t_segsz);
    273 		tp->snd_ssthresh = win * tp->t_segsz;
    274 		tp->t_dupacks = 0;
    275 		}
    276 		(void) tcp_output(tp);
    277 		break;
    278 
    279 	/*
    280 	 * Persistance timer into zero window.
    281 	 * Force a byte to be output, if possible.
    282 	 */
    283 	case TCPT_PERSIST:
    284 		/*
    285 		 * Hack: if the peer is dead/unreachable, we do not
    286 		 * time out if the window is closed.  After a full
    287 		 * backoff, drop the connection if the idle time
    288 		 * (no responses to probes) reaches the maximum
    289 		 * backoff that we would use if retransmitting.
    290 		 */
    291 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
    292 		    (tp->t_idle >= tcp_maxpersistidle ||
    293 		    tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
    294 			tcpstat.tcps_persistdrops++;
    295 			tp = tcp_drop(tp, ETIMEDOUT);
    296 			break;
    297 		}
    298 		tcpstat.tcps_persisttimeo++;
    299 		tcp_setpersist(tp);
    300 		tp->t_force = 1;
    301 		(void) tcp_output(tp);
    302 		tp->t_force = 0;
    303 		break;
    304 
    305 	/*
    306 	 * Keep-alive timer went off; send something
    307 	 * or drop connection if idle for too long.
    308 	 */
    309 	case TCPT_KEEP:
    310 		tcpstat.tcps_keeptimeo++;
    311 		if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    312 			goto dropit;
    313 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
    314 		    tp->t_state <= TCPS_CLOSE_WAIT) {
    315 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
    316 				goto dropit;
    317 			/*
    318 			 * Send a packet designed to force a response
    319 			 * if the peer is up and reachable:
    320 			 * either an ACK if the connection is still alive,
    321 			 * or an RST if the peer has closed the connection
    322 			 * due to timeout or reboot.
    323 			 * Using sequence number tp->snd_una-1
    324 			 * causes the transmitted zero-length segment
    325 			 * to lie outside the receive window;
    326 			 * by the protocol spec, this requires the
    327 			 * correspondent TCP to respond.
    328 			 */
    329 			tcpstat.tcps_keepprobe++;
    330 #ifdef TCP_COMPAT_42
    331 			/*
    332 			 * The keepalive packet must have nonzero length
    333 			 * to get a 4.2 host to respond.
    334 			 */
    335 			(void)tcp_respond(tp, tp->t_template,
    336 			    (struct mbuf *)NULL, tp->rcv_nxt - 1,
    337 			    tp->snd_una - 1, 0);
    338 #else
    339 			(void)tcp_respond(tp, tp->t_template,
    340 			    (struct mbuf *)NULL, tp->rcv_nxt,
    341 			    tp->snd_una - 1, 0);
    342 #endif
    343 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
    344 		} else
    345 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    346 		break;
    347 	dropit:
    348 		tcpstat.tcps_keepdrops++;
    349 		tp = tcp_drop(tp, ETIMEDOUT);
    350 		break;
    351 	}
    352 	return (tp);
    353 }
    354 #endif /* TUBA_INCLUDE */
    355