Home | History | Annotate | Line # | Download | only in netinet
tcp_timer.c revision 1.6
      1 /*
      2  * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California.
      3  * All rights reserved.
      4  *
      5  * Redistribution and use in source and binary forms, with or without
      6  * modification, are permitted provided that the following conditions
      7  * are met:
      8  * 1. Redistributions of source code must retain the above copyright
      9  *    notice, this list of conditions and the following disclaimer.
     10  * 2. Redistributions in binary form must reproduce the above copyright
     11  *    notice, this list of conditions and the following disclaimer in the
     12  *    documentation and/or other materials provided with the distribution.
     13  * 3. All advertising materials mentioning features or use of this software
     14  *    must display the following acknowledgement:
     15  *	This product includes software developed by the University of
     16  *	California, Berkeley and its contributors.
     17  * 4. Neither the name of the University nor the names of its contributors
     18  *    may be used to endorse or promote products derived from this software
     19  *    without specific prior written permission.
     20  *
     21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     31  * SUCH DAMAGE.
     32  *
     33  *	from: @(#)tcp_timer.c	7.18 (Berkeley) 6/28/90
     34  *	$Id: tcp_timer.c,v 1.6 1994/01/08 23:07:22 mycroft Exp $
     35  */
     36 
     37 #include <sys/param.h>
     38 #include <sys/systm.h>
     39 #include <sys/malloc.h>
     40 #include <sys/mbuf.h>
     41 #include <sys/socket.h>
     42 #include <sys/socketvar.h>
     43 #include <sys/protosw.h>
     44 #include <sys/errno.h>
     45 
     46 #include <net/if.h>
     47 #include <net/route.h>
     48 
     49 #include <netinet/in.h>
     50 #include <netinet/in_systm.h>
     51 #include <netinet/ip.h>
     52 #include <netinet/in_pcb.h>
     53 #include <netinet/ip_var.h>
     54 #include <netinet/tcp.h>
     55 #include <netinet/tcp_fsm.h>
     56 #include <netinet/tcp_seq.h>
     57 #include <netinet/tcp_timer.h>
     58 #include <netinet/tcp_var.h>
     59 #include <netinet/tcpip.h>
     60 
     61 int	tcp_keepidle = TCPTV_KEEP_IDLE;
     62 int	tcp_keepintvl = TCPTV_KEEPINTVL;
     63 int	tcp_maxidle;
     64 
     65 /*
     66  * Fast timeout routine for processing delayed acks
     67  */
     68 void
     69 tcp_fasttimo()
     70 {
     71 	register struct inpcb *inp;
     72 	register struct tcpcb *tp;
     73 	int s = splnet();
     74 
     75 	inp = tcb.inp_next;
     76 	if (inp)
     77 	for (; inp != &tcb; inp = inp->inp_next)
     78 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
     79 		    (tp->t_flags & TF_DELACK)) {
     80 			tp->t_flags &= ~TF_DELACK;
     81 			tp->t_flags |= TF_ACKNOW;
     82 			tcpstat.tcps_delack++;
     83 			(void) tcp_output(tp);
     84 		}
     85 	splx(s);
     86 }
     87 
     88 /*
     89  * Tcp protocol timeout routine called every 500 ms.
     90  * Updates the timers in all active tcb's and
     91  * causes finite state machine actions if timers expire.
     92  */
     93 void
     94 tcp_slowtimo()
     95 {
     96 	register struct inpcb *ip, *ipnxt;
     97 	register struct tcpcb *tp;
     98 	int s = splnet();
     99 	register int i;
    100 
    101 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
    102 	/*
    103 	 * Search through tcb's and update active timers.
    104 	 */
    105 	ip = tcb.inp_next;
    106 	if (ip == 0) {
    107 		splx(s);
    108 		return;
    109 	}
    110 	for (; ip != &tcb; ip = ipnxt) {
    111 		ipnxt = ip->inp_next;
    112 		tp = intotcpcb(ip);
    113 		if (tp == 0)
    114 			continue;
    115 		for (i = 0; i < TCPT_NTIMERS; i++) {
    116 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
    117 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
    118 				    PRU_SLOWTIMO, (struct mbuf *)0,
    119 				    (struct mbuf *)i, (struct mbuf *)0);
    120 				if (ipnxt->inp_prev != ip)
    121 					goto tpgone;
    122 			}
    123 		}
    124 		tp->t_idle++;
    125 		if (tp->t_rtt)
    126 			tp->t_rtt++;
    127 tpgone:
    128 		;
    129 	}
    130 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
    131 #ifdef TCP_COMPAT_42
    132 	if ((int)tcp_iss < 0)
    133 		tcp_iss = 0;				/* XXX */
    134 #endif
    135 	splx(s);
    136 }
    137 
    138 /*
    139  * Cancel all timers for TCP tp.
    140  */
    141 void
    142 tcp_canceltimers(tp)
    143 	struct tcpcb *tp;
    144 {
    145 	register int i;
    146 
    147 	for (i = 0; i < TCPT_NTIMERS; i++)
    148 		tp->t_timer[i] = 0;
    149 }
    150 
    151 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
    152     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
    153 
    154 /*
    155  * TCP timer processing.
    156  */
    157 struct tcpcb *
    158 tcp_timers(tp, timer)
    159 	register struct tcpcb *tp;
    160 	int timer;
    161 {
    162 	register int rexmt;
    163 
    164 	switch (timer) {
    165 
    166 	/*
    167 	 * 2 MSL timeout in shutdown went off.  If we're closed but
    168 	 * still waiting for peer to close and connection has been idle
    169 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
    170 	 * control block.  Otherwise, check again in a bit.
    171 	 */
    172 	case TCPT_2MSL:
    173 		if (tp->t_state != TCPS_TIME_WAIT &&
    174 		    tp->t_idle <= tcp_maxidle)
    175 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
    176 		else
    177 			tp = tcp_close(tp);
    178 		break;
    179 
    180 	/*
    181 	 * Retransmission timer went off.  Message has not
    182 	 * been acked within retransmit interval.  Back off
    183 	 * to a longer retransmit interval and retransmit one segment.
    184 	 */
    185 	case TCPT_REXMT:
    186 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
    187 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
    188 			tcpstat.tcps_timeoutdrop++;
    189 			tp = tcp_drop(tp, tp->t_softerror ?
    190 			    tp->t_softerror : ETIMEDOUT);
    191 			break;
    192 		}
    193 		tcpstat.tcps_rexmttimeo++;
    194 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
    195 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
    196 		    tp->t_rttmin, TCPTV_REXMTMAX);
    197 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
    198 		/*
    199 		 * If losing, let the lower level know and try for
    200 		 * a better route.  Also, if we backed off this far,
    201 		 * our srtt estimate is probably bogus.  Clobber it
    202 		 * so we'll take the next rtt measurement as our srtt;
    203 		 * move the current srtt into rttvar to keep the current
    204 		 * retransmit times until then.
    205 		 */
    206 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
    207 			in_losing(tp->t_inpcb);
    208 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
    209 			tp->t_srtt = 0;
    210 		}
    211 		tp->snd_nxt = tp->snd_una;
    212 		/*
    213 		 * If timing a segment in this window, stop the timer.
    214 		 */
    215 		tp->t_rtt = 0;
    216 		/*
    217 		 * Close the congestion window down to one segment
    218 		 * (we'll open it by one segment for each ack we get).
    219 		 * Since we probably have a window's worth of unacked
    220 		 * data accumulated, this "slow start" keeps us from
    221 		 * dumping all that data as back-to-back packets (which
    222 		 * might overwhelm an intermediate gateway).
    223 		 *
    224 		 * There are two phases to the opening: Initially we
    225 		 * open by one mss on each ack.  This makes the window
    226 		 * size increase exponentially with time.  If the
    227 		 * window is larger than the path can handle, this
    228 		 * exponential growth results in dropped packet(s)
    229 		 * almost immediately.  To get more time between
    230 		 * drops but still "push" the network to take advantage
    231 		 * of improving conditions, we switch from exponential
    232 		 * to linear window opening at some threshhold size.
    233 		 * For a threshhold, we use half the current window
    234 		 * size, truncated to a multiple of the mss.
    235 		 *
    236 		 * (the minimum cwnd that will give us exponential
    237 		 * growth is 2 mss.  We don't allow the threshhold
    238 		 * to go below this.)
    239 		 */
    240 		{
    241 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
    242 		if (win < 2)
    243 			win = 2;
    244 		tp->snd_cwnd = tp->t_maxseg;
    245 		tp->snd_ssthresh = win * tp->t_maxseg;
    246 		tp->t_dupacks = 0;
    247 		}
    248 		(void) tcp_output(tp);
    249 		break;
    250 
    251 	/*
    252 	 * Persistance timer into zero window.
    253 	 * Force a byte to be output, if possible.
    254 	 */
    255 	case TCPT_PERSIST:
    256 		tcpstat.tcps_persisttimeo++;
    257 		tcp_setpersist(tp);
    258 		tp->t_force = 1;
    259 		(void) tcp_output(tp);
    260 		tp->t_force = 0;
    261 		break;
    262 
    263 	/*
    264 	 * Keep-alive timer went off; send something
    265 	 * or drop connection if idle for too long.
    266 	 */
    267 	case TCPT_KEEP:
    268 		tcpstat.tcps_keeptimeo++;
    269 		if (tp->t_state < TCPS_ESTABLISHED)
    270 			goto dropit;
    271 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
    272 		    tp->t_state <= TCPS_CLOSE_WAIT) {
    273 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
    274 				goto dropit;
    275 			/*
    276 			 * Send a packet designed to force a response
    277 			 * if the peer is up and reachable:
    278 			 * either an ACK if the connection is still alive,
    279 			 * or an RST if the peer has closed the connection
    280 			 * due to timeout or reboot.
    281 			 * Using sequence number tp->snd_una-1
    282 			 * causes the transmitted zero-length segment
    283 			 * to lie outside the receive window;
    284 			 * by the protocol spec, this requires the
    285 			 * correspondent TCP to respond.
    286 			 */
    287 			tcpstat.tcps_keepprobe++;
    288 #ifdef TCP_COMPAT_42
    289 			/*
    290 			 * The keepalive packet must have nonzero length
    291 			 * to get a 4.2 host to respond.
    292 			 */
    293 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
    294 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
    295 #else
    296 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
    297 			    tp->rcv_nxt, tp->snd_una - 1, 0);
    298 #endif
    299 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
    300 		} else
    301 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    302 		break;
    303 	dropit:
    304 		tcpstat.tcps_keepdrops++;
    305 		tp = tcp_drop(tp, ETIMEDOUT);
    306 		break;
    307 	}
    308 	return (tp);
    309 }
    310