2.11BSD#431 - /usr/src/sys/netinet/tcp_timer.c

   1: /*
   2:  * Copyright (c) 1982, 1986 Regents of the University of California.
   3:  * All rights reserved.
   4:  *
   5:  * Redistribution and use in source and binary forms are permitted
   6:  * provided that this notice is preserved and that due credit is given
   7:  * to the University of California at Berkeley. The name of the University
   8:  * may not be used to endorse or promote products derived from this
   9:  * software without specific prior written permission. This software
  10:  * is provided ``as is'' without express or implied warranty.
  11:  *
  12:  *	@(#)tcp_timer.c	7.11.1.2 (Berkeley) 3/16/88
  13:  */
  14: 
  15: #include "param.h"
  16: #include "systm.h"
  17: #include "mbuf.h"
  18: #include "socket.h"
  19: #include "socketvar.h"
  20: #include "protosw.h"
  21: #include "errno.h"
  22: 
  23: #include "../net/if.h"
  24: #include "../net/route.h"
  25: 
  26: #include "domain.h"
  27: #include "in.h"
  28: #include "in_pcb.h"
  29: #include "in_systm.h"
  30: #include "ip.h"
  31: #include "ip_var.h"
  32: #include "tcp.h"
  33: #include "tcp_fsm.h"
  34: #include "tcp_seq.h"
  35: #include "tcp_timer.h"
  36: #include "tcp_var.h"
  37: #include "tcpip.h"
  38: 
  39: int tcpnodelack = 0;
  40: int tcp_keepidle = TCPTV_KEEP_IDLE;
  41: int tcp_keepintvl = TCPTV_KEEPINTVL;
  42: int tcp_maxidle;
  43: /*
  44:  * Fast timeout routine for processing delayed acks
  45:  */
  46: tcp_fasttimo()
  47: {
  48:     register struct inpcb *inp;
  49:     register struct tcpcb *tp;
  50:     int s = splnet();
  51: 
  52:     inp = tcb.inp_next;
  53:     if (inp)
  54:     for (; inp != &tcb; inp = inp->inp_next)
  55:         if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
  56:             (tp->t_flags & TF_DELACK)) {
  57:             tp->t_flags &= ~TF_DELACK;
  58:             tp->t_flags |= TF_ACKNOW;
  59:             tcpstat.tcps_delack++;
  60:             (void) tcp_output(tp);
  61:         }
  62:     splx(s);
  63: }
  64: 
  65: /*
  66:  * Tcp protocol timeout routine called every 500 ms.
  67:  * Updates the timers in all active tcb's and
  68:  * causes finite state machine actions if timers expire.
  69:  */
  70: tcp_slowtimo()
  71: {
  72:     register struct inpcb *ip, *ipnxt;
  73:     register struct tcpcb *tp;
  74:     int s = splnet();
  75:     register int i;
  76: 
  77:     tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
  78:     /*
  79: 	 * Search through tcb's and update active timers.
  80: 	 */
  81:     ip = tcb.inp_next;
  82:     if (ip == 0) {
  83:         splx(s);
  84:         return;
  85:     }
  86:     for (; ip != &tcb; ip = ipnxt) {
  87:         ipnxt = ip->inp_next;
  88:         tp = intotcpcb(ip);
  89:         if (tp == 0)
  90:             continue;
  91:         for (i = 0; i < TCPT_NTIMERS; i++) {
  92:             if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
  93:                 (void) tcp_usrreq(tp->t_inpcb->inp_socket,
  94:                     PRU_SLOWTIMO, (struct mbuf *)0,
  95:                     (struct mbuf *)i, (struct mbuf *)0);
  96:                 if (ipnxt->inp_prev != ip)
  97:                     goto tpgone;
  98:             }
  99:         }
 100:         tp->t_idle++;
 101:         if (tp->t_rtt)
 102:             tp->t_rtt++;
 103: tpgone:
 104:         ;
 105:     }
 106:     tcp_iss += TCP_ISSINCR/PR_SLOWHZ;       /* increment iss */
 107: #ifdef TCP_COMPAT_42
 108:     if ((long)tcp_iss < 0)
 109:         tcp_iss = 0;                /* XXX */
 110: #endif
 111:     splx(s);
 112: }
 113: 
 114: /*
 115:  * Cancel all timers for TCP tp.
 116:  */
 117: tcp_canceltimers(tp)
 118:     struct tcpcb *tp;
 119: {
 120:     register int i;
 121: 
 122:     for (i = 0; i < TCPT_NTIMERS; i++)
 123:         tp->t_timer[i] = 0;
 124: }
 125: 
 126: int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 127:     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 128: 
 129: /*
 130:  * TCP timer processing.
 131:  */
 132: struct tcpcb *
 133: tcp_timers(tp, timer)
 134:     register struct tcpcb *tp;
 135:     int timer;
 136: {
 137:     register int rexmt;
 138: 
 139:     switch (timer) {
 140: 
 141:     /*
 142: 	 * 2 MSL timeout in shutdown went off.  If we're closed but
 143: 	 * still waiting for peer to close and connection has been idle
 144: 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 145: 	 * control block.  Otherwise, check again in a bit.
 146: 	 */
 147:     case TCPT_2MSL:
 148:         if (tp->t_state != TCPS_TIME_WAIT &&
 149:             tp->t_idle <= tcp_maxidle)
 150:             tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
 151:         else
 152:             tp = tcp_close(tp);
 153:         break;
 154: 
 155:     /*
 156: 	 * Retransmission timer went off.  Message has not
 157: 	 * been acked within retransmit interval.  Back off
 158: 	 * to a longer retransmit interval and retransmit one segment.
 159: 	 */
 160:     case TCPT_REXMT:
 161:         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 162:             tp->t_rxtshift = TCP_MAXRXTSHIFT;
 163:             tcpstat.tcps_timeoutdrop++;
 164:             tp = tcp_drop(tp, ETIMEDOUT);
 165:             break;
 166:         }
 167:         tcpstat.tcps_rexmttimeo++;
 168:         rexmt = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 169:         rexmt *= tcp_backoff[tp->t_rxtshift];
 170:         TCPT_RANGESET(tp->t_rxtcur, rexmt, TCPTV_MIN, TCPTV_REXMTMAX);
 171:         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 172:         /*
 173: 		 * If losing, let the lower level know and try for
 174: 		 * a better route.  Also, if we backed off this far,
 175: 		 * our srtt estimate is probably bogus.  Clobber it
 176: 		 * so we'll take the next rtt measurement as our srtt;
 177: 		 * move the current srtt into rttvar to keep the current
 178: 		 * retransmit times until then.
 179: 		 */
 180:         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 181: #if BSD>=43
 182:             in_losing(tp->t_inpcb);
 183: #endif
 184:             tp->t_rttvar += (tp->t_srtt >> 2);
 185:             tp->t_srtt = 0;
 186:         }
 187:         tp->snd_nxt = tp->snd_una;
 188:         /*
 189: 		 * If timing a segment in this window, stop the timer.
 190: 		 */
 191:         tp->t_rtt = 0;
 192:         /*
 193: 		 * Close the congestion window down to one segment
 194: 		 * (we'll open it by one segment for each ack we get).
 195: 		 * Since we probably have a window's worth of unacked
 196: 		 * data accumulated, this "slow start" keeps us from
 197: 		 * dumping all that data as back-to-back packets (which
 198: 		 * might overwhelm an intermediate gateway).
 199: 		 *
 200: 		 * There are two phases to the opening: Initially we
 201: 		 * open by one mss on each ack.  This makes the window
 202: 		 * size increase exponentially with time.  If the
 203: 		 * window is larger than the path can handle, this
 204: 		 * exponential growth results in dropped packet(s)
 205: 		 * almost immediately.  To get more time between
 206: 		 * drops but still "push" the network to take advantage
 207: 		 * of improving conditions, we switch from exponential
 208: 		 * to linear window opening at some threshhold size.
 209: 		 * For a threshhold, we use half the current window
 210: 		 * size, truncated to a multiple of the mss.
 211: 		 *
 212: 		 * (the minimum cwnd that will give us exponential
 213: 		 * growth is 2 mss.  We don't allow the threshhold
 214: 		 * to go below this.)
 215: 		 */
 216:         {
 217:         u_int win = MIN(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 218:         if (win < 2)
 219:             win = 2;
 220:         tp->snd_cwnd = tp->t_maxseg;
 221:         tp->snd_ssthresh = win * tp->t_maxseg;
 222:         }
 223:         (void) tcp_output(tp);
 224:         break;
 225: 
 226:     /*
 227: 	 * Persistance timer into zero window.
 228: 	 * Force a byte to be output, if possible.
 229: 	 */
 230:     case TCPT_PERSIST:
 231:         tcpstat.tcps_persisttimeo++;
 232:         tcp_setpersist(tp);
 233:         tp->t_force = 1;
 234:         (void) tcp_output(tp);
 235:         tp->t_force = 0;
 236:         break;
 237: 
 238:     /*
 239: 	 * Keep-alive timer went off; send something
 240: 	 * or drop connection if idle for too long.
 241: 	 */
 242:     case TCPT_KEEP:
 243:         tcpstat.tcps_keeptimeo++;
 244:         if (tp->t_state < TCPS_ESTABLISHED)
 245:             goto dropit;
 246:         if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
 247:             tp->t_state <= TCPS_CLOSE_WAIT) {
 248:                 if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
 249:                 goto dropit;
 250:             /*
 251: 			 * Send a packet designed to force a response
 252: 			 * if the peer is up and reachable:
 253: 			 * either an ACK if the connection is still alive,
 254: 			 * or an RST if the peer has closed the connection
 255: 			 * due to timeout or reboot.
 256: 			 * Using sequence number tp->snd_una-1
 257: 			 * causes the transmitted zero-length segment
 258: 			 * to lie outside the receive window;
 259: 			 * by the protocol spec, this requires the
 260: 			 * correspondent TCP to respond.
 261: 			 */
 262:             tcpstat.tcps_keepprobe++;
 263: #ifdef TCP_COMPAT_42
 264:             /*
 265: 			 * The keepalive packet must have nonzero length
 266: 			 * to get a 4.2 host to respond.
 267: 			 */
 268:             tcp_respond(tp, tp->t_template,
 269:                 tp->rcv_nxt - 1, tp->snd_una - 1, 0);
 270: #else
 271:             tcp_respond(tp, tp->t_template,
 272:                 tp->rcv_nxt, tp->snd_una - 1, 0);
 273: #endif
 274:             tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
 275:         } else
 276:             tp->t_timer[TCPT_KEEP] = tcp_keepidle;
 277:         break;
 278:     dropit:
 279:         tcpstat.tcps_keepdrops++;
 280:         tp = tcp_drop(tp, ETIMEDOUT);
 281:         break;
 282:     }
 283:     return (tp);
 284: }

Defined functions

Defined variables