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 are permitted 6 * provided that the above copyright notice and this paragraph are 7 * duplicated in all such forms and that any documentation, 8 * advertising materials, and other materials related to such 9 * distribution and use acknowledge that the software was developed 10 * by the University of California, Berkeley. The name of the 11 * University may not be used to endorse or promote products derived 12 * from this software without specific prior written permission. 13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 16 * 17 * @(#)tcp_timer.c 7.17 (Berkeley) 06/28/90 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "malloc.h" 23 #include "mbuf.h" 24 #include "socket.h" 25 #include "socketvar.h" 26 #include "protosw.h" 27 #include "errno.h" 28 29 #include "../net/if.h" 30 #include "../net/route.h" 31 32 #include "in.h" 33 #include "in_systm.h" 34 #include "ip.h" 35 #include "in_pcb.h" 36 #include "ip_var.h" 37 #include "tcp.h" 38 #include "tcp_fsm.h" 39 #include "tcp_seq.h" 40 #include "tcp_timer.h" 41 #include "tcp_var.h" 42 #include "tcpip.h" 43 44 int tcp_keepidle = TCPTV_KEEP_IDLE; 45 int tcp_keepintvl = TCPTV_KEEPINTVL; 46 int tcp_maxidle; 47 /* 48 * Fast timeout routine for processing delayed acks 49 */ 50 tcp_fasttimo() 51 { 52 register struct inpcb *inp; 53 register struct tcpcb *tp; 54 int s = splnet(); 55 56 inp = tcb.inp_next; 57 if (inp) 58 for (; inp != &tcb; inp = inp->inp_next) 59 if ((tp = (struct tcpcb *)inp->inp_ppcb) && 60 (tp->t_flags & TF_DELACK)) { 61 tp->t_flags &= ~TF_DELACK; 62 tp->t_flags |= TF_ACKNOW; 63 tcpstat.tcps_delack++; 64 (void) tcp_output(tp); 65 } 66 splx(s); 67 } 68 69 /* 70 * Tcp protocol timeout routine called every 500 ms. 71 * Updates the timers in all active tcb's and 72 * causes finite state machine actions if timers expire. 73 */ 74 tcp_slowtimo() 75 { 76 register struct inpcb *ip, *ipnxt; 77 register struct tcpcb *tp; 78 int s = splnet(); 79 register int i; 80 81 tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl; 82 /* 83 * Search through tcb's and update active timers. 84 */ 85 ip = tcb.inp_next; 86 if (ip == 0) { 87 splx(s); 88 return; 89 } 90 for (; ip != &tcb; ip = ipnxt) { 91 ipnxt = ip->inp_next; 92 tp = intotcpcb(ip); 93 if (tp == 0) 94 continue; 95 for (i = 0; i < TCPT_NTIMERS; i++) { 96 if (tp->t_timer[i] && --tp->t_timer[i] == 0) { 97 (void) tcp_usrreq(tp->t_inpcb->inp_socket, 98 PRU_SLOWTIMO, (struct mbuf *)0, 99 (struct mbuf *)i, (struct mbuf *)0); 100 if (ipnxt->inp_prev != ip) 101 goto tpgone; 102 } 103 } 104 tp->t_idle++; 105 if (tp->t_rtt) 106 tp->t_rtt++; 107 tpgone: 108 ; 109 } 110 tcp_iss += TCP_ISSINCR/PR_SLOWHZ; /* increment iss */ 111 #ifdef TCP_COMPAT_42 112 if ((int)tcp_iss < 0) 113 tcp_iss = 0; /* XXX */ 114 #endif 115 splx(s); 116 } 117 118 /* 119 * Cancel all timers for TCP tp. 120 */ 121 tcp_canceltimers(tp) 122 struct tcpcb *tp; 123 { 124 register int i; 125 126 for (i = 0; i < TCPT_NTIMERS; i++) 127 tp->t_timer[i] = 0; 128 } 129 130 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 131 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 132 133 /* 134 * TCP timer processing. 135 */ 136 struct tcpcb * 137 tcp_timers(tp, timer) 138 register struct tcpcb *tp; 139 int timer; 140 { 141 register int rexmt; 142 143 switch (timer) { 144 145 /* 146 * 2 MSL timeout in shutdown went off. If we're closed but 147 * still waiting for peer to close and connection has been idle 148 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 149 * control block. Otherwise, check again in a bit. 150 */ 151 case TCPT_2MSL: 152 if (tp->t_state != TCPS_TIME_WAIT && 153 tp->t_idle <= tcp_maxidle) 154 tp->t_timer[TCPT_2MSL] = tcp_keepintvl; 155 else 156 tp = tcp_close(tp); 157 break; 158 159 /* 160 * Retransmission timer went off. Message has not 161 * been acked within retransmit interval. Back off 162 * to a longer retransmit interval and retransmit one segment. 163 */ 164 case TCPT_REXMT: 165 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 166 tp->t_rxtshift = TCP_MAXRXTSHIFT; 167 tcpstat.tcps_timeoutdrop++; 168 tp = tcp_drop(tp, tp->t_softerror ? 169 tp->t_softerror : ETIMEDOUT); 170 break; 171 } 172 tcpstat.tcps_rexmttimeo++; 173 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 174 TCPT_RANGESET(tp->t_rxtcur, rexmt, 175 tp->t_rttmin, TCPTV_REXMTMAX); 176 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 177 /* 178 * If losing, let the lower level know and try for 179 * a better route. Also, if we backed off this far, 180 * our srtt estimate is probably bogus. Clobber it 181 * so we'll take the next rtt measurement as our srtt; 182 * move the current srtt into rttvar to keep the current 183 * retransmit times until then. 184 */ 185 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 186 in_losing(tp->t_inpcb); 187 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 188 tp->t_srtt = 0; 189 } 190 tp->snd_nxt = tp->snd_una; 191 /* 192 * If timing a segment in this window, stop the timer. 193 */ 194 tp->t_rtt = 0; 195 /* 196 * Close the congestion window down to one segment 197 * (we'll open it by one segment for each ack we get). 198 * Since we probably have a window's worth of unacked 199 * data accumulated, this "slow start" keeps us from 200 * dumping all that data as back-to-back packets (which 201 * might overwhelm an intermediate gateway). 202 * 203 * There are two phases to the opening: Initially we 204 * open by one mss on each ack. This makes the window 205 * size increase exponentially with time. If the 206 * window is larger than the path can handle, this 207 * exponential growth results in dropped packet(s) 208 * almost immediately. To get more time between 209 * drops but still "push" the network to take advantage 210 * of improving conditions, we switch from exponential 211 * to linear window opening at some threshhold size. 212 * For a threshhold, we use half the current window 213 * size, truncated to a multiple of the mss. 214 * 215 * (the minimum cwnd that will give us exponential 216 * growth is 2 mss. We don't allow the threshhold 217 * to go below this.) 218 */ 219 { 220 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 221 if (win < 2) 222 win = 2; 223 tp->snd_cwnd = tp->t_maxseg; 224 tp->snd_ssthresh = win * tp->t_maxseg; 225 tp->t_dupacks = 0; 226 } 227 (void) tcp_output(tp); 228 break; 229 230 /* 231 * Persistance timer into zero window. 232 * Force a byte to be output, if possible. 233 */ 234 case TCPT_PERSIST: 235 tcpstat.tcps_persisttimeo++; 236 tcp_setpersist(tp); 237 tp->t_force = 1; 238 (void) tcp_output(tp); 239 tp->t_force = 0; 240 break; 241 242 /* 243 * Keep-alive timer went off; send something 244 * or drop connection if idle for too long. 245 */ 246 case TCPT_KEEP: 247 tcpstat.tcps_keeptimeo++; 248 if (tp->t_state < TCPS_ESTABLISHED) 249 goto dropit; 250 if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE && 251 tp->t_state <= TCPS_CLOSE_WAIT) { 252 if (tp->t_idle >= tcp_keepidle + tcp_maxidle) 253 goto dropit; 254 /* 255 * Send a packet designed to force a response 256 * if the peer is up and reachable: 257 * either an ACK if the connection is still alive, 258 * or an RST if the peer has closed the connection 259 * due to timeout or reboot. 260 * Using sequence number tp->snd_una-1 261 * causes the transmitted zero-length segment 262 * to lie outside the receive window; 263 * by the protocol spec, this requires the 264 * correspondent TCP to respond. 265 */ 266 tcpstat.tcps_keepprobe++; 267 #ifdef TCP_COMPAT_42 268 /* 269 * The keepalive packet must have nonzero length 270 * to get a 4.2 host to respond. 271 */ 272 tcp_respond(tp, tp->t_template, (struct mbuf *)NULL, 273 tp->rcv_nxt - 1, tp->snd_una - 1, 0); 274 #else 275 tcp_respond(tp, tp->t_template, (struct mbuf *)NULL, 276 tp->rcv_nxt, tp->snd_una - 1, 0); 277 #endif 278 tp->t_timer[TCPT_KEEP] = tcp_keepintvl; 279 } else 280 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 281 break; 282 dropit: 283 tcpstat.tcps_keepdrops++; 284 tp = tcp_drop(tp, ETIMEDOUT); 285 break; 286 } 287 return (tp); 288 } 289