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 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 7.11.1.3 (Berkeley) 02/15/89 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "mbuf.h" 23 #include "socket.h" 24 #include "socketvar.h" 25 #include "protosw.h" 26 #include "errno.h" 27 28 #include "../net/if.h" 29 #include "../net/route.h" 30 31 #include "in.h" 32 #include "in_pcb.h" 33 #include "in_systm.h" 34 #include "ip.h" 35 #include "ip_var.h" 36 #include "tcp.h" 37 #include "tcp_fsm.h" 38 #include "tcp_seq.h" 39 #include "tcp_timer.h" 40 #include "tcp_var.h" 41 #include "tcpip.h" 42 43 int tcp_keepidle = TCPTV_KEEP_IDLE; 44 int tcp_keepintvl = TCPTV_KEEPINTVL; 45 int tcp_maxidle; 46 /* 47 * Fast timeout routine for processing delayed acks 48 */ 49 tcp_fasttimo() 50 { 51 register struct inpcb *inp; 52 register struct tcpcb *tp; 53 int s = splnet(); 54 55 inp = tcb.inp_next; 56 if (inp) 57 for (; inp != &tcb; inp = inp->inp_next) 58 if ((tp = (struct tcpcb *)inp->inp_ppcb) && 59 (tp->t_flags & TF_DELACK)) { 60 tp->t_flags &= ~TF_DELACK; 61 tp->t_flags |= TF_ACKNOW; 62 tcpstat.tcps_delack++; 63 (void) tcp_output(tp); 64 } 65 splx(s); 66 } 67 68 /* 69 * Tcp protocol timeout routine called every 500 ms. 70 * Updates the timers in all active tcb's and 71 * causes finite state machine actions if timers expire. 72 */ 73 tcp_slowtimo() 74 { 75 register struct inpcb *ip, *ipnxt; 76 register struct tcpcb *tp; 77 int s = splnet(); 78 register int i; 79 80 tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl; 81 /* 82 * Search through tcb's and update active timers. 83 */ 84 ip = tcb.inp_next; 85 if (ip == 0) { 86 splx(s); 87 return; 88 } 89 for (; ip != &tcb; ip = ipnxt) { 90 ipnxt = ip->inp_next; 91 tp = intotcpcb(ip); 92 if (tp == 0) 93 continue; 94 for (i = 0; i < TCPT_NTIMERS; i++) { 95 if (tp->t_timer[i] && --tp->t_timer[i] == 0) { 96 (void) tcp_usrreq(tp->t_inpcb->inp_socket, 97 PRU_SLOWTIMO, (struct mbuf *)0, 98 (struct mbuf *)i, (struct mbuf *)0); 99 if (ipnxt->inp_prev != ip) 100 goto tpgone; 101 } 102 } 103 tp->t_idle++; 104 if (tp->t_rtt) 105 tp->t_rtt++; 106 tpgone: 107 ; 108 } 109 tcp_iss += TCP_ISSINCR/PR_SLOWHZ; /* increment iss */ 110 #ifdef TCP_COMPAT_42 111 if ((int)tcp_iss < 0) 112 tcp_iss = 0; /* XXX */ 113 #endif 114 splx(s); 115 } 116 117 /* 118 * Cancel all timers for TCP tp. 119 */ 120 tcp_canceltimers(tp) 121 struct tcpcb *tp; 122 { 123 register int i; 124 125 for (i = 0; i < TCPT_NTIMERS; i++) 126 tp->t_timer[i] = 0; 127 } 128 129 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 130 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 131 132 /* 133 * TCP timer processing. 134 */ 135 struct tcpcb * 136 tcp_timers(tp, timer) 137 register struct tcpcb *tp; 138 int timer; 139 { 140 register int rexmt; 141 142 switch (timer) { 143 144 /* 145 * 2 MSL timeout in shutdown went off. If we're closed but 146 * still waiting for peer to close and connection has been idle 147 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 148 * control block. Otherwise, check again in a bit. 149 */ 150 case TCPT_2MSL: 151 if (tp->t_state != TCPS_TIME_WAIT && 152 tp->t_idle <= tcp_maxidle) 153 tp->t_timer[TCPT_2MSL] = tcp_keepintvl; 154 else 155 tp = tcp_close(tp); 156 break; 157 158 /* 159 * Retransmission timer went off. Message has not 160 * been acked within retransmit interval. Back off 161 * to a longer retransmit interval and retransmit one segment. 162 */ 163 case TCPT_REXMT: 164 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 165 tp->t_rxtshift = TCP_MAXRXTSHIFT; 166 tcpstat.tcps_timeoutdrop++; 167 tp = tcp_drop(tp, ETIMEDOUT); 168 break; 169 } 170 tcpstat.tcps_rexmttimeo++; 171 rexmt = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; 172 rexmt *= tcp_backoff[tp->t_rxtshift]; 173 TCPT_RANGESET(tp->t_rxtcur, rexmt, TCPTV_MIN, TCPTV_REXMTMAX); 174 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 175 /* 176 * If losing, let the lower level know and try for 177 * a better route. Also, if we backed off this far, 178 * our srtt estimate is probably bogus. Clobber it 179 * so we'll take the next rtt measurement as our srtt; 180 * move the current srtt into rttvar to keep the current 181 * retransmit times until then. 182 */ 183 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 184 #if BSD>=43 185 in_losing(tp->t_inpcb); 186 #endif 187 tp->t_rttvar += (tp->t_srtt >> 2); 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 } 226 (void) tcp_output(tp); 227 break; 228 229 /* 230 * Persistance timer into zero window. 231 * Force a byte to be output, if possible. 232 */ 233 case TCPT_PERSIST: 234 tcpstat.tcps_persisttimeo++; 235 tcp_setpersist(tp); 236 tp->t_force = 1; 237 (void) tcp_output(tp); 238 tp->t_force = 0; 239 break; 240 241 /* 242 * Keep-alive timer went off; send something 243 * or drop connection if idle for too long. 244 */ 245 case TCPT_KEEP: 246 tcpstat.tcps_keeptimeo++; 247 if (tp->t_state < TCPS_ESTABLISHED) 248 goto dropit; 249 if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE && 250 tp->t_state <= TCPS_CLOSE_WAIT) { 251 if (tp->t_idle >= tcp_keepidle + tcp_maxidle) 252 goto dropit; 253 /* 254 * Send a packet designed to force a response 255 * if the peer is up and reachable: 256 * either an ACK if the connection is still alive, 257 * or an RST if the peer has closed the connection 258 * due to timeout or reboot. 259 * Using sequence number tp->snd_una-1 260 * causes the transmitted zero-length segment 261 * to lie outside the receive window; 262 * by the protocol spec, this requires the 263 * correspondent TCP to respond. 264 */ 265 tcpstat.tcps_keepprobe++; 266 #ifdef TCP_COMPAT_42 267 /* 268 * The keepalive packet must have nonzero length 269 * to get a 4.2 host to respond. 270 */ 271 tcp_respond(tp, tp->t_template, 272 tp->rcv_nxt - 1, tp->snd_una - 1, 0); 273 #else 274 tcp_respond(tp, tp->t_template, 275 tp->rcv_nxt, tp->snd_una - 1, 0); 276 #endif 277 tp->t_timer[TCPT_KEEP] = tcp_keepintvl; 278 } else 279 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 280 break; 281 dropit: 282 tcpstat.tcps_keepdrops++; 283 tp = tcp_drop(tp, ETIMEDOUT); 284 break; 285 } 286 return (tp); 287 } 288