1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_timer.c 8.2 (Berkeley) 05/24/95 8 */ 9 10 #ifndef TUBA_INCLUDE 11 #include <sys/param.h> 12 #include <sys/systm.h> 13 #include <sys/malloc.h> 14 #include <sys/mbuf.h> 15 #include <sys/socket.h> 16 #include <sys/socketvar.h> 17 #include <sys/protosw.h> 18 #include <sys/errno.h> 19 20 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 21 22 #include <net/if.h> 23 #include <net/route.h> 24 25 #include <netinet/in.h> 26 #include <netinet/in_systm.h> 27 #include <netinet/ip.h> 28 #include <netinet/in_pcb.h> 29 #include <netinet/ip_var.h> 30 #include <netinet/tcp.h> 31 #include <netinet/tcp_fsm.h> 32 #include <netinet/tcp_seq.h> 33 #include <netinet/tcp_timer.h> 34 #include <netinet/tcp_var.h> 35 #include <netinet/tcpip.h> 36 37 int tcp_keepidle = TCPTV_KEEP_IDLE; 38 int tcp_keepintvl = TCPTV_KEEPINTVL; 39 int tcp_keepcnt = TCPTV_KEEPCNT; /* max idle probes */ 40 int tcp_maxpersistidle = TCPTV_KEEP_IDLE; /* max idle time in persist */ 41 int tcp_maxidle; 42 #else /* TUBA_INCLUDE */ 43 44 extern int tcp_maxpersistidle; 45 #endif /* TUBA_INCLUDE */ 46 47 /* 48 * Fast timeout routine for processing delayed acks 49 */ 50 void 51 tcp_fasttimo() 52 { 53 register struct inpcb *inp; 54 register struct tcpcb *tp; 55 int s = splnet(); 56 57 inp = tcb.inp_next; 58 if (inp) 59 for (; inp != &tcb; inp = inp->inp_next) 60 if ((tp = (struct tcpcb *)inp->inp_ppcb) && 61 (tp->t_flags & TF_DELACK)) { 62 tp->t_flags &= ~TF_DELACK; 63 tp->t_flags |= TF_ACKNOW; 64 tcpstat.tcps_delack++; 65 (void) tcp_output(tp); 66 } 67 splx(s); 68 } 69 70 /* 71 * Tcp protocol timeout routine called every 500 ms. 72 * Updates the timers in all active tcb's and 73 * causes finite state machine actions if timers expire. 74 */ 75 void 76 tcp_slowtimo() 77 { 78 register struct inpcb *ip, *ipnxt; 79 register struct tcpcb *tp; 80 int s = splnet(); 81 register int i; 82 83 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 84 /* 85 * Search through tcb's and update active timers. 86 */ 87 ip = tcb.inp_next; 88 if (ip == 0) { 89 splx(s); 90 return; 91 } 92 for (; ip != &tcb; ip = ipnxt) { 93 ipnxt = ip->inp_next; 94 tp = intotcpcb(ip); 95 if (tp == 0 || tp->t_state == TCPS_LISTEN) 96 continue; 97 for (i = 0; i < TCPT_NTIMERS; i++) { 98 if (tp->t_timer[i] && --tp->t_timer[i] == 0) { 99 (void) tcp_usrreq(tp->t_inpcb->inp_socket, 100 PRU_SLOWTIMO, (struct mbuf *)0, 101 (struct mbuf *)i, (struct mbuf *)0); 102 if (ipnxt->inp_prev != ip) 103 goto tpgone; 104 } 105 } 106 tp->t_idle++; 107 if (tp->t_rtt) 108 tp->t_rtt++; 109 tpgone: 110 ; 111 } 112 tcp_iss += TCP_ISSINCR/PR_SLOWHZ; /* increment iss */ 113 #ifdef TCP_COMPAT_42 114 if ((int)tcp_iss < 0) 115 tcp_iss = TCP_ISSINCR; /* XXX */ 116 #endif 117 tcp_now++; /* for timestamps */ 118 splx(s); 119 } 120 #ifndef TUBA_INCLUDE 121 122 /* 123 * Cancel all timers for TCP tp. 124 */ 125 void 126 tcp_canceltimers(tp) 127 struct tcpcb *tp; 128 { 129 register int i; 130 131 for (i = 0; i < TCPT_NTIMERS; i++) 132 tp->t_timer[i] = 0; 133 } 134 135 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 136 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 137 138 int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 139 140 /* 141 * TCP timer processing. 142 */ 143 struct tcpcb * 144 tcp_timers(tp, timer) 145 register struct tcpcb *tp; 146 int timer; 147 { 148 register int rexmt; 149 150 switch (timer) { 151 152 /* 153 * 2 MSL timeout in shutdown went off. If we're closed but 154 * still waiting for peer to close and connection has been idle 155 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 156 * control block. Otherwise, check again in a bit. 157 */ 158 case TCPT_2MSL: 159 if (tp->t_state != TCPS_TIME_WAIT && 160 tp->t_idle <= tcp_maxidle) 161 tp->t_timer[TCPT_2MSL] = tcp_keepintvl; 162 else 163 tp = tcp_close(tp); 164 break; 165 166 /* 167 * Retransmission timer went off. Message has not 168 * been acked within retransmit interval. Back off 169 * to a longer retransmit interval and retransmit one segment. 170 */ 171 case TCPT_REXMT: 172 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 173 tp->t_rxtshift = TCP_MAXRXTSHIFT; 174 tcpstat.tcps_timeoutdrop++; 175 tp = tcp_drop(tp, tp->t_softerror ? 176 tp->t_softerror : ETIMEDOUT); 177 break; 178 } 179 tcpstat.tcps_rexmttimeo++; 180 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 181 TCPT_RANGESET(tp->t_rxtcur, rexmt, 182 tp->t_rttmin, TCPTV_REXMTMAX); 183 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 184 /* 185 * If losing, let the lower level know and try for 186 * a better route. Also, if we backed off this far, 187 * our srtt estimate is probably bogus. Clobber it 188 * so we'll take the next rtt measurement as our srtt; 189 * move the current srtt into rttvar to keep the current 190 * retransmit times until then. 191 */ 192 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 193 in_losing(tp->t_inpcb); 194 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 195 tp->t_srtt = 0; 196 } 197 tp->snd_nxt = tp->snd_una; 198 /* 199 * If timing a segment in this window, stop the timer. 200 */ 201 tp->t_rtt = 0; 202 /* 203 * Close the congestion window down to one segment 204 * (we'll open it by one segment for each ack we get). 205 * Since we probably have a window's worth of unacked 206 * data accumulated, this "slow start" keeps us from 207 * dumping all that data as back-to-back packets (which 208 * might overwhelm an intermediate gateway). 209 * 210 * There are two phases to the opening: Initially we 211 * open by one mss on each ack. This makes the window 212 * size increase exponentially with time. If the 213 * window is larger than the path can handle, this 214 * exponential growth results in dropped packet(s) 215 * almost immediately. To get more time between 216 * drops but still "push" the network to take advantage 217 * of improving conditions, we switch from exponential 218 * to linear window opening at some threshhold size. 219 * For a threshhold, we use half the current window 220 * size, truncated to a multiple of the mss. 221 * 222 * (the minimum cwnd that will give us exponential 223 * growth is 2 mss. We don't allow the threshhold 224 * to go below this.) 225 */ 226 { 227 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 228 if (win < 2) 229 win = 2; 230 tp->snd_cwnd = tp->t_maxseg; 231 tp->snd_ssthresh = win * tp->t_maxseg; 232 tp->t_dupacks = 0; 233 } 234 (void) tcp_output(tp); 235 break; 236 237 /* 238 * Persistance timer into zero window. 239 * Force a byte to be output, if possible. 240 */ 241 case TCPT_PERSIST: 242 tcpstat.tcps_persisttimeo++; 243 /* 244 * Hack: if the peer is dead/unreachable, we do not 245 * time out if the window is closed. After a full 246 * backoff, drop the connection if the idle time 247 * (no responses to probes) reaches the maximum 248 * backoff that we would use if retransmitting. 249 */ 250 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 251 (tp->t_idle >= tcp_maxpersistidle || 252 tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { 253 tcpstat.tcps_persistdrop++; 254 tp = tcp_drop(tp, ETIMEDOUT); 255 break; 256 } 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 #endif /* TUBA_INCLUDE */ 311