tcp_timer.c revision 1.38
1/* $NetBSD: tcp_timer.c,v 1.38 1998/05/11 19:57:24 thorpej Exp $ */ 2 3/*- 4 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 9 * Facility, NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40/* 41 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 42 * The Regents of the University of California. All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by the University of 55 * California, Berkeley and its contributors. 56 * 4. Neither the name of the University nor the names of its contributors 57 * may be used to endorse or promote products derived from this software 58 * without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 63 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 * 72 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 73 */ 74 75#include <sys/param.h> 76#include <sys/systm.h> 77#include <sys/malloc.h> 78#include <sys/mbuf.h> 79#include <sys/socket.h> 80#include <sys/socketvar.h> 81#include <sys/protosw.h> 82#include <sys/errno.h> 83 84#include <net/if.h> 85#include <net/route.h> 86 87#include <netinet/in.h> 88#include <netinet/in_systm.h> 89#include <netinet/ip.h> 90#include <netinet/in_pcb.h> 91#include <netinet/ip_var.h> 92#include <netinet/tcp.h> 93#include <netinet/tcp_fsm.h> 94#include <netinet/tcp_seq.h> 95#include <netinet/tcp_timer.h> 96#include <netinet/tcp_var.h> 97#include <netinet/tcpip.h> 98 99int tcp_keepidle = TCPTV_KEEP_IDLE; 100int tcp_keepintvl = TCPTV_KEEPINTVL; 101int tcp_keepcnt = TCPTV_KEEPCNT; /* max idle probes */ 102int tcp_maxpersistidle = TCPTV_KEEP_IDLE; /* max idle time in persist */ 103int tcp_maxidle; 104 105struct tcp_delack_head tcp_delacks; 106 107/* 108 * Fast timeout routine for processing delayed acks 109 */ 110void 111tcp_fasttimo() 112{ 113 register struct tcpcb *tp, *ntp; 114 int s; 115 116 s = splsoftnet(); 117 for (tp = tcp_delacks.lh_first; tp != NULL; tp = ntp) { 118 /* 119 * If tcp_output() can't transmit the ACK for whatever 120 * reason, it will remain on the queue for the next 121 * time the heartbeat ticks. 122 */ 123 ntp = tp->t_delack.le_next; 124 tp->t_flags |= TF_ACKNOW; 125 (void) tcp_output(tp); 126 } 127 splx(s); 128} 129 130/* 131 * Tcp protocol timeout routine called every 500 ms. 132 * Updates the timers in all active tcb's and 133 * causes finite state machine actions if timers expire. 134 */ 135void 136tcp_slowtimo() 137{ 138 register struct inpcb *inp, *ninp; 139 register struct tcpcb *tp; 140 int s; 141 register long i; 142 static int syn_cache_last = 0; 143 144 s = splsoftnet(); 145 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 146 /* 147 * Search through tcb's and update active timers. 148 */ 149 inp = tcbtable.inpt_queue.cqh_first; 150 if (inp == (struct inpcb *)0) { /* XXX */ 151 splx(s); 152 return; 153 } 154 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) { 155 ninp = inp->inp_queue.cqe_next; 156 tp = intotcpcb(inp); 157 if (tp == 0 || tp->t_state == TCPS_LISTEN) 158 continue; 159 for (i = 0; i < TCPT_NTIMERS; i++) { 160 if (TCP_TIMER_ISEXPIRED(tp, i)) { 161 (void) tcp_usrreq(tp->t_inpcb->inp_socket, 162 PRU_SLOWTIMO, (struct mbuf *)0, 163 (struct mbuf *)i, (struct mbuf *)0, 164 (struct proc *)0); 165 /* XXX NOT MP SAFE */ 166 if ((ninp == (void *)&tcbtable.inpt_queue && 167 tcbtable.inpt_queue.cqh_last != inp) || 168 ninp->inp_queue.cqe_prev != inp) 169 goto tpgone; 170 } 171 } 172 tp->t_idle++; 173 if (tp->t_rtt) 174 tp->t_rtt++; 175tpgone: 176 ; 177 } 178#if NRND == 0 /* Do we need to do this when using random() ? */ 179 tcp_iss_seq += TCP_ISSINCR; /* increment iss */ 180 if (tcp_compat_42) 181 if ((int)tcp_iss_seq < 0) 182 tcp_iss_seq = 0; /* XXX */ 183#endif 184 tcp_now++; /* for timestamps */ 185 if (++syn_cache_last >= tcp_syn_cache_interval) { 186 syn_cache_timer(); 187 syn_cache_last = 0; 188 } 189 splx(s); 190} 191 192/* 193 * Cancel all timers for TCP tp. 194 */ 195void 196tcp_canceltimers(tp) 197 struct tcpcb *tp; 198{ 199 register int i; 200 201 for (i = 0; i < TCPT_NTIMERS; i++) 202 TCP_TIMER_DISARM(tp, i); 203} 204 205int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 206 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 207 208int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 209 210/* 211 * TCP timer processing. 212 */ 213struct tcpcb * 214tcp_timers(tp, timer) 215 register struct tcpcb *tp; 216 int timer; 217{ 218 short rto; 219 220 switch (timer) { 221 222 /* 223 * 2 MSL timeout in shutdown went off. If we're closed but 224 * still waiting for peer to close and connection has been idle 225 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 226 * control block. Otherwise, check again in a bit. 227 */ 228 case TCPT_2MSL: 229 if (tp->t_state != TCPS_TIME_WAIT && 230 tp->t_idle <= tcp_maxidle) 231 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 232 else 233 tp = tcp_close(tp); 234 break; 235 236 /* 237 * Retransmission timer went off. Message has not 238 * been acked within retransmit interval. Back off 239 * to a longer retransmit interval and retransmit one segment. 240 */ 241 case TCPT_REXMT: 242 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 243 tp->t_rxtshift = TCP_MAXRXTSHIFT; 244 tcpstat.tcps_timeoutdrop++; 245 tp = tcp_drop(tp, tp->t_softerror ? 246 tp->t_softerror : ETIMEDOUT); 247 break; 248 } 249 tcpstat.tcps_rexmttimeo++; 250 rto = TCP_REXMTVAL(tp); 251 if (rto < tp->t_rttmin) 252 rto = tp->t_rttmin; 253 TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift], 254 tp->t_rttmin, TCPTV_REXMTMAX); 255 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 256#if 0 257 /* 258 * If we are losing and we are trying path MTU discovery, 259 * try turning it off. This will avoid black holes in 260 * the network which suppress or fail to send "packet 261 * too big" ICMP messages. We should ideally do 262 * lots more sophisticated searching to find the right 263 * value here... 264 */ 265 if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 266 struct inpcb *inp = tp->t_inpcb; 267 struct rtentry *rt = in_pcbrtentry(inp); 268 269 /* XXX: Black hole recovery code goes here */ 270 } 271#endif 272 /* 273 * If losing, let the lower level know and try for 274 * a better route. Also, if we backed off this far, 275 * our srtt estimate is probably bogus. Clobber it 276 * so we'll take the next rtt measurement as our srtt; 277 * move the current srtt into rttvar to keep the current 278 * retransmit times until then. 279 */ 280 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 281 in_losing(tp->t_inpcb); 282 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 283 tp->t_srtt = 0; 284 } 285 tp->snd_nxt = tp->snd_una; 286 /* 287 * If timing a segment in this window, stop the timer. 288 */ 289 tp->t_rtt = 0; 290 /* 291 * Remember if we are retransmitting a SYN, because if 292 * we do, set the initial congestion window must be set 293 * to 1 segment. 294 */ 295 if (tp->t_state == TCPS_SYN_SENT) 296 tp->t_flags |= TF_SYN_REXMT; 297 /* 298 * Close the congestion window down to the initial window 299 * (we'll open it by one segment for each ack we get). 300 * Since we probably have a window's worth of unacked 301 * data accumulated, this "slow start" keeps us from 302 * dumping all that data as back-to-back packets (which 303 * might overwhelm an intermediate gateway). 304 * 305 * There are two phases to the opening: Initially we 306 * open by one mss on each ack. This makes the window 307 * size increase exponentially with time. If the 308 * window is larger than the path can handle, this 309 * exponential growth results in dropped packet(s) 310 * almost immediately. To get more time between 311 * drops but still "push" the network to take advantage 312 * of improving conditions, we switch from exponential 313 * to linear window opening at some threshhold size. 314 * For a threshhold, we use half the current window 315 * size, truncated to a multiple of the mss. 316 * 317 * (the minimum cwnd that will give us exponential 318 * growth is 2 mss. We don't allow the threshhold 319 * to go below this.) 320 */ 321 { 322 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz; 323 if (win < 2) 324 win = 2; 325 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_segsz); 326 tp->snd_ssthresh = win * tp->t_segsz; 327 tp->t_dupacks = 0; 328 } 329 (void) tcp_output(tp); 330 break; 331 332 /* 333 * Persistance timer into zero window. 334 * Force a byte to be output, if possible. 335 */ 336 case TCPT_PERSIST: 337 /* 338 * Hack: if the peer is dead/unreachable, we do not 339 * time out if the window is closed. After a full 340 * backoff, drop the connection if the idle time 341 * (no responses to probes) reaches the maximum 342 * backoff that we would use if retransmitting. 343 */ 344 rto = TCP_REXMTVAL(tp); 345 if (rto < tp->t_rttmin) 346 rto = tp->t_rttmin; 347 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 348 (tp->t_idle >= tcp_maxpersistidle || 349 tp->t_idle >= rto * tcp_totbackoff)) { 350 tcpstat.tcps_persistdrops++; 351 tp = tcp_drop(tp, ETIMEDOUT); 352 break; 353 } 354 tcpstat.tcps_persisttimeo++; 355 tcp_setpersist(tp); 356 tp->t_force = 1; 357 (void) tcp_output(tp); 358 tp->t_force = 0; 359 break; 360 361 /* 362 * Keep-alive timer went off; send something 363 * or drop connection if idle for too long. 364 */ 365 case TCPT_KEEP: 366 tcpstat.tcps_keeptimeo++; 367 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 368 goto dropit; 369 if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE && 370 tp->t_state <= TCPS_CLOSE_WAIT) { 371 if (tp->t_idle >= tcp_keepidle + tcp_maxidle) 372 goto dropit; 373 /* 374 * Send a packet designed to force a response 375 * if the peer is up and reachable: 376 * either an ACK if the connection is still alive, 377 * or an RST if the peer has closed the connection 378 * due to timeout or reboot. 379 * Using sequence number tp->snd_una-1 380 * causes the transmitted zero-length segment 381 * to lie outside the receive window; 382 * by the protocol spec, this requires the 383 * correspondent TCP to respond. 384 */ 385 tcpstat.tcps_keepprobe++; 386 if (tcp_compat_42) { 387 /* 388 * The keepalive packet must have nonzero 389 * length to get a 4.2 host to respond. 390 */ 391 (void)tcp_respond(tp, tp->t_template, 392 (struct mbuf *)NULL, tp->rcv_nxt - 1, 393 tp->snd_una - 1, 0); 394 } else { 395 (void)tcp_respond(tp, tp->t_template, 396 (struct mbuf *)NULL, tp->rcv_nxt, 397 tp->snd_una - 1, 0); 398 } 399 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 400 } else 401 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 402 break; 403 dropit: 404 tcpstat.tcps_keepdrops++; 405 tp = tcp_drop(tp, ETIMEDOUT); 406 break; 407 } 408 return (tp); 409} 410