tcp_timer.c revision 1.49
1 /* $NetBSD: tcp_timer.c,v 1.49 2001/09/10 04:24:25 thorpej Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc. 34 * All rights reserved. 35 * 36 * This code is derived from software contributed to The NetBSD Foundation 37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 38 * Facility, NASA Ames Research Center. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed by the NetBSD 51 * Foundation, Inc. and its contributors. 52 * 4. Neither the name of The NetBSD Foundation nor the names of its 53 * contributors may be used to endorse or promote products derived 54 * from this software without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 57 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 /* 70 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 71 * The Regents of the University of California. All rights reserved. 72 * 73 * Redistribution and use in source and binary forms, with or without 74 * modification, are permitted provided that the following conditions 75 * are met: 76 * 1. Redistributions of source code must retain the above copyright 77 * notice, this list of conditions and the following disclaimer. 78 * 2. Redistributions in binary form must reproduce the above copyright 79 * notice, this list of conditions and the following disclaimer in the 80 * documentation and/or other materials provided with the distribution. 81 * 3. All advertising materials mentioning features or use of this software 82 * must display the following acknowledgement: 83 * This product includes software developed by the University of 84 * California, Berkeley and its contributors. 85 * 4. Neither the name of the University nor the names of its contributors 86 * may be used to endorse or promote products derived from this software 87 * without specific prior written permission. 88 * 89 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 90 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 91 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 92 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 93 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 94 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 95 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 96 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 97 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 98 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 99 * SUCH DAMAGE. 100 * 101 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 102 */ 103 104 #include "opt_inet.h" 105 106 #include <sys/param.h> 107 #include <sys/systm.h> 108 #include <sys/malloc.h> 109 #include <sys/mbuf.h> 110 #include <sys/socket.h> 111 #include <sys/socketvar.h> 112 #include <sys/protosw.h> 113 #include <sys/errno.h> 114 #include <sys/kernel.h> 115 116 #include <net/if.h> 117 #include <net/route.h> 118 119 #include <netinet/in.h> 120 #include <netinet/in_systm.h> 121 #include <netinet/ip.h> 122 #include <netinet/in_pcb.h> 123 #include <netinet/ip_var.h> 124 125 #ifdef INET6 126 #ifndef INET 127 #include <netinet/in.h> 128 #endif 129 #include <netinet/ip6.h> 130 #include <netinet6/in6_pcb.h> 131 #endif 132 133 #include <netinet/tcp.h> 134 #include <netinet/tcp_fsm.h> 135 #include <netinet/tcp_seq.h> 136 #include <netinet/tcp_timer.h> 137 #include <netinet/tcp_var.h> 138 #include <netinet/tcpip.h> 139 140 int tcp_keepidle = TCPTV_KEEP_IDLE; 141 int tcp_keepintvl = TCPTV_KEEPINTVL; 142 int tcp_keepcnt = TCPTV_KEEPCNT; /* max idle probes */ 143 int tcp_maxpersistidle = TCPTV_KEEP_IDLE; /* max idle time in persist */ 144 int tcp_maxidle; 145 146 /* 147 * Time to delay the ACK. This is initialized in tcp_init(), unless 148 * its patched. 149 */ 150 int tcp_delack_ticks = 0; 151 152 /* 153 * Callout to process delayed ACKs for a TCPCB. 154 */ 155 void 156 tcp_delack(void *arg) 157 { 158 struct tcpcb *tp = arg; 159 int s; 160 161 /* 162 * If tcp_output() wasn't able to transmit the ACK 163 * for whatever reason, it will restart the delayed 164 * ACK callout. 165 */ 166 167 s = splsoftnet(); 168 tp->t_flags |= TF_ACKNOW; 169 (void) tcp_output(tp); 170 splx(s); 171 } 172 173 /* 174 * Tcp protocol timeout routine called every 500 ms. 175 * Updates the timers in all active tcb's and 176 * causes finite state machine actions if timers expire. 177 */ 178 void 179 tcp_slowtimo() 180 { 181 struct inpcb *inp, *ninp; 182 struct tcpcb *tp; 183 #ifdef INET6 184 struct in6pcb *in6p, *nin6p; 185 #endif 186 int s; 187 long i; 188 static int syn_cache_last = 0; 189 int skip, mask; 190 191 skip = mask = 0; 192 193 s = splsoftnet(); 194 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 195 /* 196 * Search through tcb's and update active timers. 197 */ 198 mask |= 1; 199 inp = tcbtable.inpt_queue.cqh_first; 200 if (inp == (struct inpcb *)0) { /* XXX */ 201 skip |= 1; 202 goto dotcb6; 203 } 204 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) { 205 ninp = inp->inp_queue.cqe_next; 206 tp = intotcpcb(inp); 207 if (tp == 0 || tp->t_state == TCPS_LISTEN) 208 continue; 209 for (i = 0; i < TCPT_NTIMERS; i++) { 210 if (TCP_TIMER_ISEXPIRED(tp, i)) { 211 TCP_TIMER_DISARM(tp, i); 212 (void) tcp_usrreq(tp->t_inpcb->inp_socket, 213 PRU_SLOWTIMO, (struct mbuf *)0, 214 (struct mbuf *)i, (struct mbuf *)0, 215 (struct proc *)0); 216 /* XXX NOT MP SAFE */ 217 if ((ninp == (void *)&tcbtable.inpt_queue && 218 tcbtable.inpt_queue.cqh_last != inp) || 219 ninp->inp_queue.cqe_prev != inp) 220 goto tpgone; 221 } 222 } 223 tp->t_idle++; 224 if (tp->t_rtt) 225 tp->t_rtt++; 226 tpgone: 227 ; 228 } 229 dotcb6: 230 #ifdef INET6 231 mask |= 2; 232 in6p = tcb6.in6p_next; 233 if (in6p == (struct in6pcb *)0) { /* XXX */ 234 skip |= 2; 235 goto doiss; 236 } 237 for (; in6p != (struct in6pcb *)&tcb6; in6p = nin6p) { 238 nin6p = in6p->in6p_next; 239 tp = in6totcpcb(in6p); 240 if (tp == 0 || tp->t_state == TCPS_LISTEN) 241 continue; 242 for (i = 0; i < TCPT_NTIMERS; i++) { 243 if (TCP_TIMER_ISEXPIRED(tp, i)) { 244 TCP_TIMER_DISARM(tp, i); 245 (void) tcp_usrreq(tp->t_in6pcb->in6p_socket, 246 PRU_SLOWTIMO, (struct mbuf *)0, 247 (struct mbuf *)i, (struct mbuf *)0, 248 (struct proc *)0); 249 /* XXX NOT MP SAFE */ 250 if ((nin6p == (void *)&tcb6 && 251 tcb6.in6p_prev != in6p) || 252 nin6p->in6p_prev != in6p) 253 goto tp6gone; 254 } 255 } 256 tp->t_idle++; 257 if (tp->t_rtt) 258 tp->t_rtt++; 259 tp6gone: 260 ; 261 } 262 263 doiss: 264 #endif 265 if (mask == skip) 266 goto done; 267 tcp_iss_seq += TCP_ISSINCR; /* increment iss */ 268 tcp_now++; /* for timestamps */ 269 if (++syn_cache_last >= tcp_syn_cache_interval) { 270 syn_cache_timer(); 271 syn_cache_last = 0; 272 } 273 done: 274 splx(s); 275 } 276 277 /* 278 * Cancel all timers for TCP tp. 279 */ 280 void 281 tcp_canceltimers(tp) 282 struct tcpcb *tp; 283 { 284 int i; 285 286 for (i = 0; i < TCPT_NTIMERS; i++) 287 TCP_TIMER_DISARM(tp, i); 288 } 289 290 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 291 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 292 293 int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 294 295 /* 296 * TCP timer processing. 297 */ 298 struct tcpcb * 299 tcp_timers(tp, timer) 300 struct tcpcb *tp; 301 int timer; 302 { 303 short rto; 304 305 #ifdef DIAGNOSTIC 306 if (tp->t_inpcb && tp->t_in6pcb) 307 panic("tcp_timers: both t_inpcb and t_in6pcb are set"); 308 #endif 309 310 switch (timer) { 311 312 /* 313 * 2 MSL timeout in shutdown went off. If we're closed but 314 * still waiting for peer to close and connection has been idle 315 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 316 * control block. Otherwise, check again in a bit. 317 */ 318 case TCPT_2MSL: 319 if (tp->t_state != TCPS_TIME_WAIT && 320 ((tcp_maxidle == 0) || (tp->t_idle <= tcp_maxidle))) 321 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 322 else 323 tp = tcp_close(tp); 324 break; 325 326 /* 327 * Retransmission timer went off. Message has not 328 * been acked within retransmit interval. Back off 329 * to a longer retransmit interval and retransmit one segment. 330 */ 331 case TCPT_REXMT: 332 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 333 tp->t_rxtshift = TCP_MAXRXTSHIFT; 334 tcpstat.tcps_timeoutdrop++; 335 tp = tcp_drop(tp, tp->t_softerror ? 336 tp->t_softerror : ETIMEDOUT); 337 break; 338 } 339 tcpstat.tcps_rexmttimeo++; 340 rto = TCP_REXMTVAL(tp); 341 if (rto < tp->t_rttmin) 342 rto = tp->t_rttmin; 343 TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift], 344 tp->t_rttmin, TCPTV_REXMTMAX); 345 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 346 #if 0 347 /* 348 * If we are losing and we are trying path MTU discovery, 349 * try turning it off. This will avoid black holes in 350 * the network which suppress or fail to send "packet 351 * too big" ICMP messages. We should ideally do 352 * lots more sophisticated searching to find the right 353 * value here... 354 */ 355 if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 356 struct rtentry *rt = NULL; 357 358 #ifdef INET 359 if (tp->t_inpcb) 360 rt = in_pcbrtentry(tp->t_inpcb); 361 #endif 362 #ifdef INET6 363 if (tp->t_in6pcb) 364 rt = in6_pcbrtentry(tp->t_in6pcb); 365 #endif 366 367 /* XXX: Black hole recovery code goes here */ 368 } 369 #endif 370 /* 371 * If losing, let the lower level know and try for 372 * a better route. Also, if we backed off this far, 373 * our srtt estimate is probably bogus. Clobber it 374 * so we'll take the next rtt measurement as our srtt; 375 * move the current srtt into rttvar to keep the current 376 * retransmit times until then. 377 */ 378 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 379 #ifdef INET 380 if (tp->t_inpcb) 381 in_losing(tp->t_inpcb); 382 #endif 383 #ifdef INET6 384 if (tp->t_in6pcb) 385 in6_losing(tp->t_in6pcb); 386 #endif 387 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 388 tp->t_srtt = 0; 389 } 390 tp->snd_nxt = tp->snd_una; 391 /* 392 * If timing a segment in this window, stop the timer. 393 */ 394 tp->t_rtt = 0; 395 /* 396 * Remember if we are retransmitting a SYN, because if 397 * we do, set the initial congestion window must be set 398 * to 1 segment. 399 */ 400 if (tp->t_state == TCPS_SYN_SENT) 401 tp->t_flags |= TF_SYN_REXMT; 402 /* 403 * Close the congestion window down to one segment 404 * (we'll open it by one segment for each ack we get). 405 * Since we probably have a window's worth of unacked 406 * data accumulated, this "slow start" keeps us from 407 * dumping all that data as back-to-back packets (which 408 * might overwhelm an intermediate gateway). 409 * 410 * There are two phases to the opening: Initially we 411 * open by one mss on each ack. This makes the window 412 * size increase exponentially with time. If the 413 * window is larger than the path can handle, this 414 * exponential growth results in dropped packet(s) 415 * almost immediately. To get more time between 416 * drops but still "push" the network to take advantage 417 * of improving conditions, we switch from exponential 418 * to linear window opening at some threshhold size. 419 * For a threshhold, we use half the current window 420 * size, truncated to a multiple of the mss. 421 * 422 * (the minimum cwnd that will give us exponential 423 * growth is 2 mss. We don't allow the threshhold 424 * to go below this.) 425 */ 426 { 427 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz; 428 if (win < 2) 429 win = 2; 430 /* Loss Window MUST be one segment. */ 431 tp->snd_cwnd = tp->t_segsz; 432 tp->snd_ssthresh = win * tp->t_segsz; 433 tp->t_dupacks = 0; 434 } 435 (void) tcp_output(tp); 436 break; 437 438 /* 439 * Persistance timer into zero window. 440 * Force a byte to be output, if possible. 441 */ 442 case TCPT_PERSIST: 443 /* 444 * Hack: if the peer is dead/unreachable, we do not 445 * time out if the window is closed. After a full 446 * backoff, drop the connection if the idle time 447 * (no responses to probes) reaches the maximum 448 * backoff that we would use if retransmitting. 449 */ 450 rto = TCP_REXMTVAL(tp); 451 if (rto < tp->t_rttmin) 452 rto = tp->t_rttmin; 453 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 454 (tp->t_idle >= tcp_maxpersistidle || 455 tp->t_idle >= rto * tcp_totbackoff)) { 456 tcpstat.tcps_persistdrops++; 457 tp = tcp_drop(tp, ETIMEDOUT); 458 break; 459 } 460 tcpstat.tcps_persisttimeo++; 461 tcp_setpersist(tp); 462 tp->t_force = 1; 463 (void) tcp_output(tp); 464 tp->t_force = 0; 465 break; 466 467 /* 468 * Keep-alive timer went off; send something 469 * or drop connection if idle for too long. 470 */ 471 case TCPT_KEEP: 472 { 473 struct socket *so = NULL; 474 475 tcpstat.tcps_keeptimeo++; 476 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 477 goto dropit; 478 #ifdef INET 479 if (tp->t_inpcb) 480 so = tp->t_inpcb->inp_socket; 481 #endif 482 #ifdef INET6 483 if (tp->t_in6pcb) 484 so = tp->t_in6pcb->in6p_socket; 485 #endif 486 if (so->so_options & SO_KEEPALIVE && 487 tp->t_state <= TCPS_CLOSE_WAIT) { 488 if ((tcp_maxidle > 0) && 489 (tp->t_idle >= tcp_keepidle + tcp_maxidle)) 490 goto dropit; 491 /* 492 * Send a packet designed to force a response 493 * if the peer is up and reachable: 494 * either an ACK if the connection is still alive, 495 * or an RST if the peer has closed the connection 496 * due to timeout or reboot. 497 * Using sequence number tp->snd_una-1 498 * causes the transmitted zero-length segment 499 * to lie outside the receive window; 500 * by the protocol spec, this requires the 501 * correspondent TCP to respond. 502 */ 503 tcpstat.tcps_keepprobe++; 504 if (tcp_compat_42) { 505 /* 506 * The keepalive packet must have nonzero 507 * length to get a 4.2 host to respond. 508 */ 509 (void)tcp_respond(tp, tp->t_template, 510 (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1, 511 tp->snd_una - 1, 0); 512 } else { 513 (void)tcp_respond(tp, tp->t_template, 514 (struct mbuf *)NULL, NULL, tp->rcv_nxt, 515 tp->snd_una - 1, 0); 516 } 517 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 518 } else 519 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 520 break; 521 } 522 dropit: 523 tcpstat.tcps_keepdrops++; 524 tp = tcp_drop(tp, ETIMEDOUT); 525 break; 526 } 527 return (tp); 528 } 529
Indexes created Mon Sep 22 13:09:51 GMT 2025