1 /* $NetBSD: tcp_timer.c,v 1.62 2003/02/03 23:51:04 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, 2001 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 <sys/cdefs.h> 105 __KERNEL_RCSID(0, "$NetBSD: tcp_timer.c,v 1.62 2003/02/03 23:51:04 thorpej Exp $"); 106 107 #include "opt_inet.h" 108 #include "opt_tcp_debug.h" 109 110 #include <sys/param.h> 111 #include <sys/systm.h> 112 #include <sys/malloc.h> 113 #include <sys/mbuf.h> 114 #include <sys/socket.h> 115 #include <sys/socketvar.h> 116 #include <sys/protosw.h> 117 #include <sys/errno.h> 118 #include <sys/kernel.h> 119 120 #include <net/if.h> 121 #include <net/route.h> 122 123 #include <netinet/in.h> 124 #include <netinet/in_systm.h> 125 #include <netinet/ip.h> 126 #include <netinet/in_pcb.h> 127 #include <netinet/ip_var.h> 128 129 #ifdef INET6 130 #ifndef INET 131 #include <netinet/in.h> 132 #endif 133 #include <netinet/ip6.h> 134 #include <netinet6/in6_pcb.h> 135 #endif 136 137 #include <netinet/tcp.h> 138 #include <netinet/tcp_fsm.h> 139 #include <netinet/tcp_seq.h> 140 #include <netinet/tcp_timer.h> 141 #include <netinet/tcp_var.h> 142 #include <netinet/tcpip.h> 143 #ifdef TCP_DEBUG 144 #include <netinet/tcp_debug.h> 145 #endif 146 147 /* 148 * Various tunable timer parameters. These are initialized in tcp_init(), 149 * unless they are patched. 150 */ 151 int tcp_keepidle = 0; 152 int tcp_keepintvl = 0; 153 int tcp_keepcnt = 0; /* max idle probes */ 154 int tcp_maxpersistidle = 0; /* max idle time in persist */ 155 int tcp_maxidle; /* computed in tcp_slowtimo() */ 156 157 /* 158 * Time to delay the ACK. This is initialized in tcp_init(), unless 159 * its patched. 160 */ 161 int tcp_delack_ticks = 0; 162 163 void tcp_timer_rexmt(void *); 164 void tcp_timer_persist(void *); 165 void tcp_timer_keep(void *); 166 void tcp_timer_2msl(void *); 167 168 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = { 169 tcp_timer_rexmt, 170 tcp_timer_persist, 171 tcp_timer_keep, 172 tcp_timer_2msl, 173 }; 174 175 /* 176 * Timer state initialization, called from tcp_init(). 177 */ 178 void 179 tcp_timer_init(void) 180 { 181 182 if (tcp_keepidle == 0) 183 tcp_keepidle = TCPTV_KEEP_IDLE; 184 185 if (tcp_keepintvl == 0) 186 tcp_keepintvl = TCPTV_KEEPINTVL; 187 188 if (tcp_keepcnt == 0) 189 tcp_keepcnt = TCPTV_KEEPCNT; 190 191 if (tcp_maxpersistidle == 0) 192 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 193 194 if (tcp_delack_ticks == 0) 195 tcp_delack_ticks = TCP_DELACK_TICKS; 196 } 197 198 /* 199 * Callout to process delayed ACKs for a TCPCB. 200 */ 201 void 202 tcp_delack(void *arg) 203 { 204 struct tcpcb *tp = arg; 205 int s; 206 207 /* 208 * If tcp_output() wasn't able to transmit the ACK 209 * for whatever reason, it will restart the delayed 210 * ACK callout. 211 */ 212 213 s = splsoftnet(); 214 tp->t_flags |= TF_ACKNOW; 215 (void) tcp_output(tp); 216 splx(s); 217 } 218 219 /* 220 * Tcp protocol timeout routine called every 500 ms. 221 * Updates the timers in all active tcb's and 222 * causes finite state machine actions if timers expire. 223 */ 224 void 225 tcp_slowtimo() 226 { 227 int s; 228 229 s = splsoftnet(); 230 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 231 tcp_iss_seq += TCP_ISSINCR; /* increment iss */ 232 tcp_now++; /* for timestamps */ 233 splx(s); 234 } 235 236 /* 237 * Cancel all timers for TCP tp. 238 */ 239 void 240 tcp_canceltimers(tp) 241 struct tcpcb *tp; 242 { 243 int i; 244 245 for (i = 0; i < TCPT_NTIMERS; i++) 246 TCP_TIMER_DISARM(tp, i); 247 } 248 249 const int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 250 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 251 252 const int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 253 254 /* 255 * TCP timer processing. 256 */ 257 258 void 259 tcp_timer_rexmt(void *arg) 260 { 261 struct tcpcb *tp = arg; 262 uint32_t rto; 263 int s; 264 #ifdef TCP_DEBUG 265 struct socket *so; 266 short ostate; 267 #endif 268 269 s = splsoftnet(); 270 271 #ifdef TCP_DEBUG 272 #ifdef INET 273 if (tp->t_inpcb) 274 so = tp->t_inpcb->inp_socket; 275 #endif 276 #ifdef INET6 277 if (tp->t_in6pcb) 278 so = tp->t_in6pcb->in6p_socket; 279 #endif 280 ostate = tp->t_state; 281 #endif /* TCP_DEBUG */ 282 283 /* 284 * Retransmission timer went off. Message has not 285 * been acked within retransmit interval. Back off 286 * to a longer retransmit interval and retransmit one segment. 287 */ 288 289 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 290 tp->t_rxtshift = TCP_MAXRXTSHIFT; 291 tcpstat.tcps_timeoutdrop++; 292 tp = tcp_drop(tp, tp->t_softerror ? 293 tp->t_softerror : ETIMEDOUT); 294 goto out; 295 } 296 tcpstat.tcps_rexmttimeo++; 297 rto = TCP_REXMTVAL(tp); 298 if (rto < tp->t_rttmin) 299 rto = tp->t_rttmin; 300 TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift], 301 tp->t_rttmin, TCPTV_REXMTMAX); 302 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 303 304 /* 305 * If we are losing and we are trying path MTU discovery, 306 * try turning it off. This will avoid black holes in 307 * the network which suppress or fail to send "packet 308 * too big" ICMP messages. We should ideally do 309 * lots more sophisticated searching to find the right 310 * value here... 311 */ 312 if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 313 tcpstat.tcps_pmtublackhole++; 314 315 #ifdef INET 316 /* try turning PMTUD off */ 317 if (tp->t_inpcb) 318 tp->t_mtudisc = 0; 319 #endif 320 #ifdef INET6 321 /* try using IPv6 minimum MTU */ 322 if (tp->t_in6pcb) 323 tp->t_mtudisc = 0; 324 #endif 325 326 /* XXX: more sophisticated Black hole recovery code? */ 327 } 328 329 /* 330 * If losing, let the lower level know and try for 331 * a better route. Also, if we backed off this far, 332 * our srtt estimate is probably bogus. Clobber it 333 * so we'll take the next rtt measurement as our srtt; 334 * move the current srtt into rttvar to keep the current 335 * retransmit times until then. 336 */ 337 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 338 #ifdef INET 339 if (tp->t_inpcb) 340 in_losing(tp->t_inpcb); 341 #endif 342 #ifdef INET6 343 if (tp->t_in6pcb) 344 in6_losing(tp->t_in6pcb); 345 #endif 346 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 347 tp->t_srtt = 0; 348 } 349 tp->snd_nxt = tp->snd_una; 350 /* 351 * If timing a segment in this window, stop the timer. 352 */ 353 tp->t_rtttime = 0; 354 /* 355 * Remember if we are retransmitting a SYN, because if 356 * we do, set the initial congestion window must be set 357 * to 1 segment. 358 */ 359 if (tp->t_state == TCPS_SYN_SENT) 360 tp->t_flags |= TF_SYN_REXMT; 361 /* 362 * Close the congestion window down to one segment 363 * (we'll open it by one segment for each ack we get). 364 * Since we probably have a window's worth of unacked 365 * data accumulated, this "slow start" keeps us from 366 * dumping all that data as back-to-back packets (which 367 * might overwhelm an intermediate gateway). 368 * 369 * There are two phases to the opening: Initially we 370 * open by one mss on each ack. This makes the window 371 * size increase exponentially with time. If the 372 * window is larger than the path can handle, this 373 * exponential growth results in dropped packet(s) 374 * almost immediately. To get more time between 375 * drops but still "push" the network to take advantage 376 * of improving conditions, we switch from exponential 377 * to linear window opening at some threshhold size. 378 * For a threshhold, we use half the current window 379 * size, truncated to a multiple of the mss. 380 * 381 * (the minimum cwnd that will give us exponential 382 * growth is 2 mss. We don't allow the threshhold 383 * to go below this.) 384 */ 385 { 386 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz; 387 if (win < 2) 388 win = 2; 389 /* Loss Window MUST be one segment. */ 390 tp->snd_cwnd = tp->t_segsz; 391 tp->snd_ssthresh = win * tp->t_segsz; 392 tp->t_dupacks = 0; 393 } 394 (void) tcp_output(tp); 395 396 out: 397 #ifdef TCP_DEBUG 398 if (tp && so->so_options & SO_DEBUG) 399 tcp_trace(TA_USER, ostate, tp, NULL, 400 PRU_SLOWTIMO | (TCPT_REXMT << 8)); 401 #endif 402 splx(s); 403 } 404 405 void 406 tcp_timer_persist(void *arg) 407 { 408 struct tcpcb *tp = arg; 409 uint32_t rto; 410 int s; 411 #ifdef TCP_DEBUG 412 struct socket *so; 413 short ostate; 414 #endif 415 416 s = splsoftnet(); 417 418 #ifdef TCP_DEBUG 419 #ifdef INET 420 if (tp->t_inpcb) 421 so = tp->t_inpcb->inp_socket; 422 #endif 423 #ifdef INET6 424 if (tp->t_in6pcb) 425 so = tp->t_in6pcb->in6p_socket; 426 #endif 427 428 ostate = tp->t_state; 429 #endif /* TCP_DEBUG */ 430 431 /* 432 * Persistance timer into zero window. 433 * Force a byte to be output, if possible. 434 */ 435 436 /* 437 * Hack: if the peer is dead/unreachable, we do not 438 * time out if the window is closed. After a full 439 * backoff, drop the connection if the idle time 440 * (no responses to probes) reaches the maximum 441 * backoff that we would use if retransmitting. 442 */ 443 rto = TCP_REXMTVAL(tp); 444 if (rto < tp->t_rttmin) 445 rto = tp->t_rttmin; 446 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 447 ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle || 448 (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) { 449 tcpstat.tcps_persistdrops++; 450 tp = tcp_drop(tp, ETIMEDOUT); 451 goto out; 452 } 453 tcpstat.tcps_persisttimeo++; 454 tcp_setpersist(tp); 455 tp->t_force = 1; 456 (void) tcp_output(tp); 457 tp->t_force = 0; 458 459 out: 460 #ifdef TCP_DEBUG 461 if (tp && so->so_options & SO_DEBUG) 462 tcp_trace(TA_USER, ostate, tp, NULL, 463 PRU_SLOWTIMO | (TCPT_PERSIST << 8)); 464 #endif 465 splx(s); 466 } 467 468 void 469 tcp_timer_keep(void *arg) 470 { 471 struct tcpcb *tp = arg; 472 struct socket *so = NULL; /* Quell compiler warning */ 473 int s; 474 #ifdef TCP_DEBUG 475 short ostate; 476 #endif 477 478 s = splsoftnet(); 479 480 #ifdef TCP_DEBUG 481 ostate = tp->t_state; 482 #endif /* TCP_DEBUG */ 483 484 /* 485 * Keep-alive timer went off; send something 486 * or drop connection if idle for too long. 487 */ 488 489 tcpstat.tcps_keeptimeo++; 490 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 491 goto dropit; 492 #ifdef INET 493 if (tp->t_inpcb) 494 so = tp->t_inpcb->inp_socket; 495 #endif 496 #ifdef INET6 497 if (tp->t_in6pcb) 498 so = tp->t_in6pcb->in6p_socket; 499 #endif 500 if (so->so_options & SO_KEEPALIVE && 501 tp->t_state <= TCPS_CLOSE_WAIT) { 502 if ((tcp_maxidle > 0) && 503 ((tcp_now - tp->t_rcvtime) >= 504 tcp_keepidle + tcp_maxidle)) 505 goto dropit; 506 /* 507 * Send a packet designed to force a response 508 * if the peer is up and reachable: 509 * either an ACK if the connection is still alive, 510 * or an RST if the peer has closed the connection 511 * due to timeout or reboot. 512 * Using sequence number tp->snd_una-1 513 * causes the transmitted zero-length segment 514 * to lie outside the receive window; 515 * by the protocol spec, this requires the 516 * correspondent TCP to respond. 517 */ 518 tcpstat.tcps_keepprobe++; 519 if (tcp_compat_42) { 520 /* 521 * The keepalive packet must have nonzero 522 * length to get a 4.2 host to respond. 523 */ 524 (void)tcp_respond(tp, tp->t_template, 525 (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1, 526 tp->snd_una - 1, 0); 527 } else { 528 (void)tcp_respond(tp, tp->t_template, 529 (struct mbuf *)NULL, NULL, tp->rcv_nxt, 530 tp->snd_una - 1, 0); 531 } 532 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 533 } else 534 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 535 536 #ifdef TCP_DEBUG 537 if (tp && so->so_options & SO_DEBUG) 538 tcp_trace(TA_USER, ostate, tp, NULL, 539 PRU_SLOWTIMO | (TCPT_KEEP << 8)); 540 #endif 541 splx(s); 542 return; 543 544 dropit: 545 tcpstat.tcps_keepdrops++; 546 (void) tcp_drop(tp, ETIMEDOUT); 547 splx(s); 548 } 549 550 void 551 tcp_timer_2msl(void *arg) 552 { 553 struct tcpcb *tp = arg; 554 int s; 555 #ifdef TCP_DEBUG 556 struct socket *so; 557 short ostate; 558 #endif 559 560 s = splsoftnet(); 561 562 #ifdef TCP_DEBUG 563 #ifdef INET 564 if (tp->t_inpcb) 565 so = tp->t_inpcb->inp_socket; 566 #endif 567 #ifdef INET6 568 if (tp->t_in6pcb) 569 so = tp->t_in6pcb->in6p_socket; 570 #endif 571 572 ostate = tp->t_state; 573 #endif /* TCP_DEBUG */ 574 575 /* 576 * 2 MSL timeout in shutdown went off. If we're closed but 577 * still waiting for peer to close and connection has been idle 578 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 579 * control block. Otherwise, check again in a bit. 580 */ 581 if (tp->t_state != TCPS_TIME_WAIT && 582 ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle))) 583 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 584 else 585 tp = tcp_close(tp); 586 587 #ifdef TCP_DEBUG 588 if (tp && so->so_options & SO_DEBUG) 589 tcp_trace(TA_USER, ostate, tp, NULL, 590 PRU_SLOWTIMO | (TCPT_2MSL << 8)); 591 #endif 592 splx(s); 593 } 594