1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_inet6.h" 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/mbuf.h> 44 #include <sys/mutex.h> 45 #include <sys/protosw.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #include <sys/sysctl.h> 49 #include <sys/systm.h> 50 51 #include <net/route.h> 52 53 #include <netinet/in.h> 54 #include <netinet/in_pcb.h> 55 #include <netinet/in_systm.h> 56 #ifdef INET6 57 #include <netinet6/in6_pcb.h> 58 #endif 59 #include <netinet/ip_var.h> 60 #include <netinet/tcp.h> 61 #include <netinet/tcp_fsm.h> 62 #include <netinet/tcp_timer.h> 63 #include <netinet/tcp_var.h> 64 #include <netinet/tcpip.h> 65 #ifdef TCPDEBUG 66 #include <netinet/tcp_debug.h> 67 #endif 68 69 static int 70 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS) 71 { 72 int error, s, tt; 73 74 tt = *(int *)oidp->oid_arg1; 75 s = (int)((int64_t)tt * 1000 / hz); 76 77 error = sysctl_handle_int(oidp, &s, 0, req); 78 if (error || !req->newptr) 79 return (error); 80 81 tt = (int)((int64_t)s * hz / 1000); 82 if (tt < 1) 83 return (EINVAL); 84 85 *(int *)oidp->oid_arg1 = tt; 86 return (0); 87 } 88 89 int tcp_keepinit; 90 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW, 91 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", ""); 92 93 int tcp_keepidle; 94 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW, 95 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", ""); 96 97 int tcp_keepintvl; 98 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW, 99 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", ""); 100 101 int tcp_delacktime; 102 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, 103 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I", 104 "Time before a delayed ACK is sent"); 105 106 int tcp_msl; 107 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW, 108 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime"); 109 110 int tcp_rexmit_min; 111 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW, 112 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout"); 113 114 int tcp_rexmit_slop; 115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW, 116 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop"); 117 118 static int always_keepalive = 1; 119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, 120 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections"); 121 122 static int tcp_keepcnt = TCPTV_KEEPCNT; 123 /* max idle probes */ 124 int tcp_maxpersistidle; 125 /* max idle time in persist */ 126 int tcp_maxidle; 127 128 /* 129 * Tcp protocol timeout routine called every 500 ms. 130 * Updates timestamps used for TCP 131 * causes finite state machine actions if timers expire. 132 */ 133 void 134 tcp_slowtimo() 135 { 136 int s; 137 138 s = splnet(); 139 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 140 splx(s); 141 INP_INFO_WLOCK(&tcbinfo); 142 (void) tcp_timer_2msl_tw(0); 143 INP_INFO_WUNLOCK(&tcbinfo); 144 } 145 146 /* 147 * Cancel all timers for TCP tp. 148 */ 149 void 150 tcp_canceltimers(tp) 151 struct tcpcb *tp; 152 { 153 callout_stop(tp->tt_2msl); 154 callout_stop(tp->tt_persist); 155 callout_stop(tp->tt_keep); 156 callout_stop(tp->tt_rexmt); 157 } 158 159 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] = 160 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 }; 161 162 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 163 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 164 165 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 166 167 /* 168 * TCP timer processing. 169 */ 170 171 void 172 tcp_timer_delack(xtp) 173 void *xtp; 174 { 175 struct tcpcb *tp = xtp; 176 int s; 177 struct inpcb *inp; 178 179 s = splnet(); 180 INP_INFO_RLOCK(&tcbinfo); 181 inp = tp->t_inpcb; 182 if (!inp) { 183 INP_INFO_RUNLOCK(&tcbinfo); 184 splx(s); 185 return; 186 } 187 INP_LOCK(inp); 188 INP_INFO_RUNLOCK(&tcbinfo); 189 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) { 190 INP_UNLOCK(inp); 191 splx(s); 192 return; 193 } 194 callout_deactivate(tp->tt_delack); 195 196 tp->t_flags |= TF_ACKNOW; 197 tcpstat.tcps_delack++; 198 (void) tcp_output(tp); 199 INP_UNLOCK(inp); 200 splx(s); 201 } 202 203 void 204 tcp_timer_2msl(xtp) 205 void *xtp; 206 { 207 struct tcpcb *tp = xtp; 208 int s; 209 struct inpcb *inp; 210 #ifdef TCPDEBUG 211 int ostate; 212 213 ostate = tp->t_state; 214 #endif 215 s = splnet(); 216 INP_INFO_WLOCK(&tcbinfo); 217 inp = tp->t_inpcb; 218 if (!inp) { 219 INP_INFO_WUNLOCK(&tcbinfo); 220 splx(s); 221 return; 222 } 223 INP_LOCK(inp); 224 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) { 225 INP_UNLOCK(tp->t_inpcb); 226 INP_INFO_WUNLOCK(&tcbinfo); 227 splx(s); 228 return; 229 } 230 callout_deactivate(tp->tt_2msl); 231 /* 232 * 2 MSL timeout in shutdown went off. If we're closed but 233 * still waiting for peer to close and connection has been idle 234 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 235 * control block. Otherwise, check again in a bit. 236 */ 237 if (tp->t_state != TCPS_TIME_WAIT && 238 (ticks - tp->t_rcvtime) <= tcp_maxidle) 239 callout_reset(tp->tt_2msl, tcp_keepintvl, 240 tcp_timer_2msl, tp); 241 else 242 tp = tcp_close(tp); 243 244 #ifdef TCPDEBUG 245 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 246 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 247 PRU_SLOWTIMO); 248 #endif 249 if (tp) 250 INP_UNLOCK(inp); 251 INP_INFO_WUNLOCK(&tcbinfo); 252 splx(s); 253 } 254 255 struct twlist { 256 LIST_HEAD(, tcptw) tw_list; 257 struct tcptw tw_tail; 258 }; 259 #define TWLIST_NLISTS 2 260 static struct twlist twl_2msl[TWLIST_NLISTS]; 261 static struct twlist *tw_2msl_list[] = { &twl_2msl[0], &twl_2msl[1], NULL }; 262 263 void 264 tcp_timer_init(void) 265 { 266 int i; 267 struct twlist *twl; 268 269 for (i = 0; i < TWLIST_NLISTS; i++) { 270 twl = &twl_2msl[i]; 271 LIST_INIT(&twl->tw_list); 272 LIST_INSERT_HEAD(&twl->tw_list, &twl->tw_tail, tw_2msl); 273 } 274 } 275 276 void 277 tcp_timer_2msl_reset(struct tcptw *tw, int timeo) 278 { 279 int i; 280 struct tcptw *tw_tail; 281 282 if (tw->tw_time != 0) 283 LIST_REMOVE(tw, tw_2msl); 284 tw->tw_time = timeo + ticks; 285 i = timeo > tcp_msl ? 1 : 0; 286 tw_tail = &twl_2msl[i].tw_tail; 287 LIST_INSERT_BEFORE(tw_tail, tw, tw_2msl); 288 } 289 290 void 291 tcp_timer_2msl_stop(struct tcptw *tw) 292 { 293 294 if (tw->tw_time != 0) 295 LIST_REMOVE(tw, tw_2msl); 296 } 297 298 struct tcptw * 299 tcp_timer_2msl_tw(int reuse) 300 { 301 struct tcptw *tw, *tw_tail; 302 struct twlist *twl; 303 int i; 304 305 for (i = 0; i < 2; i++) { 306 twl = tw_2msl_list[i]; 307 tw_tail = &twl->tw_tail; 308 for (;;) { 309 tw = LIST_FIRST(&twl->tw_list); 310 if (tw == tw_tail || (!reuse && tw->tw_time > ticks)) 311 break; 312 INP_LOCK(tw->tw_inpcb); 313 if (tcp_twclose(tw, reuse) != NULL) 314 return (tw); 315 } 316 } 317 return (NULL); 318 } 319 320 void 321 tcp_timer_keep(xtp) 322 void *xtp; 323 { 324 struct tcpcb *tp = xtp; 325 struct tcptemp *t_template; 326 int s; 327 struct inpcb *inp; 328 #ifdef TCPDEBUG 329 int ostate; 330 331 ostate = tp->t_state; 332 #endif 333 s = splnet(); 334 INP_INFO_WLOCK(&tcbinfo); 335 inp = tp->t_inpcb; 336 if (!inp) { 337 INP_INFO_WUNLOCK(&tcbinfo); 338 splx(s); 339 return; 340 } 341 INP_LOCK(inp); 342 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) { 343 INP_UNLOCK(inp); 344 INP_INFO_WUNLOCK(&tcbinfo); 345 splx(s); 346 return; 347 } 348 callout_deactivate(tp->tt_keep); 349 /* 350 * Keep-alive timer went off; send something 351 * or drop connection if idle for too long. 352 */ 353 tcpstat.tcps_keeptimeo++; 354 if (tp->t_state < TCPS_ESTABLISHED) 355 goto dropit; 356 if ((always_keepalive || 357 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) && 358 tp->t_state <= TCPS_CLOSING) { 359 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle) 360 goto dropit; 361 /* 362 * Send a packet designed to force a response 363 * if the peer is up and reachable: 364 * either an ACK if the connection is still alive, 365 * or an RST if the peer has closed the connection 366 * due to timeout or reboot. 367 * Using sequence number tp->snd_una-1 368 * causes the transmitted zero-length segment 369 * to lie outside the receive window; 370 * by the protocol spec, this requires the 371 * correspondent TCP to respond. 372 */ 373 tcpstat.tcps_keepprobe++; 374 t_template = tcpip_maketemplate(inp); 375 if (t_template) { 376 tcp_respond(tp, t_template->tt_ipgen, 377 &t_template->tt_t, (struct mbuf *)NULL, 378 tp->rcv_nxt, tp->snd_una - 1, 0); 379 (void) m_free(dtom(t_template)); 380 } 381 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp); 382 } else 383 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 384 385 #ifdef TCPDEBUG 386 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 387 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 388 PRU_SLOWTIMO); 389 #endif 390 INP_UNLOCK(inp); 391 INP_INFO_WUNLOCK(&tcbinfo); 392 splx(s); 393 return; 394 395 dropit: 396 tcpstat.tcps_keepdrops++; 397 tp = tcp_drop(tp, ETIMEDOUT); 398 399 #ifdef TCPDEBUG 400 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 401 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 402 PRU_SLOWTIMO); 403 #endif 404 if (tp) 405 INP_UNLOCK(tp->t_inpcb); 406 INP_INFO_WUNLOCK(&tcbinfo); 407 splx(s); 408 } 409 410 void 411 tcp_timer_persist(xtp) 412 void *xtp; 413 { 414 struct tcpcb *tp = xtp; 415 int s; 416 struct inpcb *inp; 417 #ifdef TCPDEBUG 418 int ostate; 419 420 ostate = tp->t_state; 421 #endif 422 s = splnet(); 423 INP_INFO_WLOCK(&tcbinfo); 424 inp = tp->t_inpcb; 425 if (!inp) { 426 INP_INFO_WUNLOCK(&tcbinfo); 427 splx(s); 428 return; 429 } 430 INP_LOCK(inp); 431 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){ 432 INP_UNLOCK(inp); 433 INP_INFO_WUNLOCK(&tcbinfo); 434 splx(s); 435 return; 436 } 437 callout_deactivate(tp->tt_persist); 438 /* 439 * Persistance timer into zero window. 440 * Force a byte to be output, if possible. 441 */ 442 tcpstat.tcps_persisttimeo++; 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 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 451 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle || 452 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { 453 tcpstat.tcps_persistdrop++; 454 tp = tcp_drop(tp, ETIMEDOUT); 455 goto out; 456 } 457 tcp_setpersist(tp); 458 tp->t_force = 1; 459 (void) tcp_output(tp); 460 tp->t_force = 0; 461 462 out: 463 #ifdef TCPDEBUG 464 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 465 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 466 PRU_SLOWTIMO); 467 #endif 468 if (tp) 469 INP_UNLOCK(inp); 470 INP_INFO_WUNLOCK(&tcbinfo); 471 splx(s); 472 } 473 474 void 475 tcp_timer_rexmt(xtp) 476 void *xtp; 477 { 478 struct tcpcb *tp = xtp; 479 int s; 480 int rexmt; 481 int headlocked; 482 struct inpcb *inp; 483 #ifdef TCPDEBUG 484 int ostate; 485 486 ostate = tp->t_state; 487 #endif 488 s = splnet(); 489 INP_INFO_WLOCK(&tcbinfo); 490 headlocked = 1; 491 inp = tp->t_inpcb; 492 if (!inp) { 493 INP_INFO_WUNLOCK(&tcbinfo); 494 splx(s); 495 return; 496 } 497 INP_LOCK(inp); 498 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) { 499 INP_UNLOCK(inp); 500 INP_INFO_WUNLOCK(&tcbinfo); 501 splx(s); 502 return; 503 } 504 callout_deactivate(tp->tt_rexmt); 505 /* 506 * Retransmission timer went off. Message has not 507 * been acked within retransmit interval. Back off 508 * to a longer retransmit interval and retransmit one segment. 509 */ 510 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 511 tp->t_rxtshift = TCP_MAXRXTSHIFT; 512 tcpstat.tcps_timeoutdrop++; 513 tp = tcp_drop(tp, tp->t_softerror ? 514 tp->t_softerror : ETIMEDOUT); 515 goto out; 516 } 517 INP_INFO_WUNLOCK(&tcbinfo); 518 headlocked = 0; 519 if (tp->t_rxtshift == 1) { 520 /* 521 * first retransmit; record ssthresh and cwnd so they can 522 * be recovered if this turns out to be a "bad" retransmit. 523 * A retransmit is considered "bad" if an ACK for this 524 * segment is received within RTT/2 interval; the assumption 525 * here is that the ACK was already in flight. See 526 * "On Estimating End-to-End Network Path Properties" by 527 * Allman and Paxson for more details. 528 */ 529 tp->snd_cwnd_prev = tp->snd_cwnd; 530 tp->snd_ssthresh_prev = tp->snd_ssthresh; 531 tp->snd_high_prev = tp->snd_high; 532 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); 533 } 534 tcpstat.tcps_rexmttimeo++; 535 if (tp->t_state == TCPS_SYN_SENT) 536 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift]; 537 else 538 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 539 TCPT_RANGESET(tp->t_rxtcur, rexmt, 540 tp->t_rttmin, TCPTV_REXMTMAX); 541 /* 542 * Disable rfc1323 and rfc1644 if we havn't got any response to 543 * our third SYN to work-around some broken terminal servers 544 * (most of which have hopefully been retired) that have bad VJ 545 * header compression code which trashes TCP segments containing 546 * unknown-to-them TCP options. 547 */ 548 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3)) 549 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC); 550 /* 551 * If losing, let the lower level know and try for 552 * a better route. Also, if we backed off this far, 553 * our srtt estimate is probably bogus. Clobber it 554 * so we'll take the next rtt measurement as our srtt; 555 * move the current srtt into rttvar to keep the current 556 * retransmit times until then. 557 */ 558 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 559 #ifdef INET6 560 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 561 in6_losing(tp->t_inpcb); 562 else 563 #endif 564 in_losing(tp->t_inpcb); 565 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 566 tp->t_srtt = 0; 567 } 568 tp->snd_nxt = tp->snd_una; 569 tp->snd_high = tp->snd_max; 570 /* 571 * Force a segment to be sent. 572 */ 573 tp->t_flags |= TF_ACKNOW; 574 /* 575 * If timing a segment in this window, stop the timer. 576 */ 577 tp->t_rtttime = 0; 578 /* 579 * Close the congestion window down to one segment 580 * (we'll open it by one segment for each ack we get). 581 * Since we probably have a window's worth of unacked 582 * data accumulated, this "slow start" keeps us from 583 * dumping all that data as back-to-back packets (which 584 * might overwhelm an intermediate gateway). 585 * 586 * There are two phases to the opening: Initially we 587 * open by one mss on each ack. This makes the window 588 * size increase exponentially with time. If the 589 * window is larger than the path can handle, this 590 * exponential growth results in dropped packet(s) 591 * almost immediately. To get more time between 592 * drops but still "push" the network to take advantage 593 * of improving conditions, we switch from exponential 594 * to linear window opening at some threshhold size. 595 * For a threshhold, we use half the current window 596 * size, truncated to a multiple of the mss. 597 * 598 * (the minimum cwnd that will give us exponential 599 * growth is 2 mss. We don't allow the threshhold 600 * to go below this.) 601 */ 602 { 603 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 604 if (win < 2) 605 win = 2; 606 tp->snd_cwnd = tp->t_maxseg; 607 tp->snd_ssthresh = win * tp->t_maxseg; 608 tp->t_dupacks = 0; 609 } 610 (void) tcp_output(tp); 611 612 out: 613 #ifdef TCPDEBUG 614 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 615 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 616 PRU_SLOWTIMO); 617 #endif 618 if (tp) 619 INP_UNLOCK(inp); 620 if (headlocked) 621 INP_INFO_WUNLOCK(&tcbinfo); 622 splx(s); 623 } 624