1 /* $OpenBSD: tcp_timer.c,v 1.74 2024/01/11 13:49:49 bluhm Exp $ */ 2 /* $NetBSD: tcp_timer.c,v 1.14 1996/02/13 23:44:09 christos Exp $ */ 3 4 /* 5 * Copyright (c) 1982, 1986, 1988, 1990, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)tcp_timer.c 8.1 (Berkeley) 6/10/93 33 */ 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/mbuf.h> 38 #include <sys/socket.h> 39 #include <sys/socketvar.h> 40 #include <sys/protosw.h> 41 #include <sys/kernel.h> 42 #include <sys/pool.h> 43 44 #include <net/route.h> 45 46 #include <netinet/in.h> 47 #include <netinet/ip.h> 48 #include <netinet/in_pcb.h> 49 #include <netinet/ip_var.h> 50 #include <netinet/tcp.h> 51 #include <netinet/tcp_fsm.h> 52 #include <netinet/tcp_timer.h> 53 #include <netinet/tcp_var.h> 54 #include <netinet/tcp_debug.h> 55 #include <netinet/ip_icmp.h> 56 #include <netinet/tcp_seq.h> 57 58 /* 59 * Locks used to protect struct members in this file: 60 * T tcp_timer_mtx global tcp timer data structures 61 */ 62 63 int tcp_always_keepalive; 64 int tcp_keepidle; 65 int tcp_keepintvl; 66 int tcp_maxpersistidle; /* max idle time in persist */ 67 int tcp_maxidle; /* [T] max idle time for keep alive */ 68 69 /* 70 * Time to delay the ACK. This is initialized in tcp_init(), unless 71 * its patched. 72 */ 73 int tcp_delack_msecs; 74 75 void tcp_timer_rexmt(void *); 76 void tcp_timer_persist(void *); 77 void tcp_timer_keep(void *); 78 void tcp_timer_2msl(void *); 79 void tcp_timer_reaper(void *); 80 void tcp_timer_delack(void *); 81 82 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = { 83 tcp_timer_rexmt, 84 tcp_timer_persist, 85 tcp_timer_keep, 86 tcp_timer_2msl, 87 tcp_timer_reaper, 88 tcp_timer_delack, 89 }; 90 91 /* 92 * Timer state initialization, called from tcp_init(). 93 */ 94 void 95 tcp_timer_init(void) 96 { 97 98 if (tcp_keepidle == 0) 99 tcp_keepidle = TCPTV_KEEP_IDLE; 100 101 if (tcp_keepintvl == 0) 102 tcp_keepintvl = TCPTV_KEEPINTVL; 103 104 if (tcp_maxpersistidle == 0) 105 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 106 107 if (tcp_delack_msecs == 0) 108 tcp_delack_msecs = TCP_DELACK_MSECS; 109 } 110 111 /* 112 * Callout to process delayed ACKs for a TCPCB. 113 */ 114 void 115 tcp_timer_delack(void *arg) 116 { 117 struct tcpcb *otp = NULL, *tp = arg; 118 short ostate; 119 120 /* 121 * If tcp_output() wasn't able to transmit the ACK 122 * for whatever reason, it will restart the delayed 123 * ACK callout. 124 */ 125 NET_LOCK(); 126 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 127 if (!ISSET((tp)->t_flags, TF_TMR_DELACK) || 128 timeout_pending(&tp->t_timer[TCPT_DELACK])) 129 goto out; 130 CLR((tp)->t_flags, TF_TMR_DELACK); 131 132 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { 133 otp = tp; 134 ostate = tp->t_state; 135 } 136 tp->t_flags |= TF_ACKNOW; 137 (void) tcp_output(tp); 138 if (otp) 139 tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_DELACK, 0); 140 out: 141 NET_UNLOCK(); 142 } 143 144 /* 145 * Tcp protocol timeout routine called every 500 ms. 146 * Updates the timers in all active tcb's and 147 * causes finite state machine actions if timers expire. 148 */ 149 void 150 tcp_slowtimo(void) 151 { 152 mtx_enter(&tcp_timer_mtx); 153 tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl; 154 tcp_iss += TCP_ISSINCR2/PR_SLOWHZ; /* increment iss */ 155 mtx_leave(&tcp_timer_mtx); 156 } 157 158 /* 159 * Cancel all timers for TCP tp. 160 */ 161 void 162 tcp_canceltimers(struct tcpcb *tp) 163 { 164 int i; 165 166 for (i = 0; i < TCPT_NTIMERS; i++) 167 TCP_TIMER_DISARM(tp, i); 168 } 169 170 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 171 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 172 173 int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 174 175 /* 176 * TCP timer processing. 177 */ 178 179 void tcp_timer_freesack(struct tcpcb *); 180 181 void 182 tcp_timer_freesack(struct tcpcb *tp) 183 { 184 struct sackhole *p, *q; 185 /* 186 * Free SACK holes for 2MSL and REXMT timers. 187 */ 188 q = tp->snd_holes; 189 while (q != NULL) { 190 p = q; 191 q = q->next; 192 pool_put(&sackhl_pool, p); 193 } 194 tp->snd_holes = 0; 195 } 196 197 void 198 tcp_timer_rexmt(void *arg) 199 { 200 struct tcpcb *otp = NULL, *tp = arg; 201 uint32_t rto; 202 short ostate; 203 204 NET_LOCK(); 205 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 206 if (!ISSET((tp)->t_flags, TF_TMR_REXMT) || 207 timeout_pending(&tp->t_timer[TCPT_REXMT])) 208 goto out; 209 CLR((tp)->t_flags, TF_TMR_REXMT); 210 211 if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb && 212 SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) && 213 SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) { 214 struct sockaddr_in sin; 215 struct icmp icmp; 216 217 tp->t_flags &= ~TF_PMTUD_PEND; 218 219 /* XXX create fake icmp message with relevant entries */ 220 icmp.icmp_nextmtu = tp->t_pmtud_nextmtu; 221 icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len; 222 icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl; 223 icmp.icmp_ip.ip_dst = tp->t_inpcb->inp_faddr; 224 icmp_mtudisc(&icmp, tp->t_inpcb->inp_rtableid); 225 226 /* 227 * Notify all connections to the same peer about 228 * new mss and trigger retransmit. 229 */ 230 bzero(&sin, sizeof(sin)); 231 sin.sin_len = sizeof(sin); 232 sin.sin_family = AF_INET; 233 sin.sin_addr = tp->t_inpcb->inp_faddr; 234 in_pcbnotifyall(&tcbtable, sintosa(&sin), 235 tp->t_inpcb->inp_rtableid, EMSGSIZE, tcp_mtudisc); 236 goto out; 237 } 238 239 tcp_timer_freesack(tp); 240 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 241 tp->t_rxtshift = TCP_MAXRXTSHIFT; 242 tcpstat_inc(tcps_timeoutdrop); 243 tp = tcp_drop(tp, tp->t_softerror ? 244 tp->t_softerror : ETIMEDOUT); 245 goto out; 246 } 247 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { 248 otp = tp; 249 ostate = tp->t_state; 250 } 251 tcpstat_inc(tcps_rexmttimeo); 252 rto = TCP_REXMTVAL(tp); 253 if (rto < tp->t_rttmin) 254 rto = tp->t_rttmin; 255 TCPT_RANGESET(tp->t_rxtcur, 256 rto * tcp_backoff[tp->t_rxtshift], 257 tp->t_rttmin, TCPTV_REXMTMAX); 258 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 259 260 /* 261 * If we are losing and we are trying path MTU discovery, 262 * try turning it off. This will avoid black holes in 263 * the network which suppress or fail to send "packet 264 * too big" ICMP messages. We should ideally do 265 * lots more sophisticated searching to find the right 266 * value here... 267 */ 268 if (ip_mtudisc && tp->t_inpcb && 269 TCPS_HAVEESTABLISHED(tp->t_state) && 270 tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 271 struct inpcb *inp = tp->t_inpcb; 272 struct rtentry *rt = NULL; 273 274 /* No data to send means path mtu is not a problem */ 275 if (!inp->inp_socket->so_snd.sb_cc) 276 goto leave; 277 278 rt = in_pcbrtentry(inp); 279 /* Check if path MTU discovery is disabled already */ 280 if (rt && (rt->rt_flags & RTF_HOST) && 281 (rt->rt_locks & RTV_MTU)) 282 goto leave; 283 284 rt = NULL; 285 switch(tp->pf) { 286 #ifdef INET6 287 case PF_INET6: 288 /* 289 * We can not turn off path MTU for IPv6. 290 * Do nothing for now, maybe lower to 291 * minimum MTU. 292 */ 293 break; 294 #endif 295 case PF_INET: 296 rt = icmp_mtudisc_clone(inp->inp_faddr, 297 inp->inp_rtableid, 0); 298 break; 299 } 300 if (rt != NULL) { 301 /* Disable path MTU discovery */ 302 if ((rt->rt_locks & RTV_MTU) == 0) { 303 rt->rt_locks |= RTV_MTU; 304 in_rtchange(inp, 0); 305 } 306 307 rtfree(rt); 308 } 309 leave: 310 ; 311 } 312 313 /* 314 * If losing, let the lower level know and try for 315 * a better route. Also, if we backed off this far, 316 * our srtt estimate is probably bogus. Clobber it 317 * so we'll take the next rtt measurement as our srtt; 318 * move the current srtt into rttvar to keep the current 319 * retransmit times until then. 320 */ 321 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 322 in_losing(tp->t_inpcb); 323 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 324 tp->t_srtt = 0; 325 } 326 tp->snd_nxt = tp->snd_una; 327 /* 328 * Note: We overload snd_last to function also as the 329 * snd_last variable described in RFC 2582 330 */ 331 tp->snd_last = tp->snd_max; 332 /* 333 * If timing a segment in this window, stop the timer. 334 */ 335 tp->t_rtttime = 0; 336 #ifdef TCP_ECN 337 /* 338 * if ECN is enabled, there might be a broken firewall which 339 * blocks ecn packets. fall back to non-ecn. 340 */ 341 if ((tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) 342 && tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 343 tp->t_flags |= TF_DISABLE_ECN; 344 #endif 345 /* 346 * Close the congestion window down to one segment 347 * (we'll open it by one segment for each ack we get). 348 * Since we probably have a window's worth of unacked 349 * data accumulated, this "slow start" keeps us from 350 * dumping all that data as back-to-back packets (which 351 * might overwhelm an intermediate gateway). 352 * 353 * There are two phases to the opening: Initially we 354 * open by one mss on each ack. This makes the window 355 * size increase exponentially with time. If the 356 * window is larger than the path can handle, this 357 * exponential growth results in dropped packet(s) 358 * almost immediately. To get more time between 359 * drops but still "push" the network to take advantage 360 * of improving conditions, we switch from exponential 361 * to linear window opening at some threshold size. 362 * For a threshold, we use half the current window 363 * size, truncated to a multiple of the mss. 364 * 365 * (the minimum cwnd that will give us exponential 366 * growth is 2 mss. We don't allow the threshold 367 * to go below this.) 368 */ 369 { 370 u_long win; 371 372 win = ulmin(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 373 if (win < 2) 374 win = 2; 375 tp->snd_cwnd = tp->t_maxseg; 376 tp->snd_ssthresh = win * tp->t_maxseg; 377 tp->t_dupacks = 0; 378 #ifdef TCP_ECN 379 tp->snd_last = tp->snd_max; 380 tp->t_flags |= TF_SEND_CWR; 381 #endif 382 #if 1 /* TCP_ECN */ 383 tcpstat_inc(tcps_cwr_timeout); 384 #endif 385 } 386 (void) tcp_output(tp); 387 if (otp) 388 tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_REXMT, 0); 389 out: 390 NET_UNLOCK(); 391 } 392 393 void 394 tcp_timer_persist(void *arg) 395 { 396 struct tcpcb *otp = NULL, *tp = arg; 397 uint32_t rto; 398 short ostate; 399 uint64_t now; 400 401 NET_LOCK(); 402 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 403 if (!ISSET((tp)->t_flags, TF_TMR_PERSIST) || 404 timeout_pending(&tp->t_timer[TCPT_PERSIST])) 405 goto out; 406 CLR((tp)->t_flags, TF_TMR_PERSIST); 407 408 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 409 goto out; 410 411 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { 412 otp = tp; 413 ostate = tp->t_state; 414 } 415 tcpstat_inc(tcps_persisttimeo); 416 /* 417 * Hack: if the peer is dead/unreachable, we do not 418 * time out if the window is closed. After a full 419 * backoff, drop the connection if the idle time 420 * (no responses to probes) reaches the maximum 421 * backoff that we would use if retransmitting. 422 */ 423 rto = TCP_REXMTVAL(tp); 424 if (rto < tp->t_rttmin) 425 rto = tp->t_rttmin; 426 now = tcp_now(); 427 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 428 ((now - tp->t_rcvtime) >= tcp_maxpersistidle || 429 (now - tp->t_rcvtime) >= rto * tcp_totbackoff)) { 430 tcpstat_inc(tcps_persistdrop); 431 tp = tcp_drop(tp, ETIMEDOUT); 432 goto out; 433 } 434 tcp_setpersist(tp); 435 tp->t_force = 1; 436 (void) tcp_output(tp); 437 tp->t_force = 0; 438 if (otp) 439 tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_PERSIST, 0); 440 out: 441 NET_UNLOCK(); 442 } 443 444 void 445 tcp_timer_keep(void *arg) 446 { 447 struct tcpcb *otp = NULL, *tp = arg; 448 short ostate; 449 450 NET_LOCK(); 451 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 452 if (!ISSET((tp)->t_flags, TF_TMR_KEEP) || 453 timeout_pending(&tp->t_timer[TCPT_KEEP])) 454 goto out; 455 CLR((tp)->t_flags, TF_TMR_KEEP); 456 457 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { 458 otp = tp; 459 ostate = tp->t_state; 460 } 461 tcpstat_inc(tcps_keeptimeo); 462 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 463 goto dropit; 464 if ((tcp_always_keepalive || 465 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) && 466 tp->t_state <= TCPS_CLOSING) { 467 int maxidle; 468 uint64_t now; 469 470 maxidle = READ_ONCE(tcp_maxidle); 471 now = tcp_now(); 472 if ((maxidle > 0) && 473 ((now - tp->t_rcvtime) >= tcp_keepidle + maxidle)) 474 goto dropit; 475 /* 476 * Send a packet designed to force a response 477 * if the peer is up and reachable: 478 * either an ACK if the connection is still alive, 479 * or an RST if the peer has closed the connection 480 * due to timeout or reboot. 481 * Using sequence number tp->snd_una-1 482 * causes the transmitted zero-length segment 483 * to lie outside the receive window; 484 * by the protocol spec, this requires the 485 * correspondent TCP to respond. 486 */ 487 tcpstat_inc(tcps_keepprobe); 488 tcp_respond(tp, mtod(tp->t_template, caddr_t), 489 NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0, now); 490 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 491 } else 492 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 493 if (otp) 494 tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_KEEP, 0); 495 out: 496 NET_UNLOCK(); 497 return; 498 499 dropit: 500 tcpstat_inc(tcps_keepdrops); 501 tp = tcp_drop(tp, ETIMEDOUT); 502 NET_UNLOCK(); 503 } 504 505 void 506 tcp_timer_2msl(void *arg) 507 { 508 struct tcpcb *otp = NULL, *tp = arg; 509 short ostate; 510 int maxidle; 511 uint64_t now; 512 513 NET_LOCK(); 514 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 515 if (!ISSET((tp)->t_flags, TF_TMR_2MSL) || 516 timeout_pending(&tp->t_timer[TCPT_2MSL])) 517 goto out; 518 CLR((tp)->t_flags, TF_TMR_2MSL); 519 520 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { 521 otp = tp; 522 ostate = tp->t_state; 523 } 524 tcp_timer_freesack(tp); 525 526 maxidle = READ_ONCE(tcp_maxidle); 527 now = tcp_now(); 528 if (tp->t_state != TCPS_TIME_WAIT && 529 ((maxidle == 0) || ((now - tp->t_rcvtime) <= maxidle))) 530 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 531 else 532 tp = tcp_close(tp); 533 if (otp) 534 tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_2MSL, 0); 535 out: 536 NET_UNLOCK(); 537 } 538 539 void 540 tcp_timer_reaper(void *arg) 541 { 542 struct tcpcb *tp = arg; 543 544 /* 545 * This timer is necessary to delay the pool_put() after all timers 546 * have finished, even if they were sleeping to grab the net lock. 547 * Putting the pool_put() in a timer is sufficient as all timers run 548 * from the same timeout thread. Note that neither softnet thread nor 549 * user process may access the tcpcb after arming the reaper timer. 550 * Freeing may run in parallel as it does not grab the net lock. 551 */ 552 pool_put(&tcpcb_pool, tp); 553 tcpstat_inc(tcps_closed); 554 } 555