1 /* $OpenBSD: tcp_timer.c,v 1.66 2018/05/10 13:30:25 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 int tcp_always_keepalive; 59 int tcp_keepidle; 60 int tcp_keepintvl; 61 int tcp_maxpersistidle; /* max idle time in persist */ 62 int tcp_maxidle; 63 64 /* 65 * Time to delay the ACK. This is initialized in tcp_init(), unless 66 * its patched. 67 */ 68 int tcp_delack_msecs; 69 70 void tcp_timer_rexmt(void *); 71 void tcp_timer_persist(void *); 72 void tcp_timer_keep(void *); 73 void tcp_timer_2msl(void *); 74 void tcp_timer_reaper(void *); 75 void tcp_timer_delack(void *); 76 77 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = { 78 tcp_timer_rexmt, 79 tcp_timer_persist, 80 tcp_timer_keep, 81 tcp_timer_2msl, 82 tcp_timer_reaper, 83 tcp_timer_delack, 84 }; 85 86 /* 87 * Timer state initialization, called from tcp_init(). 88 */ 89 void 90 tcp_timer_init(void) 91 { 92 93 if (tcp_keepidle == 0) 94 tcp_keepidle = TCPTV_KEEP_IDLE; 95 96 if (tcp_keepintvl == 0) 97 tcp_keepintvl = TCPTV_KEEPINTVL; 98 99 if (tcp_maxpersistidle == 0) 100 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 101 102 if (tcp_delack_msecs == 0) 103 tcp_delack_msecs = TCP_DELACK_MSECS; 104 } 105 106 /* 107 * Callout to process delayed ACKs for a TCPCB. 108 */ 109 void 110 tcp_timer_delack(void *arg) 111 { 112 struct tcpcb *tp = arg; 113 short ostate; 114 115 /* 116 * If tcp_output() wasn't able to transmit the ACK 117 * for whatever reason, it will restart the delayed 118 * ACK callout. 119 */ 120 NET_LOCK(); 121 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 122 if (!ISSET((tp)->t_flags, TF_TMR_DELACK) || 123 timeout_pending(&tp->t_timer[TCPT_DELACK])) 124 goto out; 125 CLR((tp)->t_flags, TF_TMR_DELACK); 126 127 ostate = tp->t_state; 128 tp->t_flags |= TF_ACKNOW; 129 (void) tcp_output(tp); 130 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 131 tcp_trace(TA_TIMER, ostate, tp, (caddr_t)0, TCPT_DELACK, 0); 132 out: 133 NET_UNLOCK(); 134 } 135 136 /* 137 * Tcp protocol timeout routine called every 500 ms. 138 * Updates the timers in all active tcb's and 139 * causes finite state machine actions if timers expire. 140 */ 141 void 142 tcp_slowtimo(void) 143 { 144 NET_LOCK(); 145 146 tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl; 147 tcp_iss += TCP_ISSINCR2/PR_SLOWHZ; /* increment iss */ 148 tcp_now++; /* for timestamps */ 149 150 NET_UNLOCK(); 151 } 152 153 /* 154 * Cancel all timers for TCP tp. 155 */ 156 void 157 tcp_canceltimers(struct tcpcb *tp) 158 { 159 int i; 160 161 for (i = 0; i < TCPT_NTIMERS; i++) 162 TCP_TIMER_DISARM(tp, i); 163 } 164 165 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 166 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 167 168 int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 169 170 /* 171 * TCP timer processing. 172 */ 173 174 void tcp_timer_freesack(struct tcpcb *); 175 176 void 177 tcp_timer_freesack(struct tcpcb *tp) 178 { 179 struct sackhole *p, *q; 180 /* 181 * Free SACK holes for 2MSL and REXMT timers. 182 */ 183 q = tp->snd_holes; 184 while (q != NULL) { 185 p = q; 186 q = q->next; 187 pool_put(&sackhl_pool, p); 188 } 189 tp->snd_holes = 0; 190 } 191 192 void 193 tcp_timer_rexmt(void *arg) 194 { 195 struct tcpcb *tp = arg; 196 uint32_t rto; 197 short ostate; 198 199 NET_LOCK(); 200 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 201 if (!ISSET((tp)->t_flags, TF_TMR_REXMT) || 202 timeout_pending(&tp->t_timer[TCPT_REXMT])) 203 goto out; 204 CLR((tp)->t_flags, TF_TMR_REXMT); 205 206 if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb && 207 SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) && 208 SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) { 209 struct sockaddr_in sin; 210 struct icmp icmp; 211 212 tp->t_flags &= ~TF_PMTUD_PEND; 213 214 /* XXX create fake icmp message with relevant entries */ 215 icmp.icmp_nextmtu = tp->t_pmtud_nextmtu; 216 icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len; 217 icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl; 218 icmp.icmp_ip.ip_dst = tp->t_inpcb->inp_faddr; 219 icmp_mtudisc(&icmp, tp->t_inpcb->inp_rtableid); 220 221 /* 222 * Notify all connections to the same peer about 223 * new mss and trigger retransmit. 224 */ 225 bzero(&sin, sizeof(sin)); 226 sin.sin_len = sizeof(sin); 227 sin.sin_family = AF_INET; 228 sin.sin_addr = tp->t_inpcb->inp_faddr; 229 in_pcbnotifyall(&tcbtable, sintosa(&sin), 230 tp->t_inpcb->inp_rtableid, EMSGSIZE, tcp_mtudisc); 231 goto out; 232 } 233 234 tcp_timer_freesack(tp); 235 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 236 tp->t_rxtshift = TCP_MAXRXTSHIFT; 237 tcpstat_inc(tcps_timeoutdrop); 238 tp = tcp_drop(tp, tp->t_softerror ? 239 tp->t_softerror : ETIMEDOUT); 240 goto out; 241 } 242 ostate = tp->t_state; 243 tcpstat_inc(tcps_rexmttimeo); 244 rto = TCP_REXMTVAL(tp); 245 if (rto < tp->t_rttmin) 246 rto = tp->t_rttmin; 247 TCPT_RANGESET(tp->t_rxtcur, 248 rto * tcp_backoff[tp->t_rxtshift], 249 tp->t_rttmin, TCPTV_REXMTMAX); 250 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 251 252 /* 253 * If we are losing and we are trying path MTU discovery, 254 * try turning it off. This will avoid black holes in 255 * the network which suppress or fail to send "packet 256 * too big" ICMP messages. We should ideally do 257 * lots more sophisticated searching to find the right 258 * value here... 259 */ 260 if (ip_mtudisc && tp->t_inpcb && 261 TCPS_HAVEESTABLISHED(tp->t_state) && 262 tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 263 struct inpcb *inp = tp->t_inpcb; 264 struct rtentry *rt = NULL; 265 266 /* No data to send means path mtu is not a problem */ 267 if (!inp->inp_socket->so_snd.sb_cc) 268 goto leave; 269 270 rt = in_pcbrtentry(inp); 271 /* Check if path MTU discovery is disabled already */ 272 if (rt && (rt->rt_flags & RTF_HOST) && 273 (rt->rt_locks & RTV_MTU)) 274 goto leave; 275 276 rt = NULL; 277 switch(tp->pf) { 278 #ifdef INET6 279 case PF_INET6: 280 /* 281 * We can not turn off path MTU for IPv6. 282 * Do nothing for now, maybe lower to 283 * minimum MTU. 284 */ 285 break; 286 #endif 287 case PF_INET: 288 rt = icmp_mtudisc_clone(inp->inp_faddr, 289 inp->inp_rtableid); 290 break; 291 } 292 if (rt != NULL) { 293 /* Disable path MTU discovery */ 294 if ((rt->rt_locks & RTV_MTU) == 0) { 295 rt->rt_locks |= RTV_MTU; 296 in_rtchange(inp, 0); 297 } 298 299 rtfree(rt); 300 } 301 leave: 302 ; 303 } 304 305 /* 306 * If losing, let the lower level know and try for 307 * a better route. Also, if we backed off this far, 308 * our srtt estimate is probably bogus. Clobber it 309 * so we'll take the next rtt measurement as our srtt; 310 * move the current srtt into rttvar to keep the current 311 * retransmit times until then. 312 */ 313 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 314 in_losing(tp->t_inpcb); 315 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 316 tp->t_srtt = 0; 317 } 318 tp->snd_nxt = tp->snd_una; 319 /* 320 * Note: We overload snd_last to function also as the 321 * snd_last variable described in RFC 2582 322 */ 323 tp->snd_last = tp->snd_max; 324 /* 325 * If timing a segment in this window, stop the timer. 326 */ 327 tp->t_rtttime = 0; 328 #ifdef TCP_ECN 329 /* 330 * if ECN is enabled, there might be a broken firewall which 331 * blocks ecn packets. fall back to non-ecn. 332 */ 333 if ((tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) 334 && tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 335 tp->t_flags |= TF_DISABLE_ECN; 336 #endif 337 /* 338 * Close the congestion window down to one segment 339 * (we'll open it by one segment for each ack we get). 340 * Since we probably have a window's worth of unacked 341 * data accumulated, this "slow start" keeps us from 342 * dumping all that data as back-to-back packets (which 343 * might overwhelm an intermediate gateway). 344 * 345 * There are two phases to the opening: Initially we 346 * open by one mss on each ack. This makes the window 347 * size increase exponentially with time. If the 348 * window is larger than the path can handle, this 349 * exponential growth results in dropped packet(s) 350 * almost immediately. To get more time between 351 * drops but still "push" the network to take advantage 352 * of improving conditions, we switch from exponential 353 * to linear window opening at some threshold size. 354 * For a threshold, we use half the current window 355 * size, truncated to a multiple of the mss. 356 * 357 * (the minimum cwnd that will give us exponential 358 * growth is 2 mss. We don't allow the threshold 359 * to go below this.) 360 */ 361 { 362 u_long win = ulmin(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 363 if (win < 2) 364 win = 2; 365 tp->snd_cwnd = tp->t_maxseg; 366 tp->snd_ssthresh = win * tp->t_maxseg; 367 tp->t_dupacks = 0; 368 #ifdef TCP_ECN 369 tp->snd_last = tp->snd_max; 370 tp->t_flags |= TF_SEND_CWR; 371 #endif 372 #if 1 /* TCP_ECN */ 373 tcpstat_inc(tcps_cwr_timeout); 374 #endif 375 } 376 (void) tcp_output(tp); 377 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 378 tcp_trace(TA_TIMER, ostate, tp, (caddr_t)0, TCPT_REXMT, 0); 379 out: 380 NET_UNLOCK(); 381 } 382 383 void 384 tcp_timer_persist(void *arg) 385 { 386 struct tcpcb *tp = arg; 387 uint32_t rto; 388 short ostate; 389 390 NET_LOCK(); 391 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 392 if (!ISSET((tp)->t_flags, TF_TMR_PERSIST) || 393 timeout_pending(&tp->t_timer[TCPT_PERSIST])) 394 goto out; 395 CLR((tp)->t_flags, TF_TMR_PERSIST); 396 397 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT)) 398 goto out; 399 400 ostate = tp->t_state; 401 tcpstat_inc(tcps_persisttimeo); 402 /* 403 * Hack: if the peer is dead/unreachable, we do not 404 * time out if the window is closed. After a full 405 * backoff, drop the connection if the idle time 406 * (no responses to probes) reaches the maximum 407 * backoff that we would use if retransmitting. 408 */ 409 rto = TCP_REXMTVAL(tp); 410 if (rto < tp->t_rttmin) 411 rto = tp->t_rttmin; 412 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 413 ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle || 414 (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) { 415 tcpstat_inc(tcps_persistdrop); 416 tp = tcp_drop(tp, ETIMEDOUT); 417 goto out; 418 } 419 tcp_setpersist(tp); 420 tp->t_force = 1; 421 (void) tcp_output(tp); 422 tp->t_force = 0; 423 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 424 tcp_trace(TA_TIMER, ostate, tp, (caddr_t)0, TCPT_PERSIST, 0); 425 out: 426 NET_UNLOCK(); 427 } 428 429 void 430 tcp_timer_keep(void *arg) 431 { 432 struct tcpcb *tp = arg; 433 short ostate; 434 435 NET_LOCK(); 436 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 437 if (!ISSET((tp)->t_flags, TF_TMR_KEEP) || 438 timeout_pending(&tp->t_timer[TCPT_KEEP])) 439 goto out; 440 CLR((tp)->t_flags, TF_TMR_KEEP); 441 442 ostate = tp->t_state; 443 tcpstat_inc(tcps_keeptimeo); 444 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 445 goto dropit; 446 if ((tcp_always_keepalive || 447 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) && 448 tp->t_state <= TCPS_CLOSING) { 449 if ((tcp_maxidle > 0) && 450 ((tcp_now - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)) 451 goto dropit; 452 /* 453 * Send a packet designed to force a response 454 * if the peer is up and reachable: 455 * either an ACK if the connection is still alive, 456 * or an RST if the peer has closed the connection 457 * due to timeout or reboot. 458 * Using sequence number tp->snd_una-1 459 * causes the transmitted zero-length segment 460 * to lie outside the receive window; 461 * by the protocol spec, this requires the 462 * correspondent TCP to respond. 463 */ 464 tcpstat_inc(tcps_keepprobe); 465 tcp_respond(tp, mtod(tp->t_template, caddr_t), 466 NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0); 467 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 468 } else 469 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 470 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 471 tcp_trace(TA_TIMER, ostate, tp, (caddr_t)0, TCPT_KEEP, 0); 472 out: 473 NET_UNLOCK(); 474 return; 475 476 dropit: 477 tcpstat_inc(tcps_keepdrops); 478 tp = tcp_drop(tp, ETIMEDOUT); 479 NET_UNLOCK(); 480 } 481 482 void 483 tcp_timer_2msl(void *arg) 484 { 485 struct tcpcb *tp = arg; 486 short ostate; 487 488 NET_LOCK(); 489 /* Ignore canceled timeouts or timeouts that have been rescheduled. */ 490 if (!ISSET((tp)->t_flags, TF_TMR_2MSL) || 491 timeout_pending(&tp->t_timer[TCPT_2MSL])) 492 goto out; 493 CLR((tp)->t_flags, TF_TMR_2MSL); 494 495 ostate = tp->t_state; 496 tcp_timer_freesack(tp); 497 498 if (tp->t_state != TCPS_TIME_WAIT && 499 ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle))) 500 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 501 else 502 tp = tcp_close(tp); 503 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 504 tcp_trace(TA_TIMER, ostate, tp, (caddr_t)0, TCPT_2MSL, 0); 505 out: 506 NET_UNLOCK(); 507 } 508 509 void 510 tcp_timer_reaper(void *arg) 511 { 512 struct tcpcb *tp = arg; 513 514 /* 515 * This timer is necessary to delay the pool_put() after all timers 516 * have finished, even if they were sleeping to grab the net lock. 517 * Putting the pool_put() in a timer is sufficinet as all timers run 518 * from the same timeout thread. Note that neither softnet thread nor 519 * user process may access the tcpcb after arming the reaper timer. 520 * Freeing may run in parallel as it does not grab the net lock. 521 */ 522 pool_put(&tcpcb_pool, tp); 523 tcpstat_inc(tcps_closed); 524 } 525