1 /* $OpenBSD: tcp_timer.c,v 1.46 2011/07/06 23:44:20 sthen 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/in_systm.h> 48 #include <netinet/ip.h> 49 #include <netinet/in_pcb.h> 50 #include <netinet/ip_var.h> 51 #include <netinet/tcp.h> 52 #include <netinet/tcp_fsm.h> 53 #include <netinet/tcp_timer.h> 54 #include <netinet/tcp_var.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_ticks; 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 75 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = { 76 tcp_timer_rexmt, 77 tcp_timer_persist, 78 tcp_timer_keep, 79 tcp_timer_2msl, 80 }; 81 82 /* 83 * Timer state initialization, called from tcp_init(). 84 */ 85 void 86 tcp_timer_init(void) 87 { 88 89 if (tcp_keepidle == 0) 90 tcp_keepidle = TCPTV_KEEP_IDLE; 91 92 if (tcp_keepintvl == 0) 93 tcp_keepintvl = TCPTV_KEEPINTVL; 94 95 if (tcp_maxpersistidle == 0) 96 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 97 98 if (tcp_delack_ticks == 0) 99 tcp_delack_ticks = TCP_DELACK_TICKS; 100 } 101 102 /* 103 * Callout to process delayed ACKs for a TCPCB. 104 */ 105 void 106 tcp_delack(void *arg) 107 { 108 struct tcpcb *tp = arg; 109 int s; 110 111 /* 112 * If tcp_output() wasn't able to transmit the ACK 113 * for whatever reason, it will restart the delayed 114 * ACK callout. 115 */ 116 117 s = splsoftnet(); 118 if (tp->t_flags & TF_DEAD) { 119 splx(s); 120 return; 121 } 122 tp->t_flags |= TF_ACKNOW; 123 (void) tcp_output(tp); 124 splx(s); 125 } 126 127 /* 128 * Tcp protocol timeout routine called every 500 ms. 129 * Updates the timers in all active tcb's and 130 * causes finite state machine actions if timers expire. 131 */ 132 void 133 tcp_slowtimo() 134 { 135 int s; 136 137 s = splsoftnet(); 138 tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl; 139 tcp_iss += TCP_ISSINCR2/PR_SLOWHZ; /* increment iss */ 140 tcp_now++; /* for timestamps */ 141 splx(s); 142 } 143 144 /* 145 * Cancel all timers for TCP tp. 146 */ 147 void 148 tcp_canceltimers(tp) 149 struct tcpcb *tp; 150 { 151 int i; 152 153 for (i = 0; i < TCPT_NTIMERS; i++) 154 TCP_TIMER_DISARM(tp, i); 155 } 156 157 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 158 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 159 160 int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 161 162 /* 163 * TCP timer processing. 164 */ 165 166 #ifdef TCP_SACK 167 void tcp_timer_freesack(struct tcpcb *); 168 169 void 170 tcp_timer_freesack(struct tcpcb *tp) 171 { 172 struct sackhole *p, *q; 173 /* 174 * Free SACK holes for 2MSL and REXMT timers. 175 */ 176 q = tp->snd_holes; 177 while (q != NULL) { 178 p = q; 179 q = q->next; 180 pool_put(&sackhl_pool, p); 181 } 182 tp->snd_holes = 0; 183 #ifdef TCP_FACK 184 tp->snd_fack = tp->snd_una; 185 tp->retran_data = 0; 186 tp->snd_awnd = 0; 187 #endif /* TCP_FACK */ 188 } 189 #endif /* TCP_SACK */ 190 191 void 192 tcp_timer_rexmt(void *arg) 193 { 194 struct tcpcb *tp = arg; 195 uint32_t rto; 196 int s; 197 198 s = splsoftnet(); 199 if (tp->t_flags & TF_DEAD) { 200 splx(s); 201 return; 202 } 203 204 if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb && 205 SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) && 206 SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) { 207 extern struct sockaddr_in icmpsrc; 208 struct icmp icmp; 209 210 tp->t_flags &= ~TF_PMTUD_PEND; 211 212 /* XXX create fake icmp message with relevant entries */ 213 icmp.icmp_nextmtu = tp->t_pmtud_nextmtu; 214 icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len; 215 icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl; 216 icmpsrc.sin_addr = tp->t_inpcb->inp_faddr; 217 icmp_mtudisc(&icmp, tp->t_inpcb->inp_rtableid); 218 219 /* 220 * Notify all connections to the same peer about 221 * new mss and trigger retransmit. 222 */ 223 in_pcbnotifyall(&tcbtable, sintosa(&icmpsrc), 224 tp->t_inpcb->inp_rtableid, EMSGSIZE, tcp_mtudisc); 225 splx(s); 226 return; 227 } 228 229 #ifdef TCP_SACK 230 tcp_timer_freesack(tp); 231 #endif 232 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 233 tp->t_rxtshift = TCP_MAXRXTSHIFT; 234 tcpstat.tcps_timeoutdrop++; 235 (void)tcp_drop(tp, tp->t_softerror ? 236 tp->t_softerror : ETIMEDOUT); 237 goto out; 238 } 239 tcpstat.tcps_rexmttimeo++; 240 rto = TCP_REXMTVAL(tp); 241 if (rto < tp->t_rttmin) 242 rto = tp->t_rttmin; 243 TCPT_RANGESET(tp->t_rxtcur, 244 rto * tcp_backoff[tp->t_rxtshift], 245 tp->t_rttmin, TCPTV_REXMTMAX); 246 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 247 248 /* 249 * If we are losing and we are trying path MTU discovery, 250 * try turning it off. This will avoid black holes in 251 * the network which suppress or fail to send "packet 252 * too big" ICMP messages. We should ideally do 253 * lots more sophisticated searching to find the right 254 * value here... 255 */ 256 if (ip_mtudisc && tp->t_inpcb && 257 TCPS_HAVEESTABLISHED(tp->t_state) && 258 tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) { 259 struct inpcb *inp = tp->t_inpcb; 260 struct rtentry *rt = NULL; 261 struct sockaddr_in sin; 262 263 /* No data to send means path mtu is not a problem */ 264 if (!inp->inp_socket->so_snd.sb_cc) 265 goto leave; 266 267 rt = in_pcbrtentry(inp); 268 /* Check if path MTU discovery is disabled already */ 269 if (rt && (rt->rt_flags & RTF_HOST) && 270 (rt->rt_rmx.rmx_locks & RTV_MTU)) 271 goto leave; 272 273 rt = NULL; 274 switch(tp->pf) { 275 #ifdef INET6 276 case PF_INET6: 277 /* 278 * We can not turn off path MTU for IPv6. 279 * Do nothing for now, maybe lower to 280 * minimum MTU. 281 */ 282 break; 283 #endif 284 case PF_INET: 285 bzero(&sin, sizeof(struct sockaddr_in)); 286 sin.sin_family = AF_INET; 287 sin.sin_len = sizeof(struct sockaddr_in); 288 sin.sin_addr = inp->inp_faddr; 289 rt = icmp_mtudisc_clone(sintosa(&sin), 290 inp->inp_rtableid); 291 break; 292 } 293 if (rt != NULL) { 294 /* Disable path MTU discovery */ 295 if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) { 296 rt->rt_rmx.rmx_locks |= RTV_MTU; 297 in_rtchange(inp, 0); 298 } 299 300 rtfree(rt); 301 } 302 leave: 303 ; 304 } 305 306 /* 307 * If losing, let the lower level know and try for 308 * a better route. Also, if we backed off this far, 309 * our srtt estimate is probably bogus. Clobber it 310 * so we'll take the next rtt measurement as our srtt; 311 * move the current srtt into rttvar to keep the current 312 * retransmit times until then. 313 */ 314 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 315 in_losing(tp->t_inpcb); 316 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 317 tp->t_srtt = 0; 318 } 319 tp->snd_nxt = tp->snd_una; 320 #if defined(TCP_SACK) 321 /* 322 * Note: We overload snd_last to function also as the 323 * snd_last variable described in RFC 2582 324 */ 325 tp->snd_last = tp->snd_max; 326 #endif /* TCP_SACK */ 327 /* 328 * If timing a segment in this window, stop the timer. 329 */ 330 tp->t_rtttime = 0; 331 #ifdef TCP_ECN 332 /* 333 * if ECN is enabled, there might be a broken firewall which 334 * blocks ecn packets. fall back to non-ecn. 335 */ 336 if ((tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) 337 && tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 338 tp->t_flags |= TF_DISABLE_ECN; 339 #endif 340 /* 341 * Close the congestion window down to one segment 342 * (we'll open it by one segment for each ack we get). 343 * Since we probably have a window's worth of unacked 344 * data accumulated, this "slow start" keeps us from 345 * dumping all that data as back-to-back packets (which 346 * might overwhelm an intermediate gateway). 347 * 348 * There are two phases to the opening: Initially we 349 * open by one mss on each ack. This makes the window 350 * size increase exponentially with time. If the 351 * window is larger than the path can handle, this 352 * exponential growth results in dropped packet(s) 353 * almost immediately. To get more time between 354 * drops but still "push" the network to take advantage 355 * of improving conditions, we switch from exponential 356 * to linear window opening at some threshold size. 357 * For a threshold, we use half the current window 358 * size, truncated to a multiple of the mss. 359 * 360 * (the minimum cwnd that will give us exponential 361 * growth is 2 mss. We don't allow the threshold 362 * to go below this.) 363 */ 364 { 365 u_long win = ulmin(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 366 if (win < 2) 367 win = 2; 368 tp->snd_cwnd = tp->t_maxseg; 369 tp->snd_ssthresh = win * tp->t_maxseg; 370 tp->t_dupacks = 0; 371 #ifdef TCP_ECN 372 tp->snd_last = tp->snd_max; 373 tp->t_flags |= TF_SEND_CWR; 374 #endif 375 #if 1 /* TCP_ECN */ 376 tcpstat.tcps_cwr_timeout++; 377 #endif 378 } 379 (void) tcp_output(tp); 380 381 out: 382 splx(s); 383 } 384 385 void 386 tcp_timer_persist(void *arg) 387 { 388 struct tcpcb *tp = arg; 389 uint32_t rto; 390 int s; 391 392 s = splsoftnet(); 393 if ((tp->t_flags & TF_DEAD) || 394 TCP_TIMER_ISARMED(tp, TCPT_REXMT)) { 395 splx(s); 396 return; 397 } 398 tcpstat.tcps_persisttimeo++; 399 /* 400 * Hack: if the peer is dead/unreachable, we do not 401 * time out if the window is closed. After a full 402 * backoff, drop the connection if the idle time 403 * (no responses to probes) reaches the maximum 404 * backoff that we would use if retransmitting. 405 */ 406 rto = TCP_REXMTVAL(tp); 407 if (rto < tp->t_rttmin) 408 rto = tp->t_rttmin; 409 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 410 ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle || 411 (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) { 412 tcpstat.tcps_persistdrop++; 413 tp = tcp_drop(tp, ETIMEDOUT); 414 goto out; 415 } 416 tcp_setpersist(tp); 417 tp->t_force = 1; 418 (void) tcp_output(tp); 419 tp->t_force = 0; 420 out: 421 splx(s); 422 } 423 424 void 425 tcp_timer_keep(void *arg) 426 { 427 struct tcpcb *tp = arg; 428 int s; 429 430 s = splsoftnet(); 431 if (tp->t_flags & TF_DEAD) { 432 splx(s); 433 return; 434 } 435 436 tcpstat.tcps_keeptimeo++; 437 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 438 goto dropit; 439 if ((tcp_always_keepalive || 440 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) && 441 tp->t_state <= TCPS_CLOSING) { 442 if ((tcp_maxidle > 0) && 443 ((tcp_now - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)) 444 goto dropit; 445 /* 446 * Send a packet designed to force a response 447 * if the peer is up and reachable: 448 * either an ACK if the connection is still alive, 449 * or an RST if the peer has closed the connection 450 * due to timeout or reboot. 451 * Using sequence number tp->snd_una-1 452 * causes the transmitted zero-length segment 453 * to lie outside the receive window; 454 * by the protocol spec, this requires the 455 * correspondent TCP to respond. 456 */ 457 tcpstat.tcps_keepprobe++; 458 tcp_respond(tp, mtod(tp->t_template, caddr_t), 459 NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0); 460 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl); 461 } else 462 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 463 464 splx(s); 465 return; 466 467 dropit: 468 tcpstat.tcps_keepdrops++; 469 tp = tcp_drop(tp, ETIMEDOUT); 470 471 splx(s); 472 } 473 474 void 475 tcp_timer_2msl(void *arg) 476 { 477 struct tcpcb *tp = arg; 478 int s; 479 480 s = splsoftnet(); 481 if (tp->t_flags & TF_DEAD) { 482 splx(s); 483 return; 484 } 485 486 #ifdef TCP_SACK 487 tcp_timer_freesack(tp); 488 #endif 489 490 if (tp->t_state != TCPS_TIME_WAIT && 491 ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle))) 492 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl); 493 else 494 tp = tcp_close(tp); 495 496 splx(s); 497 } 498