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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_inet6.h" 36 #include "opt_tcpdebug.h" 37 38 #include <sys/param.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/mbuf.h> 42 #include <sys/mutex.h> 43 #include <sys/protosw.h> 44 #include <sys/smp.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/sysctl.h> 48 #include <sys/systm.h> 49 50 #include <net/if.h> 51 #include <net/route.h> 52 #include <net/vnet.h> 53 54 #include <netinet/in.h> 55 #include <netinet/in_pcb.h> 56 #include <netinet/in_systm.h> 57 #ifdef INET6 58 #include <netinet6/in6_pcb.h> 59 #endif 60 #include <netinet/ip_var.h> 61 #include <netinet/tcp.h> 62 #include <netinet/tcp_fsm.h> 63 #include <netinet/tcp_timer.h> 64 #include <netinet/tcp_var.h> 65 #include <netinet/tcpip.h> 66 #ifdef TCPDEBUG 67 #include <netinet/tcp_debug.h> 68 #endif 69 70 int tcp_keepinit; 71 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW, 72 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection"); 73 74 int tcp_keepidle; 75 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW, 76 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin"); 77 78 int tcp_keepintvl; 79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW, 80 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes"); 81 82 int tcp_delacktime; 83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW, 84 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I", 85 "Time before a delayed ACK is sent"); 86 87 int tcp_msl; 88 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW, 89 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime"); 90 91 int tcp_rexmit_min; 92 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW, 93 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", 94 "Minimum Retransmission Timeout"); 95 96 int tcp_rexmit_slop; 97 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW, 98 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", 99 "Retransmission Timer Slop"); 100 101 static int always_keepalive = 1; 102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, 103 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections"); 104 105 int tcp_fast_finwait2_recycle = 0; 106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW, 107 &tcp_fast_finwait2_recycle, 0, 108 "Recycle closed FIN_WAIT_2 connections faster"); 109 110 int tcp_finwait2_timeout; 111 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW, 112 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout"); 113 114 115 static int tcp_keepcnt = TCPTV_KEEPCNT; 116 /* max idle probes */ 117 int tcp_maxpersistidle; 118 /* max idle time in persist */ 119 int tcp_maxidle; 120 121 static int per_cpu_timers = 0; 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW, 123 &per_cpu_timers , 0, "run tcp timers on all cpus"); 124 125 #define INP_CPU(inp) (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \ 126 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0) 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(void) 135 { 136 VNET_ITERATOR_DECL(vnet_iter); 137 138 VNET_LIST_RLOCK_NOSLEEP(); 139 VNET_FOREACH(vnet_iter) { 140 CURVNET_SET(vnet_iter); 141 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 142 INP_INFO_WLOCK(&V_tcbinfo); 143 (void) tcp_tw_2msl_scan(0); 144 INP_INFO_WUNLOCK(&V_tcbinfo); 145 CURVNET_RESTORE(); 146 } 147 VNET_LIST_RUNLOCK_NOSLEEP(); 148 } 149 150 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] = 151 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 }; 152 153 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 154 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 }; 155 156 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */ 157 158 static int tcp_timer_race; 159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race, 160 0, "Count of t_inpcb races on tcp_discardcb"); 161 162 /* 163 * TCP timer processing. 164 */ 165 166 void 167 tcp_timer_delack(void *xtp) 168 { 169 struct tcpcb *tp = xtp; 170 struct inpcb *inp; 171 CURVNET_SET(tp->t_vnet); 172 173 inp = tp->t_inpcb; 174 /* 175 * XXXRW: While this assert is in fact correct, bugs in the tcpcb 176 * tear-down mean we need it as a work-around for races between 177 * timers and tcp_discardcb(). 178 * 179 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL")); 180 */ 181 if (inp == NULL) { 182 tcp_timer_race++; 183 CURVNET_RESTORE(); 184 return; 185 } 186 INP_WLOCK(inp); 187 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack) 188 || !callout_active(&tp->t_timers->tt_delack)) { 189 INP_WUNLOCK(inp); 190 CURVNET_RESTORE(); 191 return; 192 } 193 callout_deactivate(&tp->t_timers->tt_delack); 194 195 tp->t_flags |= TF_ACKNOW; 196 TCPSTAT_INC(tcps_delack); 197 (void) tcp_output(tp); 198 INP_WUNLOCK(inp); 199 CURVNET_RESTORE(); 200 } 201 202 void 203 tcp_timer_2msl(void *xtp) 204 { 205 struct tcpcb *tp = xtp; 206 struct inpcb *inp; 207 CURVNET_SET(tp->t_vnet); 208 #ifdef TCPDEBUG 209 int ostate; 210 211 ostate = tp->t_state; 212 #endif 213 /* 214 * XXXRW: Does this actually happen? 215 */ 216 INP_INFO_WLOCK(&V_tcbinfo); 217 inp = tp->t_inpcb; 218 /* 219 * XXXRW: While this assert is in fact correct, bugs in the tcpcb 220 * tear-down mean we need it as a work-around for races between 221 * timers and tcp_discardcb(). 222 * 223 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL")); 224 */ 225 if (inp == NULL) { 226 tcp_timer_race++; 227 INP_INFO_WUNLOCK(&V_tcbinfo); 228 CURVNET_RESTORE(); 229 return; 230 } 231 INP_WLOCK(inp); 232 tcp_free_sackholes(tp); 233 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) || 234 !callout_active(&tp->t_timers->tt_2msl)) { 235 INP_WUNLOCK(tp->t_inpcb); 236 INP_INFO_WUNLOCK(&V_tcbinfo); 237 CURVNET_RESTORE(); 238 return; 239 } 240 callout_deactivate(&tp->t_timers->tt_2msl); 241 /* 242 * 2 MSL timeout in shutdown went off. If we're closed but 243 * still waiting for peer to close and connection has been idle 244 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 245 * control block. Otherwise, check again in a bit. 246 * 247 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed, 248 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it. 249 * Ignore fact that there were recent incoming segments. 250 */ 251 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 && 252 tp->t_inpcb && tp->t_inpcb->inp_socket && 253 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) { 254 TCPSTAT_INC(tcps_finwait2_drops); 255 tp = tcp_close(tp); 256 } else { 257 if (tp->t_state != TCPS_TIME_WAIT && 258 ticks - tp->t_rcvtime <= tcp_maxidle) 259 callout_reset_on(&tp->t_timers->tt_2msl, tcp_keepintvl, 260 tcp_timer_2msl, tp, INP_CPU(inp)); 261 else 262 tp = tcp_close(tp); 263 } 264 265 #ifdef TCPDEBUG 266 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 267 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 268 PRU_SLOWTIMO); 269 #endif 270 if (tp != NULL) 271 INP_WUNLOCK(inp); 272 INP_INFO_WUNLOCK(&V_tcbinfo); 273 CURVNET_RESTORE(); 274 } 275 276 void 277 tcp_timer_keep(void *xtp) 278 { 279 struct tcpcb *tp = xtp; 280 struct tcptemp *t_template; 281 struct inpcb *inp; 282 CURVNET_SET(tp->t_vnet); 283 #ifdef TCPDEBUG 284 int ostate; 285 286 ostate = tp->t_state; 287 #endif 288 INP_INFO_WLOCK(&V_tcbinfo); 289 inp = tp->t_inpcb; 290 /* 291 * XXXRW: While this assert is in fact correct, bugs in the tcpcb 292 * tear-down mean we need it as a work-around for races between 293 * timers and tcp_discardcb(). 294 * 295 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL")); 296 */ 297 if (inp == NULL) { 298 tcp_timer_race++; 299 INP_INFO_WUNLOCK(&V_tcbinfo); 300 CURVNET_RESTORE(); 301 return; 302 } 303 INP_WLOCK(inp); 304 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep) 305 || !callout_active(&tp->t_timers->tt_keep)) { 306 INP_WUNLOCK(inp); 307 INP_INFO_WUNLOCK(&V_tcbinfo); 308 CURVNET_RESTORE(); 309 return; 310 } 311 callout_deactivate(&tp->t_timers->tt_keep); 312 /* 313 * Keep-alive timer went off; send something 314 * or drop connection if idle for too long. 315 */ 316 TCPSTAT_INC(tcps_keeptimeo); 317 if (tp->t_state < TCPS_ESTABLISHED) 318 goto dropit; 319 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) && 320 tp->t_state <= TCPS_CLOSING) { 321 if (ticks - tp->t_rcvtime >= tcp_keepidle + tcp_maxidle) 322 goto dropit; 323 /* 324 * Send a packet designed to force a response 325 * if the peer is up and reachable: 326 * either an ACK if the connection is still alive, 327 * or an RST if the peer has closed the connection 328 * due to timeout or reboot. 329 * Using sequence number tp->snd_una-1 330 * causes the transmitted zero-length segment 331 * to lie outside the receive window; 332 * by the protocol spec, this requires the 333 * correspondent TCP to respond. 334 */ 335 TCPSTAT_INC(tcps_keepprobe); 336 t_template = tcpip_maketemplate(inp); 337 if (t_template) { 338 tcp_respond(tp, t_template->tt_ipgen, 339 &t_template->tt_t, (struct mbuf *)NULL, 340 tp->rcv_nxt, tp->snd_una - 1, 0); 341 free(t_template, M_TEMP); 342 } 343 callout_reset_on(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp, INP_CPU(inp)); 344 } else 345 callout_reset_on(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp, INP_CPU(inp)); 346 347 #ifdef TCPDEBUG 348 if (inp->inp_socket->so_options & SO_DEBUG) 349 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 350 PRU_SLOWTIMO); 351 #endif 352 INP_WUNLOCK(inp); 353 INP_INFO_WUNLOCK(&V_tcbinfo); 354 CURVNET_RESTORE(); 355 return; 356 357 dropit: 358 TCPSTAT_INC(tcps_keepdrops); 359 tp = tcp_drop(tp, ETIMEDOUT); 360 361 #ifdef TCPDEBUG 362 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 363 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 364 PRU_SLOWTIMO); 365 #endif 366 if (tp != NULL) 367 INP_WUNLOCK(tp->t_inpcb); 368 INP_INFO_WUNLOCK(&V_tcbinfo); 369 CURVNET_RESTORE(); 370 } 371 372 void 373 tcp_timer_persist(void *xtp) 374 { 375 struct tcpcb *tp = xtp; 376 struct inpcb *inp; 377 CURVNET_SET(tp->t_vnet); 378 #ifdef TCPDEBUG 379 int ostate; 380 381 ostate = tp->t_state; 382 #endif 383 INP_INFO_WLOCK(&V_tcbinfo); 384 inp = tp->t_inpcb; 385 /* 386 * XXXRW: While this assert is in fact correct, bugs in the tcpcb 387 * tear-down mean we need it as a work-around for races between 388 * timers and tcp_discardcb(). 389 * 390 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL")); 391 */ 392 if (inp == NULL) { 393 tcp_timer_race++; 394 INP_INFO_WUNLOCK(&V_tcbinfo); 395 CURVNET_RESTORE(); 396 return; 397 } 398 INP_WLOCK(inp); 399 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist) 400 || !callout_active(&tp->t_timers->tt_persist)) { 401 INP_WUNLOCK(inp); 402 INP_INFO_WUNLOCK(&V_tcbinfo); 403 CURVNET_RESTORE(); 404 return; 405 } 406 callout_deactivate(&tp->t_timers->tt_persist); 407 /* 408 * Persistance timer into zero window. 409 * Force a byte to be output, if possible. 410 */ 411 TCPSTAT_INC(tcps_persisttimeo); 412 /* 413 * Hack: if the peer is dead/unreachable, we do not 414 * time out if the window is closed. After a full 415 * backoff, drop the connection if the idle time 416 * (no responses to probes) reaches the maximum 417 * backoff that we would use if retransmitting. 418 */ 419 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 420 (ticks - tp->t_rcvtime >= tcp_maxpersistidle || 421 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { 422 TCPSTAT_INC(tcps_persistdrop); 423 tp = tcp_drop(tp, ETIMEDOUT); 424 goto out; 425 } 426 tcp_setpersist(tp); 427 tp->t_flags |= TF_FORCEDATA; 428 (void) tcp_output(tp); 429 tp->t_flags &= ~TF_FORCEDATA; 430 431 out: 432 #ifdef TCPDEBUG 433 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 434 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); 435 #endif 436 if (tp != NULL) 437 INP_WUNLOCK(inp); 438 INP_INFO_WUNLOCK(&V_tcbinfo); 439 CURVNET_RESTORE(); 440 } 441 442 void 443 tcp_timer_rexmt(void * xtp) 444 { 445 struct tcpcb *tp = xtp; 446 CURVNET_SET(tp->t_vnet); 447 int rexmt; 448 int headlocked; 449 struct inpcb *inp; 450 #ifdef TCPDEBUG 451 int ostate; 452 453 ostate = tp->t_state; 454 #endif 455 INP_INFO_RLOCK(&V_tcbinfo); 456 inp = tp->t_inpcb; 457 /* 458 * XXXRW: While this assert is in fact correct, bugs in the tcpcb 459 * tear-down mean we need it as a work-around for races between 460 * timers and tcp_discardcb(). 461 * 462 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL")); 463 */ 464 if (inp == NULL) { 465 tcp_timer_race++; 466 INP_INFO_RUNLOCK(&V_tcbinfo); 467 CURVNET_RESTORE(); 468 return; 469 } 470 INP_WLOCK(inp); 471 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt) 472 || !callout_active(&tp->t_timers->tt_rexmt)) { 473 INP_WUNLOCK(inp); 474 INP_INFO_RUNLOCK(&V_tcbinfo); 475 CURVNET_RESTORE(); 476 return; 477 } 478 callout_deactivate(&tp->t_timers->tt_rexmt); 479 tcp_free_sackholes(tp); 480 /* 481 * Retransmission timer went off. Message has not 482 * been acked within retransmit interval. Back off 483 * to a longer retransmit interval and retransmit one segment. 484 */ 485 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 486 tp->t_rxtshift = TCP_MAXRXTSHIFT; 487 TCPSTAT_INC(tcps_timeoutdrop); 488 in_pcbref(inp); 489 INP_INFO_RUNLOCK(&V_tcbinfo); 490 INP_WUNLOCK(inp); 491 INP_INFO_WLOCK(&V_tcbinfo); 492 INP_WLOCK(inp); 493 if (in_pcbrele(inp)) { 494 INP_INFO_WUNLOCK(&V_tcbinfo); 495 CURVNET_RESTORE(); 496 return; 497 } 498 tp = tcp_drop(tp, tp->t_softerror ? 499 tp->t_softerror : ETIMEDOUT); 500 headlocked = 1; 501 goto out; 502 } 503 INP_INFO_RUNLOCK(&V_tcbinfo); 504 headlocked = 0; 505 if (tp->t_rxtshift == 1) { 506 /* 507 * first retransmit; record ssthresh and cwnd so they can 508 * be recovered if this turns out to be a "bad" retransmit. 509 * A retransmit is considered "bad" if an ACK for this 510 * segment is received within RTT/2 interval; the assumption 511 * here is that the ACK was already in flight. See 512 * "On Estimating End-to-End Network Path Properties" by 513 * Allman and Paxson for more details. 514 */ 515 tp->snd_cwnd_prev = tp->snd_cwnd; 516 tp->snd_ssthresh_prev = tp->snd_ssthresh; 517 tp->snd_recover_prev = tp->snd_recover; 518 if (IN_FASTRECOVERY(tp)) 519 tp->t_flags |= TF_WASFRECOVERY; 520 else 521 tp->t_flags &= ~TF_WASFRECOVERY; 522 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); 523 } 524 TCPSTAT_INC(tcps_rexmttimeo); 525 if (tp->t_state == TCPS_SYN_SENT) 526 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift]; 527 else 528 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 529 TCPT_RANGESET(tp->t_rxtcur, rexmt, 530 tp->t_rttmin, TCPTV_REXMTMAX); 531 /* 532 * Disable rfc1323 if we havn't got any response to 533 * our third SYN to work-around some broken terminal servers 534 * (most of which have hopefully been retired) that have bad VJ 535 * header compression code which trashes TCP segments containing 536 * unknown-to-them TCP options. 537 */ 538 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3)) 539 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP); 540 /* 541 * If we backed off this far, our srtt estimate is probably bogus. 542 * Clobber it so we'll take the next rtt measurement as our srtt; 543 * move the current srtt into rttvar to keep the current 544 * retransmit times until then. 545 */ 546 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 547 #ifdef INET6 548 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 549 in6_losing(tp->t_inpcb); 550 else 551 #endif 552 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 553 tp->t_srtt = 0; 554 } 555 tp->snd_nxt = tp->snd_una; 556 tp->snd_recover = tp->snd_max; 557 /* 558 * Force a segment to be sent. 559 */ 560 tp->t_flags |= TF_ACKNOW; 561 /* 562 * If timing a segment in this window, stop the timer. 563 */ 564 tp->t_rtttime = 0; 565 /* 566 * Close the congestion window down to one segment 567 * (we'll open it by one segment for each ack we get). 568 * Since we probably have a window's worth of unacked 569 * data accumulated, this "slow start" keeps us from 570 * dumping all that data as back-to-back packets (which 571 * might overwhelm an intermediate gateway). 572 * 573 * There are two phases to the opening: Initially we 574 * open by one mss on each ack. This makes the window 575 * size increase exponentially with time. If the 576 * window is larger than the path can handle, this 577 * exponential growth results in dropped packet(s) 578 * almost immediately. To get more time between 579 * drops but still "push" the network to take advantage 580 * of improving conditions, we switch from exponential 581 * to linear window opening at some threshhold size. 582 * For a threshhold, we use half the current window 583 * size, truncated to a multiple of the mss. 584 * 585 * (the minimum cwnd that will give us exponential 586 * growth is 2 mss. We don't allow the threshhold 587 * to go below this.) 588 */ 589 { 590 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 591 if (win < 2) 592 win = 2; 593 tp->snd_cwnd = tp->t_maxseg; 594 tp->snd_ssthresh = win * tp->t_maxseg; 595 tp->t_dupacks = 0; 596 } 597 EXIT_FASTRECOVERY(tp); 598 tp->t_bytes_acked = 0; 599 (void) tcp_output(tp); 600 601 out: 602 #ifdef TCPDEBUG 603 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 604 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 605 PRU_SLOWTIMO); 606 #endif 607 if (tp != NULL) 608 INP_WUNLOCK(inp); 609 if (headlocked) 610 INP_INFO_WUNLOCK(&V_tcbinfo); 611 CURVNET_RESTORE(); 612 } 613 614 void 615 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta) 616 { 617 struct callout *t_callout; 618 void *f_callout; 619 struct inpcb *inp = tp->t_inpcb; 620 int cpu = INP_CPU(inp); 621 622 switch (timer_type) { 623 case TT_DELACK: 624 t_callout = &tp->t_timers->tt_delack; 625 f_callout = tcp_timer_delack; 626 break; 627 case TT_REXMT: 628 t_callout = &tp->t_timers->tt_rexmt; 629 f_callout = tcp_timer_rexmt; 630 break; 631 case TT_PERSIST: 632 t_callout = &tp->t_timers->tt_persist; 633 f_callout = tcp_timer_persist; 634 break; 635 case TT_KEEP: 636 t_callout = &tp->t_timers->tt_keep; 637 f_callout = tcp_timer_keep; 638 break; 639 case TT_2MSL: 640 t_callout = &tp->t_timers->tt_2msl; 641 f_callout = tcp_timer_2msl; 642 break; 643 default: 644 panic("bad timer_type"); 645 } 646 if (delta == 0) { 647 callout_stop(t_callout); 648 } else { 649 callout_reset_on(t_callout, delta, f_callout, tp, cpu); 650 } 651 } 652 653 int 654 tcp_timer_active(struct tcpcb *tp, int timer_type) 655 { 656 struct callout *t_callout; 657 658 switch (timer_type) { 659 case TT_DELACK: 660 t_callout = &tp->t_timers->tt_delack; 661 break; 662 case TT_REXMT: 663 t_callout = &tp->t_timers->tt_rexmt; 664 break; 665 case TT_PERSIST: 666 t_callout = &tp->t_timers->tt_persist; 667 break; 668 case TT_KEEP: 669 t_callout = &tp->t_timers->tt_keep; 670 break; 671 case TT_2MSL: 672 t_callout = &tp->t_timers->tt_2msl; 673 break; 674 default: 675 panic("bad timer_type"); 676 } 677 return callout_active(t_callout); 678 } 679 680 #define ticks_to_msecs(t) (1000*(t) / hz) 681 682 void 683 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer, struct xtcp_timer *xtimer) 684 { 685 bzero(xtimer, sizeof(struct xtcp_timer)); 686 if (timer == NULL) 687 return; 688 if (callout_active(&timer->tt_delack)) 689 xtimer->tt_delack = ticks_to_msecs(timer->tt_delack.c_time - ticks); 690 if (callout_active(&timer->tt_rexmt)) 691 xtimer->tt_rexmt = ticks_to_msecs(timer->tt_rexmt.c_time - ticks); 692 if (callout_active(&timer->tt_persist)) 693 xtimer->tt_persist = ticks_to_msecs(timer->tt_persist.c_time - ticks); 694 if (callout_active(&timer->tt_keep)) 695 xtimer->tt_keep = ticks_to_msecs(timer->tt_keep.c_time - ticks); 696 if (callout_active(&timer->tt_2msl)) 697 xtimer->tt_2msl = ticks_to_msecs(timer->tt_2msl.c_time - ticks); 698 xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime); 699 } 700