sys/netinet/tcp_timer.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   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 DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. All advertising materials mentioning features or use of this software
  47  *    must display the following acknowledgement:
  48  *      This product includes software developed by the University of
  49  *      California, Berkeley and its contributors.
  50  * 4. Neither the name of the University nor the names of its contributors
  51  *    may be used to endorse or promote products derived from this software
  52  *    without specific prior written permission.
  53  *
  54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  64  * SUCH DAMAGE.
  65  *
  66  *      @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
  67  * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
  68  * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.17 2008/03/30 20:39:01 dillon Exp $
  69  */
  70
  71 #include "opt_compat.h"
  72 #include "opt_inet6.h"
  73 #include "opt_tcpdebug.h"
  74
  75 #include <sys/param.h>
  76 #include <sys/systm.h>
  77 #include <sys/kernel.h>
  78 #include <sys/mbuf.h>
  79 #include <sys/sysctl.h>
  80 #include <sys/socket.h>
  81 #include <sys/socketvar.h>
  82 #include <sys/protosw.h>
  83 #include <sys/thread.h>
  84 #include <sys/globaldata.h>
  85 #include <sys/thread2.h>
  86 #include <sys/msgport2.h>
  87
  88 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
  89
  90 #include <net/route.h>
  91 #include <net/netmsg2.h>
  92
  93 #include <netinet/in.h>
  94 #include <netinet/in_systm.h>
  95 #include <netinet/in_pcb.h>
  96 #ifdef INET6
  97 #include <netinet6/in6_pcb.h>
  98 #endif
  99 #include <netinet/ip_var.h>
 100 #include <netinet/tcp.h>
 101 #include <netinet/tcp_fsm.h>
 102 #include <netinet/tcp_seq.h>
 103 #include <netinet/tcp_timer.h>
 104 #include <netinet/tcp_timer2.h>
 105 #include <netinet/tcp_var.h>
 106 #include <netinet/tcpip.h>
 107 #ifdef TCPDEBUG
 108 #include <netinet/tcp_debug.h>
 109 #endif
 110
 111 #define TCP_TIMER_REXMT         0x01
 112 #define TCP_TIMER_PERSIST       0x02
 113 #define TCP_TIMER_KEEP          0x04
 114 #define TCP_TIMER_2MSL          0x08
 115 #define TCP_TIMER_DELACK        0x10
 116
 117 static struct tcpcb     *tcp_timer_rexmt_handler(struct tcpcb *);
 118 static struct tcpcb     *tcp_timer_persist_handler(struct tcpcb *);
 119 static struct tcpcb     *tcp_timer_keep_handler(struct tcpcb *);
 120 static struct tcpcb     *tcp_timer_2msl_handler(struct tcpcb *);
 121 static struct tcpcb     *tcp_timer_delack_handler(struct tcpcb *);
 122
 123 static const struct tcp_timer {
 124         uint32_t        tt_task;
 125         struct tcpcb    *(*tt_handler)(struct tcpcb *);
 126 } tcp_timer_handlers[] = {
 127         { TCP_TIMER_DELACK,     tcp_timer_delack_handler },
 128         { TCP_TIMER_REXMT,      tcp_timer_rexmt_handler },
 129         { TCP_TIMER_PERSIST,    tcp_timer_persist_handler },
 130         { TCP_TIMER_KEEP,       tcp_timer_keep_handler },
 131         { TCP_TIMER_2MSL,       tcp_timer_2msl_handler },
 132         { 0, NULL }
 133 };
 134
 135 static int
 136 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
 137 {
 138         int error, s, tt;
 139
 140         tt = *(int *)oidp->oid_arg1;
 141         s = (int)((int64_t)tt * 1000 / hz);
 142
 143         error = sysctl_handle_int(oidp, &s, 0, req);
 144         if (error || !req->newptr)
 145                 return (error);
 146
 147         tt = (int)((int64_t)s * hz / 1000);
 148         if (tt < 1)
 149                 return (EINVAL);
 150
 151         *(int *)oidp->oid_arg1 = tt;
 152         return (0);
 153 }
 154
 155 int     tcp_keepinit;
 156 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
 157     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
 158
 159 int     tcp_keepidle;
 160 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
 161     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
 162
 163 int     tcp_keepintvl;
 164 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
 165     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
 166
 167 int     tcp_delacktime;
 168 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
 169     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 170     "Time before a delayed ACK is sent");
 171
 172 int     tcp_msl;
 173 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 174     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 175
 176 int     tcp_rexmit_min;
 177 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 178     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
 179
 180 int     tcp_rexmit_slop;
 181 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 182     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 183     "Retransmission Timer Slop");
 184
 185 static int      always_keepalive = 1;
 186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 187     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 188
 189 static int      tcp_keepcnt = TCPTV_KEEPCNT;
 190         /* max idle probes */
 191 int     tcp_maxpersistidle;
 192         /* max idle time in persist */
 193 int     tcp_maxidle;
 194
 195 /*
 196  * Tcp protocol timeout routine called every 500 ms.
 197  * Updates timestamps used for TCP
 198  * causes finite state machine actions if timers expire.
 199  */
 200 void
 201 tcp_slowtimo(void)
 202 {
 203         crit_enter();
 204         tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
 205         crit_exit();
 206 }
 207
 208 /*
 209  * Cancel all timers for TCP tp.
 210  */
 211 void
 212 tcp_canceltimers(struct tcpcb *tp)
 213 {
 214         tcp_callout_stop(tp, tp->tt_2msl);
 215         tcp_callout_stop(tp, tp->tt_persist);
 216         tcp_callout_stop(tp, tp->tt_keep);
 217         tcp_callout_stop(tp, tp->tt_rexmt);
 218 }
 219
 220 /*
 221  * Caller should be in critical section
 222  */
 223 static void
 224 tcp_send_timermsg(struct tcpcb *tp, uint32_t task)
 225 {
 226         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 227
 228         KKASSERT(tmsg != NULL && tmsg->tt_cpuid == mycpuid &&
 229                  tmsg->tt_tcb != NULL);
 230
 231         tmsg->tt_tasks |= task;
 232         if (tmsg->tt_nmsg.nm_lmsg.ms_flags & MSGF_DONE)
 233                 lwkt_sendmsg(tmsg->tt_msgport, &tmsg->tt_nmsg.nm_lmsg);
 234 }
 235
 236 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 237     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 238
 239 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 240     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 241
 242 static int tcp_totbackoff = 511;        /* sum of tcp_backoff[] */
 243
 244 /* Caller should be in critical section */
 245 static struct tcpcb *
 246 tcp_timer_delack_handler(struct tcpcb *tp)
 247 {
 248         tp->t_flags |= TF_ACKNOW;
 249         tcpstat.tcps_delack++;
 250         tcp_output(tp);
 251         return tp;
 252 }
 253
 254 /*
 255  * TCP timer processing.
 256  */
 257 void
 258 tcp_timer_delack(void *xtp)
 259 {
 260         struct tcpcb *tp = xtp;
 261         struct callout *co = &tp->tt_delack->tc_callout;
 262
 263         crit_enter();
 264         if (callout_pending(co) || !callout_active(co)) {
 265                 crit_exit();
 266                 return;
 267         }
 268         callout_deactivate(co);
 269         tcp_send_timermsg(tp, TCP_TIMER_DELACK);
 270         crit_exit();
 271 }
 272
 273 /* Caller should be in critical section */
 274 static struct tcpcb *
 275 tcp_timer_2msl_handler(struct tcpcb *tp)
 276 {
 277 #ifdef TCPDEBUG
 278         int ostate;
 279 #endif
 280
 281 #ifdef TCPDEBUG
 282         ostate = tp->t_state;
 283 #endif
 284         /*
 285          * 2 MSL timeout in shutdown went off.  If we're closed but
 286          * still waiting for peer to close and connection has been idle
 287          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 288          * control block.  Otherwise, check again in a bit.
 289          */
 290         if (tp->t_state != TCPS_TIME_WAIT &&
 291             (ticks - tp->t_rcvtime) <= tcp_maxidle) {
 292                 tcp_callout_reset(tp, tp->tt_2msl, tcp_keepintvl,
 293                                   tcp_timer_2msl);
 294         } else {
 295                 tp = tcp_close(tp);
 296         }
 297
 298 #ifdef TCPDEBUG
 299         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 300                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 301 #endif
 302         return tp;
 303 }
 304
 305 void
 306 tcp_timer_2msl(void *xtp)
 307 {
 308         struct tcpcb *tp = xtp;
 309         struct callout *co = &tp->tt_2msl->tc_callout;
 310
 311         crit_enter();
 312         if (callout_pending(co) || !callout_active(co)) {
 313                 crit_exit();
 314                 return;
 315         }
 316         callout_deactivate(co);
 317         tcp_send_timermsg(tp, TCP_TIMER_2MSL);
 318         crit_exit();
 319 }
 320
 321 /* Caller should be in critical section */
 322 static struct tcpcb *
 323 tcp_timer_keep_handler(struct tcpcb *tp)
 324 {
 325         struct tcptemp *t_template;
 326 #ifdef TCPDEBUG
 327         int ostate;
 328 #endif
 329
 330 #ifdef TCPDEBUG
 331         ostate = tp->t_state;
 332 #endif
 333         /*
 334          * Keep-alive timer went off; send something
 335          * or drop connection if idle for too long.
 336          */
 337         tcpstat.tcps_keeptimeo++;
 338         if (tp->t_state < TCPS_ESTABLISHED)
 339                 goto dropit;
 340         if ((always_keepalive || (tp->t_flags & TF_KEEPALIVE) ||
 341              (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE)) &&
 342             tp->t_state <= TCPS_CLOSING) {
 343                 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
 344                         goto dropit;
 345                 /*
 346                  * Send a packet designed to force a response
 347                  * if the peer is up and reachable:
 348                  * either an ACK if the connection is still alive,
 349                  * or an RST if the peer has closed the connection
 350                  * due to timeout or reboot.
 351                  * Using sequence number tp->snd_una-1
 352                  * causes the transmitted zero-length segment
 353                  * to lie outside the receive window;
 354                  * by the protocol spec, this requires the
 355                  * correspondent TCP to respond.
 356                  */
 357                 tcpstat.tcps_keepprobe++;
 358                 t_template = tcp_maketemplate(tp);
 359                 if (t_template) {
 360                         tcp_respond(tp, t_template->tt_ipgen,
 361                                     &t_template->tt_t, NULL,
 362                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 363                         tcp_freetemplate(t_template);
 364                 }
 365                 tcp_callout_reset(tp, tp->tt_keep, tcp_keepintvl,
 366                                   tcp_timer_keep);
 367         } else {
 368                 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
 369                                   tcp_timer_keep);
 370         }
 371
 372 #ifdef TCPDEBUG
 373         if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 374                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 375 #endif
 376         return tp;
 377
 378 dropit:
 379         tcpstat.tcps_keepdrops++;
 380         tp = tcp_drop(tp, ETIMEDOUT);
 381
 382 #ifdef TCPDEBUG
 383         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 384                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 385 #endif
 386         return tp;
 387 }
 388
 389 void
 390 tcp_timer_keep(void *xtp)
 391 {
 392         struct tcpcb *tp = xtp;
 393         struct callout *co = &tp->tt_keep->tc_callout;
 394
 395         crit_enter();
 396         if (callout_pending(co) || !callout_active(co)) {
 397                 crit_exit();
 398                 return;
 399         }
 400         callout_deactivate(co);
 401         tcp_send_timermsg(tp, TCP_TIMER_KEEP);
 402         crit_exit();
 403 }
 404
 405 /* Caller should be in critical section */
 406 static struct tcpcb *
 407 tcp_timer_persist_handler(struct tcpcb *tp)
 408 {
 409 #ifdef TCPDEBUG
 410         int ostate;
 411 #endif
 412
 413 #ifdef TCPDEBUG
 414         ostate = tp->t_state;
 415 #endif
 416         /*
 417          * Persistance timer into zero window.
 418          * Force a byte to be output, if possible.
 419          */
 420         tcpstat.tcps_persisttimeo++;
 421         /*
 422          * Hack: if the peer is dead/unreachable, we do not
 423          * time out if the window is closed.  After a full
 424          * backoff, drop the connection if the idle time
 425          * (no responses to probes) reaches the maximum
 426          * backoff that we would use if retransmitting.
 427          */
 428         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 429             ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
 430              (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 431                 tcpstat.tcps_persistdrop++;
 432                 tp = tcp_drop(tp, ETIMEDOUT);
 433                 goto out;
 434         }
 435         tcp_setpersist(tp);
 436         tp->t_flags |= TF_FORCE;
 437         tcp_output(tp);
 438         tp->t_flags &= ~TF_FORCE;
 439
 440 out:
 441 #ifdef TCPDEBUG
 442         if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 443                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 444 #endif
 445         return tp;
 446 }
 447
 448 void
 449 tcp_timer_persist(void *xtp)
 450 {
 451         struct tcpcb *tp = xtp;
 452         struct callout *co = &tp->tt_persist->tc_callout;
 453
 454         crit_enter();
 455         if (callout_pending(co) || !callout_active(co)){
 456                 crit_exit();
 457                 return;
 458         }
 459         callout_deactivate(co);
 460         tcp_send_timermsg(tp, TCP_TIMER_PERSIST);
 461         crit_exit();
 462 }
 463
 464 void
 465 tcp_save_congestion_state(struct tcpcb *tp)
 466 {
 467         tp->snd_cwnd_prev = tp->snd_cwnd;
 468         tp->snd_wacked_prev = tp->snd_wacked;
 469         tp->snd_ssthresh_prev = tp->snd_ssthresh;
 470         tp->snd_recover_prev = tp->snd_recover;
 471         if (IN_FASTRECOVERY(tp))
 472                 tp->t_flags |= TF_WASFRECOVERY;
 473         else
 474                 tp->t_flags &= ~TF_WASFRECOVERY;
 475         if (tp->t_flags & TF_RCVD_TSTMP) {
 476                 tp->t_rexmtTS = ticks;
 477                 tp->t_flags |= TF_FIRSTACCACK;
 478         }
 479 #ifdef later
 480         tcp_sack_save_scoreboard(&tp->scb);
 481 #endif
 482 }
 483
 484 void
 485 tcp_revert_congestion_state(struct tcpcb *tp)
 486 {
 487         tp->snd_cwnd = tp->snd_cwnd_prev;
 488         tp->snd_wacked = tp->snd_wacked_prev;
 489         tp->snd_ssthresh = tp->snd_ssthresh_prev;
 490         tp->snd_recover = tp->snd_recover_prev;
 491         if (tp->t_flags & TF_WASFRECOVERY)
 492                 ENTER_FASTRECOVERY(tp);
 493         if (tp->t_flags & TF_FASTREXMT) {
 494                 ++tcpstat.tcps_sndfastrexmitbad;
 495                 if (tp->t_flags & TF_EARLYREXMT)
 496                         ++tcpstat.tcps_sndearlyrexmitbad;
 497         } else
 498                 ++tcpstat.tcps_sndrtobad;
 499         tp->t_badrxtwin = 0;
 500         tp->t_rxtshift = 0;
 501         tp->snd_nxt = tp->snd_max;
 502 #ifdef later
 503         tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una);
 504 #endif
 505 }
 506
 507 /* Caller should be in critical section */
 508 static struct tcpcb *
 509 tcp_timer_rexmt_handler(struct tcpcb *tp)
 510 {
 511         int rexmt;
 512 #ifdef TCPDEBUG
 513         int ostate;
 514 #endif
 515
 516 #ifdef TCPDEBUG
 517         ostate = tp->t_state;
 518 #endif
 519         /*
 520          * Retransmission timer went off.  Message has not
 521          * been acked within retransmit interval.  Back off
 522          * to a longer retransmit interval and retransmit one segment.
 523          */
 524         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 525                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 526                 tcpstat.tcps_timeoutdrop++;
 527                 tp = tcp_drop(tp, tp->t_softerror ?
 528                               tp->t_softerror : ETIMEDOUT);
 529                 goto out;
 530         }
 531         if (tp->t_rxtshift == 1) {
 532                 /*
 533                  * first retransmit; record ssthresh and cwnd so they can
 534                  * be recovered if this turns out to be a "bad" retransmit.
 535                  * A retransmit is considered "bad" if an ACK for this
 536                  * segment is received within RTT/2 interval; the assumption
 537                  * here is that the ACK was already in flight.  See
 538                  * "On Estimating End-to-End Network Path Properties" by
 539                  * Allman and Paxson for more details.
 540                  */
 541                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 542                 tcp_save_congestion_state(tp);
 543                 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
 544         }
 545         /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
 546         tcp_sack_cleanup(&tp->scb);
 547         tcpstat.tcps_rexmttimeo++;
 548         if (tp->t_state == TCPS_SYN_SENT)
 549                 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
 550         else
 551                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 552         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 553                       tp->t_rttmin, TCPTV_REXMTMAX);
 554         /*
 555          * Disable rfc1323 if we havn't got any response to
 556          * our third SYN to work-around some broken terminal servers
 557          * (most of which have hopefully been retired) that have bad VJ
 558          * header compression code which trashes TCP segments containing
 559          * unknown-to-them TCP options.
 560          */
 561         if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
 562                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
 563         /*
 564          * If losing, let the lower level know and try for
 565          * a better route.  Also, if we backed off this far,
 566          * our srtt estimate is probably bogus.  Clobber it
 567          * so we'll take the next rtt measurement as our srtt;
 568          * move the current srtt into rttvar to keep the current
 569          * retransmit times until then.
 570          */
 571         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 572 #ifdef INET6
 573                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 574                         in6_losing(tp->t_inpcb);
 575                 else
 576 #endif
 577                 in_losing(tp->t_inpcb);
 578                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 579                 tp->t_srtt = 0;
 580         }
 581         tp->snd_nxt = tp->snd_una;
 582         tp->rexmt_high = tp->snd_una;
 583         tp->snd_recover = tp->snd_max;
 584         /*
 585          * Force a segment to be sent.
 586          */
 587         tp->t_flags |= TF_ACKNOW;
 588         /*
 589          * If timing a segment in this window, stop the timer.
 590          */
 591         tp->t_rtttime = 0;
 592         /*
 593          * Close the congestion window down to one segment
 594          * (we'll open it by one segment for each ack we get).
 595          * Since we probably have a window's worth of unacked
 596          * data accumulated, this "slow start" keeps us from
 597          * dumping all that data as back-to-back packets (which
 598          * might overwhelm an intermediate gateway).
 599          *
 600          * There are two phases to the opening: Initially we
 601          * open by one mss on each ack.  This makes the window
 602          * size increase exponentially with time.  If the
 603          * window is larger than the path can handle, this
 604          * exponential growth results in dropped packet(s)
 605          * almost immediately.  To get more time between
 606          * drops but still "push" the network to take advantage
 607          * of improving conditions, we switch from exponential
 608          * to linear window opening at some threshhold size.
 609          * For a threshhold, we use half the current window
 610          * size, truncated to a multiple of the mss.
 611          *
 612          * (the minimum cwnd that will give us exponential
 613          * growth is 2 mss.  We don't allow the threshhold
 614          * to go below this.)
 615          */
 616         {
 617                 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 618
 619                 if (win < 2)
 620                         win = 2;
 621                 tp->snd_cwnd = tp->t_maxseg;
 622                 tp->snd_wacked = 0;
 623                 tp->snd_ssthresh = win * tp->t_maxseg;
 624                 tp->t_dupacks = 0;
 625         }
 626         EXIT_FASTRECOVERY(tp);
 627         tcp_output(tp);
 628
 629 out:
 630 #ifdef TCPDEBUG
 631         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 632                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 633 #endif
 634         return tp;
 635 }
 636
 637 void
 638 tcp_timer_rexmt(void *xtp)
 639 {
 640         struct tcpcb *tp = xtp;
 641         struct callout *co = &tp->tt_rexmt->tc_callout;
 642
 643         crit_enter();
 644         if (callout_pending(co) || !callout_active(co)) {
 645                 crit_exit();
 646                 return;
 647         }
 648         callout_deactivate(co);
 649         tcp_send_timermsg(tp, TCP_TIMER_REXMT);
 650         crit_exit();
 651 }
 652
 653 static void
 654 tcp_timer_handler(struct netmsg *nmsg)
 655 {
 656         struct netmsg_tcp_timer *tmsg = (struct netmsg_tcp_timer *)nmsg;
 657         const struct tcp_timer *tt;
 658         struct tcpcb *tp;
 659
 660         crit_enter();
 661
 662         KKASSERT(tmsg->tt_cpuid == mycpuid && tmsg->tt_tcb != NULL);
 663         tp = tmsg->tt_tcb;
 664
 665         /* Save pending tasks and reset the tasks in message */
 666         tmsg->tt_running_tasks = tmsg->tt_tasks;
 667         tmsg->tt_prev_tasks = tmsg->tt_tasks;
 668         tmsg->tt_tasks = 0;
 669
 670         /* Reply ASAP */
 671         lwkt_replymsg(&tmsg->tt_nmsg.nm_lmsg, 0);
 672
 673         if (tmsg->tt_running_tasks == 0) {
 674                 /*
 675                  * All of the timers are cancelled when the message
 676                  * is pending; bail out.
 677                  */
 678                 crit_exit();
 679                 return;
 680         }
 681
 682         for (tt = tcp_timer_handlers; tt->tt_handler != NULL; ++tt) {
 683                 if ((tmsg->tt_running_tasks & tt->tt_task) == 0)
 684                         continue;
 685
 686                 tmsg->tt_running_tasks &= ~tt->tt_task;
 687                 tp = tt->tt_handler(tp);
 688                 if (tp == NULL)
 689                         break;
 690
 691                 if (tmsg->tt_running_tasks == 0) /* nothing left to do */
 692                         break;
 693         }
 694
 695         crit_exit();
 696 }
 697
 698 void
 699 tcp_create_timermsg(struct tcpcb *tp, struct lwkt_port *msgport)
 700 {
 701         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 702
 703         netmsg_init(&tmsg->tt_nmsg, NULL, &netisr_adone_rport,
 704                     MSGF_DROPABLE | MSGF_PRIORITY, tcp_timer_handler);
 705         tmsg->tt_cpuid = mycpuid;
 706         tmsg->tt_msgport = msgport;
 707         tmsg->tt_tcb = tp;
 708         tmsg->tt_tasks = 0;
 709 }
 710
 711 void
 712 tcp_destroy_timermsg(struct tcpcb *tp)
 713 {
 714         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 715
 716         if (tmsg == NULL ||             /* listen socket */
 717             tmsg->tt_tcb == NULL)       /* only tcp_attach() is called */
 718                 return;
 719
 720         KKASSERT(tmsg->tt_cpuid == mycpuid);
 721         crit_enter();
 722         if ((tmsg->tt_nmsg.nm_lmsg.ms_flags & MSGF_DONE) == 0) {
 723                 /*
 724                  * This message is still pending to be processed;
 725                  * drop it.
 726                  */
 727                 lwkt_dropmsg(&tmsg->tt_nmsg.nm_lmsg);
 728         }
 729         crit_exit();
 730 }
 731
 732 static __inline void
 733 tcp_callout_init(struct tcp_callout *tc, uint32_t task)
 734 {
 735         callout_init(&tc->tc_callout);
 736         tc->tc_task = task;
 737 }
 738
 739 void
 740 tcp_inittimers(struct tcpcb *tp)
 741 {
 742         tcp_callout_init(tp->tt_rexmt, TCP_TIMER_REXMT);
 743         tcp_callout_init(tp->tt_persist, TCP_TIMER_PERSIST);
 744         tcp_callout_init(tp->tt_keep, TCP_TIMER_KEEP);
 745         tcp_callout_init(tp->tt_2msl, TCP_TIMER_2MSL);
 746         tcp_callout_init(tp->tt_delack, TCP_TIMER_DELACK);
 747 }