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. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
  63  * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
  64  */
  65
  66 #include "opt_inet6.h"
  67 #include "opt_tcpdebug.h"
  68
  69 #include <sys/param.h>
  70 #include <sys/systm.h>
  71 #include <sys/kernel.h>
  72 #include <sys/mbuf.h>
  73 #include <sys/sysctl.h>
  74 #include <sys/socket.h>
  75 #include <sys/socketvar.h>
  76 #include <sys/protosw.h>
  77 #include <sys/thread.h>
  78 #include <sys/globaldata.h>
  79 #include <sys/thread2.h>
  80 #include <sys/msgport2.h>
  81
  82 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
  83
  84 #include <net/route.h>
  85 #include <net/netmsg2.h>
  86
  87 #include <netinet/in.h>
  88 #include <netinet/in_systm.h>
  89 #include <netinet/in_pcb.h>
  90 #ifdef INET6
  91 #include <netinet6/in6_pcb.h>
  92 #endif
  93 #include <netinet/ip_var.h>
  94 #include <netinet/tcp.h>
  95 #include <netinet/tcp_fsm.h>
  96 #include <netinet/tcp_seq.h>
  97 #include <netinet/tcp_timer.h>
  98 #include <netinet/tcp_timer2.h>
  99 #include <netinet/tcp_var.h>
 100 #include <netinet/tcpip.h>
 101 #ifdef TCPDEBUG
 102 #include <netinet/tcp_debug.h>
 103 #endif
 104
 105 #define TCP_TIMER_REXMT         0x01
 106 #define TCP_TIMER_PERSIST       0x02
 107 #define TCP_TIMER_KEEP          0x04
 108 #define TCP_TIMER_2MSL          0x08
 109 #define TCP_TIMER_DELACK        0x10
 110
 111 static struct tcpcb     *tcp_timer_rexmt_handler(struct tcpcb *);
 112 static struct tcpcb     *tcp_timer_persist_handler(struct tcpcb *);
 113 static struct tcpcb     *tcp_timer_keep_handler(struct tcpcb *);
 114 static struct tcpcb     *tcp_timer_2msl_handler(struct tcpcb *);
 115 static struct tcpcb     *tcp_timer_delack_handler(struct tcpcb *);
 116
 117 static const struct tcp_timer {
 118         uint32_t        tt_task;
 119         struct tcpcb    *(*tt_handler)(struct tcpcb *);
 120 } tcp_timer_handlers[] = {
 121         { TCP_TIMER_DELACK,     tcp_timer_delack_handler },
 122         { TCP_TIMER_REXMT,      tcp_timer_rexmt_handler },
 123         { TCP_TIMER_PERSIST,    tcp_timer_persist_handler },
 124         { TCP_TIMER_KEEP,       tcp_timer_keep_handler },
 125         { TCP_TIMER_2MSL,       tcp_timer_2msl_handler },
 126         { 0, NULL }
 127 };
 128
 129 static int
 130 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
 131 {
 132         int error, s, tt;
 133
 134         tt = *(int *)oidp->oid_arg1;
 135         s = (int)((int64_t)tt * 1000 / hz);
 136
 137         error = sysctl_handle_int(oidp, &s, 0, req);
 138         if (error || !req->newptr)
 139                 return (error);
 140
 141         tt = (int)((int64_t)s * hz / 1000);
 142         if (tt < 1)
 143                 return (EINVAL);
 144
 145         *(int *)oidp->oid_arg1 = tt;
 146         return (0);
 147 }
 148
 149 int     tcp_keepinit;
 150 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
 151     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "Time to establish TCP connection");
 152
 153 int     tcp_keepidle;
 154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
 155     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "Time before TCP keepalive probes begin");
 156
 157 int     tcp_keepintvl;
 158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
 159     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "Time between TCP keepalive probes");
 160
 161 int     tcp_delacktime;
 162 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
 163     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 164     "Time before a delayed ACK is sent");
 165
 166 int     tcp_msl;
 167 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 168     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 169
 170 int     tcp_rexmit_min;
 171 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 172     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
 173
 174 int     tcp_rexmit_slop;
 175 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 176     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 177     "Retransmission Timer Slop");
 178
 179 static int      always_keepalive = 1;
 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 181     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 182
 183 /* max idle probes */
 184 int     tcp_keepcnt = TCPTV_KEEPCNT;
 185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW,
 186     &tcp_keepcnt, 0, "Maximum number of keepalive probes to be sent");
 187
 188 static int tcp_do_eifel_response = 1;
 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel_response, CTLFLAG_RW,
 190     &tcp_do_eifel_response, 0, "Eifel response algorithm (RFC 4015)");
 191
 192 int tcp_eifel_rtoinc = 2;
 193 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, eifel_rtoinc, CTLTYPE_INT|CTLFLAG_RW,
 194     &tcp_eifel_rtoinc, 0, sysctl_msec_to_ticks, "I",
 195     "Eifel response RTO increment");
 196
 197 /* max idle time in persist */
 198 int     tcp_maxpersistidle;
 199
 200 /*
 201  * Cancel all timers for TCP tp.
 202  */
 203 void
 204 tcp_canceltimers(struct tcpcb *tp)
 205 {
 206         tcp_callout_stop(tp, tp->tt_2msl);
 207         tcp_callout_stop(tp, tp->tt_persist);
 208         tcp_callout_stop(tp, tp->tt_keep);
 209         tcp_callout_stop(tp, tp->tt_rexmt);
 210 }
 211
 212 /*
 213  * Caller should be in critical section
 214  */
 215 static void
 216 tcp_send_timermsg(struct tcpcb *tp, uint32_t task)
 217 {
 218         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 219
 220         KKASSERT(tmsg != NULL && tmsg->tt_cpuid == mycpuid &&
 221                  tmsg->tt_tcb != NULL);
 222
 223         tmsg->tt_tasks |= task;
 224         if (tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE)
 225                 lwkt_sendmsg_oncpu(tmsg->tt_msgport, &tmsg->tt_msg.lmsg);
 226 }
 227
 228 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 229     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 230
 231 int     tcp_syn_backoff_low[TCP_MAXRXTSHIFT + 1] =
 232     { 1, 1, 2, 4, 8, 8, 16, 16, 32, 64, 64, 64, 64 };
 233
 234 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 235     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 236
 237 static int tcp_totbackoff = 511;        /* sum of tcp_backoff[] */
 238
 239 /* Caller should be in critical section */
 240 static struct tcpcb *
 241 tcp_timer_delack_handler(struct tcpcb *tp)
 242 {
 243         tp->t_flags |= TF_ACKNOW;
 244         tcpstat.tcps_delack++;
 245         tcp_output(tp);
 246         return tp;
 247 }
 248
 249 /*
 250  * TCP timer processing.
 251  */
 252 void
 253 tcp_timer_delack(void *xtp)
 254 {
 255         struct tcpcb *tp = xtp;
 256         struct callout *co = &tp->tt_delack->tc_callout;
 257
 258         crit_enter();
 259         if (callout_pending(co) || !callout_active(co)) {
 260                 crit_exit();
 261                 return;
 262         }
 263         callout_deactivate(co);
 264         tcp_send_timermsg(tp, TCP_TIMER_DELACK);
 265         crit_exit();
 266 }
 267
 268 /* Caller should be in critical section */
 269 static struct tcpcb *
 270 tcp_timer_2msl_handler(struct tcpcb *tp)
 271 {
 272 #ifdef TCPDEBUG
 273         int ostate;
 274 #endif
 275
 276 #ifdef TCPDEBUG
 277         ostate = tp->t_state;
 278 #endif
 279         /*
 280          * 2 MSL timeout in shutdown went off.  If we're closed but
 281          * still waiting for peer to close and connection has been idle
 282          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 283          * control block.  Otherwise, check again in a bit.
 284          */
 285         if (tp->t_state != TCPS_TIME_WAIT &&
 286             (ticks - tp->t_rcvtime) <= tp->t_maxidle) {
 287                 tcp_callout_reset(tp, tp->tt_2msl, tp->t_keepintvl,
 288                                   tcp_timer_2msl);
 289         } else {
 290                 tp = tcp_close(tp);
 291         }
 292
 293 #ifdef TCPDEBUG
 294         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 295                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 296 #endif
 297         return tp;
 298 }
 299
 300 void
 301 tcp_timer_2msl(void *xtp)
 302 {
 303         struct tcpcb *tp = xtp;
 304         struct callout *co = &tp->tt_2msl->tc_callout;
 305
 306         crit_enter();
 307         if (callout_pending(co) || !callout_active(co)) {
 308                 crit_exit();
 309                 return;
 310         }
 311         callout_deactivate(co);
 312         tcp_send_timermsg(tp, TCP_TIMER_2MSL);
 313         crit_exit();
 314 }
 315
 316 /* Caller should be in critical section */
 317 static struct tcpcb *
 318 tcp_timer_keep_handler(struct tcpcb *tp)
 319 {
 320         struct tcptemp *t_template;
 321 #ifdef TCPDEBUG
 322         int ostate = tp->t_state;
 323 #endif
 324
 325         /*
 326          * Keep-alive timer went off; send something
 327          * or drop connection if idle for too long.
 328          */
 329         tcpstat.tcps_keeptimeo++;
 330         if (tp->t_state < TCPS_ESTABLISHED)
 331                 goto dropit;
 332         if ((always_keepalive || (tp->t_flags & TF_KEEPALIVE) ||
 333              (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE)) &&
 334             tp->t_state <= TCPS_CLOSING) {
 335                 if ((ticks - tp->t_rcvtime) >= tp->t_keepidle + tp->t_maxidle)
 336                         goto dropit;
 337                 /*
 338                  * Send a packet designed to force a response
 339                  * if the peer is up and reachable:
 340                  * either an ACK if the connection is still alive,
 341                  * or an RST if the peer has closed the connection
 342                  * due to timeout or reboot.
 343                  * Using sequence number tp->snd_una-1
 344                  * causes the transmitted zero-length segment
 345                  * to lie outside the receive window;
 346                  * by the protocol spec, this requires the
 347                  * correspondent TCP to respond.
 348                  */
 349                 tcpstat.tcps_keepprobe++;
 350                 t_template = tcp_maketemplate(tp);
 351                 if (t_template) {
 352                         tcp_respond(tp, t_template->tt_ipgen,
 353                                     &t_template->tt_t, NULL,
 354                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 355                         tcp_freetemplate(t_template);
 356                 }
 357                 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepintvl,
 358                                   tcp_timer_keep);
 359         } else {
 360                 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle,
 361                                   tcp_timer_keep);
 362         }
 363
 364 #ifdef TCPDEBUG
 365         if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 366                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 367 #endif
 368         return tp;
 369
 370 dropit:
 371         tcpstat.tcps_keepdrops++;
 372         tp = tcp_drop(tp, ETIMEDOUT);
 373
 374 #ifdef TCPDEBUG
 375         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 376                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 377 #endif
 378         return tp;
 379 }
 380
 381 void
 382 tcp_timer_keep(void *xtp)
 383 {
 384         struct tcpcb *tp = xtp;
 385         struct callout *co = &tp->tt_keep->tc_callout;
 386
 387         crit_enter();
 388         if (callout_pending(co) || !callout_active(co)) {
 389                 crit_exit();
 390                 return;
 391         }
 392         callout_deactivate(co);
 393         tcp_send_timermsg(tp, TCP_TIMER_KEEP);
 394         crit_exit();
 395 }
 396
 397 /* Caller should be in critical section */
 398 static struct tcpcb *
 399 tcp_timer_persist_handler(struct tcpcb *tp)
 400 {
 401 #ifdef TCPDEBUG
 402         int ostate;
 403 #endif
 404
 405 #ifdef TCPDEBUG
 406         ostate = tp->t_state;
 407 #endif
 408         /*
 409          * Persistance timer into zero window.
 410          * Force a byte to be output, if possible.
 411          */
 412         tcpstat.tcps_persisttimeo++;
 413         /*
 414          * Hack: if the peer is dead/unreachable, we do not
 415          * time out if the window is closed.  After a full
 416          * backoff, drop the connection if the idle time
 417          * (no responses to probes) reaches the maximum
 418          * backoff that we would use if retransmitting.
 419          */
 420         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 421             ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
 422              (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 423                 tcpstat.tcps_persistdrop++;
 424                 tp = tcp_drop(tp, ETIMEDOUT);
 425                 goto out;
 426         }
 427         tcp_setpersist(tp);
 428         tp->t_flags |= TF_FORCE;
 429         tcp_output(tp);
 430         tp->t_flags &= ~TF_FORCE;
 431
 432 out:
 433 #ifdef TCPDEBUG
 434         if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 435                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 436 #endif
 437         return tp;
 438 }
 439
 440 void
 441 tcp_timer_persist(void *xtp)
 442 {
 443         struct tcpcb *tp = xtp;
 444         struct callout *co = &tp->tt_persist->tc_callout;
 445
 446         crit_enter();
 447         if (callout_pending(co) || !callout_active(co)){
 448                 crit_exit();
 449                 return;
 450         }
 451         callout_deactivate(co);
 452         tcp_send_timermsg(tp, TCP_TIMER_PERSIST);
 453         crit_exit();
 454 }
 455
 456 void
 457 tcp_save_congestion_state(struct tcpcb *tp)
 458 {
 459         /*
 460          * Record connection's current states so that they could be
 461          * recovered, if this turns out to be a spurious retransmit.
 462          */
 463         tp->snd_cwnd_prev = tp->snd_cwnd;
 464         tp->snd_wacked_prev = tp->snd_wacked;
 465         tp->snd_ssthresh_prev = tp->snd_ssthresh;
 466         tp->snd_recover_prev = tp->snd_recover;
 467
 468         /*
 469          * State for Eifel response after spurious timeout retransmit
 470          * is detected.  We save the current value of snd_max even if
 471          * we are called from fast retransmit code, so if RTO needs
 472          * rebase, it will be rebased using the RTT of segment that
 473          * is not sent during possible congestion.
 474          */
 475         tp->snd_max_prev = tp->snd_max;
 476
 477         if (IN_FASTRECOVERY(tp))
 478                 tp->rxt_flags |= TRXT_F_WASFRECOVERY;
 479         else
 480                 tp->rxt_flags &= ~TRXT_F_WASFRECOVERY;
 481         if (tp->t_flags & TF_RCVD_TSTMP) {
 482                 /* States for Eifel detection */
 483                 tp->t_rexmtTS = ticks;
 484                 tp->rxt_flags |= TRXT_F_FIRSTACCACK;
 485         }
 486 #ifdef later
 487         tcp_sack_save_scoreboard(&tp->scb);
 488 #endif
 489 }
 490
 491 void
 492 tcp_revert_congestion_state(struct tcpcb *tp)
 493 {
 494         tp->snd_cwnd = tp->snd_cwnd_prev;
 495         tp->snd_wacked = tp->snd_wacked_prev;
 496         tp->snd_ssthresh = tp->snd_ssthresh_prev;
 497         tp->snd_recover = tp->snd_recover_prev;
 498         if (tp->rxt_flags & TRXT_F_WASFRECOVERY)
 499                 ENTER_FASTRECOVERY(tp);
 500         if (tp->rxt_flags & TRXT_F_FASTREXMT) {
 501                 ++tcpstat.tcps_sndfastrexmitbad;
 502                 if (tp->rxt_flags & TRXT_F_EARLYREXMT)
 503                         ++tcpstat.tcps_sndearlyrexmitbad;
 504         } else {
 505                 ++tcpstat.tcps_sndrtobad;
 506                 tp->snd_last = ticks;
 507                 if (tcp_do_eifel_response)
 508                         tp->rxt_flags |= TRXT_F_REBASERTO;
 509         }
 510         tp->t_badrxtwin = 0;
 511         tp->t_rxtshift = 0;
 512         tp->snd_nxt = tp->snd_max;
 513 #ifdef later
 514         tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una);
 515 #endif
 516 }
 517
 518 /* Caller should be in critical section */
 519 static struct tcpcb *
 520 tcp_timer_rexmt_handler(struct tcpcb *tp)
 521 {
 522         int rexmt;
 523 #ifdef TCPDEBUG
 524         int ostate;
 525 #endif
 526
 527 #ifdef TCPDEBUG
 528         ostate = tp->t_state;
 529 #endif
 530         /*
 531          * Retransmission timer went off.  Message has not
 532          * been acked within retransmit interval.  Back off
 533          * to a longer retransmit interval and retransmit one segment.
 534          */
 535         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 536                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 537                 tcpstat.tcps_timeoutdrop++;
 538                 tp = tcp_drop(tp, tp->t_softerror ?
 539                               tp->t_softerror : ETIMEDOUT);
 540                 goto out;
 541         }
 542         if (tp->t_rxtshift == 1) {
 543                 /*
 544                  * First retransmit.
 545                  */
 546
 547                 /*
 548                  * State for "RTT based spurious timeout retransmit detection"
 549                  *
 550                  * RTT based spurious timeout retransmit detection:
 551                  * A retransmit is considered spurious if an ACK for this
 552                  * segment is received within RTT/2 interval; the assumption
 553                  * here is that the ACK was already in flight.  See
 554                  * "On Estimating End-to-End Network Path Properties" by
 555                  * Allman and Paxson for more details.
 556                  */
 557                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 558
 559                 /*
 560                  * States for Eifel response after spurious timeout retransmit
 561                  * is detected.
 562                  */
 563                 tp->t_rxtcur_prev = tp->t_rxtcur;
 564                 tp->t_srtt_prev = tp->t_srtt +
 565                     (tcp_eifel_rtoinc << TCP_RTT_SHIFT);
 566                 tp->t_rttvar_prev = tp->t_rttvar;
 567
 568                 tcp_save_congestion_state(tp);
 569                 tp->rxt_flags &= ~(TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT |
 570                     TRXT_F_REBASERTO);
 571         }
 572         if (tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) {
 573                 /*
 574                  * Record the time that we spent in SYN or SYN|ACK
 575                  * retransmition.
 576                  *
 577                  * Needed by RFC3390 and RFC6298.
 578                  */
 579                 tp->t_rxtsyn += tp->t_rxtcur;
 580         }
 581         /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
 582         tcp_sack_discard(tp);
 583         tcpstat.tcps_rexmttimeo++;
 584         if (tp->t_state == TCPS_SYN_SENT) {
 585                 if (tcp_low_rtobase) {
 586                         rexmt = TCP_REXMTVAL(tp) *
 587                                 tcp_syn_backoff_low[tp->t_rxtshift];
 588                 } else {
 589                         rexmt = TCP_REXMTVAL(tp) *
 590                                 tcp_syn_backoff[tp->t_rxtshift];
 591                 }
 592         } else {
 593                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 594         }
 595         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 596                       tp->t_rttmin, TCPTV_REXMTMAX);
 597         /*
 598          * If losing, let the lower level know and try for
 599          * a better route.  Also, if we backed off this far,
 600          * our srtt estimate is probably bogus.  Clobber it
 601          * so we'll take the next rtt measurement as our srtt;
 602          * move the current srtt into rttvar to keep the current
 603          * retransmit times until then.
 604          */
 605         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 606 #ifdef INET6
 607                 if (INP_ISIPV6(tp->t_inpcb))
 608                         in6_losing(tp->t_inpcb);
 609                 else
 610 #endif
 611                 in_losing(tp->t_inpcb);
 612                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 613                 tp->t_srtt = 0;
 614         }
 615         tp->snd_nxt = tp->snd_una;
 616         tp->snd_recover = tp->snd_max;
 617         /*
 618          * Force a segment to be sent.
 619          */
 620         tp->t_flags |= TF_ACKNOW;
 621         /*
 622          * If timing a segment in this window, stop the timer.
 623          */
 624         tp->t_rtttime = 0;
 625         /*
 626          * Close the congestion window down to one segment
 627          * (we'll open it by one segment for each ack we get).
 628          * Since we probably have a window's worth of unacked
 629          * data accumulated, this "slow start" keeps us from
 630          * dumping all that data as back-to-back packets (which
 631          * might overwhelm an intermediate gateway).
 632          *
 633          * There are two phases to the opening: Initially we
 634          * open by one mss on each ack.  This makes the window
 635          * size increase exponentially with time.  If the
 636          * window is larger than the path can handle, this
 637          * exponential growth results in dropped packet(s)
 638          * almost immediately.  To get more time between
 639          * drops but still "push" the network to take advantage
 640          * of improving conditions, we switch from exponential
 641          * to linear window opening at some threshhold size.
 642          * For a threshhold, we use half the current window
 643          * size, truncated to a multiple of the mss.
 644          *
 645          * (the minimum cwnd that will give us exponential
 646          * growth is 2 mss.  We don't allow the threshhold
 647          * to go below this.)
 648          */
 649         {
 650                 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 651
 652                 if (win < 2)
 653                         win = 2;
 654                 tp->snd_cwnd = tp->t_maxseg;
 655                 tp->snd_wacked = 0;
 656                 tp->snd_ssthresh = win * tp->t_maxseg;
 657                 tp->t_dupacks = 0;
 658         }
 659         EXIT_FASTRECOVERY(tp);
 660         tcp_output(tp);
 661
 662 out:
 663 #ifdef TCPDEBUG
 664         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 665                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 666 #endif
 667         return tp;
 668 }
 669
 670 void
 671 tcp_timer_rexmt(void *xtp)
 672 {
 673         struct tcpcb *tp = xtp;
 674         struct callout *co = &tp->tt_rexmt->tc_callout;
 675
 676         crit_enter();
 677         if (callout_pending(co) || !callout_active(co)) {
 678                 crit_exit();
 679                 return;
 680         }
 681         callout_deactivate(co);
 682         tcp_send_timermsg(tp, TCP_TIMER_REXMT);
 683         crit_exit();
 684 }
 685
 686 static void
 687 tcp_timer_handler(netmsg_t msg)
 688 {
 689         struct netmsg_tcp_timer *tmsg = (struct netmsg_tcp_timer *)msg;
 690         const struct tcp_timer *tt;
 691         struct tcpcb *tp;
 692
 693         crit_enter();
 694
 695         KKASSERT(tmsg->tt_cpuid == mycpuid && tmsg->tt_tcb != NULL);
 696         tp = tmsg->tt_tcb;
 697
 698         /* Save pending tasks and reset the tasks in message */
 699         tmsg->tt_running_tasks = tmsg->tt_tasks;
 700         tmsg->tt_prev_tasks = tmsg->tt_tasks;
 701         tmsg->tt_tasks = 0;
 702
 703         /* Reply ASAP */
 704         lwkt_replymsg(&tmsg->tt_msg.lmsg, 0);
 705
 706         if (tmsg->tt_running_tasks == 0) {
 707                 /*
 708                  * All of the timers are cancelled when the message
 709                  * is pending; bail out.
 710                  */
 711                 crit_exit();
 712                 return;
 713         }
 714
 715         for (tt = tcp_timer_handlers; tt->tt_handler != NULL; ++tt) {
 716                 if ((tmsg->tt_running_tasks & tt->tt_task) == 0)
 717                         continue;
 718
 719                 tmsg->tt_running_tasks &= ~tt->tt_task;
 720                 tp = tt->tt_handler(tp);
 721                 if (tp == NULL)
 722                         break;
 723
 724                 if (tmsg->tt_running_tasks == 0) /* nothing left to do */
 725                         break;
 726         }
 727
 728         crit_exit();
 729 }
 730
 731 void
 732 tcp_create_timermsg(struct tcpcb *tp, struct lwkt_port *msgport)
 733 {
 734         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 735
 736         netmsg_init(&tmsg->tt_msg, NULL, &netisr_adone_rport,
 737                     MSGF_DROPABLE | MSGF_PRIORITY, tcp_timer_handler);
 738         tmsg->tt_cpuid = mycpuid;
 739         tmsg->tt_msgport = msgport;
 740         tmsg->tt_tcb = tp;
 741         tmsg->tt_tasks = 0;
 742 }
 743
 744 void
 745 tcp_destroy_timermsg(struct tcpcb *tp)
 746 {
 747         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 748
 749         if (tmsg == NULL ||             /* listen socket */
 750             tmsg->tt_tcb == NULL)       /* only tcp_attach() is called */
 751                 return;
 752
 753         KKASSERT(tmsg->tt_cpuid == mycpuid);
 754
 755         /*
 756          * This message is still pending to be processed;
 757          * drop it.  Optimized.
 758          */
 759         crit_enter();
 760         if ((tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE) == 0) {
 761                 lwkt_dropmsg(&tmsg->tt_msg.lmsg);
 762         }
 763         crit_exit();
 764 }
 765
 766 static __inline void
 767 tcp_callout_init(struct tcp_callout *tc, uint32_t task)
 768 {
 769         callout_init_mp(&tc->tc_callout);
 770         tc->tc_task = task;
 771 }
 772
 773 void
 774 tcp_inittimers(struct tcpcb *tp)
 775 {
 776         tcp_callout_init(tp->tt_rexmt, TCP_TIMER_REXMT);
 777         tcp_callout_init(tp->tt_persist, TCP_TIMER_PERSIST);
 778         tcp_callout_init(tp->tt_keep, TCP_TIMER_KEEP);
 779         tcp_callout_init(tp->tt_2msl, TCP_TIMER_2MSL);
 780         tcp_callout_init(tp->tt_delack, TCP_TIMER_DELACK);
 781 }