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  * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.17 2008/03/30 20:39:01 dillon Exp $
  65  */
  66
  67 #include "opt_compat.h"
  68 #include "opt_inet6.h"
  69 #include "opt_tcpdebug.h"
  70
  71 #include <sys/param.h>
  72 #include <sys/systm.h>
  73 #include <sys/kernel.h>
  74 #include <sys/mbuf.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/socket.h>
  77 #include <sys/socketvar.h>
  78 #include <sys/protosw.h>
  79 #include <sys/thread.h>
  80 #include <sys/globaldata.h>
  81 #include <sys/thread2.h>
  82 #include <sys/msgport2.h>
  83
  84 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
  85
  86 #include <net/route.h>
  87 #include <net/netmsg2.h>
  88
  89 #include <netinet/in.h>
  90 #include <netinet/in_systm.h>
  91 #include <netinet/in_pcb.h>
  92 #ifdef INET6
  93 #include <netinet6/in6_pcb.h>
  94 #endif
  95 #include <netinet/ip_var.h>
  96 #include <netinet/tcp.h>
  97 #include <netinet/tcp_fsm.h>
  98 #include <netinet/tcp_seq.h>
  99 #include <netinet/tcp_timer.h>
 100 #include <netinet/tcp_timer2.h>
 101 #include <netinet/tcp_var.h>
 102 #include <netinet/tcpip.h>
 103 #ifdef TCPDEBUG
 104 #include <netinet/tcp_debug.h>
 105 #endif
 106
 107 #define TCP_TIMER_REXMT         0x01
 108 #define TCP_TIMER_PERSIST       0x02
 109 #define TCP_TIMER_KEEP          0x04
 110 #define TCP_TIMER_2MSL          0x08
 111 #define TCP_TIMER_DELACK        0x10
 112
 113 static struct tcpcb     *tcp_timer_rexmt_handler(struct tcpcb *);
 114 static struct tcpcb     *tcp_timer_persist_handler(struct tcpcb *);
 115 static struct tcpcb     *tcp_timer_keep_handler(struct tcpcb *);
 116 static struct tcpcb     *tcp_timer_2msl_handler(struct tcpcb *);
 117 static struct tcpcb     *tcp_timer_delack_handler(struct tcpcb *);
 118
 119 static const struct tcp_timer {
 120         uint32_t        tt_task;
 121         struct tcpcb    *(*tt_handler)(struct tcpcb *);
 122 } tcp_timer_handlers[] = {
 123         { TCP_TIMER_DELACK,     tcp_timer_delack_handler },
 124         { TCP_TIMER_REXMT,      tcp_timer_rexmt_handler },
 125         { TCP_TIMER_PERSIST,    tcp_timer_persist_handler },
 126         { TCP_TIMER_KEEP,       tcp_timer_keep_handler },
 127         { TCP_TIMER_2MSL,       tcp_timer_2msl_handler },
 128         { 0, NULL }
 129 };
 130
 131 static int
 132 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
 133 {
 134         int error, s, tt;
 135
 136         tt = *(int *)oidp->oid_arg1;
 137         s = (int)((int64_t)tt * 1000 / hz);
 138
 139         error = sysctl_handle_int(oidp, &s, 0, req);
 140         if (error || !req->newptr)
 141                 return (error);
 142
 143         tt = (int)((int64_t)s * hz / 1000);
 144         if (tt < 1)
 145                 return (EINVAL);
 146
 147         *(int *)oidp->oid_arg1 = tt;
 148         return (0);
 149 }
 150
 151 int     tcp_keepinit;
 152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
 153     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "Time to establish TCP connection");
 154
 155 int     tcp_keepidle;
 156 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
 157     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "Time before TCP keepalive probes begin");
 158
 159 int     tcp_keepintvl;
 160 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
 161     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "Time between TCP keepalive probes");
 162
 163 int     tcp_delacktime;
 164 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
 165     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 166     "Time before a delayed ACK is sent");
 167
 168 int     tcp_msl;
 169 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 170     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 171
 172 int     tcp_rexmit_min;
 173 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 174     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
 175
 176 int     tcp_rexmit_slop;
 177 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 178     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 179     "Retransmission Timer Slop");
 180
 181 static int      always_keepalive = 1;
 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 183     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 184
 185 /* max idle probes */
 186 int     tcp_keepcnt = TCPTV_KEEPCNT;
 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW,
 188     &tcp_keepcnt, 0, "Maximum number of keepalive probes to be sent");
 189
 190 static int tcp_do_eifel_response = 1;
 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel_response, CTLFLAG_RW,
 192     &tcp_do_eifel_response, 0, "Eifel response algorithm (RFC 4015)");
 193
 194 int tcp_eifel_rtoinc = 2;
 195 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, eifel_rtoinc, CTLTYPE_INT|CTLFLAG_RW,
 196     &tcp_eifel_rtoinc, 0, sysctl_msec_to_ticks, "I",
 197     "Eifel response RTO increment");
 198
 199 /* max idle time in persist */
 200 int     tcp_maxpersistidle;
 201
 202 /*
 203  * Cancel all timers for TCP tp.
 204  */
 205 void
 206 tcp_canceltimers(struct tcpcb *tp)
 207 {
 208         tcp_callout_stop(tp, tp->tt_2msl);
 209         tcp_callout_stop(tp, tp->tt_persist);
 210         tcp_callout_stop(tp, tp->tt_keep);
 211         tcp_callout_stop(tp, tp->tt_rexmt);
 212 }
 213
 214 /*
 215  * Caller should be in critical section
 216  */
 217 static void
 218 tcp_send_timermsg(struct tcpcb *tp, uint32_t task)
 219 {
 220         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 221
 222         KKASSERT(tmsg != NULL && tmsg->tt_cpuid == mycpuid &&
 223                  tmsg->tt_tcb != NULL);
 224
 225         tmsg->tt_tasks |= task;
 226         if (tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE)
 227                 lwkt_sendmsg_oncpu(tmsg->tt_msgport, &tmsg->tt_msg.lmsg);
 228 }
 229
 230 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 231     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 232
 233 int     tcp_syn_backoff_low[TCP_MAXRXTSHIFT + 1] =
 234     { 1, 1, 2, 4, 8, 8, 16, 16, 32, 64, 64, 64, 64 };
 235
 236 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 237     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 238
 239 static int tcp_totbackoff = 511;        /* sum of tcp_backoff[] */
 240
 241 /* Caller should be in critical section */
 242 static struct tcpcb *
 243 tcp_timer_delack_handler(struct tcpcb *tp)
 244 {
 245         tp->t_flags |= TF_ACKNOW;
 246         tcpstat.tcps_delack++;
 247         tcp_output(tp);
 248         return tp;
 249 }
 250
 251 /*
 252  * TCP timer processing.
 253  */
 254 void
 255 tcp_timer_delack(void *xtp)
 256 {
 257         struct tcpcb *tp = xtp;
 258         struct callout *co = &tp->tt_delack->tc_callout;
 259
 260         crit_enter();
 261         if (callout_pending(co) || !callout_active(co)) {
 262                 crit_exit();
 263                 return;
 264         }
 265         callout_deactivate(co);
 266         tcp_send_timermsg(tp, TCP_TIMER_DELACK);
 267         crit_exit();
 268 }
 269
 270 /* Caller should be in critical section */
 271 static struct tcpcb *
 272 tcp_timer_2msl_handler(struct tcpcb *tp)
 273 {
 274 #ifdef TCPDEBUG
 275         int ostate;
 276 #endif
 277
 278 #ifdef TCPDEBUG
 279         ostate = tp->t_state;
 280 #endif
 281         /*
 282          * 2 MSL timeout in shutdown went off.  If we're closed but
 283          * still waiting for peer to close and connection has been idle
 284          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 285          * control block.  Otherwise, check again in a bit.
 286          */
 287         if (tp->t_state != TCPS_TIME_WAIT &&
 288             (ticks - tp->t_rcvtime) <= tp->t_maxidle) {
 289                 tcp_callout_reset(tp, tp->tt_2msl, tp->t_keepintvl,
 290                                   tcp_timer_2msl);
 291         } else {
 292                 tp = tcp_close(tp);
 293         }
 294
 295 #ifdef TCPDEBUG
 296         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 297                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 298 #endif
 299         return tp;
 300 }
 301
 302 void
 303 tcp_timer_2msl(void *xtp)
 304 {
 305         struct tcpcb *tp = xtp;
 306         struct callout *co = &tp->tt_2msl->tc_callout;
 307
 308         crit_enter();
 309         if (callout_pending(co) || !callout_active(co)) {
 310                 crit_exit();
 311                 return;
 312         }
 313         callout_deactivate(co);
 314         tcp_send_timermsg(tp, TCP_TIMER_2MSL);
 315         crit_exit();
 316 }
 317
 318 /* Caller should be in critical section */
 319 static struct tcpcb *
 320 tcp_timer_keep_handler(struct tcpcb *tp)
 321 {
 322         struct tcptemp *t_template;
 323 #ifdef TCPDEBUG
 324         int ostate = tp->t_state;
 325 #endif
 326
 327         /*
 328          * Keep-alive timer went off; send something
 329          * or drop connection if idle for too long.
 330          */
 331         tcpstat.tcps_keeptimeo++;
 332         if (tp->t_state < TCPS_ESTABLISHED)
 333                 goto dropit;
 334         if ((always_keepalive || (tp->t_flags & TF_KEEPALIVE) ||
 335              (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE)) &&
 336             tp->t_state <= TCPS_CLOSING) {
 337                 if ((ticks - tp->t_rcvtime) >= tp->t_keepidle + tp->t_maxidle)
 338                         goto dropit;
 339                 /*
 340                  * Send a packet designed to force a response
 341                  * if the peer is up and reachable:
 342                  * either an ACK if the connection is still alive,
 343                  * or an RST if the peer has closed the connection
 344                  * due to timeout or reboot.
 345                  * Using sequence number tp->snd_una-1
 346                  * causes the transmitted zero-length segment
 347                  * to lie outside the receive window;
 348                  * by the protocol spec, this requires the
 349                  * correspondent TCP to respond.
 350                  */
 351                 tcpstat.tcps_keepprobe++;
 352                 t_template = tcp_maketemplate(tp);
 353                 if (t_template) {
 354                         tcp_respond(tp, t_template->tt_ipgen,
 355                                     &t_template->tt_t, NULL,
 356                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 357                         tcp_freetemplate(t_template);
 358                 }
 359                 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepintvl,
 360                                   tcp_timer_keep);
 361         } else {
 362                 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle,
 363                                   tcp_timer_keep);
 364         }
 365
 366 #ifdef TCPDEBUG
 367         if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 368                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 369 #endif
 370         return tp;
 371
 372 dropit:
 373         tcpstat.tcps_keepdrops++;
 374         tp = tcp_drop(tp, ETIMEDOUT);
 375
 376 #ifdef TCPDEBUG
 377         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 378                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 379 #endif
 380         return tp;
 381 }
 382
 383 void
 384 tcp_timer_keep(void *xtp)
 385 {
 386         struct tcpcb *tp = xtp;
 387         struct callout *co = &tp->tt_keep->tc_callout;
 388
 389         crit_enter();
 390         if (callout_pending(co) || !callout_active(co)) {
 391                 crit_exit();
 392                 return;
 393         }
 394         callout_deactivate(co);
 395         tcp_send_timermsg(tp, TCP_TIMER_KEEP);
 396         crit_exit();
 397 }
 398
 399 /* Caller should be in critical section */
 400 static struct tcpcb *
 401 tcp_timer_persist_handler(struct tcpcb *tp)
 402 {
 403 #ifdef TCPDEBUG
 404         int ostate;
 405 #endif
 406
 407 #ifdef TCPDEBUG
 408         ostate = tp->t_state;
 409 #endif
 410         /*
 411          * Persistance timer into zero window.
 412          * Force a byte to be output, if possible.
 413          */
 414         tcpstat.tcps_persisttimeo++;
 415         /*
 416          * Hack: if the peer is dead/unreachable, we do not
 417          * time out if the window is closed.  After a full
 418          * backoff, drop the connection if the idle time
 419          * (no responses to probes) reaches the maximum
 420          * backoff that we would use if retransmitting.
 421          */
 422         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 423             ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
 424              (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 425                 tcpstat.tcps_persistdrop++;
 426                 tp = tcp_drop(tp, ETIMEDOUT);
 427                 goto out;
 428         }
 429         tcp_setpersist(tp);
 430         tp->t_flags |= TF_FORCE;
 431         tcp_output(tp);
 432         tp->t_flags &= ~TF_FORCE;
 433
 434 out:
 435 #ifdef TCPDEBUG
 436         if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 437                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 438 #endif
 439         return tp;
 440 }
 441
 442 void
 443 tcp_timer_persist(void *xtp)
 444 {
 445         struct tcpcb *tp = xtp;
 446         struct callout *co = &tp->tt_persist->tc_callout;
 447
 448         crit_enter();
 449         if (callout_pending(co) || !callout_active(co)){
 450                 crit_exit();
 451                 return;
 452         }
 453         callout_deactivate(co);
 454         tcp_send_timermsg(tp, TCP_TIMER_PERSIST);
 455         crit_exit();
 456 }
 457
 458 void
 459 tcp_save_congestion_state(struct tcpcb *tp)
 460 {
 461         /*
 462          * Record connection's current states so that they could be
 463          * recovered, if this turns out to be a spurious retransmit.
 464          */
 465         tp->snd_cwnd_prev = tp->snd_cwnd;
 466         tp->snd_wacked_prev = tp->snd_wacked;
 467         tp->snd_ssthresh_prev = tp->snd_ssthresh;
 468         tp->snd_recover_prev = tp->snd_recover;
 469
 470         /*
 471          * State for Eifel response after spurious timeout retransmit
 472          * is detected.  We save the current value of snd_max even if
 473          * we are called from fast retransmit code, so if RTO needs
 474          * rebase, it will be rebased using the RTT of segment that
 475          * is not sent during possible congestion.
 476          */
 477         tp->snd_max_prev = tp->snd_max;
 478
 479         if (IN_FASTRECOVERY(tp))
 480                 tp->rxt_flags |= TRXT_F_WASFRECOVERY;
 481         else
 482                 tp->rxt_flags &= ~TRXT_F_WASFRECOVERY;
 483         if (tp->t_flags & TF_RCVD_TSTMP) {
 484                 /* States for Eifel detection */
 485                 tp->t_rexmtTS = ticks;
 486                 tp->rxt_flags |= TRXT_F_FIRSTACCACK;
 487         }
 488 #ifdef later
 489         tcp_sack_save_scoreboard(&tp->scb);
 490 #endif
 491 }
 492
 493 void
 494 tcp_revert_congestion_state(struct tcpcb *tp)
 495 {
 496         tp->snd_cwnd = tp->snd_cwnd_prev;
 497         tp->snd_wacked = tp->snd_wacked_prev;
 498         tp->snd_ssthresh = tp->snd_ssthresh_prev;
 499         tp->snd_recover = tp->snd_recover_prev;
 500         if (tp->rxt_flags & TRXT_F_WASFRECOVERY)
 501                 ENTER_FASTRECOVERY(tp);
 502         if (tp->rxt_flags & TRXT_F_FASTREXMT) {
 503                 ++tcpstat.tcps_sndfastrexmitbad;
 504                 if (tp->rxt_flags & TRXT_F_EARLYREXMT)
 505                         ++tcpstat.tcps_sndearlyrexmitbad;
 506         } else {
 507                 ++tcpstat.tcps_sndrtobad;
 508                 tp->snd_last = ticks;
 509                 if (tcp_do_eifel_response)
 510                         tp->rxt_flags |= TRXT_F_REBASERTO;
 511         }
 512         tp->t_badrxtwin = 0;
 513         tp->t_rxtshift = 0;
 514         tp->snd_nxt = tp->snd_max;
 515 #ifdef later
 516         tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una);
 517 #endif
 518 }
 519
 520 /* Caller should be in critical section */
 521 static struct tcpcb *
 522 tcp_timer_rexmt_handler(struct tcpcb *tp)
 523 {
 524         int rexmt;
 525 #ifdef TCPDEBUG
 526         int ostate;
 527 #endif
 528
 529 #ifdef TCPDEBUG
 530         ostate = tp->t_state;
 531 #endif
 532         /*
 533          * Retransmission timer went off.  Message has not
 534          * been acked within retransmit interval.  Back off
 535          * to a longer retransmit interval and retransmit one segment.
 536          */
 537         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 538                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 539                 tcpstat.tcps_timeoutdrop++;
 540                 tp = tcp_drop(tp, tp->t_softerror ?
 541                               tp->t_softerror : ETIMEDOUT);
 542                 goto out;
 543         }
 544         if (tp->t_rxtshift == 1) {
 545                 /*
 546                  * First retransmit.
 547                  */
 548
 549                 /*
 550                  * State for "RTT based spurious timeout retransmit detection"
 551                  *
 552                  * RTT based spurious timeout retransmit detection:
 553                  * A retransmit is considered spurious if an ACK for this
 554                  * segment is received within RTT/2 interval; the assumption
 555                  * here is that the ACK was already in flight.  See
 556                  * "On Estimating End-to-End Network Path Properties" by
 557                  * Allman and Paxson for more details.
 558                  */
 559                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 560
 561                 /*
 562                  * States for Eifel response after spurious timeout retransmit
 563                  * is detected.
 564                  */
 565                 tp->t_rxtcur_prev = tp->t_rxtcur;
 566                 tp->t_srtt_prev = tp->t_srtt +
 567                     (tcp_eifel_rtoinc << TCP_RTT_SHIFT);
 568                 tp->t_rttvar_prev = tp->t_rttvar;
 569
 570                 tcp_save_congestion_state(tp);
 571                 tp->rxt_flags &= ~(TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT |
 572                     TRXT_F_REBASERTO);
 573         }
 574         if (tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) {
 575                 /*
 576                  * Record the time that we spent in SYN or SYN|ACK
 577                  * retransmition.
 578                  *
 579                  * Needed by RFC3390 and RFC6298.
 580                  */
 581                 tp->t_rxtsyn += tp->t_rxtcur;
 582         }
 583         /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
 584         tcp_sack_discard(tp);
 585         tcpstat.tcps_rexmttimeo++;
 586         if (tp->t_state == TCPS_SYN_SENT) {
 587                 if (tcp_low_rtobase) {
 588                         rexmt = TCP_REXMTVAL(tp) *
 589                                 tcp_syn_backoff_low[tp->t_rxtshift];
 590                 } else {
 591                         rexmt = TCP_REXMTVAL(tp) *
 592                                 tcp_syn_backoff[tp->t_rxtshift];
 593                 }
 594         } else {
 595                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 596         }
 597         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 598                       tp->t_rttmin, TCPTV_REXMTMAX);
 599         /*
 600          * If losing, let the lower level know and try for
 601          * a better route.  Also, if we backed off this far,
 602          * our srtt estimate is probably bogus.  Clobber it
 603          * so we'll take the next rtt measurement as our srtt;
 604          * move the current srtt into rttvar to keep the current
 605          * retransmit times until then.
 606          */
 607         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 608 #ifdef INET6
 609                 if (INP_ISIPV6(tp->t_inpcb))
 610                         in6_losing(tp->t_inpcb);
 611                 else
 612 #endif
 613                 in_losing(tp->t_inpcb);
 614                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 615                 tp->t_srtt = 0;
 616         }
 617         tp->snd_nxt = tp->snd_una;
 618         tp->snd_recover = tp->snd_max;
 619         /*
 620          * Force a segment to be sent.
 621          */
 622         tp->t_flags |= TF_ACKNOW;
 623         /*
 624          * If timing a segment in this window, stop the timer.
 625          */
 626         tp->t_rtttime = 0;
 627         /*
 628          * Close the congestion window down to one segment
 629          * (we'll open it by one segment for each ack we get).
 630          * Since we probably have a window's worth of unacked
 631          * data accumulated, this "slow start" keeps us from
 632          * dumping all that data as back-to-back packets (which
 633          * might overwhelm an intermediate gateway).
 634          *
 635          * There are two phases to the opening: Initially we
 636          * open by one mss on each ack.  This makes the window
 637          * size increase exponentially with time.  If the
 638          * window is larger than the path can handle, this
 639          * exponential growth results in dropped packet(s)
 640          * almost immediately.  To get more time between
 641          * drops but still "push" the network to take advantage
 642          * of improving conditions, we switch from exponential
 643          * to linear window opening at some threshhold size.
 644          * For a threshhold, we use half the current window
 645          * size, truncated to a multiple of the mss.
 646          *
 647          * (the minimum cwnd that will give us exponential
 648          * growth is 2 mss.  We don't allow the threshhold
 649          * to go below this.)
 650          */
 651         {
 652                 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 653
 654                 if (win < 2)
 655                         win = 2;
 656                 tp->snd_cwnd = tp->t_maxseg;
 657                 tp->snd_wacked = 0;
 658                 tp->snd_ssthresh = win * tp->t_maxseg;
 659                 tp->t_dupacks = 0;
 660         }
 661         EXIT_FASTRECOVERY(tp);
 662         tcp_output(tp);
 663
 664 out:
 665 #ifdef TCPDEBUG
 666         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 667                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 668 #endif
 669         return tp;
 670 }
 671
 672 void
 673 tcp_timer_rexmt(void *xtp)
 674 {
 675         struct tcpcb *tp = xtp;
 676         struct callout *co = &tp->tt_rexmt->tc_callout;
 677
 678         crit_enter();
 679         if (callout_pending(co) || !callout_active(co)) {
 680                 crit_exit();
 681                 return;
 682         }
 683         callout_deactivate(co);
 684         tcp_send_timermsg(tp, TCP_TIMER_REXMT);
 685         crit_exit();
 686 }
 687
 688 static void
 689 tcp_timer_handler(netmsg_t msg)
 690 {
 691         struct netmsg_tcp_timer *tmsg = (struct netmsg_tcp_timer *)msg;
 692         const struct tcp_timer *tt;
 693         struct tcpcb *tp;
 694
 695         crit_enter();
 696
 697         KKASSERT(tmsg->tt_cpuid == mycpuid && tmsg->tt_tcb != NULL);
 698         tp = tmsg->tt_tcb;
 699
 700         /* Save pending tasks and reset the tasks in message */
 701         tmsg->tt_running_tasks = tmsg->tt_tasks;
 702         tmsg->tt_prev_tasks = tmsg->tt_tasks;
 703         tmsg->tt_tasks = 0;
 704
 705         /* Reply ASAP */
 706         lwkt_replymsg(&tmsg->tt_msg.lmsg, 0);
 707
 708         if (tmsg->tt_running_tasks == 0) {
 709                 /*
 710                  * All of the timers are cancelled when the message
 711                  * is pending; bail out.
 712                  */
 713                 crit_exit();
 714                 return;
 715         }
 716
 717         for (tt = tcp_timer_handlers; tt->tt_handler != NULL; ++tt) {
 718                 if ((tmsg->tt_running_tasks & tt->tt_task) == 0)
 719                         continue;
 720
 721                 tmsg->tt_running_tasks &= ~tt->tt_task;
 722                 tp = tt->tt_handler(tp);
 723                 if (tp == NULL)
 724                         break;
 725
 726                 if (tmsg->tt_running_tasks == 0) /* nothing left to do */
 727                         break;
 728         }
 729
 730         crit_exit();
 731 }
 732
 733 void
 734 tcp_create_timermsg(struct tcpcb *tp, struct lwkt_port *msgport)
 735 {
 736         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 737
 738         netmsg_init(&tmsg->tt_msg, NULL, &netisr_adone_rport,
 739                     MSGF_DROPABLE | MSGF_PRIORITY, tcp_timer_handler);
 740         tmsg->tt_cpuid = mycpuid;
 741         tmsg->tt_msgport = msgport;
 742         tmsg->tt_tcb = tp;
 743         tmsg->tt_tasks = 0;
 744 }
 745
 746 void
 747 tcp_destroy_timermsg(struct tcpcb *tp)
 748 {
 749         struct netmsg_tcp_timer *tmsg = tp->tt_msg;
 750
 751         if (tmsg == NULL ||             /* listen socket */
 752             tmsg->tt_tcb == NULL)       /* only tcp_attach() is called */
 753                 return;
 754
 755         KKASSERT(tmsg->tt_cpuid == mycpuid);
 756
 757         /*
 758          * This message is still pending to be processed;
 759          * drop it.  Optimized.
 760          */
 761         crit_enter();
 762         if ((tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE) == 0) {
 763                 lwkt_dropmsg(&tmsg->tt_msg.lmsg);
 764         }
 765         crit_exit();
 766 }
 767
 768 static __inline void
 769 tcp_callout_init(struct tcp_callout *tc, uint32_t task)
 770 {
 771         callout_init_mp(&tc->tc_callout);
 772         tc->tc_task = task;
 773 }
 774
 775 void
 776 tcp_inittimers(struct tcpcb *tp)
 777 {
 778         tcp_callout_init(tp->tt_rexmt, TCP_TIMER_REXMT);
 779         tcp_callout_init(tp->tt_persist, TCP_TIMER_PERSIST);
 780         tcp_callout_init(tp->tt_keep, TCP_TIMER_KEEP);
 781         tcp_callout_init(tp->tt_2msl, TCP_TIMER_2MSL);
 782         tcp_callout_init(tp->tt_delack, TCP_TIMER_DELACK);
 783 }