sys/netinet/tcp_timer.c

   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/socket.h>
  45 #include <sys/socketvar.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/systm.h>
  48
  49 #include <net/if.h>
  50 #include <net/route.h>
  51 #include <net/vnet.h>
  52
  53 #include <netinet/in.h>
  54 #include <netinet/in_pcb.h>
  55 #include <netinet/in_systm.h>
  56 #ifdef INET6
  57 #include <netinet6/in6_pcb.h>
  58 #endif
  59 #include <netinet/ip_var.h>
  60 #include <netinet/tcp.h>
  61 #include <netinet/tcp_fsm.h>
  62 #include <netinet/tcp_timer.h>
  63 #include <netinet/tcp_var.h>
  64 #include <netinet/tcpip.h>
  65 #ifdef TCPDEBUG
  66 #include <netinet/tcp_debug.h>
  67 #endif
  68
  69 int     tcp_keepinit;
  70 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
  71     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
  72
  73 int     tcp_keepidle;
  74 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
  75     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
  76
  77 int     tcp_keepintvl;
  78 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
  79     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
  80
  81 int     tcp_delacktime;
  82 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
  83     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
  84     "Time before a delayed ACK is sent");
  85
  86 int     tcp_msl;
  87 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
  88     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
  89
  90 int     tcp_rexmit_min;
  91 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
  92     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
  93     "Minimum Retransmission Timeout");
  94
  95 int     tcp_rexmit_slop;
  96 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
  97     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
  98     "Retransmission Timer Slop");
  99
 100 static int      always_keepalive = 1;
 101 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 102     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 103
 104 int    tcp_fast_finwait2_recycle = 0;
 105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 106     &tcp_fast_finwait2_recycle, 0,
 107     "Recycle closed FIN_WAIT_2 connections faster");
 108
 109 int    tcp_finwait2_timeout;
 110 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 111     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 112
 113
 114 static int      tcp_keepcnt = TCPTV_KEEPCNT;
 115         /* max idle probes */
 116 int     tcp_maxpersistidle;
 117         /* max idle time in persist */
 118 int     tcp_maxidle;
 119
 120 /*
 121  * Tcp protocol timeout routine called every 500 ms.
 122  * Updates timestamps used for TCP
 123  * causes finite state machine actions if timers expire.
 124  */
 125 void
 126 tcp_slowtimo(void)
 127 {
 128         VNET_ITERATOR_DECL(vnet_iter);
 129
 130         VNET_LIST_RLOCK_NOSLEEP();
 131         VNET_FOREACH(vnet_iter) {
 132                 CURVNET_SET(vnet_iter);
 133                 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
 134                 INP_INFO_WLOCK(&V_tcbinfo);
 135                 (void) tcp_tw_2msl_scan(0);
 136                 INP_INFO_WUNLOCK(&V_tcbinfo);
 137                 CURVNET_RESTORE();
 138         }
 139         VNET_LIST_RUNLOCK_NOSLEEP();
 140 }
 141
 142 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 143     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 144
 145 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 146     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 147
 148 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 149
 150 static int tcp_timer_race;
 151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
 152     0, "Count of t_inpcb races on tcp_discardcb");
 153
 154 /*
 155  * TCP timer processing.
 156  */
 157
 158 void
 159 tcp_timer_delack(void *xtp)
 160 {
 161         struct tcpcb *tp = xtp;
 162         struct inpcb *inp;
 163         CURVNET_SET(tp->t_vnet);
 164
 165         inp = tp->t_inpcb;
 166         /*
 167          * XXXRW: While this assert is in fact correct, bugs in the tcpcb
 168          * tear-down mean we need it as a work-around for races between
 169          * timers and tcp_discardcb().
 170          *
 171          * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
 172          */
 173         if (inp == NULL) {
 174                 tcp_timer_race++;
 175                 CURVNET_RESTORE();
 176                 return;
 177         }
 178         INP_WLOCK(inp);
 179         if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack)
 180             || !callout_active(&tp->t_timers->tt_delack)) {
 181                 INP_WUNLOCK(inp);
 182                 CURVNET_RESTORE();
 183                 return;
 184         }
 185         callout_deactivate(&tp->t_timers->tt_delack);
 186
 187         tp->t_flags |= TF_ACKNOW;
 188         TCPSTAT_INC(tcps_delack);
 189         (void) tcp_output(tp);
 190         INP_WUNLOCK(inp);
 191         CURVNET_RESTORE();
 192 }
 193
 194 void
 195 tcp_timer_2msl(void *xtp)
 196 {
 197         struct tcpcb *tp = xtp;
 198         struct inpcb *inp;
 199         CURVNET_SET(tp->t_vnet);
 200 #ifdef TCPDEBUG
 201         int ostate;
 202
 203         ostate = tp->t_state;
 204 #endif
 205         /*
 206          * XXXRW: Does this actually happen?
 207          */
 208         INP_INFO_WLOCK(&V_tcbinfo);
 209         inp = tp->t_inpcb;
 210         /*
 211          * XXXRW: While this assert is in fact correct, bugs in the tcpcb
 212          * tear-down mean we need it as a work-around for races between
 213          * timers and tcp_discardcb().
 214          *
 215          * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
 216          */
 217         if (inp == NULL) {
 218                 tcp_timer_race++;
 219                 INP_INFO_WUNLOCK(&V_tcbinfo);
 220                 CURVNET_RESTORE();
 221                 return;
 222         }
 223         INP_WLOCK(inp);
 224         tcp_free_sackholes(tp);
 225         if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) ||
 226             !callout_active(&tp->t_timers->tt_2msl)) {
 227                 INP_WUNLOCK(tp->t_inpcb);
 228                 INP_INFO_WUNLOCK(&V_tcbinfo);
 229                 CURVNET_RESTORE();
 230                 return;
 231         }
 232         callout_deactivate(&tp->t_timers->tt_2msl);
 233         /*
 234          * 2 MSL timeout in shutdown went off.  If we're closed but
 235          * still waiting for peer to close and connection has been idle
 236          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 237          * control block.  Otherwise, check again in a bit.
 238          *
 239          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 240          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 241          * Ignore fact that there were recent incoming segments.
 242          */
 243         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 244             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 245             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 246                 TCPSTAT_INC(tcps_finwait2_drops);
 247                 tp = tcp_close(tp);
 248         } else {
 249                 if (tp->t_state != TCPS_TIME_WAIT &&
 250                    ticks - tp->t_rcvtime <= tcp_maxidle)
 251                        callout_reset(&tp->t_timers->tt_2msl, tcp_keepintvl,
 252                                      tcp_timer_2msl, tp);
 253                else
 254                        tp = tcp_close(tp);
 255        }
 256
 257 #ifdef TCPDEBUG
 258         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 259                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 260                           PRU_SLOWTIMO);
 261 #endif
 262         if (tp != NULL)
 263                 INP_WUNLOCK(inp);
 264         INP_INFO_WUNLOCK(&V_tcbinfo);
 265         CURVNET_RESTORE();
 266 }
 267
 268 void
 269 tcp_timer_keep(void *xtp)
 270 {
 271         struct tcpcb *tp = xtp;
 272         struct tcptemp *t_template;
 273         struct inpcb *inp;
 274         CURVNET_SET(tp->t_vnet);
 275 #ifdef TCPDEBUG
 276         int ostate;
 277
 278         ostate = tp->t_state;
 279 #endif
 280         INP_INFO_WLOCK(&V_tcbinfo);
 281         inp = tp->t_inpcb;
 282         /*
 283          * XXXRW: While this assert is in fact correct, bugs in the tcpcb
 284          * tear-down mean we need it as a work-around for races between
 285          * timers and tcp_discardcb().
 286          *
 287          * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
 288          */
 289         if (inp == NULL) {
 290                 tcp_timer_race++;
 291                 INP_INFO_WUNLOCK(&V_tcbinfo);
 292                 CURVNET_RESTORE();
 293                 return;
 294         }
 295         INP_WLOCK(inp);
 296         if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep)
 297             || !callout_active(&tp->t_timers->tt_keep)) {
 298                 INP_WUNLOCK(inp);
 299                 INP_INFO_WUNLOCK(&V_tcbinfo);
 300                 CURVNET_RESTORE();
 301                 return;
 302         }
 303         callout_deactivate(&tp->t_timers->tt_keep);
 304         /*
 305          * Keep-alive timer went off; send something
 306          * or drop connection if idle for too long.
 307          */
 308         TCPSTAT_INC(tcps_keeptimeo);
 309         if (tp->t_state < TCPS_ESTABLISHED)
 310                 goto dropit;
 311         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
 312             tp->t_state <= TCPS_CLOSING) {
 313                 if (ticks - tp->t_rcvtime >= tcp_keepidle + tcp_maxidle)
 314                         goto dropit;
 315                 /*
 316                  * Send a packet designed to force a response
 317                  * if the peer is up and reachable:
 318                  * either an ACK if the connection is still alive,
 319                  * or an RST if the peer has closed the connection
 320                  * due to timeout or reboot.
 321                  * Using sequence number tp->snd_una-1
 322                  * causes the transmitted zero-length segment
 323                  * to lie outside the receive window;
 324                  * by the protocol spec, this requires the
 325                  * correspondent TCP to respond.
 326                  */
 327                 TCPSTAT_INC(tcps_keepprobe);
 328                 t_template = tcpip_maketemplate(inp);
 329                 if (t_template) {
 330                         tcp_respond(tp, t_template->tt_ipgen,
 331                                     &t_template->tt_t, (struct mbuf *)NULL,
 332                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 333                         free(t_template, M_TEMP);
 334                 }
 335                 callout_reset(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
 336         } else
 337                 callout_reset(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
 338
 339 #ifdef TCPDEBUG
 340         if (inp->inp_socket->so_options & SO_DEBUG)
 341                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 342                           PRU_SLOWTIMO);
 343 #endif
 344         INP_WUNLOCK(inp);
 345         INP_INFO_WUNLOCK(&V_tcbinfo);
 346         CURVNET_RESTORE();
 347         return;
 348
 349 dropit:
 350         TCPSTAT_INC(tcps_keepdrops);
 351         tp = tcp_drop(tp, ETIMEDOUT);
 352
 353 #ifdef TCPDEBUG
 354         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 355                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 356                           PRU_SLOWTIMO);
 357 #endif
 358         if (tp != NULL)
 359                 INP_WUNLOCK(tp->t_inpcb);
 360         INP_INFO_WUNLOCK(&V_tcbinfo);
 361         CURVNET_RESTORE();
 362 }
 363
 364 void
 365 tcp_timer_persist(void *xtp)
 366 {
 367         struct tcpcb *tp = xtp;
 368         struct inpcb *inp;
 369         CURVNET_SET(tp->t_vnet);
 370 #ifdef TCPDEBUG
 371         int ostate;
 372
 373         ostate = tp->t_state;
 374 #endif
 375         INP_INFO_WLOCK(&V_tcbinfo);
 376         inp = tp->t_inpcb;
 377         /*
 378          * XXXRW: While this assert is in fact correct, bugs in the tcpcb
 379          * tear-down mean we need it as a work-around for races between
 380          * timers and tcp_discardcb().
 381          *
 382          * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
 383          */
 384         if (inp == NULL) {
 385                 tcp_timer_race++;
 386                 INP_INFO_WUNLOCK(&V_tcbinfo);
 387                 CURVNET_RESTORE();
 388                 return;
 389         }
 390         INP_WLOCK(inp);
 391         if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
 392             || !callout_active(&tp->t_timers->tt_persist)) {
 393                 INP_WUNLOCK(inp);
 394                 INP_INFO_WUNLOCK(&V_tcbinfo);
 395                 CURVNET_RESTORE();
 396                 return;
 397         }
 398         callout_deactivate(&tp->t_timers->tt_persist);
 399         /*
 400          * Persistance timer into zero window.
 401          * Force a byte to be output, if possible.
 402          */
 403         TCPSTAT_INC(tcps_persisttimeo);
 404         /*
 405          * Hack: if the peer is dead/unreachable, we do not
 406          * time out if the window is closed.  After a full
 407          * backoff, drop the connection if the idle time
 408          * (no responses to probes) reaches the maximum
 409          * backoff that we would use if retransmitting.
 410          */
 411         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 412             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 413              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 414                 TCPSTAT_INC(tcps_persistdrop);
 415                 tp = tcp_drop(tp, ETIMEDOUT);
 416                 goto out;
 417         }
 418         tcp_setpersist(tp);
 419         tp->t_flags |= TF_FORCEDATA;
 420         (void) tcp_output(tp);
 421         tp->t_flags &= ~TF_FORCEDATA;
 422
 423 out:
 424 #ifdef TCPDEBUG
 425         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 426                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 427 #endif
 428         if (tp != NULL)
 429                 INP_WUNLOCK(inp);
 430         INP_INFO_WUNLOCK(&V_tcbinfo);
 431         CURVNET_RESTORE();
 432 }
 433
 434 void
 435 tcp_timer_rexmt(void * xtp)
 436 {
 437         struct tcpcb *tp = xtp;
 438         CURVNET_SET(tp->t_vnet);
 439         int rexmt;
 440         int headlocked;
 441         struct inpcb *inp;
 442 #ifdef TCPDEBUG
 443         int ostate;
 444
 445         ostate = tp->t_state;
 446 #endif
 447         INP_INFO_WLOCK(&V_tcbinfo);
 448         headlocked = 1;
 449         inp = tp->t_inpcb;
 450         /*
 451          * XXXRW: While this assert is in fact correct, bugs in the tcpcb
 452          * tear-down mean we need it as a work-around for races between
 453          * timers and tcp_discardcb().
 454          *
 455          * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
 456          */
 457         if (inp == NULL) {
 458                 tcp_timer_race++;
 459                 INP_INFO_WUNLOCK(&V_tcbinfo);
 460                 CURVNET_RESTORE();
 461                 return;
 462         }
 463         INP_WLOCK(inp);
 464         if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt)
 465             || !callout_active(&tp->t_timers->tt_rexmt)) {
 466                 INP_WUNLOCK(inp);
 467                 INP_INFO_WUNLOCK(&V_tcbinfo);
 468                 CURVNET_RESTORE();
 469                 return;
 470         }
 471         callout_deactivate(&tp->t_timers->tt_rexmt);
 472         tcp_free_sackholes(tp);
 473         /*
 474          * Retransmission timer went off.  Message has not
 475          * been acked within retransmit interval.  Back off
 476          * to a longer retransmit interval and retransmit one segment.
 477          */
 478         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 479                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 480                 TCPSTAT_INC(tcps_timeoutdrop);
 481                 tp = tcp_drop(tp, tp->t_softerror ?
 482                               tp->t_softerror : ETIMEDOUT);
 483                 goto out;
 484         }
 485         INP_INFO_WUNLOCK(&V_tcbinfo);
 486         headlocked = 0;
 487         if (tp->t_rxtshift == 1) {
 488                 /*
 489                  * first retransmit; record ssthresh and cwnd so they can
 490                  * be recovered if this turns out to be a "bad" retransmit.
 491                  * A retransmit is considered "bad" if an ACK for this
 492                  * segment is received within RTT/2 interval; the assumption
 493                  * here is that the ACK was already in flight.  See
 494                  * "On Estimating End-to-End Network Path Properties" by
 495                  * Allman and Paxson for more details.
 496                  */
 497                 tp->snd_cwnd_prev = tp->snd_cwnd;
 498                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 499                 tp->snd_recover_prev = tp->snd_recover;
 500                 if (IN_FASTRECOVERY(tp))
 501                   tp->t_flags |= TF_WASFRECOVERY;
 502                 else
 503                   tp->t_flags &= ~TF_WASFRECOVERY;
 504                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 505         }
 506         TCPSTAT_INC(tcps_rexmttimeo);
 507         if (tp->t_state == TCPS_SYN_SENT)
 508                 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
 509         else
 510                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 511         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 512                       tp->t_rttmin, TCPTV_REXMTMAX);
 513         /*
 514          * Disable rfc1323 if we havn't got any response to
 515          * our third SYN to work-around some broken terminal servers
 516          * (most of which have hopefully been retired) that have bad VJ
 517          * header compression code which trashes TCP segments containing
 518          * unknown-to-them TCP options.
 519          */
 520         if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
 521                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
 522         /*
 523          * If we backed off this far, our srtt estimate is probably bogus.
 524          * Clobber it so we'll take the next rtt measurement as our srtt;
 525          * move the current srtt into rttvar to keep the current
 526          * retransmit times until then.
 527          */
 528         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 529 #ifdef INET6
 530                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 531                         in6_losing(tp->t_inpcb);
 532                 else
 533 #endif
 534                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 535                 tp->t_srtt = 0;
 536         }
 537         tp->snd_nxt = tp->snd_una;
 538         tp->snd_recover = tp->snd_max;
 539         /*
 540          * Force a segment to be sent.
 541          */
 542         tp->t_flags |= TF_ACKNOW;
 543         /*
 544          * If timing a segment in this window, stop the timer.
 545          */
 546         tp->t_rtttime = 0;
 547         /*
 548          * Close the congestion window down to one segment
 549          * (we'll open it by one segment for each ack we get).
 550          * Since we probably have a window's worth of unacked
 551          * data accumulated, this "slow start" keeps us from
 552          * dumping all that data as back-to-back packets (which
 553          * might overwhelm an intermediate gateway).
 554          *
 555          * There are two phases to the opening: Initially we
 556          * open by one mss on each ack.  This makes the window
 557          * size increase exponentially with time.  If the
 558          * window is larger than the path can handle, this
 559          * exponential growth results in dropped packet(s)
 560          * almost immediately.  To get more time between
 561          * drops but still "push" the network to take advantage
 562          * of improving conditions, we switch from exponential
 563          * to linear window opening at some threshhold size.
 564          * For a threshhold, we use half the current window
 565          * size, truncated to a multiple of the mss.
 566          *
 567          * (the minimum cwnd that will give us exponential
 568          * growth is 2 mss.  We don't allow the threshhold
 569          * to go below this.)
 570          */
 571         {
 572                 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 573                 if (win < 2)
 574                         win = 2;
 575                 tp->snd_cwnd = tp->t_maxseg;
 576                 tp->snd_ssthresh = win * tp->t_maxseg;
 577                 tp->t_dupacks = 0;
 578         }
 579         EXIT_FASTRECOVERY(tp);
 580         tp->t_bytes_acked = 0;
 581         (void) tcp_output(tp);
 582
 583 out:
 584 #ifdef TCPDEBUG
 585         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 586                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 587                           PRU_SLOWTIMO);
 588 #endif
 589         if (tp != NULL)
 590                 INP_WUNLOCK(inp);
 591         if (headlocked)
 592                 INP_INFO_WUNLOCK(&V_tcbinfo);
 593         CURVNET_RESTORE();
 594 }
 595
 596 void
 597 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
 598 {
 599         struct callout *t_callout;
 600         void *f_callout;
 601
 602         switch (timer_type) {
 603                 case TT_DELACK:
 604                         t_callout = &tp->t_timers->tt_delack;
 605                         f_callout = tcp_timer_delack;
 606                         break;
 607                 case TT_REXMT:
 608                         t_callout = &tp->t_timers->tt_rexmt;
 609                         f_callout = tcp_timer_rexmt;
 610                         break;
 611                 case TT_PERSIST:
 612                         t_callout = &tp->t_timers->tt_persist;
 613                         f_callout = tcp_timer_persist;
 614                         break;
 615                 case TT_KEEP:
 616                         t_callout = &tp->t_timers->tt_keep;
 617                         f_callout = tcp_timer_keep;
 618                         break;
 619                 case TT_2MSL:
 620                         t_callout = &tp->t_timers->tt_2msl;
 621                         f_callout = tcp_timer_2msl;
 622                         break;
 623                 default:
 624                         panic("bad timer_type");
 625                 }
 626         if (delta == 0) {
 627                 callout_stop(t_callout);
 628         } else {
 629                 callout_reset(t_callout, delta, f_callout, tp);
 630         }
 631 }
 632
 633 int
 634 tcp_timer_active(struct tcpcb *tp, int timer_type)
 635 {
 636         struct callout *t_callout;
 637
 638         switch (timer_type) {
 639                 case TT_DELACK:
 640                         t_callout = &tp->t_timers->tt_delack;
 641                         break;
 642                 case TT_REXMT:
 643                         t_callout = &tp->t_timers->tt_rexmt;
 644                         break;
 645                 case TT_PERSIST:
 646                         t_callout = &tp->t_timers->tt_persist;
 647                         break;
 648                 case TT_KEEP:
 649                         t_callout = &tp->t_timers->tt_keep;
 650                         break;
 651                 case TT_2MSL:
 652                         t_callout = &tp->t_timers->tt_2msl;
 653                         break;
 654                 default:
 655                         panic("bad timer_type");
 656                 }
 657         return callout_active(t_callout);
 658 }