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) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
  36  *
  37  * License terms: all terms for the DragonFly license above plus the following:
  38  *
  39  * 4. All advertising materials mentioning features or use of this software
  40  *    must display the following acknowledgement:
  41  *
  42  *      This product includes software developed by Jeffrey M. Hsu
  43  *      for the DragonFly Project.
  44  *
  45  *    This requirement may be waived with permission from Jeffrey Hsu.
  46  *    This requirement will sunset and may be removed on July 8 2005,
  47  *    after which the standard DragonFly license (as shown above) will
  48  *    apply.
  49  */
  50
  51 /*
  52  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
  53  *      The Regents of the University of California.  All rights reserved.
  54  *
  55  * Redistribution and use in source and binary forms, with or without
  56  * modification, are permitted provided that the following conditions
  57  * are met:
  58  * 1. Redistributions of source code must retain the above copyright
  59  *    notice, this list of conditions and the following disclaimer.
  60  * 2. Redistributions in binary form must reproduce the above copyright
  61  *    notice, this list of conditions and the following disclaimer in the
  62  *    documentation and/or other materials provided with the distribution.
  63  * 3. All advertising materials mentioning features or use of this software
  64  *    must display the following acknowledgement:
  65  *      This product includes software developed by the University of
  66  *      California, Berkeley and its contributors.
  67  * 4. Neither the name of the University nor the names of its contributors
  68  *    may be used to endorse or promote products derived from this software
  69  *    without specific prior written permission.
  70  *
  71  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  72  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  73  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  74  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  75  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  76  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  77  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  78  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  79  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  80  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  81  * SUCH DAMAGE.
  82  *
  83  *      @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
  84  * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
  85  * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.15 2005/06/02 23:52:42 dillon Exp $
  86  */
  87
  88 #include "opt_compat.h"
  89 #include "opt_inet6.h"
  90 #include "opt_tcpdebug.h"
  91
  92 #include <sys/param.h>
  93 #include <sys/systm.h>
  94 #include <sys/kernel.h>
  95 #include <sys/mbuf.h>
  96 #include <sys/sysctl.h>
  97 #include <sys/socket.h>
  98 #include <sys/socketvar.h>
  99 #include <sys/protosw.h>
 100 #include <sys/thread.h>
 101 #include <sys/globaldata.h>
 102 #include <sys/thread2.h>
 103
 104 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
 105
 106 #include <net/route.h>
 107
 108 #include <netinet/in.h>
 109 #include <netinet/in_systm.h>
 110 #include <netinet/in_pcb.h>
 111 #ifdef INET6
 112 #include <netinet6/in6_pcb.h>
 113 #endif
 114 #include <netinet/ip_var.h>
 115 #include <netinet/tcp.h>
 116 #include <netinet/tcp_fsm.h>
 117 #include <netinet/tcp_seq.h>
 118 #include <netinet/tcp_timer.h>
 119 #include <netinet/tcp_var.h>
 120 #include <netinet/tcpip.h>
 121 #ifdef TCPDEBUG
 122 #include <netinet/tcp_debug.h>
 123 #endif
 124
 125 static int
 126 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
 127 {
 128         int error, s, tt;
 129
 130         tt = *(int *)oidp->oid_arg1;
 131         s = (int)((int64_t)tt * 1000 / hz);
 132
 133         error = sysctl_handle_int(oidp, &s, 0, req);
 134         if (error || !req->newptr)
 135                 return (error);
 136
 137         tt = (int)((int64_t)s * hz / 1000);
 138         if (tt < 1)
 139                 return (EINVAL);
 140
 141         *(int *)oidp->oid_arg1 = tt;
 142         return (0);
 143 }
 144
 145 int     tcp_keepinit;
 146 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
 147     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
 148
 149 int     tcp_keepidle;
 150 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
 151     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
 152
 153 int     tcp_keepintvl;
 154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
 155     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
 156
 157 int     tcp_delacktime;
 158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
 159     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 160     "Time before a delayed ACK is sent");
 161
 162 int     tcp_msl;
 163 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 164     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 165
 166 int     tcp_rexmit_min;
 167 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 168     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
 169
 170 int     tcp_rexmit_slop;
 171 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 172     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 173     "Retransmission Timer Slop");
 174
 175 static int      always_keepalive = 0;
 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 177     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 178
 179 static int      tcp_keepcnt = TCPTV_KEEPCNT;
 180         /* max idle probes */
 181 int     tcp_maxpersistidle;
 182         /* max idle time in persist */
 183 int     tcp_maxidle;
 184
 185 /*
 186  * Tcp protocol timeout routine called every 500 ms.
 187  * Updates timestamps used for TCP
 188  * causes finite state machine actions if timers expire.
 189  */
 190 void
 191 tcp_slowtimo(void)
 192 {
 193         crit_enter();
 194         tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
 195         crit_exit();
 196 }
 197
 198 /*
 199  * Cancel all timers for TCP tp.
 200  */
 201 void
 202 tcp_canceltimers(struct tcpcb *tp)
 203 {
 204         callout_stop(tp->tt_2msl);
 205         callout_stop(tp->tt_persist);
 206         callout_stop(tp->tt_keep);
 207         callout_stop(tp->tt_rexmt);
 208 }
 209
 210 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 211     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 212
 213 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 214     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 215
 216 static int tcp_totbackoff = 511;        /* sum of tcp_backoff[] */
 217
 218 /*
 219  * TCP timer processing.
 220  */
 221 void
 222 tcp_timer_delack(void *xtp)
 223 {
 224         struct tcpcb *tp = xtp;
 225
 226         crit_enter();
 227         if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
 228                 crit_exit();
 229                 return;
 230         }
 231         callout_deactivate(tp->tt_delack);
 232
 233         tp->t_flags |= TF_ACKNOW;
 234         tcpstat.tcps_delack++;
 235         tcp_output(tp);
 236         crit_exit();
 237 }
 238
 239 void
 240 tcp_timer_2msl(void *xtp)
 241 {
 242         struct tcpcb *tp = xtp;
 243 #ifdef TCPDEBUG
 244         int ostate;
 245
 246         ostate = tp->t_state;
 247 #endif
 248         crit_enter();
 249         if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
 250                 crit_exit();
 251                 return;
 252         }
 253         callout_deactivate(tp->tt_2msl);
 254         /*
 255          * 2 MSL timeout in shutdown went off.  If we're closed but
 256          * still waiting for peer to close and connection has been idle
 257          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 258          * control block.  Otherwise, check again in a bit.
 259          */
 260         if (tp->t_state != TCPS_TIME_WAIT &&
 261             (ticks - tp->t_rcvtime) <= tcp_maxidle)
 262                 callout_reset(tp->tt_2msl, tcp_keepintvl,
 263                               tcp_timer_2msl, tp);
 264         else
 265                 tp = tcp_close(tp);
 266
 267 #ifdef TCPDEBUG
 268         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 269                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 270 #endif
 271         crit_exit();
 272 }
 273
 274 void
 275 tcp_timer_keep(void *xtp)
 276 {
 277         struct tcpcb *tp = xtp;
 278         struct tcptemp *t_template;
 279 #ifdef TCPDEBUG
 280         int ostate;
 281
 282         ostate = tp->t_state;
 283 #endif
 284         crit_enter();
 285         if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
 286                 crit_exit();
 287                 return;
 288         }
 289         callout_deactivate(tp->tt_keep);
 290         /*
 291          * Keep-alive timer went off; send something
 292          * or drop connection if idle for too long.
 293          */
 294         tcpstat.tcps_keeptimeo++;
 295         if (tp->t_state < TCPS_ESTABLISHED)
 296                 goto dropit;
 297         if ((always_keepalive ||
 298              tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
 299             tp->t_state <= TCPS_CLOSING) {
 300                 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
 301                         goto dropit;
 302                 /*
 303                  * Send a packet designed to force a response
 304                  * if the peer is up and reachable:
 305                  * either an ACK if the connection is still alive,
 306                  * or an RST if the peer has closed the connection
 307                  * due to timeout or reboot.
 308                  * Using sequence number tp->snd_una-1
 309                  * causes the transmitted zero-length segment
 310                  * to lie outside the receive window;
 311                  * by the protocol spec, this requires the
 312                  * correspondent TCP to respond.
 313                  */
 314                 tcpstat.tcps_keepprobe++;
 315                 t_template = tcp_maketemplate(tp);
 316                 if (t_template) {
 317                         tcp_respond(tp, t_template->tt_ipgen,
 318                                     &t_template->tt_t, (struct mbuf *)NULL,
 319                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 320                         tcp_freetemplate(t_template);
 321                 }
 322                 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
 323         } else
 324                 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
 325
 326 #ifdef TCPDEBUG
 327         if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 328                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 329 #endif
 330         crit_exit();
 331         return;
 332
 333 dropit:
 334         tcpstat.tcps_keepdrops++;
 335         tp = tcp_drop(tp, ETIMEDOUT);
 336
 337 #ifdef TCPDEBUG
 338         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 339                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 340 #endif
 341         crit_exit();
 342 }
 343
 344 void
 345 tcp_timer_persist(void *xtp)
 346 {
 347         struct tcpcb *tp = xtp;
 348 #ifdef TCPDEBUG
 349         int ostate;
 350
 351         ostate = tp->t_state;
 352 #endif
 353         crit_enter();
 354         if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
 355                 crit_exit();
 356                 return;
 357         }
 358         callout_deactivate(tp->tt_persist);
 359         /*
 360          * Persistance timer into zero window.
 361          * Force a byte to be output, if possible.
 362          */
 363         tcpstat.tcps_persisttimeo++;
 364         /*
 365          * Hack: if the peer is dead/unreachable, we do not
 366          * time out if the window is closed.  After a full
 367          * backoff, drop the connection if the idle time
 368          * (no responses to probes) reaches the maximum
 369          * backoff that we would use if retransmitting.
 370          */
 371         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 372             ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
 373              (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 374                 tcpstat.tcps_persistdrop++;
 375                 tp = tcp_drop(tp, ETIMEDOUT);
 376                 goto out;
 377         }
 378         tcp_setpersist(tp);
 379         tp->t_flags |= TF_FORCE;
 380         tcp_output(tp);
 381         tp->t_flags &= ~TF_FORCE;
 382
 383 out:
 384 #ifdef TCPDEBUG
 385         if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 386                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 387 #endif
 388         crit_exit();
 389 }
 390
 391 void
 392 tcp_save_congestion_state(struct tcpcb *tp)
 393 {
 394         tp->snd_cwnd_prev = tp->snd_cwnd;
 395         tp->snd_wacked_prev = tp->snd_wacked;
 396         tp->snd_ssthresh_prev = tp->snd_ssthresh;
 397         tp->snd_recover_prev = tp->snd_recover;
 398         if (IN_FASTRECOVERY(tp))
 399                 tp->t_flags |= TF_WASFRECOVERY;
 400         else
 401                 tp->t_flags &= ~TF_WASFRECOVERY;
 402         if (tp->t_flags & TF_RCVD_TSTMP) {
 403                 tp->t_rexmtTS = ticks;
 404                 tp->t_flags |= TF_FIRSTACCACK;
 405         }
 406 #ifdef later
 407         tcp_sack_save_scoreboard(&tp->scb);
 408 #endif
 409 }
 410
 411 void
 412 tcp_revert_congestion_state(struct tcpcb *tp)
 413 {
 414         tp->snd_cwnd = tp->snd_cwnd_prev;
 415         tp->snd_wacked = tp->snd_wacked_prev;
 416         tp->snd_ssthresh = tp->snd_ssthresh_prev;
 417         tp->snd_recover = tp->snd_recover_prev;
 418         if (tp->t_flags & TF_WASFRECOVERY)
 419                 ENTER_FASTRECOVERY(tp);
 420         if (tp->t_flags & TF_FASTREXMT) {
 421                 ++tcpstat.tcps_sndfastrexmitbad;
 422                 if (tp->t_flags & TF_EARLYREXMT)
 423                         ++tcpstat.tcps_sndearlyrexmitbad;
 424         } else
 425                 ++tcpstat.tcps_sndrtobad;
 426         tp->t_badrxtwin = 0;
 427         tp->t_rxtshift = 0;
 428         tp->snd_nxt = tp->snd_max;
 429 #ifdef later
 430         tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una);
 431 #endif
 432 }
 433
 434 void
 435 tcp_timer_rexmt(void *xtp)
 436 {
 437         struct tcpcb *tp = xtp;
 438         int rexmt;
 439 #ifdef TCPDEBUG
 440         int ostate;
 441
 442         ostate = tp->t_state;
 443 #endif
 444         crit_enter();
 445         if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
 446                 crit_exit();
 447                 return;
 448         }
 449         callout_deactivate(tp->tt_rexmt);
 450         /*
 451          * Retransmission timer went off.  Message has not
 452          * been acked within retransmit interval.  Back off
 453          * to a longer retransmit interval and retransmit one segment.
 454          */
 455         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 456                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 457                 tcpstat.tcps_timeoutdrop++;
 458                 tp = tcp_drop(tp, tp->t_softerror ?
 459                               tp->t_softerror : ETIMEDOUT);
 460                 goto out;
 461         }
 462         if (tp->t_rxtshift == 1) {
 463                 /*
 464                  * first retransmit; record ssthresh and cwnd so they can
 465                  * be recovered if this turns out to be a "bad" retransmit.
 466                  * A retransmit is considered "bad" if an ACK for this
 467                  * segment is received within RTT/2 interval; the assumption
 468                  * here is that the ACK was already in flight.  See
 469                  * "On Estimating End-to-End Network Path Properties" by
 470                  * Allman and Paxson for more details.
 471                  */
 472                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 473                 tcp_save_congestion_state(tp);
 474                 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
 475         }
 476         /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
 477         tcp_sack_cleanup(&tp->scb);
 478         tcpstat.tcps_rexmttimeo++;
 479         if (tp->t_state == TCPS_SYN_SENT)
 480                 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
 481         else
 482                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 483         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 484                       tp->t_rttmin, TCPTV_REXMTMAX);
 485         /*
 486          * Disable rfc1323 and rfc1644 if we havn't got any response to
 487          * our third SYN to work-around some broken terminal servers
 488          * (most of which have hopefully been retired) that have bad VJ
 489          * header compression code which trashes TCP segments containing
 490          * unknown-to-them TCP options.
 491          */
 492         if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
 493                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
 494         /*
 495          * If losing, let the lower level know and try for
 496          * a better route.  Also, if we backed off this far,
 497          * our srtt estimate is probably bogus.  Clobber it
 498          * so we'll take the next rtt measurement as our srtt;
 499          * move the current srtt into rttvar to keep the current
 500          * retransmit times until then.
 501          */
 502         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 503 #ifdef INET6
 504                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 505                         in6_losing(tp->t_inpcb);
 506                 else
 507 #endif
 508                 in_losing(tp->t_inpcb);
 509                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 510                 tp->t_srtt = 0;
 511         }
 512         tp->snd_nxt = tp->snd_una;
 513         tp->rexmt_high = tp->snd_una;
 514         tp->snd_recover = tp->snd_max;
 515         /*
 516          * Force a segment to be sent.
 517          */
 518         tp->t_flags |= TF_ACKNOW;
 519         /*
 520          * If timing a segment in this window, stop the timer.
 521          */
 522         tp->t_rtttime = 0;
 523         /*
 524          * Close the congestion window down to one segment
 525          * (we'll open it by one segment for each ack we get).
 526          * Since we probably have a window's worth of unacked
 527          * data accumulated, this "slow start" keeps us from
 528          * dumping all that data as back-to-back packets (which
 529          * might overwhelm an intermediate gateway).
 530          *
 531          * There are two phases to the opening: Initially we
 532          * open by one mss on each ack.  This makes the window
 533          * size increase exponentially with time.  If the
 534          * window is larger than the path can handle, this
 535          * exponential growth results in dropped packet(s)
 536          * almost immediately.  To get more time between
 537          * drops but still "push" the network to take advantage
 538          * of improving conditions, we switch from exponential
 539          * to linear window opening at some threshhold size.
 540          * For a threshhold, we use half the current window
 541          * size, truncated to a multiple of the mss.
 542          *
 543          * (the minimum cwnd that will give us exponential
 544          * growth is 2 mss.  We don't allow the threshhold
 545          * to go below this.)
 546          */
 547         {
 548                 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
 549
 550                 if (win < 2)
 551                         win = 2;
 552                 tp->snd_cwnd = tp->t_maxseg;
 553                 tp->snd_wacked = 0;
 554                 tp->snd_ssthresh = win * tp->t_maxseg;
 555                 tp->t_dupacks = 0;
 556         }
 557         EXIT_FASTRECOVERY(tp);
 558         tcp_output(tp);
 559
 560 out:
 561 #ifdef TCPDEBUG
 562         if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 563                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 564 #endif
 565         crit_exit();
 566 }