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_inet.h"
  36 #include "opt_inet6.h"
  37 #include "opt_tcpdebug.h"
  38
  39 #include <sys/param.h>
  40 #include <sys/kernel.h>
  41 #include <sys/lock.h>
  42 #include <sys/mbuf.h>
  43 #include <sys/mutex.h>
  44 #include <sys/protosw.h>
  45 #include <sys/smp.h>
  46 #include <sys/socket.h>
  47 #include <sys/socketvar.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/systm.h>
  50
  51 #include <net/if.h>
  52 #include <net/route.h>
  53 #include <net/vnet.h>
  54
  55 #include <netinet/cc.h>
  56 #include <netinet/in.h>
  57 #include <netinet/in_pcb.h>
  58 #include <netinet/in_systm.h>
  59 #ifdef INET6
  60 #include <netinet6/in6_pcb.h>
  61 #endif
  62 #include <netinet/ip_var.h>
  63 #include <netinet/tcp_fsm.h>
  64 #include <netinet/tcp_timer.h>
  65 #include <netinet/tcp_var.h>
  66 #ifdef INET6
  67 #include <netinet6/tcp6_var.h>
  68 #endif
  69 #include <netinet/tcpip.h>
  70 #ifdef TCPDEBUG
  71 #include <netinet/tcp_debug.h>
  72 #endif
  73
  74 int     tcp_keepinit;
  75 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
  76     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
  77
  78 int     tcp_keepidle;
  79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
  80     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
  81
  82 int     tcp_keepintvl;
  83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
  84     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
  85
  86 int     tcp_delacktime;
  87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
  88     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
  89     "Time before a delayed ACK is sent");
  90
  91 int     tcp_msl;
  92 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
  93     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
  94
  95 int     tcp_rexmit_min;
  96 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
  97     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
  98     "Minimum Retransmission Timeout");
  99
 100 int     tcp_rexmit_slop;
 101 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 102     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 103     "Retransmission Timer Slop");
 104
 105 static int      always_keepalive = 1;
 106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 107     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 108
 109 int    tcp_fast_finwait2_recycle = 0;
 110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 111     &tcp_fast_finwait2_recycle, 0,
 112     "Recycle closed FIN_WAIT_2 connections faster");
 113
 114 int    tcp_finwait2_timeout;
 115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 116     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 117
 118 int     tcp_keepcnt = TCPTV_KEEPCNT;
 119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
 120     "Number of keepalive probes to send");
 121
 122         /* max idle probes */
 123 int     tcp_maxpersistidle;
 124
 125 static int      tcp_rexmit_drop_options = 0;
 126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
 127     &tcp_rexmit_drop_options, 0,
 128     "Drop TCP options from 3rd and later retransmitted SYN");
 129
 130 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
 131 #define V_tcp_pmtud_blackhole_detect    VNET(tcp_pmtud_blackhole_detect)
 132 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
 133     CTLFLAG_RW,
 134     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
 135     "Path MTU Discovery Black Hole Detection Enabled");
 136
 137 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
 138 #define V_tcp_pmtud_blackhole_activated \
 139     VNET(tcp_pmtud_blackhole_activated)
 140 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
 141     CTLFLAG_RD,
 142     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
 143     "Path MTU Discovery Black Hole Detection, Activation Count");
 144
 145 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
 146 #define V_tcp_pmtud_blackhole_activated_min_mss \
 147     VNET(tcp_pmtud_blackhole_activated_min_mss)
 148 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
 149     CTLFLAG_RD,
 150     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
 151     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
 152
 153 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
 154 #define V_tcp_pmtud_blackhole_failed    VNET(tcp_pmtud_blackhole_failed)
 155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
 156     CTLFLAG_RD,
 157     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
 158     "Path MTU Discovery Black Hole Detection, Failure Count");
 159
 160 #ifdef INET
 161 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
 162 #define V_tcp_pmtud_blackhole_mss       VNET(tcp_pmtud_blackhole_mss)
 163 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
 164     CTLFLAG_RW,
 165     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
 166     "Path MTU Discovery Black Hole Detection lowered MSS");
 167 #endif
 168
 169 #ifdef INET6
 170 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
 171 #define V_tcp_v6pmtud_blackhole_mss     VNET(tcp_v6pmtud_blackhole_mss)
 172 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
 173     CTLFLAG_RW,
 174     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
 175     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
 176 #endif
 177
 178 static int      per_cpu_timers = 0;
 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
 180     &per_cpu_timers , 0, "run tcp timers on all cpus");
 181
 182 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
 183                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
 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         VNET_ITERATOR_DECL(vnet_iter);
 194
 195         VNET_LIST_RLOCK_NOSLEEP();
 196         VNET_FOREACH(vnet_iter) {
 197                 CURVNET_SET(vnet_iter);
 198                 (void) tcp_tw_2msl_scan(0);
 199                 CURVNET_RESTORE();
 200         }
 201         VNET_LIST_RUNLOCK_NOSLEEP();
 202 }
 203
 204 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 205     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 206
 207 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 208     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 209
 210 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 211
 212 /*
 213  * TCP timer processing.
 214  */
 215
 216 void
 217 tcp_timer_delack(void *xtp)
 218 {
 219         struct tcpcb *tp = xtp;
 220         struct inpcb *inp;
 221         CURVNET_SET(tp->t_vnet);
 222
 223         inp = tp->t_inpcb;
 224         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 225         INP_WLOCK(inp);
 226         if (callout_pending(&tp->t_timers->tt_delack) ||
 227             !callout_active(&tp->t_timers->tt_delack)) {
 228                 INP_WUNLOCK(inp);
 229                 CURVNET_RESTORE();
 230                 return;
 231         }
 232         callout_deactivate(&tp->t_timers->tt_delack);
 233         if ((inp->inp_flags & INP_DROPPED) != 0) {
 234                 INP_WUNLOCK(inp);
 235                 CURVNET_RESTORE();
 236                 return;
 237         }
 238         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 239                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 240         KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
 241                 ("%s: tp %p delack callout should be running", __func__, tp));
 242
 243         tp->t_flags |= TF_ACKNOW;
 244         TCPSTAT_INC(tcps_delack);
 245         (void) tcp_output(tp);
 246         INP_WUNLOCK(inp);
 247         CURVNET_RESTORE();
 248 }
 249
 250 void
 251 tcp_timer_2msl(void *xtp)
 252 {
 253         struct tcpcb *tp = xtp;
 254         struct inpcb *inp;
 255         CURVNET_SET(tp->t_vnet);
 256 #ifdef TCPDEBUG
 257         int ostate;
 258
 259         ostate = tp->t_state;
 260 #endif
 261         INP_INFO_WLOCK(&V_tcbinfo);
 262         inp = tp->t_inpcb;
 263         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 264         INP_WLOCK(inp);
 265         tcp_free_sackholes(tp);
 266         if (callout_pending(&tp->t_timers->tt_2msl) ||
 267             !callout_active(&tp->t_timers->tt_2msl)) {
 268                 INP_WUNLOCK(tp->t_inpcb);
 269                 INP_INFO_WUNLOCK(&V_tcbinfo);
 270                 CURVNET_RESTORE();
 271                 return;
 272         }
 273         callout_deactivate(&tp->t_timers->tt_2msl);
 274         if ((inp->inp_flags & INP_DROPPED) != 0) {
 275                 INP_WUNLOCK(inp);
 276                 INP_INFO_WUNLOCK(&V_tcbinfo);
 277                 CURVNET_RESTORE();
 278                 return;
 279         }
 280         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 281                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 282         KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
 283                 ("%s: tp %p 2msl callout should be running", __func__, tp));
 284         /*
 285          * 2 MSL timeout in shutdown went off.  If we're closed but
 286          * still waiting for peer to close and connection has been idle
 287          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
 288          * control block.  Otherwise, check again in a bit.
 289          *
 290          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 291          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 292          * Ignore fact that there were recent incoming segments.
 293          */
 294         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 295             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 296             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 297                 TCPSTAT_INC(tcps_finwait2_drops);
 298                 tp = tcp_close(tp);
 299         } else {
 300                 if (tp->t_state != TCPS_TIME_WAIT &&
 301                    ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 302                         if (!callout_reset(&tp->t_timers->tt_2msl,
 303                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
 304                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
 305                         }
 306                 } else
 307                        tp = tcp_close(tp);
 308        }
 309
 310 #ifdef TCPDEBUG
 311         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 312                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 313                           PRU_SLOWTIMO);
 314 #endif
 315         if (tp != NULL)
 316                 INP_WUNLOCK(inp);
 317         INP_INFO_WUNLOCK(&V_tcbinfo);
 318         CURVNET_RESTORE();
 319 }
 320
 321 void
 322 tcp_timer_keep(void *xtp)
 323 {
 324         struct tcpcb *tp = xtp;
 325         struct tcptemp *t_template;
 326         struct inpcb *inp;
 327         CURVNET_SET(tp->t_vnet);
 328 #ifdef TCPDEBUG
 329         int ostate;
 330
 331         ostate = tp->t_state;
 332 #endif
 333         INP_INFO_WLOCK(&V_tcbinfo);
 334         inp = tp->t_inpcb;
 335         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 336         INP_WLOCK(inp);
 337         if (callout_pending(&tp->t_timers->tt_keep) ||
 338             !callout_active(&tp->t_timers->tt_keep)) {
 339                 INP_WUNLOCK(inp);
 340                 INP_INFO_WUNLOCK(&V_tcbinfo);
 341                 CURVNET_RESTORE();
 342                 return;
 343         }
 344         callout_deactivate(&tp->t_timers->tt_keep);
 345         if ((inp->inp_flags & INP_DROPPED) != 0) {
 346                 INP_WUNLOCK(inp);
 347                 INP_INFO_WUNLOCK(&V_tcbinfo);
 348                 CURVNET_RESTORE();
 349                 return;
 350         }
 351         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 352                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 353         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
 354                 ("%s: tp %p keep callout should be running", __func__, tp));
 355         /*
 356          * Keep-alive timer went off; send something
 357          * or drop connection if idle for too long.
 358          */
 359         TCPSTAT_INC(tcps_keeptimeo);
 360         if (tp->t_state < TCPS_ESTABLISHED)
 361                 goto dropit;
 362         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
 363             tp->t_state <= TCPS_CLOSING) {
 364                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 365                         goto dropit;
 366                 /*
 367                  * Send a packet designed to force a response
 368                  * if the peer is up and reachable:
 369                  * either an ACK if the connection is still alive,
 370                  * or an RST if the peer has closed the connection
 371                  * due to timeout or reboot.
 372                  * Using sequence number tp->snd_una-1
 373                  * causes the transmitted zero-length segment
 374                  * to lie outside the receive window;
 375                  * by the protocol spec, this requires the
 376                  * correspondent TCP to respond.
 377                  */
 378                 TCPSTAT_INC(tcps_keepprobe);
 379                 t_template = tcpip_maketemplate(inp);
 380                 if (t_template) {
 381                         tcp_respond(tp, t_template->tt_ipgen,
 382                                     &t_template->tt_t, (struct mbuf *)NULL,
 383                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 384                         free(t_template, M_TEMP);
 385                 }
 386                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 387                     tcp_timer_keep, tp)) {
 388                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 389                 }
 390         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 391                     tcp_timer_keep, tp)) {
 392                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 393                 }
 394
 395 #ifdef TCPDEBUG
 396         if (inp->inp_socket->so_options & SO_DEBUG)
 397                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 398                           PRU_SLOWTIMO);
 399 #endif
 400         INP_WUNLOCK(inp);
 401         INP_INFO_WUNLOCK(&V_tcbinfo);
 402         CURVNET_RESTORE();
 403         return;
 404
 405 dropit:
 406         TCPSTAT_INC(tcps_keepdrops);
 407         tp = tcp_drop(tp, ETIMEDOUT);
 408
 409 #ifdef TCPDEBUG
 410         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 411                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 412                           PRU_SLOWTIMO);
 413 #endif
 414         if (tp != NULL)
 415                 INP_WUNLOCK(tp->t_inpcb);
 416         INP_INFO_WUNLOCK(&V_tcbinfo);
 417         CURVNET_RESTORE();
 418 }
 419
 420 void
 421 tcp_timer_persist(void *xtp)
 422 {
 423         struct tcpcb *tp = xtp;
 424         struct inpcb *inp;
 425         CURVNET_SET(tp->t_vnet);
 426 #ifdef TCPDEBUG
 427         int ostate;
 428
 429         ostate = tp->t_state;
 430 #endif
 431         INP_INFO_WLOCK(&V_tcbinfo);
 432         inp = tp->t_inpcb;
 433         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 434         INP_WLOCK(inp);
 435         if (callout_pending(&tp->t_timers->tt_persist) ||
 436             !callout_active(&tp->t_timers->tt_persist)) {
 437                 INP_WUNLOCK(inp);
 438                 INP_INFO_WUNLOCK(&V_tcbinfo);
 439                 CURVNET_RESTORE();
 440                 return;
 441         }
 442         callout_deactivate(&tp->t_timers->tt_persist);
 443         if ((inp->inp_flags & INP_DROPPED) != 0) {
 444                 INP_WUNLOCK(inp);
 445                 INP_INFO_WUNLOCK(&V_tcbinfo);
 446                 CURVNET_RESTORE();
 447                 return;
 448         }
 449         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 450                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 451         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
 452                 ("%s: tp %p persist callout should be running", __func__, tp));
 453         /*
 454          * Persistance timer into zero window.
 455          * Force a byte to be output, if possible.
 456          */
 457         TCPSTAT_INC(tcps_persisttimeo);
 458         /*
 459          * Hack: if the peer is dead/unreachable, we do not
 460          * time out if the window is closed.  After a full
 461          * backoff, drop the connection if the idle time
 462          * (no responses to probes) reaches the maximum
 463          * backoff that we would use if retransmitting.
 464          */
 465         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 466             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 467              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 468                 TCPSTAT_INC(tcps_persistdrop);
 469                 tp = tcp_drop(tp, ETIMEDOUT);
 470                 goto out;
 471         }
 472         /*
 473          * If the user has closed the socket then drop a persisting
 474          * connection after a much reduced timeout.
 475          */
 476         if (tp->t_state > TCPS_CLOSE_WAIT &&
 477             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 478                 TCPSTAT_INC(tcps_persistdrop);
 479                 tp = tcp_drop(tp, ETIMEDOUT);
 480                 goto out;
 481         }
 482         tcp_setpersist(tp);
 483         tp->t_flags |= TF_FORCEDATA;
 484         (void) tcp_output(tp);
 485         tp->t_flags &= ~TF_FORCEDATA;
 486
 487 out:
 488 #ifdef TCPDEBUG
 489         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 490                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 491 #endif
 492         if (tp != NULL)
 493                 INP_WUNLOCK(inp);
 494         INP_INFO_WUNLOCK(&V_tcbinfo);
 495         CURVNET_RESTORE();
 496 }
 497
 498 void
 499 tcp_timer_rexmt(void * xtp)
 500 {
 501         struct tcpcb *tp = xtp;
 502         CURVNET_SET(tp->t_vnet);
 503         int rexmt;
 504         int headlocked;
 505         struct inpcb *inp;
 506 #ifdef TCPDEBUG
 507         int ostate;
 508
 509         ostate = tp->t_state;
 510 #endif
 511
 512         INP_INFO_RLOCK(&V_tcbinfo);
 513         inp = tp->t_inpcb;
 514         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 515         INP_WLOCK(inp);
 516         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 517             !callout_active(&tp->t_timers->tt_rexmt)) {
 518                 INP_WUNLOCK(inp);
 519                 INP_INFO_RUNLOCK(&V_tcbinfo);
 520                 CURVNET_RESTORE();
 521                 return;
 522         }
 523         callout_deactivate(&tp->t_timers->tt_rexmt);
 524         if ((inp->inp_flags & INP_DROPPED) != 0) {
 525                 INP_WUNLOCK(inp);
 526                 INP_INFO_RUNLOCK(&V_tcbinfo);
 527                 CURVNET_RESTORE();
 528                 return;
 529         }
 530         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 531                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 532         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
 533                 ("%s: tp %p rexmt callout should be running", __func__, tp));
 534         tcp_free_sackholes(tp);
 535         /*
 536          * Retransmission timer went off.  Message has not
 537          * been acked within retransmit interval.  Back off
 538          * to a longer retransmit interval and retransmit one segment.
 539          */
 540         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 541                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 542                 TCPSTAT_INC(tcps_timeoutdrop);
 543                 in_pcbref(inp);
 544                 INP_INFO_RUNLOCK(&V_tcbinfo);
 545                 INP_WUNLOCK(inp);
 546                 INP_INFO_WLOCK(&V_tcbinfo);
 547                 INP_WLOCK(inp);
 548                 if (in_pcbrele_wlocked(inp)) {
 549                         INP_INFO_WUNLOCK(&V_tcbinfo);
 550                         CURVNET_RESTORE();
 551                         return;
 552                 }
 553                 if (inp->inp_flags & INP_DROPPED) {
 554                         INP_WUNLOCK(inp);
 555                         INP_INFO_WUNLOCK(&V_tcbinfo);
 556                         CURVNET_RESTORE();
 557                         return;
 558                 }
 559
 560                 tp = tcp_drop(tp, tp->t_softerror ?
 561                               tp->t_softerror : ETIMEDOUT);
 562                 headlocked = 1;
 563                 goto out;
 564         }
 565         INP_INFO_RUNLOCK(&V_tcbinfo);
 566         headlocked = 0;
 567         if (tp->t_state == TCPS_SYN_SENT) {
 568                 /*
 569                  * If the SYN was retransmitted, indicate CWND to be
 570                  * limited to 1 segment in cc_conn_init().
 571                  */
 572                 tp->snd_cwnd = 1;
 573         } else if (tp->t_rxtshift == 1) {
 574                 /*
 575                  * first retransmit; record ssthresh and cwnd so they can
 576                  * be recovered if this turns out to be a "bad" retransmit.
 577                  * A retransmit is considered "bad" if an ACK for this
 578                  * segment is received within RTT/2 interval; the assumption
 579                  * here is that the ACK was already in flight.  See
 580                  * "On Estimating End-to-End Network Path Properties" by
 581                  * Allman and Paxson for more details.
 582                  */
 583                 tp->snd_cwnd_prev = tp->snd_cwnd;
 584                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 585                 tp->snd_recover_prev = tp->snd_recover;
 586                 if (IN_FASTRECOVERY(tp->t_flags))
 587                         tp->t_flags |= TF_WASFRECOVERY;
 588                 else
 589                         tp->t_flags &= ~TF_WASFRECOVERY;
 590                 if (IN_CONGRECOVERY(tp->t_flags))
 591                         tp->t_flags |= TF_WASCRECOVERY;
 592                 else
 593                         tp->t_flags &= ~TF_WASCRECOVERY;
 594                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 595                 tp->t_flags |= TF_PREVVALID;
 596         } else
 597                 tp->t_flags &= ~TF_PREVVALID;
 598         TCPSTAT_INC(tcps_rexmttimeo);
 599         if (tp->t_state == TCPS_SYN_SENT)
 600                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 601         else
 602                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 603         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 604                       tp->t_rttmin, TCPTV_REXMTMAX);
 605
 606         /*
 607          * We enter the path for PLMTUD if connection is established or, if
 608          * connection is FIN_WAIT_1 status, reason for the last is that if
 609          * amount of data we send is very small, we could send it in couple of
 610          * packets and process straight to FIN. In that case we won't catch
 611          * ESTABLISHED state.
 612          */
 613         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 614             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 615                 int optlen;
 616 #ifdef INET6
 617                 int isipv6;
 618 #endif
 619
 620                 /*
 621                  * Idea here is that at each stage of mtu probe (usually, 1448
 622                  * -> 1188 -> 524) should be given 2 chances to recover before
 623                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 624                  *  take care of that.
 625                  */
 626                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 627                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 628                     (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
 629                         /*
 630                          * Enter Path MTU Black-hole Detection mechanism:
 631                          * - Disable Path MTU Discovery (IP "DF" bit).
 632                          * - Reduce MTU to lower value than what we
 633                          *   negotiated with peer.
 634                          */
 635                         /* Record that we may have found a black hole. */
 636                         tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 637
 638                         /* Keep track of previous MSS. */
 639                         optlen = tp->t_maxopd - tp->t_maxseg;
 640                         tp->t_pmtud_saved_maxopd = tp->t_maxopd;
 641
 642                         /*
 643                          * Reduce the MSS to blackhole value or to the default
 644                          * in an attempt to retransmit.
 645                          */
 646 #ifdef INET6
 647                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 648                         if (isipv6 &&
 649                             tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
 650                                 /* Use the sysctl tuneable blackhole MSS. */
 651                                 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
 652                                 V_tcp_pmtud_blackhole_activated++;
 653                         } else if (isipv6) {
 654                                 /* Use the default MSS. */
 655                                 tp->t_maxopd = V_tcp_v6mssdflt;
 656                                 /*
 657                                  * Disable Path MTU Discovery when we switch to
 658                                  * minmss.
 659                                  */
 660                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 661                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 662                         }
 663 #endif
 664 #if defined(INET6) && defined(INET)
 665                         else
 666 #endif
 667 #ifdef INET
 668                         if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
 669                                 /* Use the sysctl tuneable blackhole MSS. */
 670                                 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
 671                                 V_tcp_pmtud_blackhole_activated++;
 672                         } else {
 673                                 /* Use the default MSS. */
 674                                 tp->t_maxopd = V_tcp_mssdflt;
 675                                 /*
 676                                  * Disable Path MTU Discovery when we switch to
 677                                  * minmss.
 678                                  */
 679                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 680                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 681                         }
 682 #endif
 683                         tp->t_maxseg = tp->t_maxopd - optlen;
 684                         /*
 685                          * Reset the slow-start flight size
 686                          * as it may depend on the new MSS.
 687                          */
 688                         if (CC_ALGO(tp)->conn_init != NULL)
 689                                 CC_ALGO(tp)->conn_init(tp->ccv);
 690                 } else {
 691                         /*
 692                          * If further retransmissions are still unsuccessful
 693                          * with a lowered MTU, maybe this isn't a blackhole and
 694                          * we restore the previous MSS and blackhole detection
 695                          * flags.
 696                          * The limit '6' is determined by giving each probe
 697                          * stage (1448, 1188, 524) 2 chances to recover.
 698                          */
 699                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 700                             (tp->t_rxtshift > 6)) {
 701                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 702                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 703                                 optlen = tp->t_maxopd - tp->t_maxseg;
 704                                 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
 705                                 tp->t_maxseg = tp->t_maxopd - optlen;
 706                                 V_tcp_pmtud_blackhole_failed++;
 707                                 /*
 708                                  * Reset the slow-start flight size as it
 709                                  * may depend on the new MSS.
 710                                  */
 711                                 if (CC_ALGO(tp)->conn_init != NULL)
 712                                         CC_ALGO(tp)->conn_init(tp->ccv);
 713                         }
 714                 }
 715         }
 716
 717         /*
 718          * Disable RFC1323 and SACK if we haven't got any response to
 719          * our third SYN to work-around some broken terminal servers
 720          * (most of which have hopefully been retired) that have bad VJ
 721          * header compression code which trashes TCP segments containing
 722          * unknown-to-them TCP options.
 723          */
 724         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 725             (tp->t_rxtshift == 3))
 726                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 727         /*
 728          * If we backed off this far, our srtt estimate is probably bogus.
 729          * Clobber it so we'll take the next rtt measurement as our srtt;
 730          * move the current srtt into rttvar to keep the current
 731          * retransmit times until then.
 732          */
 733         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 734 #ifdef INET6
 735                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 736                         in6_losing(tp->t_inpcb);
 737 #endif
 738                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 739                 tp->t_srtt = 0;
 740         }
 741         tp->snd_nxt = tp->snd_una;
 742         tp->snd_recover = tp->snd_max;
 743         /*
 744          * Force a segment to be sent.
 745          */
 746         tp->t_flags |= TF_ACKNOW;
 747         /*
 748          * If timing a segment in this window, stop the timer.
 749          */
 750         tp->t_rtttime = 0;
 751
 752         cc_cong_signal(tp, NULL, CC_RTO);
 753
 754         (void) tcp_output(tp);
 755
 756 out:
 757 #ifdef TCPDEBUG
 758         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 759                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 760                           PRU_SLOWTIMO);
 761 #endif
 762         if (tp != NULL)
 763                 INP_WUNLOCK(inp);
 764         if (headlocked)
 765                 INP_INFO_WUNLOCK(&V_tcbinfo);
 766         CURVNET_RESTORE();
 767 }
 768
 769 void
 770 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 771 {
 772         struct callout *t_callout;
 773         timeout_t *f_callout;
 774         struct inpcb *inp = tp->t_inpcb;
 775         int cpu = INP_CPU(inp);
 776         uint32_t f_reset;
 777
 778 #ifdef TCP_OFFLOAD
 779         if (tp->t_flags & TF_TOE)
 780                 return;
 781 #endif
 782
 783         if (tp->t_timers->tt_flags & TT_STOPPED)
 784                 return;
 785
 786         switch (timer_type) {
 787                 case TT_DELACK:
 788                         t_callout = &tp->t_timers->tt_delack;
 789                         f_callout = tcp_timer_delack;
 790                         f_reset = TT_DELACK_RST;
 791                         break;
 792                 case TT_REXMT:
 793                         t_callout = &tp->t_timers->tt_rexmt;
 794                         f_callout = tcp_timer_rexmt;
 795                         f_reset = TT_REXMT_RST;
 796                         break;
 797                 case TT_PERSIST:
 798                         t_callout = &tp->t_timers->tt_persist;
 799                         f_callout = tcp_timer_persist;
 800                         f_reset = TT_PERSIST_RST;
 801                         break;
 802                 case TT_KEEP:
 803                         t_callout = &tp->t_timers->tt_keep;
 804                         f_callout = tcp_timer_keep;
 805                         f_reset = TT_KEEP_RST;
 806                         break;
 807                 case TT_2MSL:
 808                         t_callout = &tp->t_timers->tt_2msl;
 809                         f_callout = tcp_timer_2msl;
 810                         f_reset = TT_2MSL_RST;
 811                         break;
 812                 default:
 813                         panic("tp %p bad timer_type %#x", tp, timer_type);
 814                 }
 815         if (delta == 0) {
 816                 if ((tp->t_timers->tt_flags & timer_type) &&
 817                     callout_stop(t_callout) &&
 818                     (tp->t_timers->tt_flags & f_reset)) {
 819                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 820                 }
 821         } else {
 822                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
 823                         tp->t_timers->tt_flags |= (timer_type | f_reset);
 824                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 825                 } else {
 826                         /* Reset already running callout on the same CPU. */
 827                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
 828                                 /*
 829                                  * Callout not cancelled, consider it as not
 830                                  * properly restarted. */
 831                                 tp->t_timers->tt_flags &= ~f_reset;
 832                         }
 833                 }
 834         }
 835 }
 836
 837 int
 838 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 839 {
 840         struct callout *t_callout;
 841
 842         switch (timer_type) {
 843                 case TT_DELACK:
 844                         t_callout = &tp->t_timers->tt_delack;
 845                         break;
 846                 case TT_REXMT:
 847                         t_callout = &tp->t_timers->tt_rexmt;
 848                         break;
 849                 case TT_PERSIST:
 850                         t_callout = &tp->t_timers->tt_persist;
 851                         break;
 852                 case TT_KEEP:
 853                         t_callout = &tp->t_timers->tt_keep;
 854                         break;
 855                 case TT_2MSL:
 856                         t_callout = &tp->t_timers->tt_2msl;
 857                         break;
 858                 default:
 859                         panic("tp %p bad timer_type %#x", tp, timer_type);
 860                 }
 861         return callout_active(t_callout);
 862 }
 863
 864 void
 865 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 866 {
 867         struct callout *t_callout;
 868         timeout_t *f_callout;
 869         uint32_t f_reset;
 870
 871         tp->t_timers->tt_flags |= TT_STOPPED;
 872
 873         switch (timer_type) {
 874                 case TT_DELACK:
 875                         t_callout = &tp->t_timers->tt_delack;
 876                         f_callout = tcp_timer_delack_discard;
 877                         f_reset = TT_DELACK_RST;
 878                         break;
 879                 case TT_REXMT:
 880                         t_callout = &tp->t_timers->tt_rexmt;
 881                         f_callout = tcp_timer_rexmt_discard;
 882                         f_reset = TT_REXMT_RST;
 883                         break;
 884                 case TT_PERSIST:
 885                         t_callout = &tp->t_timers->tt_persist;
 886                         f_callout = tcp_timer_persist_discard;
 887                         f_reset = TT_PERSIST_RST;
 888                         break;
 889                 case TT_KEEP:
 890                         t_callout = &tp->t_timers->tt_keep;
 891                         f_callout = tcp_timer_keep_discard;
 892                         f_reset = TT_KEEP_RST;
 893                         break;
 894                 case TT_2MSL:
 895                         t_callout = &tp->t_timers->tt_2msl;
 896                         f_callout = tcp_timer_2msl_discard;
 897                         f_reset = TT_2MSL_RST;
 898                         break;
 899                 default:
 900                         panic("tp %p bad timer_type %#x", tp, timer_type);
 901                 }
 902
 903         if (tp->t_timers->tt_flags & timer_type) {
 904                 if (callout_stop(t_callout) &&
 905                     (tp->t_timers->tt_flags & f_reset)) {
 906                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 907                 } else {
 908                         /*
 909                          * Can't stop the callout, defer tcpcb actual deletion
 910                          * to the last tcp timer discard callout.
 911                          * The TT_STOPPED flag will ensure that no tcp timer
 912                          * callouts can be restarted on our behalf, and
 913                          * past this point currently running callouts waiting
 914                          * on inp lock will return right away after the
 915                          * classical check for callout reset/stop events:
 916                          * callout_pending() || !callout_active()
 917                          */
 918                         callout_reset(t_callout, 1, f_callout, tp);
 919                 }
 920         }
 921 }
 922
 923 #define ticks_to_msecs(t)       (1000*(t) / hz)
 924
 925 void
 926 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
 927     struct xtcp_timer *xtimer)
 928 {
 929         sbintime_t now;
 930
 931         bzero(xtimer, sizeof(*xtimer));
 932         if (timer == NULL)
 933                 return;
 934         now = getsbinuptime();
 935         if (callout_active(&timer->tt_delack))
 936                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
 937         if (callout_active(&timer->tt_rexmt))
 938                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
 939         if (callout_active(&timer->tt_persist))
 940                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
 941         if (callout_active(&timer->tt_keep))
 942                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
 943         if (callout_active(&timer->tt_2msl))
 944                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
 945         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
 946 }