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_persmin;
  75 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
  76     &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
  77
  78 int    tcp_persmax;
  79 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
  80     &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
  81
  82 int     tcp_keepinit;
  83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
  84     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
  85
  86 int     tcp_keepidle;
  87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
  88     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
  89
  90 int     tcp_keepintvl;
  91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
  92     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
  93
  94 int     tcp_delacktime;
  95 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
  96     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
  97     "Time before a delayed ACK is sent");
  98
  99 int     tcp_msl;
 100 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 101     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 102
 103 int     tcp_rexmit_min;
 104 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 105     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
 106     "Minimum Retransmission Timeout");
 107
 108 int     tcp_rexmit_slop;
 109 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 110     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 111     "Retransmission Timer Slop");
 112
 113 static int      always_keepalive = 1;
 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 115     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 116
 117 int    tcp_fast_finwait2_recycle = 0;
 118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 119     &tcp_fast_finwait2_recycle, 0,
 120     "Recycle closed FIN_WAIT_2 connections faster");
 121
 122 int    tcp_finwait2_timeout;
 123 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 124     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 125
 126 int     tcp_keepcnt = TCPTV_KEEPCNT;
 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
 128     "Number of keepalive probes to send");
 129
 130         /* max idle probes */
 131 int     tcp_maxpersistidle;
 132
 133 static int      tcp_rexmit_drop_options = 0;
 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
 135     &tcp_rexmit_drop_options, 0,
 136     "Drop TCP options from 3rd and later retransmitted SYN");
 137
 138 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
 139 #define V_tcp_pmtud_blackhole_detect    VNET(tcp_pmtud_blackhole_detect)
 140 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
 141     CTLFLAG_RW,
 142     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
 143     "Path MTU Discovery Black Hole Detection Enabled");
 144
 145 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
 146 #define V_tcp_pmtud_blackhole_activated \
 147     VNET(tcp_pmtud_blackhole_activated)
 148 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
 149     CTLFLAG_RD,
 150     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
 151     "Path MTU Discovery Black Hole Detection, Activation Count");
 152
 153 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
 154 #define V_tcp_pmtud_blackhole_activated_min_mss \
 155     VNET(tcp_pmtud_blackhole_activated_min_mss)
 156 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
 157     CTLFLAG_RD,
 158     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
 159     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
 160
 161 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
 162 #define V_tcp_pmtud_blackhole_failed    VNET(tcp_pmtud_blackhole_failed)
 163 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
 164     CTLFLAG_RD,
 165     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
 166     "Path MTU Discovery Black Hole Detection, Failure Count");
 167
 168 #ifdef INET
 169 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
 170 #define V_tcp_pmtud_blackhole_mss       VNET(tcp_pmtud_blackhole_mss)
 171 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
 172     CTLFLAG_RW,
 173     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
 174     "Path MTU Discovery Black Hole Detection lowered MSS");
 175 #endif
 176
 177 #ifdef INET6
 178 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
 179 #define V_tcp_v6pmtud_blackhole_mss     VNET(tcp_v6pmtud_blackhole_mss)
 180 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
 181     CTLFLAG_RW,
 182     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
 183     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
 184 #endif
 185
 186 static int      per_cpu_timers = 0;
 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
 188     &per_cpu_timers , 0, "run tcp timers on all cpus");
 189
 190 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
 191                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
 192
 193 /*
 194  * Tcp protocol timeout routine called every 500 ms.
 195  * Updates timestamps used for TCP
 196  * causes finite state machine actions if timers expire.
 197  */
 198 void
 199 tcp_slowtimo(void)
 200 {
 201         VNET_ITERATOR_DECL(vnet_iter);
 202
 203         VNET_LIST_RLOCK_NOSLEEP();
 204         VNET_FOREACH(vnet_iter) {
 205                 CURVNET_SET(vnet_iter);
 206                 (void) tcp_tw_2msl_scan(0);
 207                 CURVNET_RESTORE();
 208         }
 209         VNET_LIST_RUNLOCK_NOSLEEP();
 210 }
 211
 212 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 213     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 214
 215 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 216     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 217
 218 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 219
 220 /*
 221  * TCP timer processing.
 222  */
 223
 224 void
 225 tcp_timer_delack(void *xtp)
 226 {
 227         struct tcpcb *tp = xtp;
 228         struct inpcb *inp;
 229         CURVNET_SET(tp->t_vnet);
 230
 231         inp = tp->t_inpcb;
 232         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 233         INP_WLOCK(inp);
 234         if (callout_pending(&tp->t_timers->tt_delack) ||
 235             !callout_active(&tp->t_timers->tt_delack)) {
 236                 INP_WUNLOCK(inp);
 237                 CURVNET_RESTORE();
 238                 return;
 239         }
 240         callout_deactivate(&tp->t_timers->tt_delack);
 241         if ((inp->inp_flags & INP_DROPPED) != 0) {
 242                 INP_WUNLOCK(inp);
 243                 CURVNET_RESTORE();
 244                 return;
 245         }
 246         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 247                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 248         KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
 249                 ("%s: tp %p delack callout should be running", __func__, tp));
 250
 251         tp->t_flags |= TF_ACKNOW;
 252         TCPSTAT_INC(tcps_delack);
 253         (void) tcp_output(tp);
 254         INP_WUNLOCK(inp);
 255         CURVNET_RESTORE();
 256 }
 257
 258 void
 259 tcp_timer_2msl(void *xtp)
 260 {
 261         struct tcpcb *tp = xtp;
 262         struct inpcb *inp;
 263         CURVNET_SET(tp->t_vnet);
 264 #ifdef TCPDEBUG
 265         int ostate;
 266
 267         ostate = tp->t_state;
 268 #endif
 269         INP_INFO_RLOCK(&V_tcbinfo);
 270         inp = tp->t_inpcb;
 271         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 272         INP_WLOCK(inp);
 273         tcp_free_sackholes(tp);
 274         if (callout_pending(&tp->t_timers->tt_2msl) ||
 275             !callout_active(&tp->t_timers->tt_2msl)) {
 276                 INP_WUNLOCK(tp->t_inpcb);
 277                 INP_INFO_RUNLOCK(&V_tcbinfo);
 278                 CURVNET_RESTORE();
 279                 return;
 280         }
 281         callout_deactivate(&tp->t_timers->tt_2msl);
 282         if ((inp->inp_flags & INP_DROPPED) != 0) {
 283                 INP_WUNLOCK(inp);
 284                 INP_INFO_RUNLOCK(&V_tcbinfo);
 285                 CURVNET_RESTORE();
 286                 return;
 287         }
 288         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 289                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 290         KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
 291                 ("%s: tp %p 2msl callout should be running", __func__, tp));
 292         /*
 293          * 2 MSL timeout in shutdown went off.  If we're closed but
 294          * still waiting for peer to close and connection has been idle
 295          * too long delete connection control block.  Otherwise, check
 296          * again in a bit.
 297          *
 298          * If in TIME_WAIT state just ignore as this timeout is handled in
 299          * tcp_tw_2msl_scan().
 300          *
 301          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 302          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 303          * Ignore fact that there were recent incoming segments.
 304          */
 305         if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
 306                 INP_WUNLOCK(inp);
 307                 INP_INFO_RUNLOCK(&V_tcbinfo);
 308                 CURVNET_RESTORE();
 309                 return;
 310         }
 311         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 312             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 313             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 314                 TCPSTAT_INC(tcps_finwait2_drops);
 315                 tp = tcp_close(tp);
 316         } else {
 317                 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 318                         if (!callout_reset(&tp->t_timers->tt_2msl,
 319                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
 320                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
 321                         }
 322                 } else
 323                        tp = tcp_close(tp);
 324        }
 325
 326 #ifdef TCPDEBUG
 327         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 328                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 329                           PRU_SLOWTIMO);
 330 #endif
 331         if (tp != NULL)
 332                 INP_WUNLOCK(inp);
 333         INP_INFO_RUNLOCK(&V_tcbinfo);
 334         CURVNET_RESTORE();
 335 }
 336
 337 void
 338 tcp_timer_keep(void *xtp)
 339 {
 340         struct tcpcb *tp = xtp;
 341         struct tcptemp *t_template;
 342         struct inpcb *inp;
 343         CURVNET_SET(tp->t_vnet);
 344 #ifdef TCPDEBUG
 345         int ostate;
 346
 347         ostate = tp->t_state;
 348 #endif
 349         INP_INFO_RLOCK(&V_tcbinfo);
 350         inp = tp->t_inpcb;
 351         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 352         INP_WLOCK(inp);
 353         if (callout_pending(&tp->t_timers->tt_keep) ||
 354             !callout_active(&tp->t_timers->tt_keep)) {
 355                 INP_WUNLOCK(inp);
 356                 INP_INFO_RUNLOCK(&V_tcbinfo);
 357                 CURVNET_RESTORE();
 358                 return;
 359         }
 360         callout_deactivate(&tp->t_timers->tt_keep);
 361         if ((inp->inp_flags & INP_DROPPED) != 0) {
 362                 INP_WUNLOCK(inp);
 363                 INP_INFO_RUNLOCK(&V_tcbinfo);
 364                 CURVNET_RESTORE();
 365                 return;
 366         }
 367         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 368                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 369         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
 370                 ("%s: tp %p keep callout should be running", __func__, tp));
 371         /*
 372          * Keep-alive timer went off; send something
 373          * or drop connection if idle for too long.
 374          */
 375         TCPSTAT_INC(tcps_keeptimeo);
 376         if (tp->t_state < TCPS_ESTABLISHED)
 377                 goto dropit;
 378         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
 379             tp->t_state <= TCPS_CLOSING) {
 380                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 381                         goto dropit;
 382                 /*
 383                  * Send a packet designed to force a response
 384                  * if the peer is up and reachable:
 385                  * either an ACK if the connection is still alive,
 386                  * or an RST if the peer has closed the connection
 387                  * due to timeout or reboot.
 388                  * Using sequence number tp->snd_una-1
 389                  * causes the transmitted zero-length segment
 390                  * to lie outside the receive window;
 391                  * by the protocol spec, this requires the
 392                  * correspondent TCP to respond.
 393                  */
 394                 TCPSTAT_INC(tcps_keepprobe);
 395                 t_template = tcpip_maketemplate(inp);
 396                 if (t_template) {
 397                         tcp_respond(tp, t_template->tt_ipgen,
 398                                     &t_template->tt_t, (struct mbuf *)NULL,
 399                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 400                         free(t_template, M_TEMP);
 401                 }
 402                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 403                     tcp_timer_keep, tp)) {
 404                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 405                 }
 406         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 407                     tcp_timer_keep, tp)) {
 408                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 409                 }
 410
 411 #ifdef TCPDEBUG
 412         if (inp->inp_socket->so_options & SO_DEBUG)
 413                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 414                           PRU_SLOWTIMO);
 415 #endif
 416         INP_WUNLOCK(inp);
 417         INP_INFO_RUNLOCK(&V_tcbinfo);
 418         CURVNET_RESTORE();
 419         return;
 420
 421 dropit:
 422         TCPSTAT_INC(tcps_keepdrops);
 423         tp = tcp_drop(tp, ETIMEDOUT);
 424
 425 #ifdef TCPDEBUG
 426         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 427                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 428                           PRU_SLOWTIMO);
 429 #endif
 430         if (tp != NULL)
 431                 INP_WUNLOCK(tp->t_inpcb);
 432         INP_INFO_RUNLOCK(&V_tcbinfo);
 433         CURVNET_RESTORE();
 434 }
 435
 436 void
 437 tcp_timer_persist(void *xtp)
 438 {
 439         struct tcpcb *tp = xtp;
 440         struct inpcb *inp;
 441         CURVNET_SET(tp->t_vnet);
 442 #ifdef TCPDEBUG
 443         int ostate;
 444
 445         ostate = tp->t_state;
 446 #endif
 447         INP_INFO_RLOCK(&V_tcbinfo);
 448         inp = tp->t_inpcb;
 449         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 450         INP_WLOCK(inp);
 451         if (callout_pending(&tp->t_timers->tt_persist) ||
 452             !callout_active(&tp->t_timers->tt_persist)) {
 453                 INP_WUNLOCK(inp);
 454                 INP_INFO_RUNLOCK(&V_tcbinfo);
 455                 CURVNET_RESTORE();
 456                 return;
 457         }
 458         callout_deactivate(&tp->t_timers->tt_persist);
 459         if ((inp->inp_flags & INP_DROPPED) != 0) {
 460                 INP_WUNLOCK(inp);
 461                 INP_INFO_RUNLOCK(&V_tcbinfo);
 462                 CURVNET_RESTORE();
 463                 return;
 464         }
 465         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 466                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 467         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
 468                 ("%s: tp %p persist callout should be running", __func__, tp));
 469         /*
 470          * Persistance timer into zero window.
 471          * Force a byte to be output, if possible.
 472          */
 473         TCPSTAT_INC(tcps_persisttimeo);
 474         /*
 475          * Hack: if the peer is dead/unreachable, we do not
 476          * time out if the window is closed.  After a full
 477          * backoff, drop the connection if the idle time
 478          * (no responses to probes) reaches the maximum
 479          * backoff that we would use if retransmitting.
 480          */
 481         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 482             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 483              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 484                 TCPSTAT_INC(tcps_persistdrop);
 485                 tp = tcp_drop(tp, ETIMEDOUT);
 486                 goto out;
 487         }
 488         /*
 489          * If the user has closed the socket then drop a persisting
 490          * connection after a much reduced timeout.
 491          */
 492         if (tp->t_state > TCPS_CLOSE_WAIT &&
 493             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 494                 TCPSTAT_INC(tcps_persistdrop);
 495                 tp = tcp_drop(tp, ETIMEDOUT);
 496                 goto out;
 497         }
 498         tcp_setpersist(tp);
 499         tp->t_flags |= TF_FORCEDATA;
 500         (void) tcp_output(tp);
 501         tp->t_flags &= ~TF_FORCEDATA;
 502
 503 out:
 504 #ifdef TCPDEBUG
 505         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 506                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 507 #endif
 508         if (tp != NULL)
 509                 INP_WUNLOCK(inp);
 510         INP_INFO_RUNLOCK(&V_tcbinfo);
 511         CURVNET_RESTORE();
 512 }
 513
 514 void
 515 tcp_timer_rexmt(void * xtp)
 516 {
 517         struct tcpcb *tp = xtp;
 518         CURVNET_SET(tp->t_vnet);
 519         int rexmt;
 520         int headlocked;
 521         struct inpcb *inp;
 522 #ifdef TCPDEBUG
 523         int ostate;
 524
 525         ostate = tp->t_state;
 526 #endif
 527
 528         INP_INFO_RLOCK(&V_tcbinfo);
 529         inp = tp->t_inpcb;
 530         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 531         INP_WLOCK(inp);
 532         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 533             !callout_active(&tp->t_timers->tt_rexmt)) {
 534                 INP_WUNLOCK(inp);
 535                 INP_INFO_RUNLOCK(&V_tcbinfo);
 536                 CURVNET_RESTORE();
 537                 return;
 538         }
 539         callout_deactivate(&tp->t_timers->tt_rexmt);
 540         if ((inp->inp_flags & INP_DROPPED) != 0) {
 541                 INP_WUNLOCK(inp);
 542                 INP_INFO_RUNLOCK(&V_tcbinfo);
 543                 CURVNET_RESTORE();
 544                 return;
 545         }
 546         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 547                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 548         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
 549                 ("%s: tp %p rexmt callout should be running", __func__, tp));
 550         tcp_free_sackholes(tp);
 551         /*
 552          * Retransmission timer went off.  Message has not
 553          * been acked within retransmit interval.  Back off
 554          * to a longer retransmit interval and retransmit one segment.
 555          */
 556         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 557                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 558                 TCPSTAT_INC(tcps_timeoutdrop);
 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             (tp->t_state == TCPS_SYN_RECEIVED))
 601                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 602         else
 603                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 604         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 605                       tp->t_rttmin, TCPTV_REXMTMAX);
 606
 607         /*
 608          * We enter the path for PLMTUD if connection is established or, if
 609          * connection is FIN_WAIT_1 status, reason for the last is that if
 610          * amount of data we send is very small, we could send it in couple of
 611          * packets and process straight to FIN. In that case we won't catch
 612          * ESTABLISHED state.
 613          */
 614         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 615             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 616                 int optlen;
 617 #ifdef INET6
 618                 int isipv6;
 619 #endif
 620
 621                 /*
 622                  * Idea here is that at each stage of mtu probe (usually, 1448
 623                  * -> 1188 -> 524) should be given 2 chances to recover before
 624                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 625                  *  take care of that.
 626                  */
 627                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 628                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 629                     (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
 630                         /*
 631                          * Enter Path MTU Black-hole Detection mechanism:
 632                          * - Disable Path MTU Discovery (IP "DF" bit).
 633                          * - Reduce MTU to lower value than what we
 634                          *   negotiated with peer.
 635                          */
 636                         /* Record that we may have found a black hole. */
 637                         tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 638
 639                         /* Keep track of previous MSS. */
 640                         optlen = tp->t_maxopd - tp->t_maxseg;
 641                         tp->t_pmtud_saved_maxopd = tp->t_maxopd;
 642
 643                         /*
 644                          * Reduce the MSS to blackhole value or to the default
 645                          * in an attempt to retransmit.
 646                          */
 647 #ifdef INET6
 648                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 649                         if (isipv6 &&
 650                             tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
 651                                 /* Use the sysctl tuneable blackhole MSS. */
 652                                 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
 653                                 V_tcp_pmtud_blackhole_activated++;
 654                         } else if (isipv6) {
 655                                 /* Use the default MSS. */
 656                                 tp->t_maxopd = V_tcp_v6mssdflt;
 657                                 /*
 658                                  * Disable Path MTU Discovery when we switch to
 659                                  * minmss.
 660                                  */
 661                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 662                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 663                         }
 664 #endif
 665 #if defined(INET6) && defined(INET)
 666                         else
 667 #endif
 668 #ifdef INET
 669                         if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
 670                                 /* Use the sysctl tuneable blackhole MSS. */
 671                                 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
 672                                 V_tcp_pmtud_blackhole_activated++;
 673                         } else {
 674                                 /* Use the default MSS. */
 675                                 tp->t_maxopd = V_tcp_mssdflt;
 676                                 /*
 677                                  * Disable Path MTU Discovery when we switch to
 678                                  * minmss.
 679                                  */
 680                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 681                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 682                         }
 683 #endif
 684                         tp->t_maxseg = tp->t_maxopd - optlen;
 685                         /*
 686                          * Reset the slow-start flight size
 687                          * as it may depend on the new MSS.
 688                          */
 689                         if (CC_ALGO(tp)->conn_init != NULL)
 690                                 CC_ALGO(tp)->conn_init(tp->ccv);
 691                 } else {
 692                         /*
 693                          * If further retransmissions are still unsuccessful
 694                          * with a lowered MTU, maybe this isn't a blackhole and
 695                          * we restore the previous MSS and blackhole detection
 696                          * flags.
 697                          * The limit '6' is determined by giving each probe
 698                          * stage (1448, 1188, 524) 2 chances to recover.
 699                          */
 700                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 701                             (tp->t_rxtshift > 6)) {
 702                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 703                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 704                                 optlen = tp->t_maxopd - tp->t_maxseg;
 705                                 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
 706                                 tp->t_maxseg = tp->t_maxopd - optlen;
 707                                 V_tcp_pmtud_blackhole_failed++;
 708                                 /*
 709                                  * Reset the slow-start flight size as it
 710                                  * may depend on the new MSS.
 711                                  */
 712                                 if (CC_ALGO(tp)->conn_init != NULL)
 713                                         CC_ALGO(tp)->conn_init(tp->ccv);
 714                         }
 715                 }
 716         }
 717
 718         /*
 719          * Disable RFC1323 and SACK if we haven't got any response to
 720          * our third SYN to work-around some broken terminal servers
 721          * (most of which have hopefully been retired) that have bad VJ
 722          * header compression code which trashes TCP segments containing
 723          * unknown-to-them TCP options.
 724          */
 725         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 726             (tp->t_rxtshift == 3))
 727                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 728         /*
 729          * If we backed off this far, our srtt estimate is probably bogus.
 730          * Clobber it so we'll take the next rtt measurement as our srtt;
 731          * move the current srtt into rttvar to keep the current
 732          * retransmit times until then.
 733          */
 734         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 735 #ifdef INET6
 736                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 737                         in6_losing(tp->t_inpcb);
 738 #endif
 739                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 740                 tp->t_srtt = 0;
 741         }
 742         tp->snd_nxt = tp->snd_una;
 743         tp->snd_recover = tp->snd_max;
 744         /*
 745          * Force a segment to be sent.
 746          */
 747         tp->t_flags |= TF_ACKNOW;
 748         /*
 749          * If timing a segment in this window, stop the timer.
 750          */
 751         tp->t_rtttime = 0;
 752
 753         cc_cong_signal(tp, NULL, CC_RTO);
 754
 755         (void) tcp_output(tp);
 756
 757 out:
 758 #ifdef TCPDEBUG
 759         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 760                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 761                           PRU_SLOWTIMO);
 762 #endif
 763         if (tp != NULL)
 764                 INP_WUNLOCK(inp);
 765         if (headlocked)
 766                 INP_INFO_RUNLOCK(&V_tcbinfo);
 767         CURVNET_RESTORE();
 768 }
 769
 770 void
 771 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 772 {
 773         struct callout *t_callout;
 774         timeout_t *f_callout;
 775         struct inpcb *inp = tp->t_inpcb;
 776         int cpu = INP_CPU(inp);
 777         uint32_t f_reset;
 778
 779 #ifdef TCP_OFFLOAD
 780         if (tp->t_flags & TF_TOE)
 781                 return;
 782 #endif
 783
 784         if (tp->t_timers->tt_flags & TT_STOPPED)
 785                 return;
 786
 787         switch (timer_type) {
 788                 case TT_DELACK:
 789                         t_callout = &tp->t_timers->tt_delack;
 790                         f_callout = tcp_timer_delack;
 791                         f_reset = TT_DELACK_RST;
 792                         break;
 793                 case TT_REXMT:
 794                         t_callout = &tp->t_timers->tt_rexmt;
 795                         f_callout = tcp_timer_rexmt;
 796                         f_reset = TT_REXMT_RST;
 797                         break;
 798                 case TT_PERSIST:
 799                         t_callout = &tp->t_timers->tt_persist;
 800                         f_callout = tcp_timer_persist;
 801                         f_reset = TT_PERSIST_RST;
 802                         break;
 803                 case TT_KEEP:
 804                         t_callout = &tp->t_timers->tt_keep;
 805                         f_callout = tcp_timer_keep;
 806                         f_reset = TT_KEEP_RST;
 807                         break;
 808                 case TT_2MSL:
 809                         t_callout = &tp->t_timers->tt_2msl;
 810                         f_callout = tcp_timer_2msl;
 811                         f_reset = TT_2MSL_RST;
 812                         break;
 813                 default:
 814                         panic("tp %p bad timer_type %#x", tp, timer_type);
 815                 }
 816         if (delta == 0) {
 817                 if ((tp->t_timers->tt_flags & timer_type) &&
 818                     callout_stop(t_callout) &&
 819                     (tp->t_timers->tt_flags & f_reset)) {
 820                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 821                 }
 822         } else {
 823                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
 824                         tp->t_timers->tt_flags |= (timer_type | f_reset);
 825                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 826                 } else {
 827                         /* Reset already running callout on the same CPU. */
 828                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
 829                                 /*
 830                                  * Callout not cancelled, consider it as not
 831                                  * properly restarted. */
 832                                 tp->t_timers->tt_flags &= ~f_reset;
 833                         }
 834                 }
 835         }
 836 }
 837
 838 int
 839 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 840 {
 841         struct callout *t_callout;
 842
 843         switch (timer_type) {
 844                 case TT_DELACK:
 845                         t_callout = &tp->t_timers->tt_delack;
 846                         break;
 847                 case TT_REXMT:
 848                         t_callout = &tp->t_timers->tt_rexmt;
 849                         break;
 850                 case TT_PERSIST:
 851                         t_callout = &tp->t_timers->tt_persist;
 852                         break;
 853                 case TT_KEEP:
 854                         t_callout = &tp->t_timers->tt_keep;
 855                         break;
 856                 case TT_2MSL:
 857                         t_callout = &tp->t_timers->tt_2msl;
 858                         break;
 859                 default:
 860                         panic("tp %p bad timer_type %#x", tp, timer_type);
 861                 }
 862         return callout_active(t_callout);
 863 }
 864
 865 void
 866 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 867 {
 868         struct callout *t_callout;
 869         timeout_t *f_callout;
 870         uint32_t f_reset;
 871
 872         tp->t_timers->tt_flags |= TT_STOPPED;
 873
 874         switch (timer_type) {
 875                 case TT_DELACK:
 876                         t_callout = &tp->t_timers->tt_delack;
 877                         f_callout = tcp_timer_delack_discard;
 878                         f_reset = TT_DELACK_RST;
 879                         break;
 880                 case TT_REXMT:
 881                         t_callout = &tp->t_timers->tt_rexmt;
 882                         f_callout = tcp_timer_rexmt_discard;
 883                         f_reset = TT_REXMT_RST;
 884                         break;
 885                 case TT_PERSIST:
 886                         t_callout = &tp->t_timers->tt_persist;
 887                         f_callout = tcp_timer_persist_discard;
 888                         f_reset = TT_PERSIST_RST;
 889                         break;
 890                 case TT_KEEP:
 891                         t_callout = &tp->t_timers->tt_keep;
 892                         f_callout = tcp_timer_keep_discard;
 893                         f_reset = TT_KEEP_RST;
 894                         break;
 895                 case TT_2MSL:
 896                         t_callout = &tp->t_timers->tt_2msl;
 897                         f_callout = tcp_timer_2msl_discard;
 898                         f_reset = TT_2MSL_RST;
 899                         break;
 900                 default:
 901                         panic("tp %p bad timer_type %#x", tp, timer_type);
 902                 }
 903
 904         if (tp->t_timers->tt_flags & timer_type) {
 905                 if (callout_stop(t_callout) &&
 906                     (tp->t_timers->tt_flags & f_reset)) {
 907                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 908                 } else {
 909                         /*
 910                          * Can't stop the callout, defer tcpcb actual deletion
 911                          * to the last tcp timer discard callout.
 912                          * The TT_STOPPED flag will ensure that no tcp timer
 913                          * callouts can be restarted on our behalf, and
 914                          * past this point currently running callouts waiting
 915                          * on inp lock will return right away after the
 916                          * classical check for callout reset/stop events:
 917                          * callout_pending() || !callout_active()
 918                          */
 919                         callout_reset(t_callout, 1, f_callout, tp);
 920                 }
 921         }
 922 }
 923
 924 #define ticks_to_msecs(t)       (1000*(t) / hz)
 925
 926 void
 927 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
 928     struct xtcp_timer *xtimer)
 929 {
 930         sbintime_t now;
 931
 932         bzero(xtimer, sizeof(*xtimer));
 933         if (timer == NULL)
 934                 return;
 935         now = getsbinuptime();
 936         if (callout_active(&timer->tt_delack))
 937                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
 938         if (callout_active(&timer->tt_rexmt))
 939                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
 940         if (callout_active(&timer->tt_persist))
 941                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
 942         if (callout_active(&timer->tt_keep))
 943                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
 944         if (callout_active(&timer->tt_2msl))
 945                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
 946         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
 947 }