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_WLOCK(&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_WUNLOCK(&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_WUNLOCK(&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, or if 2MSL time is up from TIME_WAIT, delete connection
 296          * control block.  Otherwise, check again in a bit.
 297          *
 298          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 299          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 300          * Ignore fact that there were recent incoming segments.
 301          */
 302         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 303             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 304             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 305                 TCPSTAT_INC(tcps_finwait2_drops);
 306                 tp = tcp_close(tp);
 307         } else {
 308                 if (tp->t_state != TCPS_TIME_WAIT &&
 309                    ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 310                         if (!callout_reset(&tp->t_timers->tt_2msl,
 311                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
 312                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
 313                         }
 314                 } else
 315                        tp = tcp_close(tp);
 316        }
 317
 318 #ifdef TCPDEBUG
 319         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 320                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 321                           PRU_SLOWTIMO);
 322 #endif
 323         if (tp != NULL)
 324                 INP_WUNLOCK(inp);
 325         INP_INFO_WUNLOCK(&V_tcbinfo);
 326         CURVNET_RESTORE();
 327 }
 328
 329 void
 330 tcp_timer_keep(void *xtp)
 331 {
 332         struct tcpcb *tp = xtp;
 333         struct tcptemp *t_template;
 334         struct inpcb *inp;
 335         CURVNET_SET(tp->t_vnet);
 336 #ifdef TCPDEBUG
 337         int ostate;
 338
 339         ostate = tp->t_state;
 340 #endif
 341         INP_INFO_WLOCK(&V_tcbinfo);
 342         inp = tp->t_inpcb;
 343         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 344         INP_WLOCK(inp);
 345         if (callout_pending(&tp->t_timers->tt_keep) ||
 346             !callout_active(&tp->t_timers->tt_keep)) {
 347                 INP_WUNLOCK(inp);
 348                 INP_INFO_WUNLOCK(&V_tcbinfo);
 349                 CURVNET_RESTORE();
 350                 return;
 351         }
 352         callout_deactivate(&tp->t_timers->tt_keep);
 353         if ((inp->inp_flags & INP_DROPPED) != 0) {
 354                 INP_WUNLOCK(inp);
 355                 INP_INFO_WUNLOCK(&V_tcbinfo);
 356                 CURVNET_RESTORE();
 357                 return;
 358         }
 359         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 360                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 361         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
 362                 ("%s: tp %p keep callout should be running", __func__, tp));
 363         /*
 364          * Keep-alive timer went off; send something
 365          * or drop connection if idle for too long.
 366          */
 367         TCPSTAT_INC(tcps_keeptimeo);
 368         if (tp->t_state < TCPS_ESTABLISHED)
 369                 goto dropit;
 370         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
 371             tp->t_state <= TCPS_CLOSING) {
 372                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 373                         goto dropit;
 374                 /*
 375                  * Send a packet designed to force a response
 376                  * if the peer is up and reachable:
 377                  * either an ACK if the connection is still alive,
 378                  * or an RST if the peer has closed the connection
 379                  * due to timeout or reboot.
 380                  * Using sequence number tp->snd_una-1
 381                  * causes the transmitted zero-length segment
 382                  * to lie outside the receive window;
 383                  * by the protocol spec, this requires the
 384                  * correspondent TCP to respond.
 385                  */
 386                 TCPSTAT_INC(tcps_keepprobe);
 387                 t_template = tcpip_maketemplate(inp);
 388                 if (t_template) {
 389                         tcp_respond(tp, t_template->tt_ipgen,
 390                                     &t_template->tt_t, (struct mbuf *)NULL,
 391                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 392                         free(t_template, M_TEMP);
 393                 }
 394                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 395                     tcp_timer_keep, tp)) {
 396                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 397                 }
 398         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 399                     tcp_timer_keep, tp)) {
 400                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 401                 }
 402
 403 #ifdef TCPDEBUG
 404         if (inp->inp_socket->so_options & SO_DEBUG)
 405                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 406                           PRU_SLOWTIMO);
 407 #endif
 408         INP_WUNLOCK(inp);
 409         INP_INFO_WUNLOCK(&V_tcbinfo);
 410         CURVNET_RESTORE();
 411         return;
 412
 413 dropit:
 414         TCPSTAT_INC(tcps_keepdrops);
 415         tp = tcp_drop(tp, ETIMEDOUT);
 416
 417 #ifdef TCPDEBUG
 418         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 419                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 420                           PRU_SLOWTIMO);
 421 #endif
 422         if (tp != NULL)
 423                 INP_WUNLOCK(tp->t_inpcb);
 424         INP_INFO_WUNLOCK(&V_tcbinfo);
 425         CURVNET_RESTORE();
 426 }
 427
 428 void
 429 tcp_timer_persist(void *xtp)
 430 {
 431         struct tcpcb *tp = xtp;
 432         struct inpcb *inp;
 433         CURVNET_SET(tp->t_vnet);
 434 #ifdef TCPDEBUG
 435         int ostate;
 436
 437         ostate = tp->t_state;
 438 #endif
 439         INP_INFO_WLOCK(&V_tcbinfo);
 440         inp = tp->t_inpcb;
 441         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 442         INP_WLOCK(inp);
 443         if (callout_pending(&tp->t_timers->tt_persist) ||
 444             !callout_active(&tp->t_timers->tt_persist)) {
 445                 INP_WUNLOCK(inp);
 446                 INP_INFO_WUNLOCK(&V_tcbinfo);
 447                 CURVNET_RESTORE();
 448                 return;
 449         }
 450         callout_deactivate(&tp->t_timers->tt_persist);
 451         if ((inp->inp_flags & INP_DROPPED) != 0) {
 452                 INP_WUNLOCK(inp);
 453                 INP_INFO_WUNLOCK(&V_tcbinfo);
 454                 CURVNET_RESTORE();
 455                 return;
 456         }
 457         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 458                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 459         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
 460                 ("%s: tp %p persist callout should be running", __func__, tp));
 461         /*
 462          * Persistance timer into zero window.
 463          * Force a byte to be output, if possible.
 464          */
 465         TCPSTAT_INC(tcps_persisttimeo);
 466         /*
 467          * Hack: if the peer is dead/unreachable, we do not
 468          * time out if the window is closed.  After a full
 469          * backoff, drop the connection if the idle time
 470          * (no responses to probes) reaches the maximum
 471          * backoff that we would use if retransmitting.
 472          */
 473         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 474             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 475              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 476                 TCPSTAT_INC(tcps_persistdrop);
 477                 tp = tcp_drop(tp, ETIMEDOUT);
 478                 goto out;
 479         }
 480         /*
 481          * If the user has closed the socket then drop a persisting
 482          * connection after a much reduced timeout.
 483          */
 484         if (tp->t_state > TCPS_CLOSE_WAIT &&
 485             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 486                 TCPSTAT_INC(tcps_persistdrop);
 487                 tp = tcp_drop(tp, ETIMEDOUT);
 488                 goto out;
 489         }
 490         tcp_setpersist(tp);
 491         tp->t_flags |= TF_FORCEDATA;
 492         (void) tcp_output(tp);
 493         tp->t_flags &= ~TF_FORCEDATA;
 494
 495 out:
 496 #ifdef TCPDEBUG
 497         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 498                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 499 #endif
 500         if (tp != NULL)
 501                 INP_WUNLOCK(inp);
 502         INP_INFO_WUNLOCK(&V_tcbinfo);
 503         CURVNET_RESTORE();
 504 }
 505
 506 void
 507 tcp_timer_rexmt(void * xtp)
 508 {
 509         struct tcpcb *tp = xtp;
 510         CURVNET_SET(tp->t_vnet);
 511         int rexmt;
 512         int headlocked;
 513         struct inpcb *inp;
 514 #ifdef TCPDEBUG
 515         int ostate;
 516
 517         ostate = tp->t_state;
 518 #endif
 519
 520         INP_INFO_RLOCK(&V_tcbinfo);
 521         inp = tp->t_inpcb;
 522         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 523         INP_WLOCK(inp);
 524         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 525             !callout_active(&tp->t_timers->tt_rexmt)) {
 526                 INP_WUNLOCK(inp);
 527                 INP_INFO_RUNLOCK(&V_tcbinfo);
 528                 CURVNET_RESTORE();
 529                 return;
 530         }
 531         callout_deactivate(&tp->t_timers->tt_rexmt);
 532         if ((inp->inp_flags & INP_DROPPED) != 0) {
 533                 INP_WUNLOCK(inp);
 534                 INP_INFO_RUNLOCK(&V_tcbinfo);
 535                 CURVNET_RESTORE();
 536                 return;
 537         }
 538         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 539                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 540         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
 541                 ("%s: tp %p rexmt callout should be running", __func__, tp));
 542         tcp_free_sackholes(tp);
 543         /*
 544          * Retransmission timer went off.  Message has not
 545          * been acked within retransmit interval.  Back off
 546          * to a longer retransmit interval and retransmit one segment.
 547          */
 548         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 549                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 550                 TCPSTAT_INC(tcps_timeoutdrop);
 551                 in_pcbref(inp);
 552                 INP_INFO_RUNLOCK(&V_tcbinfo);
 553                 INP_WUNLOCK(inp);
 554                 INP_INFO_WLOCK(&V_tcbinfo);
 555                 INP_WLOCK(inp);
 556                 if (in_pcbrele_wlocked(inp)) {
 557                         INP_INFO_WUNLOCK(&V_tcbinfo);
 558                         CURVNET_RESTORE();
 559                         return;
 560                 }
 561                 if (inp->inp_flags & INP_DROPPED) {
 562                         INP_WUNLOCK(inp);
 563                         INP_INFO_WUNLOCK(&V_tcbinfo);
 564                         CURVNET_RESTORE();
 565                         return;
 566                 }
 567
 568                 tp = tcp_drop(tp, tp->t_softerror ?
 569                               tp->t_softerror : ETIMEDOUT);
 570                 headlocked = 1;
 571                 goto out;
 572         }
 573         INP_INFO_RUNLOCK(&V_tcbinfo);
 574         headlocked = 0;
 575         if (tp->t_state == TCPS_SYN_SENT) {
 576                 /*
 577                  * If the SYN was retransmitted, indicate CWND to be
 578                  * limited to 1 segment in cc_conn_init().
 579                  */
 580                 tp->snd_cwnd = 1;
 581         } else if (tp->t_rxtshift == 1) {
 582                 /*
 583                  * first retransmit; record ssthresh and cwnd so they can
 584                  * be recovered if this turns out to be a "bad" retransmit.
 585                  * A retransmit is considered "bad" if an ACK for this
 586                  * segment is received within RTT/2 interval; the assumption
 587                  * here is that the ACK was already in flight.  See
 588                  * "On Estimating End-to-End Network Path Properties" by
 589                  * Allman and Paxson for more details.
 590                  */
 591                 tp->snd_cwnd_prev = tp->snd_cwnd;
 592                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 593                 tp->snd_recover_prev = tp->snd_recover;
 594                 if (IN_FASTRECOVERY(tp->t_flags))
 595                         tp->t_flags |= TF_WASFRECOVERY;
 596                 else
 597                         tp->t_flags &= ~TF_WASFRECOVERY;
 598                 if (IN_CONGRECOVERY(tp->t_flags))
 599                         tp->t_flags |= TF_WASCRECOVERY;
 600                 else
 601                         tp->t_flags &= ~TF_WASCRECOVERY;
 602                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 603                 tp->t_flags |= TF_PREVVALID;
 604         } else
 605                 tp->t_flags &= ~TF_PREVVALID;
 606         TCPSTAT_INC(tcps_rexmttimeo);
 607         if ((tp->t_state == TCPS_SYN_SENT) ||
 608             (tp->t_state == TCPS_SYN_RECEIVED))
 609                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 610         else
 611                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 612         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 613                       tp->t_rttmin, TCPTV_REXMTMAX);
 614
 615         /*
 616          * We enter the path for PLMTUD if connection is established or, if
 617          * connection is FIN_WAIT_1 status, reason for the last is that if
 618          * amount of data we send is very small, we could send it in couple of
 619          * packets and process straight to FIN. In that case we won't catch
 620          * ESTABLISHED state.
 621          */
 622         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 623             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 624                 int optlen;
 625 #ifdef INET6
 626                 int isipv6;
 627 #endif
 628
 629                 /*
 630                  * Idea here is that at each stage of mtu probe (usually, 1448
 631                  * -> 1188 -> 524) should be given 2 chances to recover before
 632                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 633                  *  take care of that.
 634                  */
 635                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 636                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 637                     (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
 638                         /*
 639                          * Enter Path MTU Black-hole Detection mechanism:
 640                          * - Disable Path MTU Discovery (IP "DF" bit).
 641                          * - Reduce MTU to lower value than what we
 642                          *   negotiated with peer.
 643                          */
 644                         /* Record that we may have found a black hole. */
 645                         tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 646
 647                         /* Keep track of previous MSS. */
 648                         optlen = tp->t_maxopd - tp->t_maxseg;
 649                         tp->t_pmtud_saved_maxopd = tp->t_maxopd;
 650
 651                         /*
 652                          * Reduce the MSS to blackhole value or to the default
 653                          * in an attempt to retransmit.
 654                          */
 655 #ifdef INET6
 656                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 657                         if (isipv6 &&
 658                             tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
 659                                 /* Use the sysctl tuneable blackhole MSS. */
 660                                 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
 661                                 V_tcp_pmtud_blackhole_activated++;
 662                         } else if (isipv6) {
 663                                 /* Use the default MSS. */
 664                                 tp->t_maxopd = V_tcp_v6mssdflt;
 665                                 /*
 666                                  * Disable Path MTU Discovery when we switch to
 667                                  * minmss.
 668                                  */
 669                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 670                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 671                         }
 672 #endif
 673 #if defined(INET6) && defined(INET)
 674                         else
 675 #endif
 676 #ifdef INET
 677                         if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
 678                                 /* Use the sysctl tuneable blackhole MSS. */
 679                                 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
 680                                 V_tcp_pmtud_blackhole_activated++;
 681                         } else {
 682                                 /* Use the default MSS. */
 683                                 tp->t_maxopd = V_tcp_mssdflt;
 684                                 /*
 685                                  * Disable Path MTU Discovery when we switch to
 686                                  * minmss.
 687                                  */
 688                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 689                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 690                         }
 691 #endif
 692                         tp->t_maxseg = tp->t_maxopd - optlen;
 693                         /*
 694                          * Reset the slow-start flight size
 695                          * as it may depend on the new MSS.
 696                          */
 697                         if (CC_ALGO(tp)->conn_init != NULL)
 698                                 CC_ALGO(tp)->conn_init(tp->ccv);
 699                 } else {
 700                         /*
 701                          * If further retransmissions are still unsuccessful
 702                          * with a lowered MTU, maybe this isn't a blackhole and
 703                          * we restore the previous MSS and blackhole detection
 704                          * flags.
 705                          * The limit '6' is determined by giving each probe
 706                          * stage (1448, 1188, 524) 2 chances to recover.
 707                          */
 708                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 709                             (tp->t_rxtshift > 6)) {
 710                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 711                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 712                                 optlen = tp->t_maxopd - tp->t_maxseg;
 713                                 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
 714                                 tp->t_maxseg = tp->t_maxopd - optlen;
 715                                 V_tcp_pmtud_blackhole_failed++;
 716                                 /*
 717                                  * Reset the slow-start flight size as it
 718                                  * may depend on the new MSS.
 719                                  */
 720                                 if (CC_ALGO(tp)->conn_init != NULL)
 721                                         CC_ALGO(tp)->conn_init(tp->ccv);
 722                         }
 723                 }
 724         }
 725
 726         /*
 727          * Disable RFC1323 and SACK if we haven't got any response to
 728          * our third SYN to work-around some broken terminal servers
 729          * (most of which have hopefully been retired) that have bad VJ
 730          * header compression code which trashes TCP segments containing
 731          * unknown-to-them TCP options.
 732          */
 733         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 734             (tp->t_rxtshift == 3))
 735                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 736         /*
 737          * If we backed off this far, our srtt estimate is probably bogus.
 738          * Clobber it so we'll take the next rtt measurement as our srtt;
 739          * move the current srtt into rttvar to keep the current
 740          * retransmit times until then.
 741          */
 742         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 743 #ifdef INET6
 744                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 745                         in6_losing(tp->t_inpcb);
 746 #endif
 747                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 748                 tp->t_srtt = 0;
 749         }
 750         tp->snd_nxt = tp->snd_una;
 751         tp->snd_recover = tp->snd_max;
 752         /*
 753          * Force a segment to be sent.
 754          */
 755         tp->t_flags |= TF_ACKNOW;
 756         /*
 757          * If timing a segment in this window, stop the timer.
 758          */
 759         tp->t_rtttime = 0;
 760
 761         cc_cong_signal(tp, NULL, CC_RTO);
 762
 763         (void) tcp_output(tp);
 764
 765 out:
 766 #ifdef TCPDEBUG
 767         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 768                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 769                           PRU_SLOWTIMO);
 770 #endif
 771         if (tp != NULL)
 772                 INP_WUNLOCK(inp);
 773         if (headlocked)
 774                 INP_INFO_WUNLOCK(&V_tcbinfo);
 775         CURVNET_RESTORE();
 776 }
 777
 778 void
 779 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 780 {
 781         struct callout *t_callout;
 782         timeout_t *f_callout;
 783         struct inpcb *inp = tp->t_inpcb;
 784         int cpu = INP_CPU(inp);
 785         uint32_t f_reset;
 786
 787 #ifdef TCP_OFFLOAD
 788         if (tp->t_flags & TF_TOE)
 789                 return;
 790 #endif
 791
 792         if (tp->t_timers->tt_flags & TT_STOPPED)
 793                 return;
 794
 795         switch (timer_type) {
 796                 case TT_DELACK:
 797                         t_callout = &tp->t_timers->tt_delack;
 798                         f_callout = tcp_timer_delack;
 799                         f_reset = TT_DELACK_RST;
 800                         break;
 801                 case TT_REXMT:
 802                         t_callout = &tp->t_timers->tt_rexmt;
 803                         f_callout = tcp_timer_rexmt;
 804                         f_reset = TT_REXMT_RST;
 805                         break;
 806                 case TT_PERSIST:
 807                         t_callout = &tp->t_timers->tt_persist;
 808                         f_callout = tcp_timer_persist;
 809                         f_reset = TT_PERSIST_RST;
 810                         break;
 811                 case TT_KEEP:
 812                         t_callout = &tp->t_timers->tt_keep;
 813                         f_callout = tcp_timer_keep;
 814                         f_reset = TT_KEEP_RST;
 815                         break;
 816                 case TT_2MSL:
 817                         t_callout = &tp->t_timers->tt_2msl;
 818                         f_callout = tcp_timer_2msl;
 819                         f_reset = TT_2MSL_RST;
 820                         break;
 821                 default:
 822                         panic("tp %p bad timer_type %#x", tp, timer_type);
 823                 }
 824         if (delta == 0) {
 825                 if ((tp->t_timers->tt_flags & timer_type) &&
 826                     callout_stop(t_callout) &&
 827                     (tp->t_timers->tt_flags & f_reset)) {
 828                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 829                 }
 830         } else {
 831                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
 832                         tp->t_timers->tt_flags |= (timer_type | f_reset);
 833                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 834                 } else {
 835                         /* Reset already running callout on the same CPU. */
 836                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
 837                                 /*
 838                                  * Callout not cancelled, consider it as not
 839                                  * properly restarted. */
 840                                 tp->t_timers->tt_flags &= ~f_reset;
 841                         }
 842                 }
 843         }
 844 }
 845
 846 int
 847 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 848 {
 849         struct callout *t_callout;
 850
 851         switch (timer_type) {
 852                 case TT_DELACK:
 853                         t_callout = &tp->t_timers->tt_delack;
 854                         break;
 855                 case TT_REXMT:
 856                         t_callout = &tp->t_timers->tt_rexmt;
 857                         break;
 858                 case TT_PERSIST:
 859                         t_callout = &tp->t_timers->tt_persist;
 860                         break;
 861                 case TT_KEEP:
 862                         t_callout = &tp->t_timers->tt_keep;
 863                         break;
 864                 case TT_2MSL:
 865                         t_callout = &tp->t_timers->tt_2msl;
 866                         break;
 867                 default:
 868                         panic("tp %p bad timer_type %#x", tp, timer_type);
 869                 }
 870         return callout_active(t_callout);
 871 }
 872
 873 void
 874 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 875 {
 876         struct callout *t_callout;
 877         timeout_t *f_callout;
 878         uint32_t f_reset;
 879
 880         tp->t_timers->tt_flags |= TT_STOPPED;
 881
 882         switch (timer_type) {
 883                 case TT_DELACK:
 884                         t_callout = &tp->t_timers->tt_delack;
 885                         f_callout = tcp_timer_delack_discard;
 886                         f_reset = TT_DELACK_RST;
 887                         break;
 888                 case TT_REXMT:
 889                         t_callout = &tp->t_timers->tt_rexmt;
 890                         f_callout = tcp_timer_rexmt_discard;
 891                         f_reset = TT_REXMT_RST;
 892                         break;
 893                 case TT_PERSIST:
 894                         t_callout = &tp->t_timers->tt_persist;
 895                         f_callout = tcp_timer_persist_discard;
 896                         f_reset = TT_PERSIST_RST;
 897                         break;
 898                 case TT_KEEP:
 899                         t_callout = &tp->t_timers->tt_keep;
 900                         f_callout = tcp_timer_keep_discard;
 901                         f_reset = TT_KEEP_RST;
 902                         break;
 903                 case TT_2MSL:
 904                         t_callout = &tp->t_timers->tt_2msl;
 905                         f_callout = tcp_timer_2msl_discard;
 906                         f_reset = TT_2MSL_RST;
 907                         break;
 908                 default:
 909                         panic("tp %p bad timer_type %#x", tp, timer_type);
 910                 }
 911
 912         if (tp->t_timers->tt_flags & timer_type) {
 913                 if (callout_stop(t_callout) &&
 914                     (tp->t_timers->tt_flags & f_reset)) {
 915                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 916                 } else {
 917                         /*
 918                          * Can't stop the callout, defer tcpcb actual deletion
 919                          * to the last tcp timer discard callout.
 920                          * The TT_STOPPED flag will ensure that no tcp timer
 921                          * callouts can be restarted on our behalf, and
 922                          * past this point currently running callouts waiting
 923                          * on inp lock will return right away after the
 924                          * classical check for callout reset/stop events:
 925                          * callout_pending() || !callout_active()
 926                          */
 927                         callout_reset(t_callout, 1, f_callout, tp);
 928                 }
 929         }
 930 }
 931
 932 #define ticks_to_msecs(t)       (1000*(t) / hz)
 933
 934 void
 935 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
 936     struct xtcp_timer *xtimer)
 937 {
 938         sbintime_t now;
 939
 940         bzero(xtimer, sizeof(*xtimer));
 941         if (timer == NULL)
 942                 return;
 943         now = getsbinuptime();
 944         if (callout_active(&timer->tt_delack))
 945                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
 946         if (callout_active(&timer->tt_rexmt))
 947                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
 948         if (callout_active(&timer->tt_persist))
 949                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
 950         if (callout_active(&timer->tt_keep))
 951                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
 952         if (callout_active(&timer->tt_2msl))
 953                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
 954         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
 955 }