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 int     tcp_always_keepalive = 1;
 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 115     &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 116 __strong_reference(tcp_always_keepalive, always_keepalive);
 117
 118 int    tcp_fast_finwait2_recycle = 0;
 119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 120     &tcp_fast_finwait2_recycle, 0,
 121     "Recycle closed FIN_WAIT_2 connections faster");
 122
 123 int    tcp_finwait2_timeout;
 124 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 125     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 126
 127 int     tcp_keepcnt = TCPTV_KEEPCNT;
 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
 129     "Number of keepalive probes to send");
 130
 131         /* max idle probes */
 132 int     tcp_maxpersistidle;
 133
 134 static int      tcp_rexmit_drop_options = 0;
 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
 136     &tcp_rexmit_drop_options, 0,
 137     "Drop TCP options from 3rd and later retransmitted SYN");
 138
 139 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
 140 #define V_tcp_pmtud_blackhole_detect    VNET(tcp_pmtud_blackhole_detect)
 141 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
 142     CTLFLAG_RW,
 143     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
 144     "Path MTU Discovery Black Hole Detection Enabled");
 145
 146 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
 147 #define V_tcp_pmtud_blackhole_activated \
 148     VNET(tcp_pmtud_blackhole_activated)
 149 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
 150     CTLFLAG_RD,
 151     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
 152     "Path MTU Discovery Black Hole Detection, Activation Count");
 153
 154 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
 155 #define V_tcp_pmtud_blackhole_activated_min_mss \
 156     VNET(tcp_pmtud_blackhole_activated_min_mss)
 157 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
 158     CTLFLAG_RD,
 159     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
 160     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
 161
 162 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
 163 #define V_tcp_pmtud_blackhole_failed    VNET(tcp_pmtud_blackhole_failed)
 164 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
 165     CTLFLAG_RD,
 166     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
 167     "Path MTU Discovery Black Hole Detection, Failure Count");
 168
 169 #ifdef INET
 170 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
 171 #define V_tcp_pmtud_blackhole_mss       VNET(tcp_pmtud_blackhole_mss)
 172 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
 173     CTLFLAG_RW,
 174     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
 175     "Path MTU Discovery Black Hole Detection lowered MSS");
 176 #endif
 177
 178 #ifdef INET6
 179 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
 180 #define V_tcp_v6pmtud_blackhole_mss     VNET(tcp_v6pmtud_blackhole_mss)
 181 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
 182     CTLFLAG_RW,
 183     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
 184     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
 185 #endif
 186
 187 static int      per_cpu_timers = 0;
 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
 189     &per_cpu_timers , 0, "run tcp timers on all cpus");
 190
 191 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
 192                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
 193
 194 /*
 195  * Tcp protocol timeout routine called every 500 ms.
 196  * Updates timestamps used for TCP
 197  * causes finite state machine actions if timers expire.
 198  */
 199 void
 200 tcp_slowtimo(void)
 201 {
 202         VNET_ITERATOR_DECL(vnet_iter);
 203
 204         VNET_LIST_RLOCK_NOSLEEP();
 205         VNET_FOREACH(vnet_iter) {
 206                 CURVNET_SET(vnet_iter);
 207                 (void) tcp_tw_2msl_scan(0);
 208                 CURVNET_RESTORE();
 209         }
 210         VNET_LIST_RUNLOCK_NOSLEEP();
 211 }
 212
 213 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 214     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 215
 216 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 217     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 218
 219 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 220
 221 /*
 222  * TCP timer processing.
 223  */
 224
 225 void
 226 tcp_timer_delack(void *xtp)
 227 {
 228         struct tcpcb *tp = xtp;
 229         struct inpcb *inp;
 230         CURVNET_SET(tp->t_vnet);
 231
 232         inp = tp->t_inpcb;
 233         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 234         INP_WLOCK(inp);
 235         if (callout_pending(&tp->t_timers->tt_delack) ||
 236             !callout_active(&tp->t_timers->tt_delack)) {
 237                 INP_WUNLOCK(inp);
 238                 CURVNET_RESTORE();
 239                 return;
 240         }
 241         callout_deactivate(&tp->t_timers->tt_delack);
 242         if ((inp->inp_flags & INP_DROPPED) != 0) {
 243                 INP_WUNLOCK(inp);
 244                 CURVNET_RESTORE();
 245                 return;
 246         }
 247         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 248                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 249         KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
 250                 ("%s: tp %p delack callout should be running", __func__, tp));
 251
 252         tp->t_flags |= TF_ACKNOW;
 253         TCPSTAT_INC(tcps_delack);
 254         (void) tcp_output(tp);
 255         INP_WUNLOCK(inp);
 256         CURVNET_RESTORE();
 257 }
 258
 259 void
 260 tcp_timer_2msl(void *xtp)
 261 {
 262         struct tcpcb *tp = xtp;
 263         struct inpcb *inp;
 264         CURVNET_SET(tp->t_vnet);
 265 #ifdef TCPDEBUG
 266         int ostate;
 267
 268         ostate = tp->t_state;
 269 #endif
 270         INP_INFO_RLOCK(&V_tcbinfo);
 271         inp = tp->t_inpcb;
 272         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 273         INP_WLOCK(inp);
 274         tcp_free_sackholes(tp);
 275         if (callout_pending(&tp->t_timers->tt_2msl) ||
 276             !callout_active(&tp->t_timers->tt_2msl)) {
 277                 INP_WUNLOCK(tp->t_inpcb);
 278                 INP_INFO_RUNLOCK(&V_tcbinfo);
 279                 CURVNET_RESTORE();
 280                 return;
 281         }
 282         callout_deactivate(&tp->t_timers->tt_2msl);
 283         if ((inp->inp_flags & INP_DROPPED) != 0) {
 284                 INP_WUNLOCK(inp);
 285                 INP_INFO_RUNLOCK(&V_tcbinfo);
 286                 CURVNET_RESTORE();
 287                 return;
 288         }
 289         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 290                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 291         KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
 292                 ("%s: tp %p 2msl callout should be running", __func__, tp));
 293         /*
 294          * 2 MSL timeout in shutdown went off.  If we're closed but
 295          * still waiting for peer to close and connection has been idle
 296          * too long delete connection control block.  Otherwise, check
 297          * again in a bit.
 298          *
 299          * If in TIME_WAIT state just ignore as this timeout is handled in
 300          * tcp_tw_2msl_scan().
 301          *
 302          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 303          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 304          * Ignore fact that there were recent incoming segments.
 305          */
 306         if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
 307                 INP_WUNLOCK(inp);
 308                 INP_INFO_RUNLOCK(&V_tcbinfo);
 309                 CURVNET_RESTORE();
 310                 return;
 311         }
 312         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 313             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 314             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 315                 TCPSTAT_INC(tcps_finwait2_drops);
 316                 tp = tcp_close(tp);
 317         } else {
 318                 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 319                         if (!callout_reset(&tp->t_timers->tt_2msl,
 320                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
 321                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
 322                         }
 323                 } else
 324                        tp = tcp_close(tp);
 325        }
 326
 327 #ifdef TCPDEBUG
 328         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 329                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 330                           PRU_SLOWTIMO);
 331 #endif
 332         if (tp != NULL)
 333                 INP_WUNLOCK(inp);
 334         INP_INFO_RUNLOCK(&V_tcbinfo);
 335         CURVNET_RESTORE();
 336 }
 337
 338 void
 339 tcp_timer_keep(void *xtp)
 340 {
 341         struct tcpcb *tp = xtp;
 342         struct tcptemp *t_template;
 343         struct inpcb *inp;
 344         CURVNET_SET(tp->t_vnet);
 345 #ifdef TCPDEBUG
 346         int ostate;
 347
 348         ostate = tp->t_state;
 349 #endif
 350         INP_INFO_RLOCK(&V_tcbinfo);
 351         inp = tp->t_inpcb;
 352         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 353         INP_WLOCK(inp);
 354         if (callout_pending(&tp->t_timers->tt_keep) ||
 355             !callout_active(&tp->t_timers->tt_keep)) {
 356                 INP_WUNLOCK(inp);
 357                 INP_INFO_RUNLOCK(&V_tcbinfo);
 358                 CURVNET_RESTORE();
 359                 return;
 360         }
 361         callout_deactivate(&tp->t_timers->tt_keep);
 362         if ((inp->inp_flags & INP_DROPPED) != 0) {
 363                 INP_WUNLOCK(inp);
 364                 INP_INFO_RUNLOCK(&V_tcbinfo);
 365                 CURVNET_RESTORE();
 366                 return;
 367         }
 368         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 369                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 370         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
 371                 ("%s: tp %p keep callout should be running", __func__, tp));
 372         /*
 373          * Keep-alive timer went off; send something
 374          * or drop connection if idle for too long.
 375          */
 376         TCPSTAT_INC(tcps_keeptimeo);
 377         if (tp->t_state < TCPS_ESTABLISHED)
 378                 goto dropit;
 379         if ((tcp_always_keepalive ||
 380             inp->inp_socket->so_options & SO_KEEPALIVE) &&
 381             tp->t_state <= TCPS_CLOSING) {
 382                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 383                         goto dropit;
 384                 /*
 385                  * Send a packet designed to force a response
 386                  * if the peer is up and reachable:
 387                  * either an ACK if the connection is still alive,
 388                  * or an RST if the peer has closed the connection
 389                  * due to timeout or reboot.
 390                  * Using sequence number tp->snd_una-1
 391                  * causes the transmitted zero-length segment
 392                  * to lie outside the receive window;
 393                  * by the protocol spec, this requires the
 394                  * correspondent TCP to respond.
 395                  */
 396                 TCPSTAT_INC(tcps_keepprobe);
 397                 t_template = tcpip_maketemplate(inp);
 398                 if (t_template) {
 399                         tcp_respond(tp, t_template->tt_ipgen,
 400                                     &t_template->tt_t, (struct mbuf *)NULL,
 401                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 402                         free(t_template, M_TEMP);
 403                 }
 404                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 405                     tcp_timer_keep, tp)) {
 406                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 407                 }
 408         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 409                     tcp_timer_keep, tp)) {
 410                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 411                 }
 412
 413 #ifdef TCPDEBUG
 414         if (inp->inp_socket->so_options & SO_DEBUG)
 415                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 416                           PRU_SLOWTIMO);
 417 #endif
 418         INP_WUNLOCK(inp);
 419         INP_INFO_RUNLOCK(&V_tcbinfo);
 420         CURVNET_RESTORE();
 421         return;
 422
 423 dropit:
 424         TCPSTAT_INC(tcps_keepdrops);
 425         tp = tcp_drop(tp, ETIMEDOUT);
 426
 427 #ifdef TCPDEBUG
 428         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 429                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 430                           PRU_SLOWTIMO);
 431 #endif
 432         if (tp != NULL)
 433                 INP_WUNLOCK(tp->t_inpcb);
 434         INP_INFO_RUNLOCK(&V_tcbinfo);
 435         CURVNET_RESTORE();
 436 }
 437
 438 void
 439 tcp_timer_persist(void *xtp)
 440 {
 441         struct tcpcb *tp = xtp;
 442         struct inpcb *inp;
 443         CURVNET_SET(tp->t_vnet);
 444 #ifdef TCPDEBUG
 445         int ostate;
 446
 447         ostate = tp->t_state;
 448 #endif
 449         INP_INFO_RLOCK(&V_tcbinfo);
 450         inp = tp->t_inpcb;
 451         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 452         INP_WLOCK(inp);
 453         if (callout_pending(&tp->t_timers->tt_persist) ||
 454             !callout_active(&tp->t_timers->tt_persist)) {
 455                 INP_WUNLOCK(inp);
 456                 INP_INFO_RUNLOCK(&V_tcbinfo);
 457                 CURVNET_RESTORE();
 458                 return;
 459         }
 460         callout_deactivate(&tp->t_timers->tt_persist);
 461         if ((inp->inp_flags & INP_DROPPED) != 0) {
 462                 INP_WUNLOCK(inp);
 463                 INP_INFO_RUNLOCK(&V_tcbinfo);
 464                 CURVNET_RESTORE();
 465                 return;
 466         }
 467         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 468                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 469         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
 470                 ("%s: tp %p persist callout should be running", __func__, tp));
 471         /*
 472          * Persistance timer into zero window.
 473          * Force a byte to be output, if possible.
 474          */
 475         TCPSTAT_INC(tcps_persisttimeo);
 476         /*
 477          * Hack: if the peer is dead/unreachable, we do not
 478          * time out if the window is closed.  After a full
 479          * backoff, drop the connection if the idle time
 480          * (no responses to probes) reaches the maximum
 481          * backoff that we would use if retransmitting.
 482          */
 483         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 484             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 485              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 486                 TCPSTAT_INC(tcps_persistdrop);
 487                 tp = tcp_drop(tp, ETIMEDOUT);
 488                 goto out;
 489         }
 490         /*
 491          * If the user has closed the socket then drop a persisting
 492          * connection after a much reduced timeout.
 493          */
 494         if (tp->t_state > TCPS_CLOSE_WAIT &&
 495             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 496                 TCPSTAT_INC(tcps_persistdrop);
 497                 tp = tcp_drop(tp, ETIMEDOUT);
 498                 goto out;
 499         }
 500         tcp_setpersist(tp);
 501         tp->t_flags |= TF_FORCEDATA;
 502         (void) tcp_output(tp);
 503         tp->t_flags &= ~TF_FORCEDATA;
 504
 505 out:
 506 #ifdef TCPDEBUG
 507         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 508                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 509 #endif
 510         if (tp != NULL)
 511                 INP_WUNLOCK(inp);
 512         INP_INFO_RUNLOCK(&V_tcbinfo);
 513         CURVNET_RESTORE();
 514 }
 515
 516 void
 517 tcp_timer_rexmt(void * xtp)
 518 {
 519         struct tcpcb *tp = xtp;
 520         CURVNET_SET(tp->t_vnet);
 521         int rexmt;
 522         int headlocked;
 523         struct inpcb *inp;
 524 #ifdef TCPDEBUG
 525         int ostate;
 526
 527         ostate = tp->t_state;
 528 #endif
 529
 530         INP_INFO_RLOCK(&V_tcbinfo);
 531         inp = tp->t_inpcb;
 532         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 533         INP_WLOCK(inp);
 534         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 535             !callout_active(&tp->t_timers->tt_rexmt)) {
 536                 INP_WUNLOCK(inp);
 537                 INP_INFO_RUNLOCK(&V_tcbinfo);
 538                 CURVNET_RESTORE();
 539                 return;
 540         }
 541         callout_deactivate(&tp->t_timers->tt_rexmt);
 542         if ((inp->inp_flags & INP_DROPPED) != 0) {
 543                 INP_WUNLOCK(inp);
 544                 INP_INFO_RUNLOCK(&V_tcbinfo);
 545                 CURVNET_RESTORE();
 546                 return;
 547         }
 548         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 549                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 550         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
 551                 ("%s: tp %p rexmt callout should be running", __func__, tp));
 552         tcp_free_sackholes(tp);
 553         /*
 554          * Retransmission timer went off.  Message has not
 555          * been acked within retransmit interval.  Back off
 556          * to a longer retransmit interval and retransmit one segment.
 557          */
 558         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 559                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 560                 TCPSTAT_INC(tcps_timeoutdrop);
 561
 562                 tp = tcp_drop(tp, tp->t_softerror ?
 563                               tp->t_softerror : ETIMEDOUT);
 564                 headlocked = 1;
 565                 goto out;
 566         }
 567         INP_INFO_RUNLOCK(&V_tcbinfo);
 568         headlocked = 0;
 569         if (tp->t_state == TCPS_SYN_SENT) {
 570                 /*
 571                  * If the SYN was retransmitted, indicate CWND to be
 572                  * limited to 1 segment in cc_conn_init().
 573                  */
 574                 tp->snd_cwnd = 1;
 575         } else if (tp->t_rxtshift == 1) {
 576                 /*
 577                  * first retransmit; record ssthresh and cwnd so they can
 578                  * be recovered if this turns out to be a "bad" retransmit.
 579                  * A retransmit is considered "bad" if an ACK for this
 580                  * segment is received within RTT/2 interval; the assumption
 581                  * here is that the ACK was already in flight.  See
 582                  * "On Estimating End-to-End Network Path Properties" by
 583                  * Allman and Paxson for more details.
 584                  */
 585                 tp->snd_cwnd_prev = tp->snd_cwnd;
 586                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 587                 tp->snd_recover_prev = tp->snd_recover;
 588                 if (IN_FASTRECOVERY(tp->t_flags))
 589                         tp->t_flags |= TF_WASFRECOVERY;
 590                 else
 591                         tp->t_flags &= ~TF_WASFRECOVERY;
 592                 if (IN_CONGRECOVERY(tp->t_flags))
 593                         tp->t_flags |= TF_WASCRECOVERY;
 594                 else
 595                         tp->t_flags &= ~TF_WASCRECOVERY;
 596                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 597                 tp->t_flags |= TF_PREVVALID;
 598         } else
 599                 tp->t_flags &= ~TF_PREVVALID;
 600         TCPSTAT_INC(tcps_rexmttimeo);
 601         if ((tp->t_state == TCPS_SYN_SENT) ||
 602             (tp->t_state == TCPS_SYN_RECEIVED))
 603                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 604         else
 605                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 606         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 607                       tp->t_rttmin, TCPTV_REXMTMAX);
 608
 609         /*
 610          * We enter the path for PLMTUD if connection is established or, if
 611          * connection is FIN_WAIT_1 status, reason for the last is that if
 612          * amount of data we send is very small, we could send it in couple of
 613          * packets and process straight to FIN. In that case we won't catch
 614          * ESTABLISHED state.
 615          */
 616         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 617             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 618                 int optlen;
 619 #ifdef INET6
 620                 int isipv6;
 621 #endif
 622
 623                 /*
 624                  * Idea here is that at each stage of mtu probe (usually, 1448
 625                  * -> 1188 -> 524) should be given 2 chances to recover before
 626                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 627                  *  take care of that.
 628                  */
 629                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 630                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 631                     (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
 632                         /*
 633                          * Enter Path MTU Black-hole Detection mechanism:
 634                          * - Disable Path MTU Discovery (IP "DF" bit).
 635                          * - Reduce MTU to lower value than what we
 636                          *   negotiated with peer.
 637                          */
 638                         /* Record that we may have found a black hole. */
 639                         tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 640
 641                         /* Keep track of previous MSS. */
 642                         optlen = tp->t_maxopd - tp->t_maxseg;
 643                         tp->t_pmtud_saved_maxopd = tp->t_maxopd;
 644
 645                         /*
 646                          * Reduce the MSS to blackhole value or to the default
 647                          * in an attempt to retransmit.
 648                          */
 649 #ifdef INET6
 650                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 651                         if (isipv6 &&
 652                             tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
 653                                 /* Use the sysctl tuneable blackhole MSS. */
 654                                 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
 655                                 V_tcp_pmtud_blackhole_activated++;
 656                         } else if (isipv6) {
 657                                 /* Use the default MSS. */
 658                                 tp->t_maxopd = V_tcp_v6mssdflt;
 659                                 /*
 660                                  * Disable Path MTU Discovery when we switch to
 661                                  * minmss.
 662                                  */
 663                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 664                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 665                         }
 666 #endif
 667 #if defined(INET6) && defined(INET)
 668                         else
 669 #endif
 670 #ifdef INET
 671                         if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
 672                                 /* Use the sysctl tuneable blackhole MSS. */
 673                                 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
 674                                 V_tcp_pmtud_blackhole_activated++;
 675                         } else {
 676                                 /* Use the default MSS. */
 677                                 tp->t_maxopd = V_tcp_mssdflt;
 678                                 /*
 679                                  * Disable Path MTU Discovery when we switch to
 680                                  * minmss.
 681                                  */
 682                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 683                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 684                         }
 685 #endif
 686                         tp->t_maxseg = tp->t_maxopd - optlen;
 687                         /*
 688                          * Reset the slow-start flight size
 689                          * as it may depend on the new MSS.
 690                          */
 691                         if (CC_ALGO(tp)->conn_init != NULL)
 692                                 CC_ALGO(tp)->conn_init(tp->ccv);
 693                 } else {
 694                         /*
 695                          * If further retransmissions are still unsuccessful
 696                          * with a lowered MTU, maybe this isn't a blackhole and
 697                          * we restore the previous MSS and blackhole detection
 698                          * flags.
 699                          * The limit '6' is determined by giving each probe
 700                          * stage (1448, 1188, 524) 2 chances to recover.
 701                          */
 702                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 703                             (tp->t_rxtshift > 6)) {
 704                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 705                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 706                                 optlen = tp->t_maxopd - tp->t_maxseg;
 707                                 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
 708                                 tp->t_maxseg = tp->t_maxopd - optlen;
 709                                 V_tcp_pmtud_blackhole_failed++;
 710                                 /*
 711                                  * Reset the slow-start flight size as it
 712                                  * may depend on the new MSS.
 713                                  */
 714                                 if (CC_ALGO(tp)->conn_init != NULL)
 715                                         CC_ALGO(tp)->conn_init(tp->ccv);
 716                         }
 717                 }
 718         }
 719
 720         /*
 721          * Disable RFC1323 and SACK if we haven't got any response to
 722          * our third SYN to work-around some broken terminal servers
 723          * (most of which have hopefully been retired) that have bad VJ
 724          * header compression code which trashes TCP segments containing
 725          * unknown-to-them TCP options.
 726          */
 727         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 728             (tp->t_rxtshift == 3))
 729                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 730         /*
 731          * If we backed off this far, our srtt estimate is probably bogus.
 732          * Clobber it so we'll take the next rtt measurement as our srtt;
 733          * move the current srtt into rttvar to keep the current
 734          * retransmit times until then.
 735          */
 736         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 737 #ifdef INET6
 738                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 739                         in6_losing(tp->t_inpcb);
 740 #endif
 741                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 742                 tp->t_srtt = 0;
 743         }
 744         tp->snd_nxt = tp->snd_una;
 745         tp->snd_recover = tp->snd_max;
 746         /*
 747          * Force a segment to be sent.
 748          */
 749         tp->t_flags |= TF_ACKNOW;
 750         /*
 751          * If timing a segment in this window, stop the timer.
 752          */
 753         tp->t_rtttime = 0;
 754
 755         cc_cong_signal(tp, NULL, CC_RTO);
 756
 757         (void) tcp_output(tp);
 758
 759 out:
 760 #ifdef TCPDEBUG
 761         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 762                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 763                           PRU_SLOWTIMO);
 764 #endif
 765         if (tp != NULL)
 766                 INP_WUNLOCK(inp);
 767         if (headlocked)
 768                 INP_INFO_RUNLOCK(&V_tcbinfo);
 769         CURVNET_RESTORE();
 770 }
 771
 772 void
 773 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 774 {
 775         struct callout *t_callout;
 776         timeout_t *f_callout;
 777         struct inpcb *inp = tp->t_inpcb;
 778         int cpu = INP_CPU(inp);
 779         uint32_t f_reset;
 780
 781 #ifdef TCP_OFFLOAD
 782         if (tp->t_flags & TF_TOE)
 783                 return;
 784 #endif
 785
 786         if (tp->t_timers->tt_flags & TT_STOPPED)
 787                 return;
 788
 789         switch (timer_type) {
 790                 case TT_DELACK:
 791                         t_callout = &tp->t_timers->tt_delack;
 792                         f_callout = tcp_timer_delack;
 793                         f_reset = TT_DELACK_RST;
 794                         break;
 795                 case TT_REXMT:
 796                         t_callout = &tp->t_timers->tt_rexmt;
 797                         f_callout = tcp_timer_rexmt;
 798                         f_reset = TT_REXMT_RST;
 799                         break;
 800                 case TT_PERSIST:
 801                         t_callout = &tp->t_timers->tt_persist;
 802                         f_callout = tcp_timer_persist;
 803                         f_reset = TT_PERSIST_RST;
 804                         break;
 805                 case TT_KEEP:
 806                         t_callout = &tp->t_timers->tt_keep;
 807                         f_callout = tcp_timer_keep;
 808                         f_reset = TT_KEEP_RST;
 809                         break;
 810                 case TT_2MSL:
 811                         t_callout = &tp->t_timers->tt_2msl;
 812                         f_callout = tcp_timer_2msl;
 813                         f_reset = TT_2MSL_RST;
 814                         break;
 815                 default:
 816                         panic("tp %p bad timer_type %#x", tp, timer_type);
 817                 }
 818         if (delta == 0) {
 819                 if ((tp->t_timers->tt_flags & timer_type) &&
 820                     callout_stop(t_callout) &&
 821                     (tp->t_timers->tt_flags & f_reset)) {
 822                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 823                 }
 824         } else {
 825                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
 826                         tp->t_timers->tt_flags |= (timer_type | f_reset);
 827                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 828                 } else {
 829                         /* Reset already running callout on the same CPU. */
 830                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
 831                                 /*
 832                                  * Callout not cancelled, consider it as not
 833                                  * properly restarted. */
 834                                 tp->t_timers->tt_flags &= ~f_reset;
 835                         }
 836                 }
 837         }
 838 }
 839
 840 int
 841 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 842 {
 843         struct callout *t_callout;
 844
 845         switch (timer_type) {
 846                 case TT_DELACK:
 847                         t_callout = &tp->t_timers->tt_delack;
 848                         break;
 849                 case TT_REXMT:
 850                         t_callout = &tp->t_timers->tt_rexmt;
 851                         break;
 852                 case TT_PERSIST:
 853                         t_callout = &tp->t_timers->tt_persist;
 854                         break;
 855                 case TT_KEEP:
 856                         t_callout = &tp->t_timers->tt_keep;
 857                         break;
 858                 case TT_2MSL:
 859                         t_callout = &tp->t_timers->tt_2msl;
 860                         break;
 861                 default:
 862                         panic("tp %p bad timer_type %#x", tp, timer_type);
 863                 }
 864         return callout_active(t_callout);
 865 }
 866
 867 void
 868 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 869 {
 870         struct callout *t_callout;
 871         timeout_t *f_callout;
 872         uint32_t f_reset;
 873
 874         tp->t_timers->tt_flags |= TT_STOPPED;
 875
 876         switch (timer_type) {
 877                 case TT_DELACK:
 878                         t_callout = &tp->t_timers->tt_delack;
 879                         f_callout = tcp_timer_delack_discard;
 880                         f_reset = TT_DELACK_RST;
 881                         break;
 882                 case TT_REXMT:
 883                         t_callout = &tp->t_timers->tt_rexmt;
 884                         f_callout = tcp_timer_rexmt_discard;
 885                         f_reset = TT_REXMT_RST;
 886                         break;
 887                 case TT_PERSIST:
 888                         t_callout = &tp->t_timers->tt_persist;
 889                         f_callout = tcp_timer_persist_discard;
 890                         f_reset = TT_PERSIST_RST;
 891                         break;
 892                 case TT_KEEP:
 893                         t_callout = &tp->t_timers->tt_keep;
 894                         f_callout = tcp_timer_keep_discard;
 895                         f_reset = TT_KEEP_RST;
 896                         break;
 897                 case TT_2MSL:
 898                         t_callout = &tp->t_timers->tt_2msl;
 899                         f_callout = tcp_timer_2msl_discard;
 900                         f_reset = TT_2MSL_RST;
 901                         break;
 902                 default:
 903                         panic("tp %p bad timer_type %#x", tp, timer_type);
 904                 }
 905
 906         if (tp->t_timers->tt_flags & timer_type) {
 907                 if (callout_stop(t_callout) &&
 908                     (tp->t_timers->tt_flags & f_reset)) {
 909                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 910                 } else {
 911                         /*
 912                          * Can't stop the callout, defer tcpcb actual deletion
 913                          * to the last tcp timer discard callout.
 914                          * The TT_STOPPED flag will ensure that no tcp timer
 915                          * callouts can be restarted on our behalf, and
 916                          * past this point currently running callouts waiting
 917                          * on inp lock will return right away after the
 918                          * classical check for callout reset/stop events:
 919                          * callout_pending() || !callout_active()
 920                          */
 921                         callout_reset(t_callout, 1, f_callout, tp);
 922                 }
 923         }
 924 }
 925
 926 #define ticks_to_msecs(t)       (1000*(t) / hz)
 927
 928 void
 929 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
 930     struct xtcp_timer *xtimer)
 931 {
 932         sbintime_t now;
 933
 934         bzero(xtimer, sizeof(*xtimer));
 935         if (timer == NULL)
 936                 return;
 937         now = getsbinuptime();
 938         if (callout_active(&timer->tt_delack))
 939                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
 940         if (callout_active(&timer->tt_rexmt))
 941                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
 942         if (callout_active(&timer->tt_persist))
 943                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
 944         if (callout_active(&timer->tt_keep))
 945                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
 946         if (callout_active(&timer->tt_2msl))
 947                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
 948         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
 949 }