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 #include "opt_rss.h"
  39
  40 #include <sys/param.h>
  41 #include <sys/kernel.h>
  42 #include <sys/lock.h>
  43 #include <sys/mbuf.h>
  44 #include <sys/mutex.h>
  45 #include <sys/protosw.h>
  46 #include <sys/smp.h>
  47 #include <sys/socket.h>
  48 #include <sys/socketvar.h>
  49 #include <sys/sysctl.h>
  50 #include <sys/systm.h>
  51
  52 #include <net/if.h>
  53 #include <net/route.h>
  54 #include <net/rss_config.h>
  55 #include <net/vnet.h>
  56 #include <net/netisr.h>
  57
  58 #include <netinet/in.h>
  59 #include <netinet/in_kdtrace.h>
  60 #include <netinet/in_pcb.h>
  61 #include <netinet/in_rss.h>
  62 #include <netinet/in_systm.h>
  63 #ifdef INET6
  64 #include <netinet6/in6_pcb.h>
  65 #endif
  66 #include <netinet/ip_var.h>
  67 #include <netinet/tcp.h>
  68 #include <netinet/tcp_fsm.h>
  69 #include <netinet/tcp_timer.h>
  70 #include <netinet/tcp_var.h>
  71 #include <netinet/cc/cc.h>
  72 #ifdef INET6
  73 #include <netinet6/tcp6_var.h>
  74 #endif
  75 #include <netinet/tcpip.h>
  76 #ifdef TCPDEBUG
  77 #include <netinet/tcp_debug.h>
  78 #endif
  79
  80 int    tcp_persmin;
  81 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
  82     &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
  83
  84 int    tcp_persmax;
  85 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
  86     &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
  87
  88 int     tcp_keepinit;
  89 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
  90     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
  91
  92 int     tcp_keepidle;
  93 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
  94     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
  95
  96 int     tcp_keepintvl;
  97 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
  98     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
  99
 100 int     tcp_delacktime;
 101 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
 102     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 103     "Time before a delayed ACK is sent");
 104
 105 int     tcp_msl;
 106 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 107     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 108
 109 int     tcp_rexmit_min;
 110 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 111     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
 112     "Minimum Retransmission Timeout");
 113
 114 int     tcp_rexmit_slop;
 115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 116     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 117     "Retransmission Timer Slop");
 118
 119 int     tcp_always_keepalive = 1;
 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 121     &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 122 __strong_reference(tcp_always_keepalive, always_keepalive);
 123
 124 int    tcp_fast_finwait2_recycle = 0;
 125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 126     &tcp_fast_finwait2_recycle, 0,
 127     "Recycle closed FIN_WAIT_2 connections faster");
 128
 129 int    tcp_finwait2_timeout;
 130 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 131     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 132
 133 int     tcp_keepcnt = TCPTV_KEEPCNT;
 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
 135     "Number of keepalive probes to send");
 136
 137         /* max idle probes */
 138 int     tcp_maxpersistidle;
 139
 140 static int      tcp_rexmit_drop_options = 0;
 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
 142     &tcp_rexmit_drop_options, 0,
 143     "Drop TCP options from 3rd and later retransmitted SYN");
 144
 145 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
 146 #define V_tcp_pmtud_blackhole_detect    VNET(tcp_pmtud_blackhole_detect)
 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
 148     CTLFLAG_RW|CTLFLAG_VNET,
 149     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
 150     "Path MTU Discovery Black Hole Detection Enabled");
 151
 152 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
 153 #define V_tcp_pmtud_blackhole_activated \
 154     VNET(tcp_pmtud_blackhole_activated)
 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
 156     CTLFLAG_RD|CTLFLAG_VNET,
 157     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
 158     "Path MTU Discovery Black Hole Detection, Activation Count");
 159
 160 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
 161 #define V_tcp_pmtud_blackhole_activated_min_mss \
 162     VNET(tcp_pmtud_blackhole_activated_min_mss)
 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
 164     CTLFLAG_RD|CTLFLAG_VNET,
 165     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
 166     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
 167
 168 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
 169 #define V_tcp_pmtud_blackhole_failed    VNET(tcp_pmtud_blackhole_failed)
 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
 171     CTLFLAG_RD|CTLFLAG_VNET,
 172     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
 173     "Path MTU Discovery Black Hole Detection, Failure Count");
 174
 175 #ifdef INET
 176 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
 177 #define V_tcp_pmtud_blackhole_mss       VNET(tcp_pmtud_blackhole_mss)
 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
 179     CTLFLAG_RW|CTLFLAG_VNET,
 180     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
 181     "Path MTU Discovery Black Hole Detection lowered MSS");
 182 #endif
 183
 184 #ifdef INET6
 185 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
 186 #define V_tcp_v6pmtud_blackhole_mss     VNET(tcp_v6pmtud_blackhole_mss)
 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
 188     CTLFLAG_RW|CTLFLAG_VNET,
 189     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
 190     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
 191 #endif
 192
 193 #ifdef  RSS
 194 static int      per_cpu_timers = 1;
 195 #else
 196 static int      per_cpu_timers = 0;
 197 #endif
 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
 199     &per_cpu_timers , 0, "run tcp timers on all cpus");
 200
 201 #if 0
 202 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
 203                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
 204 #endif
 205
 206 /*
 207  * Map the given inp to a CPU id.
 208  *
 209  * This queries RSS if it's compiled in, else it defaults to the current
 210  * CPU ID.
 211  */
 212 static inline int
 213 inp_to_cpuid(struct inpcb *inp)
 214 {
 215         u_int cpuid;
 216
 217 #ifdef  RSS
 218         if (per_cpu_timers) {
 219                 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
 220                 if (cpuid == NETISR_CPUID_NONE)
 221                         return (curcpu);        /* XXX */
 222                 else
 223                         return (cpuid);
 224         }
 225 #else
 226         /* Legacy, pre-RSS behaviour */
 227         if (per_cpu_timers) {
 228                 /*
 229                  * We don't have a flowid -> cpuid mapping, so cheat and
 230                  * just map unknown cpuids to curcpu.  Not the best, but
 231                  * apparently better than defaulting to swi 0.
 232                  */
 233                 cpuid = inp->inp_flowid % (mp_maxid + 1);
 234                 if (! CPU_ABSENT(cpuid))
 235                         return (cpuid);
 236                 return (curcpu);
 237         }
 238 #endif
 239         /* Default for RSS and non-RSS - cpuid 0 */
 240         else {
 241                 return (0);
 242         }
 243 }
 244
 245 /*
 246  * Tcp protocol timeout routine called every 500 ms.
 247  * Updates timestamps used for TCP
 248  * causes finite state machine actions if timers expire.
 249  */
 250 void
 251 tcp_slowtimo(void)
 252 {
 253         VNET_ITERATOR_DECL(vnet_iter);
 254
 255         VNET_LIST_RLOCK_NOSLEEP();
 256         VNET_FOREACH(vnet_iter) {
 257                 CURVNET_SET(vnet_iter);
 258                 (void) tcp_tw_2msl_scan(0);
 259                 CURVNET_RESTORE();
 260         }
 261         VNET_LIST_RUNLOCK_NOSLEEP();
 262 }
 263
 264 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 265     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 266
 267 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 268     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 269
 270 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 271
 272 /*
 273  * TCP timer processing.
 274  */
 275
 276 void
 277 tcp_timer_delack(void *xtp)
 278 {
 279         struct tcpcb *tp = xtp;
 280         struct inpcb *inp;
 281         CURVNET_SET(tp->t_vnet);
 282
 283         inp = tp->t_inpcb;
 284         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 285         INP_WLOCK(inp);
 286         if (callout_pending(&tp->t_timers->tt_delack) ||
 287             !callout_active(&tp->t_timers->tt_delack)) {
 288                 INP_WUNLOCK(inp);
 289                 CURVNET_RESTORE();
 290                 return;
 291         }
 292         callout_deactivate(&tp->t_timers->tt_delack);
 293         if ((inp->inp_flags & INP_DROPPED) != 0) {
 294                 INP_WUNLOCK(inp);
 295                 CURVNET_RESTORE();
 296                 return;
 297         }
 298         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 299                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 300         KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
 301                 ("%s: tp %p delack callout should be running", __func__, tp));
 302
 303         tp->t_flags |= TF_ACKNOW;
 304         TCPSTAT_INC(tcps_delack);
 305         (void) tp->t_fb->tfb_tcp_output(tp);
 306         INP_WUNLOCK(inp);
 307         CURVNET_RESTORE();
 308 }
 309
 310 void
 311 tcp_timer_2msl(void *xtp)
 312 {
 313         struct tcpcb *tp = xtp;
 314         struct inpcb *inp;
 315         CURVNET_SET(tp->t_vnet);
 316 #ifdef TCPDEBUG
 317         int ostate;
 318
 319         ostate = tp->t_state;
 320 #endif
 321         INP_INFO_RLOCK(&V_tcbinfo);
 322         inp = tp->t_inpcb;
 323         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 324         INP_WLOCK(inp);
 325         tcp_free_sackholes(tp);
 326         if (callout_pending(&tp->t_timers->tt_2msl) ||
 327             !callout_active(&tp->t_timers->tt_2msl)) {
 328                 INP_WUNLOCK(tp->t_inpcb);
 329                 INP_INFO_RUNLOCK(&V_tcbinfo);
 330                 CURVNET_RESTORE();
 331                 return;
 332         }
 333         callout_deactivate(&tp->t_timers->tt_2msl);
 334         if ((inp->inp_flags & INP_DROPPED) != 0) {
 335                 INP_WUNLOCK(inp);
 336                 INP_INFO_RUNLOCK(&V_tcbinfo);
 337                 CURVNET_RESTORE();
 338                 return;
 339         }
 340         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 341                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 342         KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
 343                 ("%s: tp %p 2msl callout should be running", __func__, tp));
 344         /*
 345          * 2 MSL timeout in shutdown went off.  If we're closed but
 346          * still waiting for peer to close and connection has been idle
 347          * too long delete connection control block.  Otherwise, check
 348          * again in a bit.
 349          *
 350          * If in TIME_WAIT state just ignore as this timeout is handled in
 351          * tcp_tw_2msl_scan().
 352          *
 353          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 354          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 355          * Ignore fact that there were recent incoming segments.
 356          */
 357         if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
 358                 INP_WUNLOCK(inp);
 359                 INP_INFO_RUNLOCK(&V_tcbinfo);
 360                 CURVNET_RESTORE();
 361                 return;
 362         }
 363         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 364             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 365             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 366                 TCPSTAT_INC(tcps_finwait2_drops);
 367                 tp = tcp_close(tp);
 368         } else {
 369                 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 370                         if (!callout_reset(&tp->t_timers->tt_2msl,
 371                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
 372                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
 373                         }
 374                 } else
 375                        tp = tcp_close(tp);
 376        }
 377
 378 #ifdef TCPDEBUG
 379         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 380                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 381                           PRU_SLOWTIMO);
 382 #endif
 383         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 384
 385         if (tp != NULL)
 386                 INP_WUNLOCK(inp);
 387         INP_INFO_RUNLOCK(&V_tcbinfo);
 388         CURVNET_RESTORE();
 389 }
 390
 391 void
 392 tcp_timer_keep(void *xtp)
 393 {
 394         struct tcpcb *tp = xtp;
 395         struct tcptemp *t_template;
 396         struct inpcb *inp;
 397         CURVNET_SET(tp->t_vnet);
 398 #ifdef TCPDEBUG
 399         int ostate;
 400
 401         ostate = tp->t_state;
 402 #endif
 403         INP_INFO_RLOCK(&V_tcbinfo);
 404         inp = tp->t_inpcb;
 405         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 406         INP_WLOCK(inp);
 407         if (callout_pending(&tp->t_timers->tt_keep) ||
 408             !callout_active(&tp->t_timers->tt_keep)) {
 409                 INP_WUNLOCK(inp);
 410                 INP_INFO_RUNLOCK(&V_tcbinfo);
 411                 CURVNET_RESTORE();
 412                 return;
 413         }
 414         callout_deactivate(&tp->t_timers->tt_keep);
 415         if ((inp->inp_flags & INP_DROPPED) != 0) {
 416                 INP_WUNLOCK(inp);
 417                 INP_INFO_RUNLOCK(&V_tcbinfo);
 418                 CURVNET_RESTORE();
 419                 return;
 420         }
 421         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 422                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 423         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
 424                 ("%s: tp %p keep callout should be running", __func__, tp));
 425         /*
 426          * Keep-alive timer went off; send something
 427          * or drop connection if idle for too long.
 428          */
 429         TCPSTAT_INC(tcps_keeptimeo);
 430         if (tp->t_state < TCPS_ESTABLISHED)
 431                 goto dropit;
 432         if ((tcp_always_keepalive ||
 433             inp->inp_socket->so_options & SO_KEEPALIVE) &&
 434             tp->t_state <= TCPS_CLOSING) {
 435                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 436                         goto dropit;
 437                 /*
 438                  * Send a packet designed to force a response
 439                  * if the peer is up and reachable:
 440                  * either an ACK if the connection is still alive,
 441                  * or an RST if the peer has closed the connection
 442                  * due to timeout or reboot.
 443                  * Using sequence number tp->snd_una-1
 444                  * causes the transmitted zero-length segment
 445                  * to lie outside the receive window;
 446                  * by the protocol spec, this requires the
 447                  * correspondent TCP to respond.
 448                  */
 449                 TCPSTAT_INC(tcps_keepprobe);
 450                 t_template = tcpip_maketemplate(inp);
 451                 if (t_template) {
 452                         tcp_respond(tp, t_template->tt_ipgen,
 453                                     &t_template->tt_t, (struct mbuf *)NULL,
 454                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 455                         free(t_template, M_TEMP);
 456                 }
 457                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 458                     tcp_timer_keep, tp)) {
 459                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 460                 }
 461         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 462                     tcp_timer_keep, tp)) {
 463                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
 464                 }
 465
 466 #ifdef TCPDEBUG
 467         if (inp->inp_socket->so_options & SO_DEBUG)
 468                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 469                           PRU_SLOWTIMO);
 470 #endif
 471         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 472         INP_WUNLOCK(inp);
 473         INP_INFO_RUNLOCK(&V_tcbinfo);
 474         CURVNET_RESTORE();
 475         return;
 476
 477 dropit:
 478         TCPSTAT_INC(tcps_keepdrops);
 479         tp = tcp_drop(tp, ETIMEDOUT);
 480
 481 #ifdef TCPDEBUG
 482         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 483                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 484                           PRU_SLOWTIMO);
 485 #endif
 486         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 487         if (tp != NULL)
 488                 INP_WUNLOCK(tp->t_inpcb);
 489         INP_INFO_RUNLOCK(&V_tcbinfo);
 490         CURVNET_RESTORE();
 491 }
 492
 493 void
 494 tcp_timer_persist(void *xtp)
 495 {
 496         struct tcpcb *tp = xtp;
 497         struct inpcb *inp;
 498         CURVNET_SET(tp->t_vnet);
 499 #ifdef TCPDEBUG
 500         int ostate;
 501
 502         ostate = tp->t_state;
 503 #endif
 504         INP_INFO_RLOCK(&V_tcbinfo);
 505         inp = tp->t_inpcb;
 506         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 507         INP_WLOCK(inp);
 508         if (callout_pending(&tp->t_timers->tt_persist) ||
 509             !callout_active(&tp->t_timers->tt_persist)) {
 510                 INP_WUNLOCK(inp);
 511                 INP_INFO_RUNLOCK(&V_tcbinfo);
 512                 CURVNET_RESTORE();
 513                 return;
 514         }
 515         callout_deactivate(&tp->t_timers->tt_persist);
 516         if ((inp->inp_flags & INP_DROPPED) != 0) {
 517                 INP_WUNLOCK(inp);
 518                 INP_INFO_RUNLOCK(&V_tcbinfo);
 519                 CURVNET_RESTORE();
 520                 return;
 521         }
 522         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 523                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 524         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
 525                 ("%s: tp %p persist callout should be running", __func__, tp));
 526         /*
 527          * Persistence timer into zero window.
 528          * Force a byte to be output, if possible.
 529          */
 530         TCPSTAT_INC(tcps_persisttimeo);
 531         /*
 532          * Hack: if the peer is dead/unreachable, we do not
 533          * time out if the window is closed.  After a full
 534          * backoff, drop the connection if the idle time
 535          * (no responses to probes) reaches the maximum
 536          * backoff that we would use if retransmitting.
 537          */
 538         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 539             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 540              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 541                 TCPSTAT_INC(tcps_persistdrop);
 542                 tp = tcp_drop(tp, ETIMEDOUT);
 543                 goto out;
 544         }
 545         /*
 546          * If the user has closed the socket then drop a persisting
 547          * connection after a much reduced timeout.
 548          */
 549         if (tp->t_state > TCPS_CLOSE_WAIT &&
 550             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 551                 TCPSTAT_INC(tcps_persistdrop);
 552                 tp = tcp_drop(tp, ETIMEDOUT);
 553                 goto out;
 554         }
 555         tcp_setpersist(tp);
 556         tp->t_flags |= TF_FORCEDATA;
 557         (void) tp->t_fb->tfb_tcp_output(tp);
 558         tp->t_flags &= ~TF_FORCEDATA;
 559
 560 out:
 561 #ifdef TCPDEBUG
 562         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 563                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 564 #endif
 565         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 566         if (tp != NULL)
 567                 INP_WUNLOCK(inp);
 568         INP_INFO_RUNLOCK(&V_tcbinfo);
 569         CURVNET_RESTORE();
 570 }
 571
 572 void
 573 tcp_timer_rexmt(void * xtp)
 574 {
 575         struct tcpcb *tp = xtp;
 576         CURVNET_SET(tp->t_vnet);
 577         int rexmt;
 578         int headlocked;
 579         struct inpcb *inp;
 580 #ifdef TCPDEBUG
 581         int ostate;
 582
 583         ostate = tp->t_state;
 584 #endif
 585
 586         INP_INFO_RLOCK(&V_tcbinfo);
 587         inp = tp->t_inpcb;
 588         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 589         INP_WLOCK(inp);
 590         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 591             !callout_active(&tp->t_timers->tt_rexmt)) {
 592                 INP_WUNLOCK(inp);
 593                 INP_INFO_RUNLOCK(&V_tcbinfo);
 594                 CURVNET_RESTORE();
 595                 return;
 596         }
 597         callout_deactivate(&tp->t_timers->tt_rexmt);
 598         if ((inp->inp_flags & INP_DROPPED) != 0) {
 599                 INP_WUNLOCK(inp);
 600                 INP_INFO_RUNLOCK(&V_tcbinfo);
 601                 CURVNET_RESTORE();
 602                 return;
 603         }
 604         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 605                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 606         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
 607                 ("%s: tp %p rexmt callout should be running", __func__, tp));
 608         tcp_free_sackholes(tp);
 609         if (tp->t_fb->tfb_tcp_rexmit_tmr) {
 610                 /* The stack has a timer action too. */
 611                 (*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
 612         }
 613         /*
 614          * Retransmission timer went off.  Message has not
 615          * been acked within retransmit interval.  Back off
 616          * to a longer retransmit interval and retransmit one segment.
 617          */
 618         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 619                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 620                 TCPSTAT_INC(tcps_timeoutdrop);
 621
 622                 tp = tcp_drop(tp, ETIMEDOUT);
 623                 headlocked = 1;
 624                 goto out;
 625         }
 626         INP_INFO_RUNLOCK(&V_tcbinfo);
 627         headlocked = 0;
 628         if (tp->t_state == TCPS_SYN_SENT) {
 629                 /*
 630                  * If the SYN was retransmitted, indicate CWND to be
 631                  * limited to 1 segment in cc_conn_init().
 632                  */
 633                 tp->snd_cwnd = 1;
 634         } else if (tp->t_rxtshift == 1) {
 635                 /*
 636                  * first retransmit; record ssthresh and cwnd so they can
 637                  * be recovered if this turns out to be a "bad" retransmit.
 638                  * A retransmit is considered "bad" if an ACK for this
 639                  * segment is received within RTT/2 interval; the assumption
 640                  * here is that the ACK was already in flight.  See
 641                  * "On Estimating End-to-End Network Path Properties" by
 642                  * Allman and Paxson for more details.
 643                  */
 644                 tp->snd_cwnd_prev = tp->snd_cwnd;
 645                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 646                 tp->snd_recover_prev = tp->snd_recover;
 647                 if (IN_FASTRECOVERY(tp->t_flags))
 648                         tp->t_flags |= TF_WASFRECOVERY;
 649                 else
 650                         tp->t_flags &= ~TF_WASFRECOVERY;
 651                 if (IN_CONGRECOVERY(tp->t_flags))
 652                         tp->t_flags |= TF_WASCRECOVERY;
 653                 else
 654                         tp->t_flags &= ~TF_WASCRECOVERY;
 655                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 656                 tp->t_flags |= TF_PREVVALID;
 657         } else
 658                 tp->t_flags &= ~TF_PREVVALID;
 659         TCPSTAT_INC(tcps_rexmttimeo);
 660         if ((tp->t_state == TCPS_SYN_SENT) ||
 661             (tp->t_state == TCPS_SYN_RECEIVED))
 662                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 663         else
 664                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 665         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 666                       tp->t_rttmin, TCPTV_REXMTMAX);
 667
 668         /*
 669          * We enter the path for PLMTUD if connection is established or, if
 670          * connection is FIN_WAIT_1 status, reason for the last is that if
 671          * amount of data we send is very small, we could send it in couple of
 672          * packets and process straight to FIN. In that case we won't catch
 673          * ESTABLISHED state.
 674          */
 675         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 676             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 677 #ifdef INET6
 678                 int isipv6;
 679 #endif
 680
 681                 /*
 682                  * Idea here is that at each stage of mtu probe (usually, 1448
 683                  * -> 1188 -> 524) should be given 2 chances to recover before
 684                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 685                  *  take care of that.
 686                  */
 687                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 688                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 689                     (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
 690                     tp->t_rxtshift % 2 == 0)) {
 691                         /*
 692                          * Enter Path MTU Black-hole Detection mechanism:
 693                          * - Disable Path MTU Discovery (IP "DF" bit).
 694                          * - Reduce MTU to lower value than what we
 695                          *   negotiated with peer.
 696                          */
 697                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
 698                                 /* Record that we may have found a black hole. */
 699                                 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 700                                 /* Keep track of previous MSS. */
 701                                 tp->t_pmtud_saved_maxseg = tp->t_maxseg;
 702                         }
 703
 704                         /*
 705                          * Reduce the MSS to blackhole value or to the default
 706                          * in an attempt to retransmit.
 707                          */
 708 #ifdef INET6
 709                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 710                         if (isipv6 &&
 711                             tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
 712                                 /* Use the sysctl tuneable blackhole MSS. */
 713                                 tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
 714                                 V_tcp_pmtud_blackhole_activated++;
 715                         } else if (isipv6) {
 716                                 /* Use the default MSS. */
 717                                 tp->t_maxseg = V_tcp_v6mssdflt;
 718                                 /*
 719                                  * Disable Path MTU Discovery when we switch to
 720                                  * minmss.
 721                                  */
 722                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 723                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 724                         }
 725 #endif
 726 #if defined(INET6) && defined(INET)
 727                         else
 728 #endif
 729 #ifdef INET
 730                         if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
 731                                 /* Use the sysctl tuneable blackhole MSS. */
 732                                 tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
 733                                 V_tcp_pmtud_blackhole_activated++;
 734                         } else {
 735                                 /* Use the default MSS. */
 736                                 tp->t_maxseg = V_tcp_mssdflt;
 737                                 /*
 738                                  * Disable Path MTU Discovery when we switch to
 739                                  * minmss.
 740                                  */
 741                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 742                                 V_tcp_pmtud_blackhole_activated_min_mss++;
 743                         }
 744 #endif
 745                         /*
 746                          * Reset the slow-start flight size
 747                          * as it may depend on the new MSS.
 748                          */
 749                         if (CC_ALGO(tp)->conn_init != NULL)
 750                                 CC_ALGO(tp)->conn_init(tp->ccv);
 751                 } else {
 752                         /*
 753                          * If further retransmissions are still unsuccessful
 754                          * with a lowered MTU, maybe this isn't a blackhole and
 755                          * we restore the previous MSS and blackhole detection
 756                          * flags.
 757                          * The limit '6' is determined by giving each probe
 758                          * stage (1448, 1188, 524) 2 chances to recover.
 759                          */
 760                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 761                             (tp->t_rxtshift >= 6)) {
 762                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 763                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 764                                 tp->t_maxseg = tp->t_pmtud_saved_maxseg;
 765                                 V_tcp_pmtud_blackhole_failed++;
 766                                 /*
 767                                  * Reset the slow-start flight size as it
 768                                  * may depend on the new MSS.
 769                                  */
 770                                 if (CC_ALGO(tp)->conn_init != NULL)
 771                                         CC_ALGO(tp)->conn_init(tp->ccv);
 772                         }
 773                 }
 774         }
 775
 776         /*
 777          * Disable RFC1323 and SACK if we haven't got any response to
 778          * our third SYN to work-around some broken terminal servers
 779          * (most of which have hopefully been retired) that have bad VJ
 780          * header compression code which trashes TCP segments containing
 781          * unknown-to-them TCP options.
 782          */
 783         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 784             (tp->t_rxtshift == 3))
 785                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 786         /*
 787          * If we backed off this far, our srtt estimate is probably bogus.
 788          * Clobber it so we'll take the next rtt measurement as our srtt;
 789          * move the current srtt into rttvar to keep the current
 790          * retransmit times until then.
 791          */
 792         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 793 #ifdef INET6
 794                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 795                         in6_losing(tp->t_inpcb);
 796                 else
 797 #endif
 798                         in_losing(tp->t_inpcb);
 799                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 800                 tp->t_srtt = 0;
 801         }
 802         tp->snd_nxt = tp->snd_una;
 803         tp->snd_recover = tp->snd_max;
 804         /*
 805          * Force a segment to be sent.
 806          */
 807         tp->t_flags |= TF_ACKNOW;
 808         /*
 809          * If timing a segment in this window, stop the timer.
 810          */
 811         tp->t_rtttime = 0;
 812
 813         cc_cong_signal(tp, NULL, CC_RTO);
 814
 815         (void) tp->t_fb->tfb_tcp_output(tp);
 816
 817 out:
 818 #ifdef TCPDEBUG
 819         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 820                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 821                           PRU_SLOWTIMO);
 822 #endif
 823         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 824         if (tp != NULL)
 825                 INP_WUNLOCK(inp);
 826         if (headlocked)
 827                 INP_INFO_RUNLOCK(&V_tcbinfo);
 828         CURVNET_RESTORE();
 829 }
 830
 831 void
 832 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 833 {
 834         struct callout *t_callout;
 835         timeout_t *f_callout;
 836         struct inpcb *inp = tp->t_inpcb;
 837         int cpu = inp_to_cpuid(inp);
 838         uint32_t f_reset;
 839
 840 #ifdef TCP_OFFLOAD
 841         if (tp->t_flags & TF_TOE)
 842                 return;
 843 #endif
 844
 845         if (tp->t_timers->tt_flags & TT_STOPPED)
 846                 return;
 847
 848         switch (timer_type) {
 849                 case TT_DELACK:
 850                         t_callout = &tp->t_timers->tt_delack;
 851                         f_callout = tcp_timer_delack;
 852                         f_reset = TT_DELACK_RST;
 853                         break;
 854                 case TT_REXMT:
 855                         t_callout = &tp->t_timers->tt_rexmt;
 856                         f_callout = tcp_timer_rexmt;
 857                         f_reset = TT_REXMT_RST;
 858                         break;
 859                 case TT_PERSIST:
 860                         t_callout = &tp->t_timers->tt_persist;
 861                         f_callout = tcp_timer_persist;
 862                         f_reset = TT_PERSIST_RST;
 863                         break;
 864                 case TT_KEEP:
 865                         t_callout = &tp->t_timers->tt_keep;
 866                         f_callout = tcp_timer_keep;
 867                         f_reset = TT_KEEP_RST;
 868                         break;
 869                 case TT_2MSL:
 870                         t_callout = &tp->t_timers->tt_2msl;
 871                         f_callout = tcp_timer_2msl;
 872                         f_reset = TT_2MSL_RST;
 873                         break;
 874                 default:
 875                         if (tp->t_fb->tfb_tcp_timer_activate) {
 876                                 tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
 877                                 return;
 878                         }
 879                         panic("tp %p bad timer_type %#x", tp, timer_type);
 880                 }
 881         if (delta == 0) {
 882                 if ((tp->t_timers->tt_flags & timer_type) &&
 883                     (callout_stop(t_callout) > 0) &&
 884                     (tp->t_timers->tt_flags & f_reset)) {
 885                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
 886                 }
 887         } else {
 888                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
 889                         tp->t_timers->tt_flags |= (timer_type | f_reset);
 890                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 891                 } else {
 892                         /* Reset already running callout on the same CPU. */
 893                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
 894                                 /*
 895                                  * Callout not cancelled, consider it as not
 896                                  * properly restarted. */
 897                                 tp->t_timers->tt_flags &= ~f_reset;
 898                         }
 899                 }
 900         }
 901 }
 902
 903 int
 904 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 905 {
 906         struct callout *t_callout;
 907
 908         switch (timer_type) {
 909                 case TT_DELACK:
 910                         t_callout = &tp->t_timers->tt_delack;
 911                         break;
 912                 case TT_REXMT:
 913                         t_callout = &tp->t_timers->tt_rexmt;
 914                         break;
 915                 case TT_PERSIST:
 916                         t_callout = &tp->t_timers->tt_persist;
 917                         break;
 918                 case TT_KEEP:
 919                         t_callout = &tp->t_timers->tt_keep;
 920                         break;
 921                 case TT_2MSL:
 922                         t_callout = &tp->t_timers->tt_2msl;
 923                         break;
 924                 default:
 925                         if (tp->t_fb->tfb_tcp_timer_active) {
 926                                 return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
 927                         }
 928                         panic("tp %p bad timer_type %#x", tp, timer_type);
 929                 }
 930         return callout_active(t_callout);
 931 }
 932
 933 void
 934 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 935 {
 936         struct callout *t_callout;
 937         uint32_t f_reset;
 938
 939         tp->t_timers->tt_flags |= TT_STOPPED;
 940
 941         switch (timer_type) {
 942                 case TT_DELACK:
 943                         t_callout = &tp->t_timers->tt_delack;
 944                         f_reset = TT_DELACK_RST;
 945                         break;
 946                 case TT_REXMT:
 947                         t_callout = &tp->t_timers->tt_rexmt;
 948                         f_reset = TT_REXMT_RST;
 949                         break;
 950                 case TT_PERSIST:
 951                         t_callout = &tp->t_timers->tt_persist;
 952                         f_reset = TT_PERSIST_RST;
 953                         break;
 954                 case TT_KEEP:
 955                         t_callout = &tp->t_timers->tt_keep;
 956                         f_reset = TT_KEEP_RST;
 957                         break;
 958                 case TT_2MSL:
 959                         t_callout = &tp->t_timers->tt_2msl;
 960                         f_reset = TT_2MSL_RST;
 961                         break;
 962                 default:
 963                         if (tp->t_fb->tfb_tcp_timer_stop) {
 964                                 /*
 965                                  * XXXrrs we need to look at this with the
 966                                  * stop case below (flags).
 967                                  */
 968                                 tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
 969                                 return;
 970                         }
 971                         panic("tp %p bad timer_type %#x", tp, timer_type);
 972                 }
 973
 974         if (tp->t_timers->tt_flags & timer_type) {
 975                 if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
 976                         /*
 977                          * Can't stop the callout, defer tcpcb actual deletion
 978                          * to the last one. We do this using the async drain
 979                          * function and incrementing the count in
 980                          */
 981                         tp->t_timers->tt_draincnt++;
 982                 }
 983         }
 984 }
 985
 986 #define ticks_to_msecs(t)       (1000*(t) / hz)
 987
 988 void
 989 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
 990     struct xtcp_timer *xtimer)
 991 {
 992         sbintime_t now;
 993
 994         bzero(xtimer, sizeof(*xtimer));
 995         if (timer == NULL)
 996                 return;
 997         now = getsbinuptime();
 998         if (callout_active(&timer->tt_delack))
 999                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
1000         if (callout_active(&timer->tt_rexmt))
1001                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
1002         if (callout_active(&timer->tt_persist))
1003                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
1004         if (callout_active(&timer->tt_keep))
1005                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
1006         if (callout_active(&timer->tt_2msl))
1007                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
1008         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
1009 }