sys/netinet/tcp_timer.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include "opt_inet.h"
  38 #include "opt_inet6.h"
  39 #include "opt_tcpdebug.h"
  40 #include "opt_rss.h"
  41
  42 #include <sys/param.h>
  43 #include <sys/kernel.h>
  44 #include <sys/lock.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/mutex.h>
  47 #include <sys/protosw.h>
  48 #include <sys/smp.h>
  49 #include <sys/socket.h>
  50 #include <sys/socketvar.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/systm.h>
  53
  54 #include <net/if.h>
  55 #include <net/route.h>
  56 #include <net/rss_config.h>
  57 #include <net/vnet.h>
  58 #include <net/netisr.h>
  59
  60 #include <netinet/in.h>
  61 #include <netinet/in_kdtrace.h>
  62 #include <netinet/in_pcb.h>
  63 #include <netinet/in_rss.h>
  64 #include <netinet/in_systm.h>
  65 #ifdef INET6
  66 #include <netinet6/in6_pcb.h>
  67 #endif
  68 #include <netinet/ip_var.h>
  69 #include <netinet/tcp.h>
  70 #include <netinet/tcp_fsm.h>
  71 #include <netinet/tcp_timer.h>
  72 #include <netinet/tcp_var.h>
  73 #include <netinet/cc/cc.h>
  74 #ifdef INET6
  75 #include <netinet6/tcp6_var.h>
  76 #endif
  77 #include <netinet/tcpip.h>
  78 #ifdef TCPDEBUG
  79 #include <netinet/tcp_debug.h>
  80 #endif
  81
  82 int    tcp_persmin;
  83 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
  84     &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
  85
  86 int    tcp_persmax;
  87 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
  88     &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
  89
  90 int     tcp_keepinit;
  91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
  92     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
  93
  94 int     tcp_keepidle;
  95 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
  96     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
  97
  98 int     tcp_keepintvl;
  99 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
 100     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
 101
 102 int     tcp_delacktime;
 103 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
 104     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
 105     "Time before a delayed ACK is sent");
 106
 107 int     tcp_msl;
 108 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
 109     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 110
 111 int     tcp_rexmit_min;
 112 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
 113     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
 114     "Minimum Retransmission Timeout");
 115
 116 int     tcp_rexmit_slop;
 117 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
 118     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
 119     "Retransmission Timer Slop");
 120
 121 static int      always_keepalive = 1;
 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
 123     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 124
 125 int    tcp_fast_finwait2_recycle = 0;
 126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
 127     &tcp_fast_finwait2_recycle, 0,
 128     "Recycle closed FIN_WAIT_2 connections faster");
 129
 130 int    tcp_finwait2_timeout;
 131 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
 132     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
 133
 134 int     tcp_keepcnt = TCPTV_KEEPCNT;
 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
 136     "Number of keepalive probes to send");
 137
 138         /* max idle probes */
 139 int     tcp_maxpersistidle;
 140
 141 static int      tcp_rexmit_drop_options = 0;
 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
 143     &tcp_rexmit_drop_options, 0,
 144     "Drop TCP options from 3rd and later retransmitted SYN");
 145
 146 VNET_DEFINE(int, 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 #ifdef INET
 153 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
 155     CTLFLAG_RW|CTLFLAG_VNET,
 156     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
 157     "Path MTU Discovery Black Hole Detection lowered MSS");
 158 #endif
 159
 160 #ifdef INET6
 161 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
 163     CTLFLAG_RW|CTLFLAG_VNET,
 164     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
 165     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
 166 #endif
 167
 168 #ifdef  RSS
 169 static int      per_cpu_timers = 1;
 170 #else
 171 static int      per_cpu_timers = 0;
 172 #endif
 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
 174     &per_cpu_timers , 0, "run tcp timers on all cpus");
 175
 176 #if 0
 177 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
 178                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
 179 #endif
 180
 181 /*
 182  * Map the given inp to a CPU id.
 183  *
 184  * This queries RSS if it's compiled in, else it defaults to the current
 185  * CPU ID.
 186  */
 187 static inline int
 188 inp_to_cpuid(struct inpcb *inp)
 189 {
 190         u_int cpuid;
 191
 192 #ifdef  RSS
 193         if (per_cpu_timers) {
 194                 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
 195                 if (cpuid == NETISR_CPUID_NONE)
 196                         return (curcpu);        /* XXX */
 197                 else
 198                         return (cpuid);
 199         }
 200 #else
 201         /* Legacy, pre-RSS behaviour */
 202         if (per_cpu_timers) {
 203                 /*
 204                  * We don't have a flowid -> cpuid mapping, so cheat and
 205                  * just map unknown cpuids to curcpu.  Not the best, but
 206                  * apparently better than defaulting to swi 0.
 207                  */
 208                 cpuid = inp->inp_flowid % (mp_maxid + 1);
 209                 if (! CPU_ABSENT(cpuid))
 210                         return (cpuid);
 211                 return (curcpu);
 212         }
 213 #endif
 214         /* Default for RSS and non-RSS - cpuid 0 */
 215         else {
 216                 return (0);
 217         }
 218 }
 219
 220 /*
 221  * Tcp protocol timeout routine called every 500 ms.
 222  * Updates timestamps used for TCP
 223  * causes finite state machine actions if timers expire.
 224  */
 225 void
 226 tcp_slowtimo(void)
 227 {
 228         VNET_ITERATOR_DECL(vnet_iter);
 229
 230         VNET_LIST_RLOCK_NOSLEEP();
 231         VNET_FOREACH(vnet_iter) {
 232                 CURVNET_SET(vnet_iter);
 233                 (void) tcp_tw_2msl_scan(0);
 234                 CURVNET_RESTORE();
 235         }
 236         VNET_LIST_RUNLOCK_NOSLEEP();
 237 }
 238
 239 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
 240     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
 241
 242 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
 243     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
 244
 245 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
 246
 247 /*
 248  * TCP timer processing.
 249  */
 250
 251 void
 252 tcp_timer_delack(void *xtp)
 253 {
 254         struct tcpcb *tp = xtp;
 255         struct inpcb *inp;
 256         CURVNET_SET(tp->t_vnet);
 257
 258         inp = tp->t_inpcb;
 259         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 260         INP_WLOCK(inp);
 261         if (callout_pending(&tp->t_timers->tt_delack) ||
 262             !callout_active(&tp->t_timers->tt_delack)) {
 263                 INP_WUNLOCK(inp);
 264                 CURVNET_RESTORE();
 265                 return;
 266         }
 267         callout_deactivate(&tp->t_timers->tt_delack);
 268         if ((inp->inp_flags & INP_DROPPED) != 0) {
 269                 INP_WUNLOCK(inp);
 270                 CURVNET_RESTORE();
 271                 return;
 272         }
 273         tp->t_flags |= TF_ACKNOW;
 274         TCPSTAT_INC(tcps_delack);
 275         (void) tp->t_fb->tfb_tcp_output(tp);
 276         INP_WUNLOCK(inp);
 277         CURVNET_RESTORE();
 278 }
 279
 280 /*
 281  * When a timer wants to remove a TCB it must
 282  * hold the INP_INFO_RLOCK(). The timer function
 283  * should only have grabbed the INP_WLOCK() when
 284  * it entered. To safely switch to holding both the
 285  * INP_INFO_RLOCK() and the INP_WLOCK() we must first
 286  * grab a reference on the inp, which will hold the inp
 287  * so that it can't be removed. We then unlock the INP_WLOCK(),
 288  * and grab the INP_INFO_RLOCK() lock. Once we have the INP_INFO_RLOCK()
 289  * we proceed again to get the INP_WLOCK() (this preserves proper
 290  * lock order). After acquiring the INP_WLOCK we must check if someone
 291  * else deleted the pcb i.e. the inp_flags check.
 292  * If so we return 1 otherwise we return 0.
 293  *
 294  * No matter what the tcp_inpinfo_lock_add() function
 295  * returns the caller must afterwards call tcp_inpinfo_lock_del()
 296  * to drop the locks and reference properly.
 297  */
 298
 299 int
 300 tcp_inpinfo_lock_add(struct inpcb *inp)
 301 {
 302         in_pcbref(inp);
 303         INP_WUNLOCK(inp);
 304         INP_INFO_RLOCK(&V_tcbinfo);
 305         INP_WLOCK(inp);
 306         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 307                 return(1);
 308         }
 309         return(0);
 310
 311 }
 312
 313 void
 314 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
 315 {
 316         INP_INFO_RUNLOCK(&V_tcbinfo);
 317         if (inp && (tp == NULL)) {
 318                 /*
 319                  * If tcp_close/drop() gets called and tp
 320                  * returns NULL, then the function dropped
 321                  * the inp lock, we hold a reference keeping
 322                  * this around, so we must re-aquire the
 323                  * INP_WLOCK() in order to proceed with
 324                  * our dropping the inp reference.
 325                  */
 326                 INP_WLOCK(inp);
 327         }
 328         if (inp && in_pcbrele_wlocked(inp) == 0)
 329                 INP_WUNLOCK(inp);
 330 }
 331
 332 void
 333 tcp_timer_2msl(void *xtp)
 334 {
 335         struct tcpcb *tp = xtp;
 336         struct inpcb *inp;
 337         CURVNET_SET(tp->t_vnet);
 338 #ifdef TCPDEBUG
 339         int ostate;
 340
 341         ostate = tp->t_state;
 342 #endif
 343         inp = tp->t_inpcb;
 344         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 345         INP_WLOCK(inp);
 346         tcp_free_sackholes(tp);
 347         if (callout_pending(&tp->t_timers->tt_2msl) ||
 348             !callout_active(&tp->t_timers->tt_2msl)) {
 349                 INP_WUNLOCK(tp->t_inpcb);
 350                 CURVNET_RESTORE();
 351                 return;
 352         }
 353         callout_deactivate(&tp->t_timers->tt_2msl);
 354         if ((inp->inp_flags & INP_DROPPED) != 0) {
 355                 INP_WUNLOCK(inp);
 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         /*
 362          * 2 MSL timeout in shutdown went off.  If we're closed but
 363          * still waiting for peer to close and connection has been idle
 364          * too long delete connection control block.  Otherwise, check
 365          * again in a bit.
 366          *
 367          * If in TIME_WAIT state just ignore as this timeout is handled in
 368          * tcp_tw_2msl_scan().
 369          *
 370          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
 371          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
 372          * Ignore fact that there were recent incoming segments.
 373          */
 374         if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
 375                 INP_WUNLOCK(inp);
 376                 CURVNET_RESTORE();
 377                 return;
 378         }
 379         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
 380             tp->t_inpcb && tp->t_inpcb->inp_socket &&
 381             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
 382                 TCPSTAT_INC(tcps_finwait2_drops);
 383                 if (tcp_inpinfo_lock_add(inp)) {
 384                         tcp_inpinfo_lock_del(inp, tp);
 385                         goto out;
 386                 }
 387                 tp = tcp_close(tp);
 388                 tcp_inpinfo_lock_del(inp, tp);
 389                 goto out;
 390         } else {
 391                 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
 392                         callout_reset(&tp->t_timers->tt_2msl,
 393                                       TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
 394                 } else {
 395                         if (tcp_inpinfo_lock_add(inp)) {
 396                                 tcp_inpinfo_lock_del(inp, tp);
 397                                 goto out;
 398                         }
 399                         tp = tcp_close(tp);
 400                         tcp_inpinfo_lock_del(inp, tp);
 401                         goto out;
 402                 }
 403        }
 404
 405 #ifdef TCPDEBUG
 406         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 407                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 408                           PRU_SLOWTIMO);
 409 #endif
 410         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 411
 412         if (tp != NULL)
 413                 INP_WUNLOCK(inp);
 414 out:
 415         CURVNET_RESTORE();
 416 }
 417
 418 void
 419 tcp_timer_keep(void *xtp)
 420 {
 421         struct tcpcb *tp = xtp;
 422         struct tcptemp *t_template;
 423         struct inpcb *inp;
 424         CURVNET_SET(tp->t_vnet);
 425 #ifdef TCPDEBUG
 426         int ostate;
 427
 428         ostate = tp->t_state;
 429 #endif
 430         inp = tp->t_inpcb;
 431         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 432         INP_WLOCK(inp);
 433         if (callout_pending(&tp->t_timers->tt_keep) ||
 434             !callout_active(&tp->t_timers->tt_keep)) {
 435                 INP_WUNLOCK(inp);
 436                 CURVNET_RESTORE();
 437                 return;
 438         }
 439         callout_deactivate(&tp->t_timers->tt_keep);
 440         if ((inp->inp_flags & INP_DROPPED) != 0) {
 441                 INP_WUNLOCK(inp);
 442                 CURVNET_RESTORE();
 443                 return;
 444         }
 445         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 446                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 447
 448         /*
 449          * Because we don't regularly reset the keepalive callout in
 450          * the ESTABLISHED state, it may be that we don't actually need
 451          * to send a keepalive yet. If that occurs, schedule another
 452          * call for the next time the keepalive timer might expire.
 453          */
 454         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
 455                 u_int idletime;
 456
 457                 idletime = ticks - tp->t_rcvtime;
 458                 if (idletime < TP_KEEPIDLE(tp)) {
 459                         callout_reset(&tp->t_timers->tt_keep,
 460                             TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
 461                         INP_WUNLOCK(inp);
 462                         CURVNET_RESTORE();
 463                         return;
 464                 }
 465         }
 466
 467         /*
 468          * Keep-alive timer went off; send something
 469          * or drop connection if idle for too long.
 470          */
 471         TCPSTAT_INC(tcps_keeptimeo);
 472         if (tp->t_state < TCPS_ESTABLISHED)
 473                 goto dropit;
 474         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
 475             tp->t_state <= TCPS_CLOSING) {
 476                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
 477                         goto dropit;
 478                 /*
 479                  * Send a packet designed to force a response
 480                  * if the peer is up and reachable:
 481                  * either an ACK if the connection is still alive,
 482                  * or an RST if the peer has closed the connection
 483                  * due to timeout or reboot.
 484                  * Using sequence number tp->snd_una-1
 485                  * causes the transmitted zero-length segment
 486                  * to lie outside the receive window;
 487                  * by the protocol spec, this requires the
 488                  * correspondent TCP to respond.
 489                  */
 490                 TCPSTAT_INC(tcps_keepprobe);
 491                 t_template = tcpip_maketemplate(inp);
 492                 if (t_template) {
 493                         tcp_respond(tp, t_template->tt_ipgen,
 494                                     &t_template->tt_t, (struct mbuf *)NULL,
 495                                     tp->rcv_nxt, tp->snd_una - 1, 0);
 496                         free(t_template, M_TEMP);
 497                 }
 498                 callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
 499                               tcp_timer_keep, tp);
 500         } else
 501                 callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
 502                               tcp_timer_keep, tp);
 503
 504 #ifdef TCPDEBUG
 505         if (inp->inp_socket->so_options & SO_DEBUG)
 506                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 507                           PRU_SLOWTIMO);
 508 #endif
 509         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 510         INP_WUNLOCK(inp);
 511         CURVNET_RESTORE();
 512         return;
 513
 514 dropit:
 515         TCPSTAT_INC(tcps_keepdrops);
 516
 517         if (tcp_inpinfo_lock_add(inp)) {
 518                 tcp_inpinfo_lock_del(inp, tp);
 519                 goto out;
 520         }
 521         tp = tcp_drop(tp, ETIMEDOUT);
 522
 523 #ifdef TCPDEBUG
 524         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 525                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 526                           PRU_SLOWTIMO);
 527 #endif
 528         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 529         tcp_inpinfo_lock_del(inp, tp);
 530 out:
 531         CURVNET_RESTORE();
 532 }
 533
 534 void
 535 tcp_timer_persist(void *xtp)
 536 {
 537         struct tcpcb *tp = xtp;
 538         struct inpcb *inp;
 539         CURVNET_SET(tp->t_vnet);
 540 #ifdef TCPDEBUG
 541         int ostate;
 542
 543         ostate = tp->t_state;
 544 #endif
 545         inp = tp->t_inpcb;
 546         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 547         INP_WLOCK(inp);
 548         if (callout_pending(&tp->t_timers->tt_persist) ||
 549             !callout_active(&tp->t_timers->tt_persist)) {
 550                 INP_WUNLOCK(inp);
 551                 CURVNET_RESTORE();
 552                 return;
 553         }
 554         callout_deactivate(&tp->t_timers->tt_persist);
 555         if ((inp->inp_flags & INP_DROPPED) != 0) {
 556                 INP_WUNLOCK(inp);
 557                 CURVNET_RESTORE();
 558                 return;
 559         }
 560         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 561                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 562         /*
 563          * Persistence timer into zero window.
 564          * Force a byte to be output, if possible.
 565          */
 566         TCPSTAT_INC(tcps_persisttimeo);
 567         /*
 568          * Hack: if the peer is dead/unreachable, we do not
 569          * time out if the window is closed.  After a full
 570          * backoff, drop the connection if the idle time
 571          * (no responses to probes) reaches the maximum
 572          * backoff that we would use if retransmitting.
 573          */
 574         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
 575             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
 576              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
 577                 TCPSTAT_INC(tcps_persistdrop);
 578                 if (tcp_inpinfo_lock_add(inp)) {
 579                         tcp_inpinfo_lock_del(inp, tp);
 580                         goto out;
 581                 }
 582                 tp = tcp_drop(tp, ETIMEDOUT);
 583                 tcp_inpinfo_lock_del(inp, tp);
 584                 goto out;
 585         }
 586         /*
 587          * If the user has closed the socket then drop a persisting
 588          * connection after a much reduced timeout.
 589          */
 590         if (tp->t_state > TCPS_CLOSE_WAIT &&
 591             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
 592                 TCPSTAT_INC(tcps_persistdrop);
 593                 if (tcp_inpinfo_lock_add(inp)) {
 594                         tcp_inpinfo_lock_del(inp, tp);
 595                         goto out;
 596                 }
 597                 tp = tcp_drop(tp, ETIMEDOUT);
 598                 tcp_inpinfo_lock_del(inp, tp);
 599                 goto out;
 600         }
 601         tcp_setpersist(tp);
 602         tp->t_flags |= TF_FORCEDATA;
 603         (void) tp->t_fb->tfb_tcp_output(tp);
 604         tp->t_flags &= ~TF_FORCEDATA;
 605
 606 #ifdef TCPDEBUG
 607         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
 608                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
 609 #endif
 610         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 611         INP_WUNLOCK(inp);
 612 out:
 613         CURVNET_RESTORE();
 614 }
 615
 616 void
 617 tcp_timer_rexmt(void * xtp)
 618 {
 619         struct tcpcb *tp = xtp;
 620         CURVNET_SET(tp->t_vnet);
 621         int rexmt;
 622         struct inpcb *inp;
 623 #ifdef TCPDEBUG
 624         int ostate;
 625
 626         ostate = tp->t_state;
 627 #endif
 628         inp = tp->t_inpcb;
 629         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
 630         INP_WLOCK(inp);
 631         if (callout_pending(&tp->t_timers->tt_rexmt) ||
 632             !callout_active(&tp->t_timers->tt_rexmt)) {
 633                 INP_WUNLOCK(inp);
 634                 CURVNET_RESTORE();
 635                 return;
 636         }
 637         callout_deactivate(&tp->t_timers->tt_rexmt);
 638         if ((inp->inp_flags & INP_DROPPED) != 0) {
 639                 INP_WUNLOCK(inp);
 640                 CURVNET_RESTORE();
 641                 return;
 642         }
 643         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
 644                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
 645         tcp_free_sackholes(tp);
 646         if (tp->t_fb->tfb_tcp_rexmit_tmr) {
 647                 /* The stack has a timer action too. */
 648                 (*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
 649         }
 650         /*
 651          * Retransmission timer went off.  Message has not
 652          * been acked within retransmit interval.  Back off
 653          * to a longer retransmit interval and retransmit one segment.
 654          */
 655         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
 656                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
 657                 TCPSTAT_INC(tcps_timeoutdrop);
 658                 if (tcp_inpinfo_lock_add(inp)) {
 659                         tcp_inpinfo_lock_del(inp, tp);
 660                         goto out;
 661                 }
 662                 tp = tcp_drop(tp, tp->t_softerror ?
 663                               tp->t_softerror : ETIMEDOUT);
 664                 tcp_inpinfo_lock_del(inp, tp);
 665                 goto out;
 666         }
 667         if (tp->t_state == TCPS_SYN_SENT) {
 668                 /*
 669                  * If the SYN was retransmitted, indicate CWND to be
 670                  * limited to 1 segment in cc_conn_init().
 671                  */
 672                 tp->snd_cwnd = 1;
 673         } else if (tp->t_rxtshift == 1) {
 674                 /*
 675                  * first retransmit; record ssthresh and cwnd so they can
 676                  * be recovered if this turns out to be a "bad" retransmit.
 677                  * A retransmit is considered "bad" if an ACK for this
 678                  * segment is received within RTT/2 interval; the assumption
 679                  * here is that the ACK was already in flight.  See
 680                  * "On Estimating End-to-End Network Path Properties" by
 681                  * Allman and Paxson for more details.
 682                  */
 683                 tp->snd_cwnd_prev = tp->snd_cwnd;
 684                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
 685                 tp->snd_recover_prev = tp->snd_recover;
 686                 if (IN_FASTRECOVERY(tp->t_flags))
 687                         tp->t_flags |= TF_WASFRECOVERY;
 688                 else
 689                         tp->t_flags &= ~TF_WASFRECOVERY;
 690                 if (IN_CONGRECOVERY(tp->t_flags))
 691                         tp->t_flags |= TF_WASCRECOVERY;
 692                 else
 693                         tp->t_flags &= ~TF_WASCRECOVERY;
 694                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
 695                 tp->t_flags |= TF_PREVVALID;
 696         } else
 697                 tp->t_flags &= ~TF_PREVVALID;
 698         TCPSTAT_INC(tcps_rexmttimeo);
 699         if ((tp->t_state == TCPS_SYN_SENT) ||
 700             (tp->t_state == TCPS_SYN_RECEIVED))
 701                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
 702         else
 703                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 704         TCPT_RANGESET(tp->t_rxtcur, rexmt,
 705                       tp->t_rttmin, TCPTV_REXMTMAX);
 706
 707         /*
 708          * We enter the path for PLMTUD if connection is established or, if
 709          * connection is FIN_WAIT_1 status, reason for the last is that if
 710          * amount of data we send is very small, we could send it in couple of
 711          * packets and process straight to FIN. In that case we won't catch
 712          * ESTABLISHED state.
 713          */
 714         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
 715             || (tp->t_state == TCPS_FIN_WAIT_1))) {
 716 #ifdef INET6
 717                 int isipv6;
 718 #endif
 719
 720                 /*
 721                  * Idea here is that at each stage of mtu probe (usually, 1448
 722                  * -> 1188 -> 524) should be given 2 chances to recover before
 723                  *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
 724                  *  take care of that.
 725                  */
 726                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
 727                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
 728                     (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
 729                     tp->t_rxtshift % 2 == 0)) {
 730                         /*
 731                          * Enter Path MTU Black-hole Detection mechanism:
 732                          * - Disable Path MTU Discovery (IP "DF" bit).
 733                          * - Reduce MTU to lower value than what we
 734                          *   negotiated with peer.
 735                          */
 736                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
 737                                 /* Record that we may have found a black hole. */
 738                                 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
 739                                 /* Keep track of previous MSS. */
 740                                 tp->t_pmtud_saved_maxseg = tp->t_maxseg;
 741                         }
 742
 743                         /*
 744                          * Reduce the MSS to blackhole value or to the default
 745                          * in an attempt to retransmit.
 746                          */
 747 #ifdef INET6
 748                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
 749                         if (isipv6 &&
 750                             tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
 751                                 /* Use the sysctl tuneable blackhole MSS. */
 752                                 tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
 753                                 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
 754                         } else if (isipv6) {
 755                                 /* Use the default MSS. */
 756                                 tp->t_maxseg = V_tcp_v6mssdflt;
 757                                 /*
 758                                  * Disable Path MTU Discovery when we switch to
 759                                  * minmss.
 760                                  */
 761                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 762                                 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
 763                         }
 764 #endif
 765 #if defined(INET6) && defined(INET)
 766                         else
 767 #endif
 768 #ifdef INET
 769                         if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
 770                                 /* Use the sysctl tuneable blackhole MSS. */
 771                                 tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
 772                                 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
 773                         } else {
 774                                 /* Use the default MSS. */
 775                                 tp->t_maxseg = V_tcp_mssdflt;
 776                                 /*
 777                                  * Disable Path MTU Discovery when we switch to
 778                                  * minmss.
 779                                  */
 780                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 781                                 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
 782                         }
 783 #endif
 784                         /*
 785                          * Reset the slow-start flight size
 786                          * as it may depend on the new MSS.
 787                          */
 788                         if (CC_ALGO(tp)->conn_init != NULL)
 789                                 CC_ALGO(tp)->conn_init(tp->ccv);
 790                 } else {
 791                         /*
 792                          * If further retransmissions are still unsuccessful
 793                          * with a lowered MTU, maybe this isn't a blackhole and
 794                          * we restore the previous MSS and blackhole detection
 795                          * flags.
 796                          * The limit '6' is determined by giving each probe
 797                          * stage (1448, 1188, 524) 2 chances to recover.
 798                          */
 799                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
 800                             (tp->t_rxtshift >= 6)) {
 801                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 802                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
 803                                 tp->t_maxseg = tp->t_pmtud_saved_maxseg;
 804                                 TCPSTAT_INC(tcps_pmtud_blackhole_failed);
 805                                 /*
 806                                  * Reset the slow-start flight size as it
 807                                  * may depend on the new MSS.
 808                                  */
 809                                 if (CC_ALGO(tp)->conn_init != NULL)
 810                                         CC_ALGO(tp)->conn_init(tp->ccv);
 811                         }
 812                 }
 813         }
 814
 815         /*
 816          * Disable RFC1323 and SACK if we haven't got any response to
 817          * our third SYN to work-around some broken terminal servers
 818          * (most of which have hopefully been retired) that have bad VJ
 819          * header compression code which trashes TCP segments containing
 820          * unknown-to-them TCP options.
 821          */
 822         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
 823             (tp->t_rxtshift == 3))
 824                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 825         /*
 826          * If we backed off this far, notify the L3 protocol that we're having
 827          * connection problems.
 828          */
 829         if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
 830 #ifdef INET6
 831                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 832                         in6_losing(tp->t_inpcb);
 833                 else
 834 #endif
 835                         in_losing(tp->t_inpcb);
 836         }
 837         tp->snd_nxt = tp->snd_una;
 838         tp->snd_recover = tp->snd_max;
 839         /*
 840          * Force a segment to be sent.
 841          */
 842         tp->t_flags |= TF_ACKNOW;
 843         /*
 844          * If timing a segment in this window, stop the timer.
 845          */
 846         tp->t_rtttime = 0;
 847
 848         cc_cong_signal(tp, NULL, CC_RTO);
 849
 850         (void) tp->t_fb->tfb_tcp_output(tp);
 851
 852 #ifdef TCPDEBUG
 853         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
 854                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
 855                           PRU_SLOWTIMO);
 856 #endif
 857         TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
 858         INP_WUNLOCK(inp);
 859 out:
 860         CURVNET_RESTORE();
 861 }
 862
 863 void
 864 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
 865 {
 866         struct callout *t_callout;
 867         timeout_t *f_callout;
 868         struct inpcb *inp = tp->t_inpcb;
 869         int cpu = inp_to_cpuid(inp);
 870
 871 #ifdef TCP_OFFLOAD
 872         if (tp->t_flags & TF_TOE)
 873                 return;
 874 #endif
 875
 876         if (tp->t_timers->tt_flags & TT_STOPPED)
 877                 return;
 878
 879         switch (timer_type) {
 880                 case TT_DELACK:
 881                         t_callout = &tp->t_timers->tt_delack;
 882                         f_callout = tcp_timer_delack;
 883                         break;
 884                 case TT_REXMT:
 885                         t_callout = &tp->t_timers->tt_rexmt;
 886                         f_callout = tcp_timer_rexmt;
 887                         break;
 888                 case TT_PERSIST:
 889                         t_callout = &tp->t_timers->tt_persist;
 890                         f_callout = tcp_timer_persist;
 891                         break;
 892                 case TT_KEEP:
 893                         t_callout = &tp->t_timers->tt_keep;
 894                         f_callout = tcp_timer_keep;
 895                         break;
 896                 case TT_2MSL:
 897                         t_callout = &tp->t_timers->tt_2msl;
 898                         f_callout = tcp_timer_2msl;
 899                         break;
 900                 default:
 901                         if (tp->t_fb->tfb_tcp_timer_activate) {
 902                                 tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
 903                                 return;
 904                         }
 905                         panic("tp %p bad timer_type %#x", tp, timer_type);
 906                 }
 907         if (delta == 0) {
 908                 callout_stop(t_callout);
 909         } else {
 910                 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
 911         }
 912 }
 913
 914 int
 915 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
 916 {
 917         struct callout *t_callout;
 918
 919         switch (timer_type) {
 920                 case TT_DELACK:
 921                         t_callout = &tp->t_timers->tt_delack;
 922                         break;
 923                 case TT_REXMT:
 924                         t_callout = &tp->t_timers->tt_rexmt;
 925                         break;
 926                 case TT_PERSIST:
 927                         t_callout = &tp->t_timers->tt_persist;
 928                         break;
 929                 case TT_KEEP:
 930                         t_callout = &tp->t_timers->tt_keep;
 931                         break;
 932                 case TT_2MSL:
 933                         t_callout = &tp->t_timers->tt_2msl;
 934                         break;
 935                 default:
 936                         if (tp->t_fb->tfb_tcp_timer_active) {
 937                                 return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
 938                         }
 939                         panic("tp %p bad timer_type %#x", tp, timer_type);
 940                 }
 941         return callout_active(t_callout);
 942 }
 943
 944 void
 945 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
 946 {
 947         struct callout *t_callout;
 948
 949         tp->t_timers->tt_flags |= TT_STOPPED;
 950         switch (timer_type) {
 951                 case TT_DELACK:
 952                         t_callout = &tp->t_timers->tt_delack;
 953                         break;
 954                 case TT_REXMT:
 955                         t_callout = &tp->t_timers->tt_rexmt;
 956                         break;
 957                 case TT_PERSIST:
 958                         t_callout = &tp->t_timers->tt_persist;
 959                         break;
 960                 case TT_KEEP:
 961                         t_callout = &tp->t_timers->tt_keep;
 962                         break;
 963                 case TT_2MSL:
 964                         t_callout = &tp->t_timers->tt_2msl;
 965                         break;
 966                 default:
 967                         if (tp->t_fb->tfb_tcp_timer_stop) {
 968                                 /*
 969                                  * XXXrrs we need to look at this with the
 970                                  * stop case below (flags).
 971                                  */
 972                                 tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
 973                                 return;
 974                         }
 975                         panic("tp %p bad timer_type %#x", tp, timer_type);
 976                 }
 977
 978         if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
 979                 /*
 980                  * Can't stop the callout, defer tcpcb actual deletion
 981                  * to the last one. We do this using the async drain
 982                  * function and incrementing the count in
 983                  */
 984                 tp->t_timers->tt_draincnt++;
 985         }
 986 }