sys/netinet/tcp_usrreq.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  64  */
  65
  66 #include "opt_ipsec.h"
  67 #include "opt_inet.h"
  68 #include "opt_inet6.h"
  69 #include "opt_tcpdebug.h"
  70
  71 #include <sys/param.h>
  72 #include <sys/systm.h>
  73 #include <sys/kernel.h>
  74 #include <sys/malloc.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/globaldata.h>
  77 #include <sys/thread.h>
  78
  79 #include <sys/mbuf.h>
  80 #ifdef INET6
  81 #include <sys/domain.h>
  82 #endif /* INET6 */
  83 #include <sys/socket.h>
  84 #include <sys/socketvar.h>
  85 #include <sys/socketops.h>
  86 #include <sys/protosw.h>
  87
  88 #include <sys/thread2.h>
  89 #include <sys/msgport2.h>
  90 #include <sys/socketvar2.h>
  91
  92 #include <net/if.h>
  93 #include <net/netisr.h>
  94 #include <net/route.h>
  95
  96 #include <net/netmsg2.h>
  97 #include <net/netisr2.h>
  98
  99 #include <netinet/in.h>
 100 #include <netinet/in_systm.h>
 101 #ifdef INET6
 102 #include <netinet/ip6.h>
 103 #endif
 104 #include <netinet/in_pcb.h>
 105 #ifdef INET6
 106 #include <netinet6/in6_pcb.h>
 107 #endif
 108 #include <netinet/in_var.h>
 109 #include <netinet/ip_var.h>
 110 #ifdef INET6
 111 #include <netinet6/ip6_var.h>
 112 #include <netinet6/tcp6_var.h>
 113 #endif
 114 #include <netinet/tcp.h>
 115 #include <netinet/tcp_fsm.h>
 116 #include <netinet/tcp_seq.h>
 117 #include <netinet/tcp_timer.h>
 118 #include <netinet/tcp_timer2.h>
 119 #include <netinet/tcp_var.h>
 120 #include <netinet/tcpip.h>
 121 #ifdef TCPDEBUG
 122 #include <netinet/tcp_debug.h>
 123 #endif
 124
 125 #ifdef IPSEC
 126 #include <netinet6/ipsec.h>
 127 #endif /*IPSEC*/
 128
 129 /*
 130  * TCP protocol interface to socket abstraction.
 131  */
 132 extern  char *tcpstates[];      /* XXX ??? */
 133
 134 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 135 static void     tcp_connect (netmsg_t msg);
 136 #ifdef INET6
 137 static void     tcp6_connect (netmsg_t msg);
 138 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 139                                 struct mbuf **mp,
 140                                 struct sockaddr_in6 *sin6,
 141                                 struct in6_addr *addr6);
 142 #endif /* INET6 */
 143 static struct tcpcb *
 144                 tcp_disconnect (struct tcpcb *);
 145 static struct tcpcb *
 146                 tcp_usrclosed (struct tcpcb *);
 147
 148 #ifdef TCPDEBUG
 149 #define TCPDEBUG0       int ostate = 0
 150 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 151 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 152                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 153 #else
 154 #define TCPDEBUG0
 155 #define TCPDEBUG1()
 156 #define TCPDEBUG2(req)
 157 #endif
 158
 159 /*
 160  * For some ill optimized programs, which try to use TCP_NOPUSH
 161  * to improve performance, will have small amount of data sits
 162  * in the sending buffer.  These small amount of data will _not_
 163  * be pushed into the network until more data are written into
 164  * the socket or the socket write side is shutdown.
 165  */
 166 static int      tcp_disable_nopush = 1;
 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
 168     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
 169
 170 /*
 171  * Allocate socket buffer space.
 172  */
 173 static int
 174 tcp_usr_preattach(struct socket *so, int proto __unused,
 175     struct pru_attach_info *ai)
 176 {
 177         int error;
 178
 179         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
 180                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
 181                                   ai->sb_rlimit);
 182                 if (error)
 183                         return (error);
 184         }
 185         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
 186         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
 187
 188         return 0;
 189 }
 190
 191 /*
 192  * TCP attaches to socket via pru_attach(), reserving space,
 193  * and an internet control block.  This socket may move to
 194  * other CPU later when we bind/connect.
 195  */
 196 static void
 197 tcp_usr_attach(netmsg_t msg)
 198 {
 199         struct socket *so = msg->base.nm_so;
 200         struct pru_attach_info *ai = msg->attach.nm_ai;
 201         int error;
 202         struct inpcb *inp;
 203         struct tcpcb *tp = NULL;
 204         TCPDEBUG0;
 205
 206         inp = so->so_pcb;
 207         KASSERT(inp == NULL, ("tcp socket attached"));
 208         TCPDEBUG1();
 209
 210         error = tcp_attach(so, ai);
 211         if (error)
 212                 goto out;
 213
 214         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 215                 so->so_linger = TCP_LINGERTIME;
 216         tp = sototcpcb(so);
 217 out:
 218         TCPDEBUG2(PRU_ATTACH);
 219         lwkt_replymsg(&msg->lmsg, error);
 220 }
 221
 222 /*
 223  * pru_detach() detaches the TCP protocol from the socket.
 224  * If the protocol state is non-embryonic, then can't
 225  * do this directly: have to initiate a pru_disconnect(),
 226  * which may finish later; embryonic TCB's can just
 227  * be discarded here.
 228  */
 229 static void
 230 tcp_usr_detach(netmsg_t msg)
 231 {
 232         struct socket *so = msg->base.nm_so;
 233         int error = 0;
 234         struct inpcb *inp;
 235         struct tcpcb *tp;
 236         TCPDEBUG0;
 237
 238         inp = so->so_pcb;
 239
 240         /*
 241          * If the inp is already detached or never attached, it may have
 242          * been due to an async close or async attach failure.  Just return
 243          * as if no error occured.
 244          */
 245         if (inp) {
 246                 tp = intotcpcb(inp);
 247                 KASSERT(tp != NULL, ("tcp_usr_detach: tp is NULL"));
 248                 TCPDEBUG1();
 249                 tp = tcp_disconnect(tp);
 250                 TCPDEBUG2(PRU_DETACH);
 251         }
 252         lwkt_replymsg(&msg->lmsg, error);
 253 }
 254
 255 /*
 256  * NOTE: ignore_error is non-zero for certain disconnection races
 257  * which we want to silently allow, otherwise close() may return
 258  * an unexpected error.
 259  *
 260  * NOTE: The variables (msg) and (tp) are assumed.
 261  */
 262 #define COMMON_START(so, inp, ignore_error)                     \
 263         TCPDEBUG0;                                              \
 264                                                                 \
 265         inp = so->so_pcb;                                       \
 266         do {                                                    \
 267                 if (inp == NULL) {                              \
 268                         error = ignore_error ? 0 : EINVAL;      \
 269                         tp = NULL;                              \
 270                         goto out;                               \
 271                 }                                               \
 272                 tp = intotcpcb(inp);                            \
 273                 TCPDEBUG1();                                    \
 274         } while(0)
 275
 276 #define COMMON_END1(req, noreply)                               \
 277         out: do {                                               \
 278                 TCPDEBUG2(req);                                 \
 279                 if (!(noreply))                                 \
 280                         lwkt_replymsg(&msg->lmsg, error);       \
 281                 return;                                         \
 282         } while(0)
 283
 284 #define COMMON_END(req)         COMMON_END1((req), 0)
 285
 286 static void
 287 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 288 {
 289         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 290 }
 291
 292 /*
 293  * Give the socket an address.
 294  */
 295 static void
 296 tcp_usr_bind(netmsg_t msg)
 297 {
 298         struct socket *so = msg->bind.base.nm_so;
 299         struct sockaddr *nam = msg->bind.nm_nam;
 300         struct thread *td = msg->bind.nm_td;
 301         int error = 0;
 302         struct inpcb *inp;
 303         struct tcpcb *tp;
 304         struct sockaddr_in *sinp;
 305         lwkt_port_t port0 = netisr_cpuport(0);
 306
 307         COMMON_START(so, inp, 0);
 308
 309         /*
 310          * Must check for multicast addresses and disallow binding
 311          * to them.
 312          */
 313         sinp = (struct sockaddr_in *)nam;
 314         if (sinp->sin_family == AF_INET &&
 315             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 316                 error = EAFNOSUPPORT;
 317                 goto out;
 318         }
 319
 320         /*
 321          * Check "already bound" here (in_pcbbind() does the same check
 322          * though), so we don't forward a connected socket to netisr0,
 323          * which would panic in the following in_pcbunlink().
 324          */
 325         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) {
 326                 error = EINVAL; /* already bound */
 327                 goto out;
 328         }
 329
 330         /*
 331          * Use netisr0 to serialize in_pcbbind(), so that pru_detach and
 332          * pru_bind for different sockets on the same local port could be
 333          * properly ordered.  The original race is illustrated here for
 334          * reference.
 335          *
 336          * s1 = socket();
 337          * bind(s1, *.PORT);
 338          * close(s1);  <----- asynchronous
 339          * s2 = socket();
 340          * bind(s2, *.PORT);
 341          *
 342          * All will expect bind(s2, *.PORT) to succeed.  However, it will
 343          * fail, if following sequence happens due to random socket initial
 344          * msgport and asynchronous close(2):
 345          *
 346          *    netisrN                  netisrM
 347          *       :                        :
 348          *       :                    pru_bind(s2) [*.PORT is used by s1]
 349          *  pru_detach(s1)                :
 350          */
 351         if (&curthread->td_msgport != port0) {
 352                 lwkt_msg_t lmsg = &msg->bind.base.lmsg;
 353
 354                 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0,
 355                     ("already asked to relink"));
 356
 357                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 358                 msg->bind.nm_flags |= PRUB_RELINK;
 359
 360                 TCP_STATE_MIGRATE_START(tp);
 361
 362                 /* See the related comment in tcp_connect() */
 363                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 364                 lwkt_forwardmsg(port0, lmsg);
 365                 /* msg invalid now */
 366                 return;
 367         }
 368         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 369
 370         if (msg->bind.nm_flags & PRUB_RELINK) {
 371                 msg->bind.nm_flags &= ~PRUB_RELINK;
 372                 TCP_STATE_MIGRATE_END(tp);
 373                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 374         }
 375         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 376
 377         error = in_pcbbind(inp, nam, td);
 378         if (error)
 379                 goto out;
 380
 381         COMMON_END(PRU_BIND);
 382 }
 383
 384 #ifdef INET6
 385
 386 static void
 387 tcp6_usr_bind(netmsg_t msg)
 388 {
 389         struct socket *so = msg->bind.base.nm_so;
 390         struct sockaddr *nam = msg->bind.nm_nam;
 391         struct thread *td = msg->bind.nm_td;
 392         int error = 0;
 393         struct inpcb *inp;
 394         struct tcpcb *tp;
 395         struct sockaddr_in6 *sin6p;
 396
 397         COMMON_START(so, inp, 0);
 398
 399         /*
 400          * Must check for multicast addresses and disallow binding
 401          * to them.
 402          */
 403         sin6p = (struct sockaddr_in6 *)nam;
 404         if (sin6p->sin6_family == AF_INET6 &&
 405             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 406                 error = EAFNOSUPPORT;
 407                 goto out;
 408         }
 409         error = in6_pcbbind(inp, nam, td);
 410         if (error)
 411                 goto out;
 412         COMMON_END(PRU_BIND);
 413 }
 414 #endif /* INET6 */
 415
 416 struct netmsg_inswildcard {
 417         struct netmsg_base      base;
 418         struct inpcb            *nm_inp;
 419 };
 420
 421 static void
 422 in_pcbinswildcardhash_handler(netmsg_t msg)
 423 {
 424         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 425         int cpu = mycpuid, nextcpu;
 426
 427         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 428
 429         nextcpu = cpu + 1;
 430         if (nextcpu < netisr_ncpus)
 431                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 432         else
 433                 lwkt_replymsg(&nm->base.lmsg, 0);
 434 }
 435
 436 /*
 437  * Prepare to accept connections.
 438  */
 439 static void
 440 tcp_usr_listen(netmsg_t msg)
 441 {
 442         struct socket *so = msg->listen.base.nm_so;
 443         struct thread *td = msg->listen.nm_td;
 444         int error = 0;
 445         struct inpcb *inp;
 446         struct tcpcb *tp;
 447         struct netmsg_inswildcard nm;
 448         lwkt_port_t port0 = netisr_cpuport(0);
 449
 450         COMMON_START(so, inp, 0);
 451
 452         if (&curthread->td_msgport != port0) {
 453                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 454
 455                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 456                     ("already asked to relink"));
 457
 458                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 459                 msg->listen.nm_flags |= PRUL_RELINK;
 460
 461                 TCP_STATE_MIGRATE_START(tp);
 462
 463                 /* See the related comment in tcp_connect() */
 464                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 465                 lwkt_forwardmsg(port0, lmsg);
 466                 /* msg invalid now */
 467                 return;
 468         }
 469         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 470
 471         if (msg->listen.nm_flags & PRUL_RELINK) {
 472                 msg->listen.nm_flags &= ~PRUL_RELINK;
 473                 TCP_STATE_MIGRATE_END(tp);
 474                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 475         }
 476         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 477
 478         if (tp->t_flags & TF_LISTEN)
 479                 goto out;
 480
 481         if (inp->inp_lport == 0) {
 482                 error = in_pcbbind(inp, NULL, td);
 483                 if (error)
 484                         goto out;
 485         }
 486
 487         TCP_STATE_CHANGE(tp, TCPS_LISTEN);
 488         tp->t_flags |= TF_LISTEN;
 489         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 490
 491         /*
 492          * Create tcpcb per-cpu port cache
 493          *
 494          * NOTE:
 495          * This _must_ be done before installing this inpcb into
 496          * wildcard hash.
 497          */
 498         tcp_pcbport_create(tp);
 499
 500         if (netisr_ncpus > 1) {
 501                 /*
 502                  * Put this inpcb into wildcard hash on other cpus.
 503                  */
 504                 ASSERT_INP_NOTINHASH(inp);
 505                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 506                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 507                 nm.nm_inp = inp;
 508                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 509         }
 510         in_pcbinswildcardhash(inp);
 511         COMMON_END(PRU_LISTEN);
 512 }
 513
 514 #ifdef INET6
 515
 516 static void
 517 tcp6_usr_listen(netmsg_t msg)
 518 {
 519         struct socket *so = msg->listen.base.nm_so;
 520         struct thread *td = msg->listen.nm_td;
 521         int error = 0;
 522         struct inpcb *inp;
 523         struct tcpcb *tp;
 524         struct netmsg_inswildcard nm;
 525
 526         COMMON_START(so, inp, 0);
 527
 528         if (tp->t_flags & TF_LISTEN)
 529                 goto out;
 530
 531         if (inp->inp_lport == 0) {
 532                 error = in6_pcbbind(inp, NULL, td);
 533                 if (error)
 534                         goto out;
 535         }
 536
 537         TCP_STATE_CHANGE(tp, TCPS_LISTEN);
 538         tp->t_flags |= TF_LISTEN;
 539         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 540
 541         /*
 542          * Create tcpcb per-cpu port cache
 543          *
 544          * NOTE:
 545          * This _must_ be done before installing this inpcb into
 546          * wildcard hash.
 547          */
 548         tcp_pcbport_create(tp);
 549
 550         if (netisr_ncpus > 1) {
 551                 /*
 552                  * Put this inpcb into wildcard hash on other cpus.
 553                  */
 554                 KKASSERT(so->so_port == netisr_cpuport(0));
 555                 ASSERT_NETISR0;
 556                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 557                 ASSERT_INP_NOTINHASH(inp);
 558
 559                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 560                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 561                 nm.nm_inp = inp;
 562                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 563         }
 564         in_pcbinswildcardhash(inp);
 565         COMMON_END(PRU_LISTEN);
 566 }
 567 #endif /* INET6 */
 568
 569 /*
 570  * Initiate connection to peer.
 571  * Create a template for use in transmissions on this connection.
 572  * Enter SYN_SENT state, and mark socket as connecting.
 573  * Start keep-alive timer, and seed output sequence space.
 574  * Send initial segment on connection.
 575  */
 576 static void
 577 tcp_usr_connect(netmsg_t msg)
 578 {
 579         struct socket *so = msg->connect.base.nm_so;
 580         struct sockaddr *nam = msg->connect.nm_nam;
 581         struct thread *td = msg->connect.nm_td;
 582         int error = 0;
 583         struct inpcb *inp;
 584         struct tcpcb *tp;
 585         struct sockaddr_in *sinp;
 586
 587         COMMON_START(so, inp, 0);
 588
 589         /*
 590          * Must disallow TCP ``connections'' to multicast addresses.
 591          */
 592         sinp = (struct sockaddr_in *)nam;
 593         if (sinp->sin_family == AF_INET
 594             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 595                 error = EAFNOSUPPORT;
 596                 goto out;
 597         }
 598
 599         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 600                 error = EAFNOSUPPORT; /* IPv6 only jail */
 601                 goto out;
 602         }
 603
 604         tcp_connect(msg);
 605         /* msg is invalid now */
 606         return;
 607 out:
 608         if (msg->connect.nm_m) {
 609                 m_freem(msg->connect.nm_m);
 610                 msg->connect.nm_m = NULL;
 611         }
 612         if (msg->connect.nm_flags & PRUC_HELDTD)
 613                 lwkt_rele(td);
 614         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 615                 so->so_error = error;
 616                 soisdisconnected(so);
 617         }
 618         lwkt_replymsg(&msg->lmsg, error);
 619 }
 620
 621 #ifdef INET6
 622
 623 static void
 624 tcp6_usr_connect(netmsg_t msg)
 625 {
 626         struct socket *so = msg->connect.base.nm_so;
 627         struct sockaddr *nam = msg->connect.nm_nam;
 628         struct thread *td = msg->connect.nm_td;
 629         int error = 0;
 630         struct inpcb *inp;
 631         struct tcpcb *tp;
 632         struct sockaddr_in6 *sin6p;
 633
 634         COMMON_START(so, inp, 0);
 635
 636         /*
 637          * Must disallow TCP ``connections'' to multicast addresses.
 638          */
 639         sin6p = (struct sockaddr_in6 *)nam;
 640         if (sin6p->sin6_family == AF_INET6
 641             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 642                 error = EAFNOSUPPORT;
 643                 goto out;
 644         }
 645
 646         if (!prison_remote_ip(td, nam)) {
 647                 error = EAFNOSUPPORT; /* IPv4 only jail */
 648                 goto out;
 649         }
 650
 651         /* Reject v4-mapped address */
 652         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 653                 error = EADDRNOTAVAIL;
 654                 goto out;
 655         }
 656
 657         inp->inp_inc.inc_isipv6 = 1;
 658         tcp6_connect(msg);
 659         /* msg is invalid now */
 660         return;
 661 out:
 662         if (msg->connect.nm_m) {
 663                 m_freem(msg->connect.nm_m);
 664                 msg->connect.nm_m = NULL;
 665         }
 666         lwkt_replymsg(&msg->lmsg, error);
 667 }
 668
 669 #endif /* INET6 */
 670
 671 /*
 672  * Initiate disconnect from peer.
 673  * If connection never passed embryonic stage, just drop;
 674  * else if don't need to let data drain, then can just drop anyways,
 675  * else have to begin TCP shutdown process: mark socket disconnecting,
 676  * drain unread data, state switch to reflect user close, and
 677  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 678  * when peer sends FIN and acks ours.
 679  *
 680  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 681  */
 682 static void
 683 tcp_usr_disconnect(netmsg_t msg)
 684 {
 685         struct socket *so = msg->disconnect.base.nm_so;
 686         int error = 0;
 687         struct inpcb *inp;
 688         struct tcpcb *tp;
 689
 690         COMMON_START(so, inp, 1);
 691         tp = tcp_disconnect(tp);
 692         COMMON_END(PRU_DISCONNECT);
 693 }
 694
 695 /*
 696  * Accept a connection.  Essentially all the work is
 697  * done at higher levels; just return the address
 698  * of the peer, storing through addr.
 699  */
 700 static void
 701 tcp_usr_accept(netmsg_t msg)
 702 {
 703         struct socket *so = msg->accept.base.nm_so;
 704         struct sockaddr **nam = msg->accept.nm_nam;
 705         int error = 0;
 706         struct inpcb *inp;
 707         struct tcpcb *tp = NULL;
 708         TCPDEBUG0;
 709
 710         inp = so->so_pcb;
 711         if (so->so_state & SS_ISDISCONNECTED) {
 712                 error = ECONNABORTED;
 713                 goto out;
 714         }
 715         if (inp == NULL) {
 716                 error = EINVAL;
 717                 goto out;
 718         }
 719
 720         tp = intotcpcb(inp);
 721         TCPDEBUG1();
 722         in_setpeeraddr(so, nam);
 723         COMMON_END(PRU_ACCEPT);
 724 }
 725
 726 #ifdef INET6
 727 static void
 728 tcp6_usr_accept(netmsg_t msg)
 729 {
 730         struct socket *so = msg->accept.base.nm_so;
 731         struct sockaddr **nam = msg->accept.nm_nam;
 732         int error = 0;
 733         struct inpcb *inp;
 734         struct tcpcb *tp = NULL;
 735         TCPDEBUG0;
 736
 737         inp = so->so_pcb;
 738
 739         if (so->so_state & SS_ISDISCONNECTED) {
 740                 error = ECONNABORTED;
 741                 goto out;
 742         }
 743         if (inp == NULL) {
 744                 error = EINVAL;
 745                 goto out;
 746         }
 747         tp = intotcpcb(inp);
 748         TCPDEBUG1();
 749         in6_setpeeraddr(so, nam);
 750         COMMON_END(PRU_ACCEPT);
 751 }
 752 #endif /* INET6 */
 753
 754 /*
 755  * Mark the connection as being incapable of further output.
 756  */
 757 static void
 758 tcp_usr_shutdown(netmsg_t msg)
 759 {
 760         struct socket *so = msg->shutdown.base.nm_so;
 761         int error = 0;
 762         struct inpcb *inp;
 763         struct tcpcb *tp;
 764
 765         COMMON_START(so, inp, 0);
 766         socantsendmore(so);
 767         tp = tcp_usrclosed(tp);
 768         if (tp)
 769                 error = tcp_output(tp);
 770         COMMON_END(PRU_SHUTDOWN);
 771 }
 772
 773 /*
 774  * After a receive, possibly send window update to peer.
 775  */
 776 static void
 777 tcp_usr_rcvd(netmsg_t msg)
 778 {
 779         struct socket *so = msg->rcvd.base.nm_so;
 780         int error = 0, noreply = 0;
 781         struct inpcb *inp;
 782         struct tcpcb *tp;
 783
 784         COMMON_START(so, inp, 0);
 785
 786         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 787                 noreply = 1;
 788                 so_async_rcvd_reply(so);
 789         }
 790         tcp_output(tp);
 791
 792         COMMON_END1(PRU_RCVD, noreply);
 793 }
 794
 795 /*
 796  * Do a send by putting data in output queue and updating urgent
 797  * marker if URG set.  Possibly send more data.  Unlike the other
 798  * pru_*() routines, the mbuf chains are our responsibility.  We
 799  * must either enqueue them or free them.  The other pru_* routines
 800  * generally are caller-frees.
 801  */
 802 static void
 803 tcp_usr_send(netmsg_t msg)
 804 {
 805         struct socket *so = msg->send.base.nm_so;
 806         int flags = msg->send.nm_flags;
 807         struct mbuf *m = msg->send.nm_m;
 808         int error = 0;
 809         struct inpcb *inp;
 810         struct tcpcb *tp;
 811         TCPDEBUG0;
 812
 813         KKASSERT(msg->send.nm_control == NULL);
 814         KKASSERT(msg->send.nm_addr == NULL);
 815         KKASSERT((flags & PRUS_FREEADDR) == 0);
 816
 817         inp = so->so_pcb;
 818
 819         if (inp == NULL) {
 820                 /*
 821                  * OOPS! we lost a race, the TCP session got reset after
 822                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 823                  * network interrupt in the non-critical section of sosend().
 824                  */
 825                 m_freem(m);
 826                 error = ECONNRESET;     /* XXX EPIPE? */
 827                 tp = NULL;
 828                 TCPDEBUG1();
 829                 goto out;
 830         }
 831         tp = intotcpcb(inp);
 832         TCPDEBUG1();
 833
 834 #ifdef foo
 835         /*
 836          * This is no longer necessary, since:
 837          * - sosendtcp() has already checked it for us
 838          * - It does not work with asynchronized send
 839          */
 840
 841         /*
 842          * Don't let too much OOB data build up
 843          */
 844         if (flags & PRUS_OOB) {
 845                 if (ssb_space(&so->so_snd) < -512) {
 846                         m_freem(m);
 847                         error = ENOBUFS;
 848                         goto out;
 849                 }
 850         }
 851 #endif
 852
 853         /*
 854          * Pump the data into the socket.
 855          */
 856         if (m) {
 857                 ssb_appendstream(&so->so_snd, m);
 858                 sowwakeup(so);
 859         }
 860         if (flags & PRUS_OOB) {
 861                 /*
 862                  * According to RFC961 (Assigned Protocols),
 863                  * the urgent pointer points to the last octet
 864                  * of urgent data.  We continue, however,
 865                  * to consider it to indicate the first octet
 866                  * of data past the urgent section.
 867                  * Otherwise, snd_up should be one lower.
 868                  */
 869                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 870                 tp->t_flags |= TF_FORCE;
 871                 error = tcp_output(tp);
 872                 tp->t_flags &= ~TF_FORCE;
 873         } else {
 874                 if (flags & PRUS_EOF) {
 875                         /*
 876                          * Close the send side of the connection after
 877                          * the data is sent.
 878                          */
 879                         socantsendmore(so);
 880                         tp = tcp_usrclosed(tp);
 881                 }
 882                 if (tp != NULL && !tcp_output_pending(tp)) {
 883                         if (flags & PRUS_MORETOCOME)
 884                                 tp->t_flags |= TF_MORETOCOME;
 885                         error = tcp_output_fair(tp);
 886                         if (flags & PRUS_MORETOCOME)
 887                                 tp->t_flags &= ~TF_MORETOCOME;
 888                 }
 889         }
 890         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 891                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 892                    (flags & PRUS_NOREPLY));
 893 }
 894
 895 /*
 896  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 897  *       will sofree() it when we return.
 898  */
 899 static void
 900 tcp_usr_abort(netmsg_t msg)
 901 {
 902         struct socket *so = msg->abort.base.nm_so;
 903         int error = 0;
 904         struct inpcb *inp;
 905         struct tcpcb *tp;
 906
 907         COMMON_START(so, inp, 1);
 908         tp = tcp_drop(tp, ECONNABORTED);
 909         COMMON_END(PRU_ABORT);
 910 }
 911
 912 /*
 913  * Receive out-of-band data.
 914  */
 915 static void
 916 tcp_usr_rcvoob(netmsg_t msg)
 917 {
 918         struct socket *so = msg->rcvoob.base.nm_so;
 919         struct mbuf *m = msg->rcvoob.nm_m;
 920         int flags = msg->rcvoob.nm_flags;
 921         int error = 0;
 922         struct inpcb *inp;
 923         struct tcpcb *tp;
 924
 925         COMMON_START(so, inp, 0);
 926         if ((so->so_oobmark == 0 &&
 927              (so->so_state & SS_RCVATMARK) == 0) ||
 928             so->so_options & SO_OOBINLINE ||
 929             tp->t_oobflags & TCPOOB_HADDATA) {
 930                 error = EINVAL;
 931                 goto out;
 932         }
 933         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 934                 error = EWOULDBLOCK;
 935                 goto out;
 936         }
 937         m->m_len = 1;
 938         *mtod(m, caddr_t) = tp->t_iobc;
 939         if ((flags & MSG_PEEK) == 0)
 940                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 941         COMMON_END(PRU_RCVOOB);
 942 }
 943
 944 static void
 945 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 946 {
 947         in_savefaddr(so, faddr);
 948 }
 949
 950 #ifdef INET6
 951 static void
 952 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 953 {
 954         in6_savefaddr(so, faddr);
 955 }
 956 #endif
 957
 958 static int
 959 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 960     struct thread *td __unused)
 961 {
 962         const struct sockaddr_in *sinp;
 963
 964         sinp = (const struct sockaddr_in *)nam;
 965         if (sinp->sin_family == AF_INET &&
 966             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 967                 return EAFNOSUPPORT;
 968
 969         soisconnecting(so);
 970         return 0;
 971 }
 972
 973 /* xxx - should be const */
 974 struct pr_usrreqs tcp_usrreqs = {
 975         .pru_abort = tcp_usr_abort,
 976         .pru_accept = tcp_usr_accept,
 977         .pru_attach = tcp_usr_attach,
 978         .pru_bind = tcp_usr_bind,
 979         .pru_connect = tcp_usr_connect,
 980         .pru_connect2 = pr_generic_notsupp,
 981         .pru_control = in_control_dispatch,
 982         .pru_detach = tcp_usr_detach,
 983         .pru_disconnect = tcp_usr_disconnect,
 984         .pru_listen = tcp_usr_listen,
 985         .pru_peeraddr = in_setpeeraddr_dispatch,
 986         .pru_rcvd = tcp_usr_rcvd,
 987         .pru_rcvoob = tcp_usr_rcvoob,
 988         .pru_send = tcp_usr_send,
 989         .pru_sense = pru_sense_null,
 990         .pru_shutdown = tcp_usr_shutdown,
 991         .pru_sockaddr = in_setsockaddr_dispatch,
 992         .pru_sosend = sosendtcp,
 993         .pru_soreceive = sorecvtcp,
 994         .pru_savefaddr = tcp_usr_savefaddr,
 995         .pru_preconnect = tcp_usr_preconnect,
 996         .pru_preattach = tcp_usr_preattach
 997 };
 998
 999 #ifdef INET6
1000 struct pr_usrreqs tcp6_usrreqs = {
1001         .pru_abort = tcp_usr_abort,
1002         .pru_accept = tcp6_usr_accept,
1003         .pru_attach = tcp_usr_attach,
1004         .pru_bind = tcp6_usr_bind,
1005         .pru_connect = tcp6_usr_connect,
1006         .pru_connect2 = pr_generic_notsupp,
1007         .pru_control = in6_control_dispatch,
1008         .pru_detach = tcp_usr_detach,
1009         .pru_disconnect = tcp_usr_disconnect,
1010         .pru_listen = tcp6_usr_listen,
1011         .pru_peeraddr = in6_setpeeraddr_dispatch,
1012         .pru_rcvd = tcp_usr_rcvd,
1013         .pru_rcvoob = tcp_usr_rcvoob,
1014         .pru_send = tcp_usr_send,
1015         .pru_sense = pru_sense_null,
1016         .pru_shutdown = tcp_usr_shutdown,
1017         .pru_sockaddr = in6_setsockaddr_dispatch,
1018         .pru_sosend = sosendtcp,
1019         .pru_soreceive = sorecvtcp,
1020         .pru_savefaddr = tcp6_usr_savefaddr
1021 };
1022 #endif /* INET6 */
1023
1024 static int
1025 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1026                   const struct sockaddr_in *sin, struct sockaddr_in *if_sin,
1027                   uint16_t hash)
1028 {
1029         struct inpcb *inp = tp->t_inpcb, *oinp;
1030         struct socket *so = inp->inp_socket;
1031         struct route *ro = &inp->inp_route;
1032
1033         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
1034             ("pcbinfo mismatch"));
1035
1036         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1037                                  sin->sin_addr, sin->sin_port,
1038                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
1039                                   inp->inp_laddr : if_sin->sin_addr),
1040                                 inp->inp_lport, 0, NULL);
1041         if (oinp != NULL) {
1042                 m_freem(m);
1043                 return (EADDRINUSE);
1044         }
1045         if (inp->inp_laddr.s_addr == INADDR_ANY)
1046                 inp->inp_laddr = if_sin->sin_addr;
1047         KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1048             ("faddr mismatch for reconnect"));
1049         KASSERT(inp->inp_fport == sin->sin_port,
1050             ("fport mismatch for reconnect"));
1051         in_pcbinsconnhash(inp);
1052
1053         inp->inp_flags |= INP_HASH;
1054         inp->inp_hashval = hash;
1055
1056         /*
1057          * We are now on the inpcb's owner CPU, if the cached route was
1058          * freed because the rtentry's owner CPU is not the current CPU
1059          * (e.g. in tcp_connect()), then we try to reallocate it here with
1060          * the hope that a rtentry may be cloned from a RTF_PRCLONING
1061          * rtentry.
1062          */
1063         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1064             ro->ro_rt == NULL) {
1065                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1066                 ro->ro_dst.sa_family = AF_INET;
1067                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1068                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1069                         sin->sin_addr;
1070                 rtalloc(ro);
1071         }
1072
1073         /*
1074          * Now that no more errors can occur, change the protocol processing
1075          * port to the current thread (which is the correct thread).
1076          *
1077          * Create TCP timer message now; we are on the tcpcb's owner
1078          * CPU/thread.
1079          */
1080         tcp_create_timermsg(tp, &curthread->td_msgport);
1081
1082         /*
1083          * Compute window scaling to request.  Use a larger scaling then
1084          * needed for the initial receive buffer in case the receive buffer
1085          * gets expanded.
1086          */
1087         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1088                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1089         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1090                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1091         ) {
1092                 tp->request_r_scale++;
1093         }
1094
1095         soisconnecting(so);
1096         tcpstat.tcps_connattempt++;
1097         TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1098         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1099         tp->iss = tcp_new_isn(tp);
1100         tcp_sendseqinit(tp);
1101         if (m) {
1102                 ssb_appendstream(&so->so_snd, m);
1103                 m = NULL;
1104                 if (flags & PRUS_OOB)
1105                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1106         }
1107
1108         /*
1109          * Close the send side of the connection after
1110          * the data is sent if flagged.
1111          */
1112         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1113                 socantsendmore(so);
1114                 tp = tcp_usrclosed(tp);
1115         }
1116         return (tcp_output(tp));
1117 }
1118
1119 /*
1120  * Common subroutine to open a TCP connection to remote host specified
1121  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1122  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1123  * a local host address (interface).
1124  * Initialize connection parameters and enter SYN-SENT state.
1125  */
1126 static void
1127 tcp_connect(netmsg_t msg)
1128 {
1129         struct socket *so = msg->connect.base.nm_so;
1130         struct sockaddr *nam = msg->connect.nm_nam;
1131         struct thread *td = msg->connect.nm_td;
1132         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1133         struct sockaddr_in *if_sin = NULL;
1134         struct inpcb *inp;
1135         struct tcpcb *tp;
1136         int error;
1137         uint16_t hash;
1138         lwkt_port_t port;
1139
1140         COMMON_START(so, inp, 0);
1141
1142         /*
1143          * Reconnect our pcb if we have to
1144          */
1145         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1146                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1147                 TCP_STATE_MIGRATE_END(tp);
1148                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1149         } else {
1150                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1151                         kprintf("inpcb %p, double-connect race\n", inp);
1152                         error = EISCONN;
1153                         if (so->so_state & SS_ISCONNECTING)
1154                                 error = EALREADY;
1155                         goto out;
1156                 }
1157                 KASSERT(inp->inp_fport == 0, ("invalid fport"));
1158         }
1159
1160         /*
1161          * Select local port, if it is not yet selected.
1162          */
1163         if (inp->inp_lport == 0) {
1164                 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1165
1166                 error = in_pcbladdr(inp, nam, &if_sin, td);
1167                 if (error)
1168                         goto out;
1169                 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1170                 msg->connect.nm_flags |= PRUC_HASLADDR;
1171
1172                 /*
1173                  * Install faddr/fport earlier, so that when this
1174                  * inpcb is installed on to the lport hash, the
1175                  * 4-tuple contains correct value.
1176                  *
1177                  * NOTE: The faddr/fport will have to be installed
1178                  * after the in_pcbladdr(), which may change them.
1179                  */
1180                 inp->inp_faddr = sin->sin_addr;
1181                 inp->inp_fport = sin->sin_port;
1182
1183                 error = in_pcbbind_remote(inp, nam, td);
1184                 if (error)
1185                         goto out;
1186         }
1187
1188         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1189                 /*
1190                  * Rarely used path:
1191                  * This inpcb was bound before this connect.
1192                  */
1193                 error = in_pcbladdr(inp, nam, &if_sin, td);
1194                 if (error)
1195                         goto out;
1196
1197                 /*
1198                  * Save or refresh the faddr/fport, since they may
1199                  * be changed by in_pcbladdr().
1200                  */
1201                 inp->inp_faddr = sin->sin_addr;
1202                 inp->inp_fport = sin->sin_port;
1203         }
1204 #ifdef INVARIANTS
1205         else {
1206                 KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1207                     ("faddr mismatch for reconnect"));
1208                 KASSERT(inp->inp_fport == sin->sin_port,
1209                     ("fport mismatch for reconnect"));
1210         }
1211 #endif
1212         KKASSERT(inp->inp_socket == so);
1213
1214         hash = tcp_addrhash(sin->sin_addr.s_addr, sin->sin_port,
1215                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1216                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1217                             inp->inp_lport);
1218         port = netisr_hashport(hash);
1219
1220         if (port != &curthread->td_msgport) {
1221                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1222
1223                 /*
1224                  * in_pcbladdr() may have allocated a route entry for us
1225                  * on the current CPU, but we need a route entry on the
1226                  * inpcb's owner CPU, so free it here.
1227                  */
1228                 in_pcbresetroute(inp);
1229
1230                 /*
1231                  * We are moving the protocol processing port the socket
1232                  * is on, we have to unlink here and re-link on the
1233                  * target cpu.
1234                  */
1235                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1236                 msg->connect.nm_flags |= PRUC_RECONNECT;
1237                 msg->connect.base.nm_dispatch = tcp_connect;
1238
1239                 TCP_STATE_MIGRATE_START(tp);
1240
1241                 /*
1242                  * Use message put done receipt to change this socket's
1243                  * so_port, i.e. _after_ this message was put onto the
1244                  * target netisr's msgport but _before_ the message could
1245                  * be pulled from the target netisr's msgport, so that:
1246                  * - The upper half (socket code) will not see the new
1247                  *   msgport before this message reaches the new msgport
1248                  *   and messages for this socket will be ordered.
1249                  * - This message will see the new msgport, when its
1250                  *   handler is called in the target netisr.
1251                  *
1252                  * NOTE:
1253                  * We MUST use messege put done receipt to change this
1254                  * socket's so_port:
1255                  * If we changed the so_port in this netisr after the
1256                  * lwkt_forwardmsg (so messages for this socket will be
1257                  * ordered) and changed the so_port in the target netisr
1258                  * at the very beginning of this message's handler, we
1259                  * would suffer so_port overwritten race, given this
1260                  * message might be forwarded again.
1261                  *
1262                  * NOTE:
1263                  * This mechanism depends on that the netisr's msgport
1264                  * is spin msgport (currently it is :).
1265                  *
1266                  * If the upper half saw the new msgport before this
1267                  * message reached the target netisr's msgport, the
1268                  * messages sent from the upper half could reach the new
1269                  * msgport before this message, thus there would be
1270                  * message reordering.  The worst case could be soclose()
1271                  * saw the new msgport and the detach message could reach
1272                  * the new msgport before this message, i.e. the inpcb
1273                  * could have been destroyed when this message was still
1274                  * pending on or on its way to the new msgport.  Other
1275                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1276                  * since we have unlinked this inpcb from the current
1277                  * pcbinfo first.
1278                  */
1279                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1280                 lwkt_forwardmsg(port, lmsg);
1281                 /* msg invalid now */
1282                 return;
1283         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1284                 /*
1285                  * The original thread is no longer needed; release it.
1286                  */
1287                 lwkt_rele(td);
1288                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1289         }
1290         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1291                                   msg->connect.nm_m, sin, if_sin, hash);
1292         msg->connect.nm_m = NULL;
1293 out:
1294         if (msg->connect.nm_m) {
1295                 m_freem(msg->connect.nm_m);
1296                 msg->connect.nm_m = NULL;
1297         }
1298         if (msg->connect.nm_flags & PRUC_HELDTD)
1299                 lwkt_rele(td);
1300         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1301                 so->so_error = error;
1302                 soisdisconnected(so);
1303         }
1304         lwkt_replymsg(&msg->connect.base.lmsg, error);
1305         /* msg invalid now */
1306 }
1307
1308 #ifdef INET6
1309
1310 static void
1311 tcp6_connect(netmsg_t msg)
1312 {
1313         struct tcpcb *tp;
1314         struct socket *so = msg->connect.base.nm_so;
1315         struct sockaddr *nam = msg->connect.nm_nam;
1316         struct thread *td = msg->connect.nm_td;
1317         struct inpcb *inp;
1318         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1319         struct in6_addr *addr6;
1320         lwkt_port_t port;
1321         int error;
1322
1323         COMMON_START(so, inp, 0);
1324
1325         /*
1326          * Reconnect our pcb if we have to
1327          */
1328         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1329                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1330                 TCP_STATE_MIGRATE_END(tp);
1331                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1332         }
1333
1334         /*
1335          * Bind if we have to
1336          */
1337         if (inp->inp_lport == 0) {
1338                 error = in6_pcbbind(inp, NULL, td);
1339                 if (error)
1340                         goto out;
1341         }
1342
1343         /*
1344          * Cannot simply call in_pcbconnect, because there might be an
1345          * earlier incarnation of this same connection still in
1346          * TIME_WAIT state, creating an ADDRINUSE error.
1347          */
1348         error = in6_pcbladdr(inp, nam, &addr6, td);
1349         if (error)
1350                 goto out;
1351
1352         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1353
1354         if (port != &curthread->td_msgport) {
1355                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1356
1357                 /*
1358                  * in_pcbladdr() may have allocated a route entry for us
1359                  * on the current CPU, but we need a route entry on the
1360                  * inpcb's owner CPU, so free it here.
1361                  */
1362                 in_pcbresetroute(inp);
1363
1364                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1365                 msg->connect.nm_flags |= PRUC_RECONNECT;
1366                 msg->connect.base.nm_dispatch = tcp6_connect;
1367
1368                 TCP_STATE_MIGRATE_START(tp);
1369
1370                 /* See the related comment in tcp_connect() */
1371                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1372                 lwkt_forwardmsg(port, lmsg);
1373                 /* msg invalid now */
1374                 return;
1375         }
1376         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1377                                    &msg->connect.nm_m, sin6, addr6);
1378         /* nm_m may still be intact */
1379 out:
1380         if (msg->connect.nm_m) {
1381                 m_freem(msg->connect.nm_m);
1382                 msg->connect.nm_m = NULL;
1383         }
1384         lwkt_replymsg(&msg->connect.base.lmsg, error);
1385         /* msg invalid now */
1386 }
1387
1388 static int
1389 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1390                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1391 {
1392         struct mbuf *m = *mp;
1393         struct inpcb *inp = tp->t_inpcb;
1394         struct socket *so = inp->inp_socket;
1395         struct inpcb *oinp;
1396
1397         /*
1398          * Cannot simply call in_pcbconnect, because there might be an
1399          * earlier incarnation of this same connection still in
1400          * TIME_WAIT state, creating an ADDRINUSE error.
1401          */
1402         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1403                                   &sin6->sin6_addr, sin6->sin6_port,
1404                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1405                                       addr6 : &inp->in6p_laddr),
1406                                   inp->inp_lport,  0, NULL);
1407         if (oinp)
1408                 return (EADDRINUSE);
1409
1410         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1411                 inp->in6p_laddr = *addr6;
1412         inp->in6p_faddr = sin6->sin6_addr;
1413         inp->inp_fport = sin6->sin6_port;
1414         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1415                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1416         in_pcbinsconnhash(inp);
1417
1418         /*
1419          * Now that no more errors can occur, change the protocol processing
1420          * port to the current thread (which is the correct thread).
1421          *
1422          * Create TCP timer message now; we are on the tcpcb's owner
1423          * CPU/thread.
1424          */
1425         tcp_create_timermsg(tp, &curthread->td_msgport);
1426
1427         /* Compute window scaling to request.  */
1428         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1429                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1430         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1431             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1432                 tp->request_r_scale++;
1433         }
1434
1435         soisconnecting(so);
1436         tcpstat.tcps_connattempt++;
1437         TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1438         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1439         tp->iss = tcp_new_isn(tp);
1440         tcp_sendseqinit(tp);
1441         if (m) {
1442                 ssb_appendstream(&so->so_snd, m);
1443                 *mp = NULL;
1444                 if (flags & PRUS_OOB)
1445                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1446         }
1447
1448         /*
1449          * Close the send side of the connection after
1450          * the data is sent if flagged.
1451          */
1452         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1453                 socantsendmore(so);
1454                 tp = tcp_usrclosed(tp);
1455         }
1456         return (tcp_output(tp));
1457 }
1458
1459 #endif /* INET6 */
1460
1461 /*
1462  * The new sockopt interface makes it possible for us to block in the
1463  * copyin/out step (if we take a page fault).  Taking a page fault while
1464  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1465  * both now use TSM, there probably isn't any need for this function to
1466  * run in a critical section any more.  This needs more examination.)
1467  */
1468 void
1469 tcp_ctloutput(netmsg_t msg)
1470 {
1471         struct socket *so = msg->base.nm_so;
1472         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1473         struct thread *td = NULL;
1474         int     error, opt, optval, opthz;
1475         struct  inpcb *inp;
1476         struct  tcpcb *tp;
1477
1478         if (msg->ctloutput.nm_flags & PRCO_HELDTD)
1479                 td = sopt->sopt_td;
1480
1481         error = 0;
1482         inp = so->so_pcb;
1483         if (inp == NULL) {
1484                 error = ECONNRESET;
1485                 goto done;
1486         }
1487         tp = intotcpcb(inp);
1488
1489         /* Get socket's owner cpuid hint */
1490         if (sopt->sopt_level == SOL_SOCKET &&
1491             sopt->sopt_dir == SOPT_GET &&
1492             sopt->sopt_name == SO_CPUHINT) {
1493                 if (tp->t_flags & TF_LISTEN) {
1494                         /*
1495                          * Listen sockets owner cpuid is always 0,
1496                          * which does not make sense if SO_REUSEPORT
1497                          * is not set.
1498                          *
1499                          * NOTE: inp_lgrpindex is _not_ assigned in jail.
1500                          */
1501                         if ((so->so_options & SO_REUSEPORT) &&
1502                             inp->inp_lgrpindex >= 0)
1503                                 optval = inp->inp_lgrpindex % netisr_ncpus;
1504                         else
1505                                 optval = -1; /* no hint */
1506                 } else {
1507                         optval = mycpuid;
1508                 }
1509                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1510                 goto done;
1511         }
1512
1513         if (sopt->sopt_level != IPPROTO_TCP) {
1514                 if (sopt->sopt_level == IPPROTO_IP) {
1515                         switch (sopt->sopt_name) {
1516                         case IP_MULTICAST_IF:
1517                         case IP_MULTICAST_VIF:
1518                         case IP_MULTICAST_TTL:
1519                         case IP_MULTICAST_LOOP:
1520                         case IP_ADD_MEMBERSHIP:
1521                         case IP_DROP_MEMBERSHIP:
1522                                 /*
1523                                  * Multicast does not make sense on
1524                                  * TCP sockets.
1525                                  */
1526                                 error = EOPNOTSUPP;
1527                                 goto done;
1528                         }
1529                 }
1530 #ifdef INET6
1531                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1532                         ip6_ctloutput_dispatch(msg);
1533                 else
1534 #endif /* INET6 */
1535                 ip_ctloutput(msg);
1536                 /* msg invalid now */
1537                 if (td != NULL)
1538                         lwkt_rele(td);
1539                 return;
1540         }
1541
1542         switch (sopt->sopt_dir) {
1543         case SOPT_SET:
1544                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1545                                       sizeof optval);
1546                 if (error)
1547                         break;
1548                 switch (sopt->sopt_name) {
1549                 case TCP_FASTKEEP:
1550                         if (optval > 0)
1551                                 tp->t_keepidle = tp->t_keepintvl;
1552                         else
1553                                 tp->t_keepidle = tcp_keepidle;
1554                         tcp_timer_keep_activity(tp, 0);
1555                         break;
1556 #ifdef TCP_SIGNATURE
1557                 case TCP_SIGNATURE_ENABLE:
1558                         if (tp->t_state == TCPS_CLOSED) {
1559                                 /*
1560                                  * This is the only safe state that this
1561                                  * option could be changed.  Some segments
1562                                  * could already have been sent in other
1563                                  * states.
1564                                  */
1565                                 if (optval > 0)
1566                                         tp->t_flags |= TF_SIGNATURE;
1567                                 else
1568                                         tp->t_flags &= ~TF_SIGNATURE;
1569                         } else {
1570                                 error = EOPNOTSUPP;
1571                         }
1572                         break;
1573 #endif /* TCP_SIGNATURE */
1574                 case TCP_NODELAY:
1575                 case TCP_NOOPT:
1576                         switch (sopt->sopt_name) {
1577                         case TCP_NODELAY:
1578                                 opt = TF_NODELAY;
1579                                 break;
1580                         case TCP_NOOPT:
1581                                 opt = TF_NOOPT;
1582                                 break;
1583                         default:
1584                                 opt = 0; /* dead code to fool gcc */
1585                                 break;
1586                         }
1587
1588                         if (optval)
1589                                 tp->t_flags |= opt;
1590                         else
1591                                 tp->t_flags &= ~opt;
1592                         break;
1593
1594                 case TCP_NOPUSH:
1595                         if (tcp_disable_nopush)
1596                                 break;
1597                         if (optval)
1598                                 tp->t_flags |= TF_NOPUSH;
1599                         else {
1600                                 tp->t_flags &= ~TF_NOPUSH;
1601                                 error = tcp_output(tp);
1602                         }
1603                         break;
1604
1605                 case TCP_MAXSEG:
1606                         /*
1607                          * Must be between 0 and maxseg.  If the requested
1608                          * maxseg is too small to satisfy the desired minmss,
1609                          * pump it up (silently so sysctl modifications of
1610                          * minmss do not create unexpected program failures).
1611                          * Handle degenerate cases.
1612                          */
1613                         if (optval > 0 && optval <= tp->t_maxseg) {
1614                                 if (optval + 40 < tcp_minmss) {
1615                                         optval = tcp_minmss - 40;
1616                                         if (optval < 0)
1617                                                 optval = 1;
1618                                 }
1619                                 tp->t_maxseg = optval;
1620                         } else {
1621                                 error = EINVAL;
1622                         }
1623                         break;
1624
1625                 case TCP_KEEPINIT:
1626                         opthz = ((int64_t)optval * hz) / 1000;
1627                         if (opthz >= 1)
1628                                 tp->t_keepinit = opthz;
1629                         else
1630                                 error = EINVAL;
1631                         break;
1632
1633                 case TCP_KEEPIDLE:
1634                         opthz = ((int64_t)optval * hz) / 1000;
1635                         if (opthz >= 1) {
1636                                 tp->t_keepidle = opthz;
1637                                 tcp_timer_keep_activity(tp, 0);
1638                         } else {
1639                                 error = EINVAL;
1640                         }
1641                         break;
1642
1643                 case TCP_KEEPINTVL:
1644                         opthz = ((int64_t)optval * hz) / 1000;
1645                         if (opthz >= 1) {
1646                                 tp->t_keepintvl = opthz;
1647                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1648                         } else {
1649                                 error = EINVAL;
1650                         }
1651                         break;
1652
1653                 case TCP_KEEPCNT:
1654                         if (optval > 0) {
1655                                 tp->t_keepcnt = optval;
1656                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1657                         } else {
1658                                 error = EINVAL;
1659                         }
1660                         break;
1661
1662                 default:
1663                         error = ENOPROTOOPT;
1664                         break;
1665                 }
1666                 break;
1667
1668         case SOPT_GET:
1669                 switch (sopt->sopt_name) {
1670 #ifdef TCP_SIGNATURE
1671                 case TCP_SIGNATURE_ENABLE:
1672                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1673                         break;
1674 #endif /* TCP_SIGNATURE */
1675                 case TCP_NODELAY:
1676                         optval = tp->t_flags & TF_NODELAY;
1677                         break;
1678                 case TCP_MAXSEG:
1679                         optval = tp->t_maxseg;
1680                         break;
1681                 case TCP_NOOPT:
1682                         optval = tp->t_flags & TF_NOOPT;
1683                         break;
1684                 case TCP_NOPUSH:
1685                         optval = tp->t_flags & TF_NOPUSH;
1686                         break;
1687                 case TCP_KEEPINIT:
1688                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1689                         break;
1690                 case TCP_KEEPIDLE:
1691                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1692                         break;
1693                 case TCP_KEEPINTVL:
1694                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1695                         break;
1696                 case TCP_KEEPCNT:
1697                         optval = tp->t_keepcnt;
1698                         break;
1699                 default:
1700                         error = ENOPROTOOPT;
1701                         break;
1702                 }
1703                 if (error == 0)
1704                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1705                 break;
1706         }
1707 done:
1708         if (td != NULL)
1709                 lwkt_rele(td);
1710         lwkt_replymsg(&msg->lmsg, error);
1711 }
1712
1713 struct netmsg_tcp_ctloutput {
1714         struct netmsg_pr_ctloutput ctloutput;
1715         struct sockopt          sopt;
1716         int                     sopt_val;
1717 };
1718
1719 /*
1720  * Allocate netmsg_pr_ctloutput for asynchronous tcp_ctloutput.
1721  */
1722 struct netmsg_pr_ctloutput *
1723 tcp_ctloutmsg(struct sockopt *sopt)
1724 {
1725         struct netmsg_tcp_ctloutput *msg;
1726         int flags = 0, error;
1727
1728         KASSERT(sopt->sopt_dir == SOPT_SET, ("not from ctloutput"));
1729
1730         /* Only small set of options allows asynchronous setting. */
1731         if (sopt->sopt_level != IPPROTO_TCP)
1732                 return NULL;
1733         switch (sopt->sopt_name) {
1734         case TCP_NODELAY:
1735         case TCP_NOOPT:
1736         case TCP_NOPUSH:
1737         case TCP_FASTKEEP:
1738                 break;
1739         default:
1740                 return NULL;
1741         }
1742
1743         msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_NULLOK);
1744         if (msg == NULL) {
1745                 /* Fallback to synchronous tcp_ctloutput */
1746                 return NULL;
1747         }
1748
1749         /* Save the sockopt */
1750         msg->sopt = *sopt;
1751
1752         /* Fixup the sopt.sopt_val ptr */
1753         error = sooptcopyin(sopt, &msg->sopt_val,
1754             sizeof(msg->sopt_val), sizeof(msg->sopt_val));
1755         if (error) {
1756                 kfree(msg, M_LWKTMSG);
1757                 return NULL;
1758         }
1759         msg->sopt.sopt_val = &msg->sopt_val;
1760
1761         /* Hold the current thread */
1762         if (msg->sopt.sopt_td != NULL) {
1763                 flags |= PRCO_HELDTD;
1764                 lwkt_hold(msg->sopt.sopt_td);
1765         }
1766
1767         msg->ctloutput.nm_flags = flags;
1768         msg->ctloutput.nm_sopt = &msg->sopt;
1769
1770         return &msg->ctloutput;
1771 }
1772
1773 /*
1774  * tcp_sendspace and tcp_recvspace are the default send and receive window
1775  * sizes, respectively.  These are obsolescent (this information should
1776  * be set by the route).
1777  *
1778  * Use a default that does not require tcp window scaling to be turned
1779  * on.  Individual programs or the administrator can increase the default.
1780  */
1781 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1782 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1783     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1784 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1785 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1786     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1787
1788 /*
1789  * Attach TCP protocol to socket, allocating internet protocol control
1790  * block, tcp control block, buffer space, and entering CLOSED state.
1791  */
1792 static int
1793 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1794 {
1795         struct inpcb *inp;
1796         int error;
1797         int cpu;
1798 #ifdef INET6
1799         boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1800 #endif
1801
1802         if (ai != NULL) {
1803                 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1804                 if (error)
1805                         return (error);
1806         } else {
1807                 /* Post attach; do nothing */
1808         }
1809
1810         cpu = mycpu->gd_cpuid;
1811
1812         /*
1813          * Set the default pcbinfo.  This will likely change when we
1814          * bind/connect.
1815          */
1816         error = in_pcballoc(so, &tcbinfo[cpu]);
1817         if (error)
1818                 return (error);
1819         inp = so->so_pcb;
1820 #ifdef INET6
1821         if (isipv6)
1822                 inp->in6p_hops = -1;    /* use kernel default */
1823 #endif
1824         tcp_newtcpcb(inp);
1825         /* Keep a reference for asynchronized pru_rcvd */
1826         soreference(so);
1827         return (0);
1828 }
1829
1830 /*
1831  * Initiate (or continue) disconnect.
1832  * If embryonic state, just send reset (once).
1833  * If in ``let data drain'' option and linger null, just drop.
1834  * Otherwise (hard), mark socket disconnecting and drop
1835  * current input data; switch states based on user close, and
1836  * send segment to peer (with FIN).
1837  */
1838 static struct tcpcb *
1839 tcp_disconnect(struct tcpcb *tp)
1840 {
1841         struct socket *so = tp->t_inpcb->inp_socket;
1842
1843         if (tp->t_state < TCPS_ESTABLISHED) {
1844                 tp = tcp_close(tp);
1845         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1846                 tp = tcp_drop(tp, 0);
1847         } else {
1848                 lwkt_gettoken(&so->so_rcv.ssb_token);
1849                 soisdisconnecting(so);
1850                 sbflush(&so->so_rcv.sb);
1851                 tp = tcp_usrclosed(tp);
1852                 if (tp)
1853                         tcp_output(tp);
1854                 lwkt_reltoken(&so->so_rcv.ssb_token);
1855         }
1856         return (tp);
1857 }
1858
1859 /*
1860  * User issued close, and wish to trail through shutdown states:
1861  * if never received SYN, just forget it.  If got a SYN from peer,
1862  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1863  * If already got a FIN from peer, then almost done; go to LAST_ACK
1864  * state.  In all other cases, have already sent FIN to peer (e.g.
1865  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1866  * for peer to send FIN or not respond to keep-alives, etc.
1867  * We can let the user exit from the close as soon as the FIN is acked.
1868  */
1869 static struct tcpcb *
1870 tcp_usrclosed(struct tcpcb *tp)
1871 {
1872
1873         switch (tp->t_state) {
1874
1875         case TCPS_CLOSED:
1876         case TCPS_LISTEN:
1877                 TCP_STATE_CHANGE(tp, TCPS_CLOSED);
1878                 tp = tcp_close(tp);
1879                 break;
1880
1881         case TCPS_SYN_SENT:
1882         case TCPS_SYN_RECEIVED:
1883                 tp->t_flags |= TF_NEEDFIN;
1884                 break;
1885
1886         case TCPS_ESTABLISHED:
1887                 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1888                 break;
1889
1890         case TCPS_CLOSE_WAIT:
1891                 TCP_STATE_CHANGE(tp, TCPS_LAST_ACK);
1892                 break;
1893         }
1894         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1895                 soisdisconnected(tp->t_inpcb->inp_socket);
1896                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1897                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1898                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1899                             tcp_timer_2msl);
1900                 }
1901         }
1902         return (tp);
1903 }