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