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. All advertising materials mentioning features or use of this software
  47  *    must display the following acknowledgement:
  48  *      This product includes software developed by the University of
  49  *      California, Berkeley and its contributors.
  50  * 4. Neither the name of the University nor the names of its contributors
  51  *    may be used to endorse or promote products derived from this software
  52  *    without specific prior written permission.
  53  *
  54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  64  * SUCH DAMAGE.
  65  *
  66  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  67  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  68  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
  69  */
  70
  71 #include "opt_ipsec.h"
  72 #include "opt_inet.h"
  73 #include "opt_inet6.h"
  74 #include "opt_tcpdebug.h"
  75
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/kernel.h>
  79 #include <sys/malloc.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/globaldata.h>
  82 #include <sys/thread.h>
  83
  84 #include <sys/mbuf.h>
  85 #ifdef INET6
  86 #include <sys/domain.h>
  87 #endif /* INET6 */
  88 #include <sys/socket.h>
  89 #include <sys/socketvar.h>
  90 #include <sys/protosw.h>
  91
  92 #include <sys/thread2.h>
  93 #include <sys/msgport2.h>
  94 #include <sys/socketvar2.h>
  95
  96 #include <net/if.h>
  97 #include <net/netisr.h>
  98 #include <net/route.h>
  99
 100 #include <net/netmsg2.h>
 101
 102 #include <netinet/in.h>
 103 #include <netinet/in_systm.h>
 104 #ifdef INET6
 105 #include <netinet/ip6.h>
 106 #endif
 107 #include <netinet/in_pcb.h>
 108 #ifdef INET6
 109 #include <netinet6/in6_pcb.h>
 110 #endif
 111 #include <netinet/in_var.h>
 112 #include <netinet/ip_var.h>
 113 #ifdef INET6
 114 #include <netinet6/ip6_var.h>
 115 #include <netinet6/tcp6_var.h>
 116 #endif
 117 #include <netinet/tcp.h>
 118 #include <netinet/tcp_fsm.h>
 119 #include <netinet/tcp_seq.h>
 120 #include <netinet/tcp_timer.h>
 121 #include <netinet/tcp_timer2.h>
 122 #include <netinet/tcp_var.h>
 123 #include <netinet/tcpip.h>
 124 #ifdef TCPDEBUG
 125 #include <netinet/tcp_debug.h>
 126 #endif
 127
 128 #ifdef IPSEC
 129 #include <netinet6/ipsec.h>
 130 #endif /*IPSEC*/
 131
 132 /*
 133  * TCP protocol interface to socket abstraction.
 134  */
 135 extern  char *tcpstates[];      /* XXX ??? */
 136
 137 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 138 static void     tcp_connect (netmsg_t msg);
 139 #ifdef INET6
 140 static void     tcp6_connect (netmsg_t msg);
 141 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 142                                 struct mbuf **mp,
 143                                 struct sockaddr_in6 *sin6,
 144                                 struct in6_addr *addr6);
 145 #endif /* INET6 */
 146 static struct tcpcb *
 147                 tcp_disconnect (struct tcpcb *);
 148 static struct tcpcb *
 149                 tcp_usrclosed (struct tcpcb *);
 150
 151 #ifdef TCPDEBUG
 152 #define TCPDEBUG0       int ostate = 0
 153 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 154 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 155                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 156 #else
 157 #define TCPDEBUG0
 158 #define TCPDEBUG1()
 159 #define TCPDEBUG2(req)
 160 #endif
 161
 162 static int      tcp_lport_extension = 1;
 163
 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 165     &tcp_lport_extension, 0, "");
 166
 167 /*
 168  * TCP attaches to socket via pru_attach(), reserving space,
 169  * and an internet control block.  This is likely occuring on
 170  * cpu0 and may have to move later when we bind/connect.
 171  */
 172 static void
 173 tcp_usr_attach(netmsg_t msg)
 174 {
 175         struct socket *so = msg->base.nm_so;
 176         struct pru_attach_info *ai = msg->attach.nm_ai;
 177         int error;
 178         struct inpcb *inp;
 179         struct tcpcb *tp = 0;
 180         TCPDEBUG0;
 181
 182         soreference(so);
 183         inp = so->so_pcb;
 184         TCPDEBUG1();
 185         if (inp) {
 186                 error = EISCONN;
 187                 goto out;
 188         }
 189
 190         error = tcp_attach(so, ai);
 191         if (error)
 192                 goto out;
 193
 194         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 195                 so->so_linger = TCP_LINGERTIME;
 196         tp = sototcpcb(so);
 197 out:
 198         sofree(so);             /* from ref above */
 199         TCPDEBUG2(PRU_ATTACH);
 200         lwkt_replymsg(&msg->lmsg, error);
 201 }
 202
 203 /*
 204  * pru_detach() detaches the TCP protocol from the socket.
 205  * If the protocol state is non-embryonic, then can't
 206  * do this directly: have to initiate a pru_disconnect(),
 207  * which may finish later; embryonic TCB's can just
 208  * be discarded here.
 209  */
 210 static void
 211 tcp_usr_detach(netmsg_t msg)
 212 {
 213         struct socket *so = msg->base.nm_so;
 214         int error = 0;
 215         struct inpcb *inp;
 216         struct tcpcb *tp;
 217         TCPDEBUG0;
 218
 219         inp = so->so_pcb;
 220
 221         /*
 222          * If the inp is already detached it may have been due to an async
 223          * close.  Just return as if no error occured.
 224          *
 225          * It's possible for the tcpcb (tp) to disconnect from the inp due
 226          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 227          * the detach message has been queued so we may find a NULL tp here.
 228          */
 229         if (inp) {
 230                 if ((tp = intotcpcb(inp)) != NULL) {
 231                         TCPDEBUG1();
 232                         tp = tcp_disconnect(tp);
 233                         TCPDEBUG2(PRU_DETACH);
 234                 }
 235         }
 236         lwkt_replymsg(&msg->lmsg, error);
 237 }
 238
 239 /*
 240  * NOTE: ignore_error is non-zero for certain disconnection races
 241  * which we want to silently allow, otherwise close() may return
 242  * an unexpected error.
 243  *
 244  * NOTE: The variables (msg) and (tp) are assumed.
 245  */
 246 #define COMMON_START(so, inp, ignore_error)                     \
 247         TCPDEBUG0;                                              \
 248                                                                 \
 249         inp = so->so_pcb;                                       \
 250         do {                                                    \
 251                  if (inp == NULL) {                             \
 252                         error = ignore_error ? 0 : EINVAL;      \
 253                         tp = NULL;                              \
 254                         goto out;                               \
 255                  }                                              \
 256                  tp = intotcpcb(inp);                           \
 257                  TCPDEBUG1();                                   \
 258         } while(0)
 259
 260 #define COMMON_END1(req, noreply)                               \
 261         out: do {                                               \
 262                 TCPDEBUG2(req);                                 \
 263                 if (!(noreply))                                 \
 264                         lwkt_replymsg(&msg->lmsg, error);       \
 265                 return;                                         \
 266         } while(0)
 267
 268 #define COMMON_END(req)         COMMON_END1((req), 0)
 269
 270 /*
 271  * Give the socket an address.
 272  */
 273 static void
 274 tcp_usr_bind(netmsg_t msg)
 275 {
 276         struct socket *so = msg->bind.base.nm_so;
 277         struct sockaddr *nam = msg->bind.nm_nam;
 278         struct thread *td = msg->bind.nm_td;
 279         int error = 0;
 280         struct inpcb *inp;
 281         struct tcpcb *tp;
 282         struct sockaddr_in *sinp;
 283
 284         COMMON_START(so, inp, 0);
 285
 286         /*
 287          * Must check for multicast addresses and disallow binding
 288          * to them.
 289          */
 290         sinp = (struct sockaddr_in *)nam;
 291         if (sinp->sin_family == AF_INET &&
 292             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 293                 error = EAFNOSUPPORT;
 294                 goto out;
 295         }
 296         error = in_pcbbind(inp, nam, td);
 297         if (error)
 298                 goto out;
 299         COMMON_END(PRU_BIND);
 300
 301 }
 302
 303 #ifdef INET6
 304
 305 static void
 306 tcp6_usr_bind(netmsg_t msg)
 307 {
 308         struct socket *so = msg->bind.base.nm_so;
 309         struct sockaddr *nam = msg->bind.nm_nam;
 310         struct thread *td = msg->bind.nm_td;
 311         int error = 0;
 312         struct inpcb *inp;
 313         struct tcpcb *tp;
 314         struct sockaddr_in6 *sin6p;
 315
 316         COMMON_START(so, inp, 0);
 317
 318         /*
 319          * Must check for multicast addresses and disallow binding
 320          * to them.
 321          */
 322         sin6p = (struct sockaddr_in6 *)nam;
 323         if (sin6p->sin6_family == AF_INET6 &&
 324             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 325                 error = EAFNOSUPPORT;
 326                 goto out;
 327         }
 328         inp->inp_vflag &= ~INP_IPV4;
 329         inp->inp_vflag |= INP_IPV6;
 330         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 331                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 332                         inp->inp_vflag |= INP_IPV4;
 333                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 334                         struct sockaddr_in sin;
 335
 336                         in6_sin6_2_sin(&sin, sin6p);
 337                         inp->inp_vflag |= INP_IPV4;
 338                         inp->inp_vflag &= ~INP_IPV6;
 339                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 340                         goto out;
 341                 }
 342         }
 343         error = in6_pcbbind(inp, nam, td);
 344         if (error)
 345                 goto out;
 346         COMMON_END(PRU_BIND);
 347 }
 348 #endif /* INET6 */
 349
 350 #ifdef SMP
 351
 352 struct netmsg_inswildcard {
 353         struct netmsg_base      base;
 354         struct inpcb            *nm_inp;
 355 };
 356
 357 static void
 358 in_pcbinswildcardhash_handler(netmsg_t msg)
 359 {
 360         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 361         int cpu = mycpuid, nextcpu;
 362
 363         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 364
 365         nextcpu = cpu + 1;
 366         if (nextcpu < ncpus2)
 367                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
 368         else
 369                 lwkt_replymsg(&nm->base.lmsg, 0);
 370 }
 371
 372 #endif
 373
 374 /*
 375  * Prepare to accept connections.
 376  */
 377 static void
 378 tcp_usr_listen(netmsg_t msg)
 379 {
 380         struct socket *so = msg->listen.base.nm_so;
 381         struct thread *td = msg->listen.nm_td;
 382         int error = 0;
 383         struct inpcb *inp;
 384         struct tcpcb *tp;
 385 #ifdef SMP
 386         struct netmsg_inswildcard nm;
 387 #endif
 388
 389         COMMON_START(so, inp, 0);
 390
 391         if (tp->t_flags & TF_LISTEN)
 392                 goto out;
 393
 394         if (inp->inp_lport == 0) {
 395                 error = in_pcbbind(inp, NULL, td);
 396                 if (error)
 397                         goto out;
 398         }
 399
 400         tp->t_state = TCPS_LISTEN;
 401         tp->t_flags |= TF_LISTEN;
 402         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 403
 404 #ifdef SMP
 405         if (ncpus > 1) {
 406                 /*
 407                  * We have to set the flag because we can't have other cpus
 408                  * messing with our inp's flags.
 409                  */
 410                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 411                         ("already on connhash\n"));
 412                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 413                         ("already on wildcardhash\n"));
 414                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 415                         ("already on MP wildcardhash\n"));
 416                 inp->inp_flags |= INP_WILDCARD_MP;
 417
 418                 KKASSERT(so->so_port == cpu_portfn(0));
 419                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 420                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 421
 422                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 423                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 424                 nm.nm_inp = inp;
 425                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 426         }
 427 #endif
 428         in_pcbinswildcardhash(inp);
 429         COMMON_END(PRU_LISTEN);
 430 }
 431
 432 #ifdef INET6
 433
 434 static void
 435 tcp6_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 #ifdef SMP
 443         struct netmsg_inswildcard nm;
 444 #endif
 445
 446         COMMON_START(so, inp, 0);
 447
 448         if (tp->t_flags & TF_LISTEN)
 449                 goto out;
 450
 451         if (inp->inp_lport == 0) {
 452                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 453                         inp->inp_vflag |= INP_IPV4;
 454                 else
 455                         inp->inp_vflag &= ~INP_IPV4;
 456                 error = in6_pcbbind(inp, NULL, td);
 457                 if (error)
 458                         goto out;
 459         }
 460
 461         tp->t_state = TCPS_LISTEN;
 462         tp->t_flags |= TF_LISTEN;
 463         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 464
 465 #ifdef SMP
 466         if (ncpus > 1) {
 467                 /*
 468                  * We have to set the flag because we can't have other cpus
 469                  * messing with our inp's flags.
 470                  */
 471                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 472                         ("already on connhash\n"));
 473                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 474                         ("already on wildcardhash\n"));
 475                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 476                         ("already on MP wildcardhash\n"));
 477                 inp->inp_flags |= INP_WILDCARD_MP;
 478
 479                 KKASSERT(so->so_port == cpu_portfn(0));
 480                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 481                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 482
 483                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 484                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 485                 nm.nm_inp = inp;
 486                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 487         }
 488 #endif
 489         in_pcbinswildcardhash(inp);
 490         COMMON_END(PRU_LISTEN);
 491 }
 492 #endif /* INET6 */
 493
 494 /*
 495  * Initiate connection to peer.
 496  * Create a template for use in transmissions on this connection.
 497  * Enter SYN_SENT state, and mark socket as connecting.
 498  * Start keep-alive timer, and seed output sequence space.
 499  * Send initial segment on connection.
 500  */
 501 static void
 502 tcp_usr_connect(netmsg_t msg)
 503 {
 504         struct socket *so = msg->connect.base.nm_so;
 505         struct sockaddr *nam = msg->connect.nm_nam;
 506         struct thread *td = msg->connect.nm_td;
 507         int error = 0;
 508         struct inpcb *inp;
 509         struct tcpcb *tp;
 510         struct sockaddr_in *sinp;
 511
 512         COMMON_START(so, inp, 0);
 513
 514         /*
 515          * Must disallow TCP ``connections'' to multicast addresses.
 516          */
 517         sinp = (struct sockaddr_in *)nam;
 518         if (sinp->sin_family == AF_INET
 519             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 520                 error = EAFNOSUPPORT;
 521                 goto out;
 522         }
 523
 524         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 525                 error = EAFNOSUPPORT; /* IPv6 only jail */
 526                 goto out;
 527         }
 528
 529         tcp_connect(msg);
 530         /* msg is invalid now */
 531         return;
 532 out:
 533         if (msg->connect.nm_m) {
 534                 m_freem(msg->connect.nm_m);
 535                 msg->connect.nm_m = NULL;
 536         }
 537         lwkt_replymsg(&msg->lmsg, error);
 538 }
 539
 540 #ifdef INET6
 541
 542 static void
 543 tcp6_usr_connect(netmsg_t msg)
 544 {
 545         struct socket *so = msg->connect.base.nm_so;
 546         struct sockaddr *nam = msg->connect.nm_nam;
 547         struct thread *td = msg->connect.nm_td;
 548         int error = 0;
 549         struct inpcb *inp;
 550         struct tcpcb *tp;
 551         struct sockaddr_in6 *sin6p;
 552
 553         COMMON_START(so, inp, 0);
 554
 555         /*
 556          * Must disallow TCP ``connections'' to multicast addresses.
 557          */
 558         sin6p = (struct sockaddr_in6 *)nam;
 559         if (sin6p->sin6_family == AF_INET6
 560             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 561                 error = EAFNOSUPPORT;
 562                 goto out;
 563         }
 564
 565         if (!prison_remote_ip(td, nam)) {
 566                 error = EAFNOSUPPORT; /* IPv4 only jail */
 567                 goto out;
 568         }
 569
 570         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 571                 struct sockaddr_in *sinp;
 572
 573                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 574                         error = EINVAL;
 575                         goto out;
 576                 }
 577                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 578                 in6_sin6_2_sin(sinp, sin6p);
 579                 inp->inp_vflag |= INP_IPV4;
 580                 inp->inp_vflag &= ~INP_IPV6;
 581                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 582                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
 583                 tcp_connect(msg);
 584                 /* msg is invalid now */
 585                 return;
 586         }
 587         inp->inp_vflag &= ~INP_IPV4;
 588         inp->inp_vflag |= INP_IPV6;
 589         inp->inp_inc.inc_isipv6 = 1;
 590
 591         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
 592         tcp6_connect(msg);
 593         /* msg is invalid now */
 594         return;
 595 out:
 596         if (msg->connect.nm_m) {
 597                 m_freem(msg->connect.nm_m);
 598                 msg->connect.nm_m = NULL;
 599         }
 600         lwkt_replymsg(&msg->lmsg, error);
 601 }
 602
 603 #endif /* INET6 */
 604
 605 /*
 606  * Initiate disconnect from peer.
 607  * If connection never passed embryonic stage, just drop;
 608  * else if don't need to let data drain, then can just drop anyways,
 609  * else have to begin TCP shutdown process: mark socket disconnecting,
 610  * drain unread data, state switch to reflect user close, and
 611  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 612  * when peer sends FIN and acks ours.
 613  *
 614  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 615  */
 616 static void
 617 tcp_usr_disconnect(netmsg_t msg)
 618 {
 619         struct socket *so = msg->disconnect.base.nm_so;
 620         int error = 0;
 621         struct inpcb *inp;
 622         struct tcpcb *tp;
 623
 624         COMMON_START(so, inp, 1);
 625         tp = tcp_disconnect(tp);
 626         COMMON_END(PRU_DISCONNECT);
 627 }
 628
 629 /*
 630  * Accept a connection.  Essentially all the work is
 631  * done at higher levels; just return the address
 632  * of the peer, storing through addr.
 633  */
 634 static void
 635 tcp_usr_accept(netmsg_t msg)
 636 {
 637         struct socket *so = msg->accept.base.nm_so;
 638         struct sockaddr **nam = msg->accept.nm_nam;
 639         int error = 0;
 640         struct inpcb *inp;
 641         struct tcpcb *tp = NULL;
 642         TCPDEBUG0;
 643
 644         inp = so->so_pcb;
 645         if (so->so_state & SS_ISDISCONNECTED) {
 646                 error = ECONNABORTED;
 647                 goto out;
 648         }
 649         if (inp == 0) {
 650                 error = EINVAL;
 651                 goto out;
 652         }
 653
 654         tp = intotcpcb(inp);
 655         TCPDEBUG1();
 656         in_setpeeraddr(so, nam);
 657         COMMON_END(PRU_ACCEPT);
 658 }
 659
 660 #ifdef INET6
 661 static void
 662 tcp6_usr_accept(netmsg_t msg)
 663 {
 664         struct socket *so = msg->accept.base.nm_so;
 665         struct sockaddr **nam = msg->accept.nm_nam;
 666         int error = 0;
 667         struct inpcb *inp;
 668         struct tcpcb *tp = NULL;
 669         TCPDEBUG0;
 670
 671         inp = so->so_pcb;
 672
 673         if (so->so_state & SS_ISDISCONNECTED) {
 674                 error = ECONNABORTED;
 675                 goto out;
 676         }
 677         if (inp == 0) {
 678                 error = EINVAL;
 679                 goto out;
 680         }
 681         tp = intotcpcb(inp);
 682         TCPDEBUG1();
 683         in6_mapped_peeraddr(so, nam);
 684         COMMON_END(PRU_ACCEPT);
 685 }
 686 #endif /* INET6 */
 687 /*
 688  * Mark the connection as being incapable of further output.
 689  */
 690 static void
 691 tcp_usr_shutdown(netmsg_t msg)
 692 {
 693         struct socket *so = msg->shutdown.base.nm_so;
 694         int error = 0;
 695         struct inpcb *inp;
 696         struct tcpcb *tp;
 697
 698         COMMON_START(so, inp, 0);
 699         socantsendmore(so);
 700         tp = tcp_usrclosed(tp);
 701         if (tp)
 702                 error = tcp_output(tp);
 703         COMMON_END(PRU_SHUTDOWN);
 704 }
 705
 706 /*
 707  * After a receive, possibly send window update to peer.
 708  */
 709 static void
 710 tcp_usr_rcvd(netmsg_t msg)
 711 {
 712         struct socket *so = msg->rcvd.base.nm_so;
 713         int error = 0;
 714         struct inpcb *inp;
 715         struct tcpcb *tp;
 716
 717         COMMON_START(so, inp, 0);
 718         tcp_output(tp);
 719         COMMON_END(PRU_RCVD);
 720 }
 721
 722 /*
 723  * Do a send by putting data in output queue and updating urgent
 724  * marker if URG set.  Possibly send more data.  Unlike the other
 725  * pru_*() routines, the mbuf chains are our responsibility.  We
 726  * must either enqueue them or free them.  The other pru_* routines
 727  * generally are caller-frees.
 728  */
 729 static void
 730 tcp_usr_send(netmsg_t msg)
 731 {
 732         struct socket *so = msg->send.base.nm_so;
 733         int flags = msg->send.nm_flags;
 734         struct mbuf *m = msg->send.nm_m;
 735         struct mbuf *control = msg->send.nm_control;
 736         int error = 0;
 737         struct inpcb *inp;
 738         struct tcpcb *tp;
 739         TCPDEBUG0;
 740
 741         KKASSERT(control == NULL);
 742
 743         inp = so->so_pcb;
 744
 745         if (inp == NULL) {
 746                 /*
 747                  * OOPS! we lost a race, the TCP session got reset after
 748                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 749                  * network interrupt in the non-critical section of sosend().
 750                  */
 751                 m_freem(m);
 752                 error = ECONNRESET;     /* XXX EPIPE? */
 753                 tp = NULL;
 754                 TCPDEBUG1();
 755                 goto out;
 756         }
 757         tp = intotcpcb(inp);
 758         TCPDEBUG1();
 759
 760         /*
 761          * Don't let too much OOB data build up
 762          */
 763         if (flags & PRUS_OOB) {
 764                 if (ssb_space(&so->so_snd) < -512) {
 765                         m_freem(m);
 766                         error = ENOBUFS;
 767                         goto out;
 768                 }
 769         }
 770
 771         /*
 772          * Pump the data into the socket.
 773          */
 774         if (m)
 775                 ssb_appendstream(&so->so_snd, m);
 776         if (flags & PRUS_OOB) {
 777                 /*
 778                  * According to RFC961 (Assigned Protocols),
 779                  * the urgent pointer points to the last octet
 780                  * of urgent data.  We continue, however,
 781                  * to consider it to indicate the first octet
 782                  * of data past the urgent section.
 783                  * Otherwise, snd_up should be one lower.
 784                  */
 785                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 786                 tp->t_flags |= TF_FORCE;
 787                 error = tcp_output(tp);
 788                 tp->t_flags &= ~TF_FORCE;
 789         } else {
 790                 if (flags & PRUS_EOF) {
 791                         /*
 792                          * Close the send side of the connection after
 793                          * the data is sent.
 794                          */
 795                         socantsendmore(so);
 796                         tp = tcp_usrclosed(tp);
 797                 }
 798                 if (tp != NULL) {
 799                         if (flags & PRUS_MORETOCOME)
 800                                 tp->t_flags |= TF_MORETOCOME;
 801                         error = tcp_output(tp);
 802                         if (flags & PRUS_MORETOCOME)
 803                                 tp->t_flags &= ~TF_MORETOCOME;
 804                 }
 805         }
 806         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 807                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 808                    (flags & PRUS_NOREPLY));
 809 }
 810
 811 /*
 812  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 813  *       will sofree() it when we return.
 814  */
 815 static void
 816 tcp_usr_abort(netmsg_t msg)
 817 {
 818         struct socket *so = msg->abort.base.nm_so;
 819         int error = 0;
 820         struct inpcb *inp;
 821         struct tcpcb *tp;
 822
 823         COMMON_START(so, inp, 1);
 824         tp = tcp_drop(tp, ECONNABORTED);
 825         COMMON_END(PRU_ABORT);
 826 }
 827
 828 /*
 829  * Receive out-of-band data.
 830  */
 831 static void
 832 tcp_usr_rcvoob(netmsg_t msg)
 833 {
 834         struct socket *so = msg->rcvoob.base.nm_so;
 835         struct mbuf *m = msg->rcvoob.nm_m;
 836         int flags = msg->rcvoob.nm_flags;
 837         int error = 0;
 838         struct inpcb *inp;
 839         struct tcpcb *tp;
 840
 841         COMMON_START(so, inp, 0);
 842         if ((so->so_oobmark == 0 &&
 843              (so->so_state & SS_RCVATMARK) == 0) ||
 844             so->so_options & SO_OOBINLINE ||
 845             tp->t_oobflags & TCPOOB_HADDATA) {
 846                 error = EINVAL;
 847                 goto out;
 848         }
 849         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 850                 error = EWOULDBLOCK;
 851                 goto out;
 852         }
 853         m->m_len = 1;
 854         *mtod(m, caddr_t) = tp->t_iobc;
 855         if ((flags & MSG_PEEK) == 0)
 856                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 857         COMMON_END(PRU_RCVOOB);
 858 }
 859
 860 /* xxx - should be const */
 861 struct pr_usrreqs tcp_usrreqs = {
 862         .pru_abort = tcp_usr_abort,
 863         .pru_accept = tcp_usr_accept,
 864         .pru_attach = tcp_usr_attach,
 865         .pru_bind = tcp_usr_bind,
 866         .pru_connect = tcp_usr_connect,
 867         .pru_connect2 = pr_generic_notsupp,
 868         .pru_control = in_control_dispatch,
 869         .pru_detach = tcp_usr_detach,
 870         .pru_disconnect = tcp_usr_disconnect,
 871         .pru_listen = tcp_usr_listen,
 872         .pru_peeraddr = in_setpeeraddr_dispatch,
 873         .pru_rcvd = tcp_usr_rcvd,
 874         .pru_rcvoob = tcp_usr_rcvoob,
 875         .pru_send = tcp_usr_send,
 876         .pru_sense = pru_sense_null,
 877         .pru_shutdown = tcp_usr_shutdown,
 878         .pru_sockaddr = in_setsockaddr_dispatch,
 879         .pru_sosend = sosendtcp,
 880         .pru_soreceive = soreceive
 881 };
 882
 883 #ifdef INET6
 884 struct pr_usrreqs tcp6_usrreqs = {
 885         .pru_abort = tcp_usr_abort,
 886         .pru_accept = tcp6_usr_accept,
 887         .pru_attach = tcp_usr_attach,
 888         .pru_bind = tcp6_usr_bind,
 889         .pru_connect = tcp6_usr_connect,
 890         .pru_connect2 = pr_generic_notsupp,
 891         .pru_control = in6_control_dispatch,
 892         .pru_detach = tcp_usr_detach,
 893         .pru_disconnect = tcp_usr_disconnect,
 894         .pru_listen = tcp6_usr_listen,
 895         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 896         .pru_rcvd = tcp_usr_rcvd,
 897         .pru_rcvoob = tcp_usr_rcvoob,
 898         .pru_send = tcp_usr_send,
 899         .pru_sense = pru_sense_null,
 900         .pru_shutdown = tcp_usr_shutdown,
 901         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 902         .pru_sosend = sosendtcp,
 903         .pru_soreceive = soreceive
 904 };
 905 #endif /* INET6 */
 906
 907 static int
 908 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 909                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 910 {
 911         struct inpcb *inp = tp->t_inpcb, *oinp;
 912         struct socket *so = inp->inp_socket;
 913         struct route *ro = &inp->inp_route;
 914
 915         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
 916                                  sin->sin_addr, sin->sin_port,
 917                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 918                                   inp->inp_laddr : if_sin->sin_addr),
 919                                 inp->inp_lport, 0, NULL);
 920         if (oinp != NULL) {
 921                 m_freem(m);
 922                 return (EADDRINUSE);
 923         }
 924         if (inp->inp_laddr.s_addr == INADDR_ANY)
 925                 inp->inp_laddr = if_sin->sin_addr;
 926         inp->inp_faddr = sin->sin_addr;
 927         inp->inp_fport = sin->sin_port;
 928         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
 929         in_pcbinsconnhash(inp);
 930
 931         /*
 932          * We are now on the inpcb's owner CPU, if the cached route was
 933          * freed because the rtentry's owner CPU is not the current CPU
 934          * (e.g. in tcp_connect()), then we try to reallocate it here with
 935          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 936          * rtentry.
 937          */
 938         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 939             ro->ro_rt == NULL) {
 940                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 941                 ro->ro_dst.sa_family = AF_INET;
 942                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 943                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 944                         sin->sin_addr;
 945                 rtalloc(ro);
 946         }
 947
 948         /*
 949          * Now that no more errors can occur, change the protocol processing
 950          * port to the current thread (which is the correct thread).
 951          *
 952          * Create TCP timer message now; we are on the tcpcb's owner
 953          * CPU/thread.
 954          */
 955         tcp_create_timermsg(tp, &curthread->td_msgport);
 956
 957         /*
 958          * Compute window scaling to request.  Use a larger scaling then
 959          * needed for the initial receive buffer in case the receive buffer
 960          * gets expanded.
 961          */
 962         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
 963                 tp->request_r_scale = TCP_MIN_WINSHIFT;
 964         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
 965                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
 966         ) {
 967                 tp->request_r_scale++;
 968         }
 969
 970         soisconnecting(so);
 971         tcpstat.tcps_connattempt++;
 972         tp->t_state = TCPS_SYN_SENT;
 973         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
 974         tp->iss = tcp_new_isn(tp);
 975         tcp_sendseqinit(tp);
 976         if (m) {
 977                 ssb_appendstream(&so->so_snd, m);
 978                 m = NULL;
 979                 if (flags & PRUS_OOB)
 980                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 981         }
 982
 983         /*
 984          * Close the send side of the connection after
 985          * the data is sent if flagged.
 986          */
 987         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
 988                 socantsendmore(so);
 989                 tp = tcp_usrclosed(tp);
 990         }
 991         return (tcp_output(tp));
 992 }
 993
 994 /*
 995  * Common subroutine to open a TCP connection to remote host specified
 996  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
 997  * port number if needed.  Call in_pcbladdr to do the routing and to choose
 998  * a local host address (interface).
 999  * Initialize connection parameters and enter SYN-SENT state.
1000  */
1001 static void
1002 tcp_connect(netmsg_t msg)
1003 {
1004         struct socket *so = msg->connect.base.nm_so;
1005         struct sockaddr *nam = msg->connect.nm_nam;
1006         struct thread *td = msg->connect.nm_td;
1007         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1008         struct sockaddr_in *if_sin;
1009         struct inpcb *inp;
1010         struct tcpcb *tp;
1011         int error, calc_laddr = 1;
1012 #ifdef SMP
1013         lwkt_port_t port;
1014 #endif
1015
1016         COMMON_START(so, inp, 0);
1017
1018         /*
1019          * Reconnect our pcb if we have to
1020          */
1021         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1022                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1023                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1024         }
1025
1026         /*
1027          * Bind if we have to
1028          */
1029         if (inp->inp_lport == 0) {
1030                 if (tcp_lport_extension) {
1031                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1032
1033                         error = in_pcbladdr(inp, nam, &if_sin, td);
1034                         if (error)
1035                                 goto out;
1036                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1037
1038                         error = in_pcbconn_bind(inp, nam, td);
1039                         if (error)
1040                                 goto out;
1041
1042                         calc_laddr = 0;
1043                 } else {
1044                         error = in_pcbbind(inp, NULL, td);
1045                         if (error)
1046                                 goto out;
1047                 }
1048         }
1049
1050         if (calc_laddr) {
1051                 /*
1052                  * Calculate the correct protocol processing thread.  The
1053                  * connect operation must run there.  Set the forwarding
1054                  * port before we forward the message or it will get bounced
1055                  * right back to us.
1056                  */
1057                 error = in_pcbladdr(inp, nam, &if_sin, td);
1058                 if (error)
1059                         goto out;
1060         }
1061         KKASSERT(inp->inp_socket == so);
1062
1063 #ifdef SMP
1064         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1065                             (inp->inp_laddr.s_addr ?
1066                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1067                             inp->inp_lport);
1068
1069         if (port != &curthread->td_msgport) {
1070                 struct route *ro = &inp->inp_route;
1071
1072                 /*
1073                  * in_pcbladdr() may have allocated a route entry for us
1074                  * on the current CPU, but we need a route entry on the
1075                  * inpcb's owner CPU, so free it here.
1076                  */
1077                 if (ro->ro_rt != NULL)
1078                         RTFREE(ro->ro_rt);
1079                 bzero(ro, sizeof(*ro));
1080
1081                 /*
1082                  * We are moving the protocol processing port the socket
1083                  * is on, we have to unlink here and re-link on the
1084                  * target cpu.
1085                  */
1086                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1087                 sosetport(so, port);
1088                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1089                 msg->connect.base.nm_dispatch = tcp_connect;
1090
1091                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1092                 /* msg invalid now */
1093                 return;
1094         }
1095 #else
1096         KKASSERT(so->so_port == &curthread->td_msgport);
1097 #endif
1098         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1099                                   msg->connect.nm_m, sin, if_sin);
1100         msg->connect.nm_m = NULL;
1101 out:
1102         if (msg->connect.nm_m) {
1103                 m_freem(msg->connect.nm_m);
1104                 msg->connect.nm_m = NULL;
1105         }
1106         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1107                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1108                 msg->connect.nm_nam = NULL;
1109         }
1110         lwkt_replymsg(&msg->connect.base.lmsg, error);
1111         /* msg invalid now */
1112 }
1113
1114 #ifdef INET6
1115
1116 static void
1117 tcp6_connect(netmsg_t msg)
1118 {
1119         struct tcpcb *tp;
1120         struct socket *so = msg->connect.base.nm_so;
1121         struct sockaddr *nam = msg->connect.nm_nam;
1122         struct thread *td = msg->connect.nm_td;
1123         struct inpcb *inp;
1124         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1125         struct in6_addr *addr6;
1126 #ifdef SMP
1127         lwkt_port_t port;
1128 #endif
1129         int error;
1130
1131         COMMON_START(so, inp, 0);
1132
1133         /*
1134          * Reconnect our pcb if we have to
1135          */
1136         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1137                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1138                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1139         }
1140
1141         /*
1142          * Bind if we have to
1143          */
1144         if (inp->inp_lport == 0) {
1145                 error = in6_pcbbind(inp, NULL, td);
1146                 if (error)
1147                         goto out;
1148         }
1149
1150         /*
1151          * Cannot simply call in_pcbconnect, because there might be an
1152          * earlier incarnation of this same connection still in
1153          * TIME_WAIT state, creating an ADDRINUSE error.
1154          */
1155         error = in6_pcbladdr(inp, nam, &addr6, td);
1156         if (error)
1157                 goto out;
1158
1159 #ifdef SMP
1160         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1161
1162         if (port != &curthread->td_msgport) {
1163                 struct route *ro = &inp->inp_route;
1164
1165                 /*
1166                  * in_pcbladdr() may have allocated a route entry for us
1167                  * on the current CPU, but we need a route entry on the
1168                  * inpcb's owner CPU, so free it here.
1169                  */
1170                 if (ro->ro_rt != NULL)
1171                         RTFREE(ro->ro_rt);
1172                 bzero(ro, sizeof(*ro));
1173
1174                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1175                 sosetport(so, port);
1176                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1177                 msg->connect.base.nm_dispatch = tcp6_connect;
1178
1179                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1180                 /* msg invalid now */
1181                 return;
1182         }
1183 #endif
1184         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1185                                    &msg->connect.nm_m, sin6, addr6);
1186         /* nm_m may still be intact */
1187 out:
1188         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1189                 tcp_connect(msg);
1190                 /* msg invalid now */
1191         } else {
1192                 if (msg->connect.nm_m) {
1193                         m_freem(msg->connect.nm_m);
1194                         msg->connect.nm_m = NULL;
1195                 }
1196                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1197                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1198                         msg->connect.nm_nam = NULL;
1199                 }
1200                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1201                 /* msg invalid now */
1202         }
1203 }
1204
1205 static int
1206 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1207                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1208 {
1209         struct mbuf *m = *mp;
1210         struct inpcb *inp = tp->t_inpcb;
1211         struct socket *so = inp->inp_socket;
1212         struct inpcb *oinp;
1213
1214         /*
1215          * Cannot simply call in_pcbconnect, because there might be an
1216          * earlier incarnation of this same connection still in
1217          * TIME_WAIT state, creating an ADDRINUSE error.
1218          */
1219         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1220                                   &sin6->sin6_addr, sin6->sin6_port,
1221                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1222                                       addr6 : &inp->in6p_laddr),
1223                                   inp->inp_lport,  0, NULL);
1224         if (oinp)
1225                 return (EADDRINUSE);
1226
1227         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1228                 inp->in6p_laddr = *addr6;
1229         inp->in6p_faddr = sin6->sin6_addr;
1230         inp->inp_fport = sin6->sin6_port;
1231         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1232                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1233         in_pcbinsconnhash(inp);
1234
1235         /*
1236          * Now that no more errors can occur, change the protocol processing
1237          * port to the current thread (which is the correct thread).
1238          *
1239          * Create TCP timer message now; we are on the tcpcb's owner
1240          * CPU/thread.
1241          */
1242         tcp_create_timermsg(tp, &curthread->td_msgport);
1243
1244         /* Compute window scaling to request.  */
1245         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1246                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1247         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1248             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1249                 tp->request_r_scale++;
1250         }
1251
1252         soisconnecting(so);
1253         tcpstat.tcps_connattempt++;
1254         tp->t_state = TCPS_SYN_SENT;
1255         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1256         tp->iss = tcp_new_isn(tp);
1257         tcp_sendseqinit(tp);
1258         if (m) {
1259                 ssb_appendstream(&so->so_snd, m);
1260                 *mp = NULL;
1261                 if (flags & PRUS_OOB)
1262                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1263         }
1264
1265         /*
1266          * Close the send side of the connection after
1267          * the data is sent if flagged.
1268          */
1269         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1270                 socantsendmore(so);
1271                 tp = tcp_usrclosed(tp);
1272         }
1273         return (tcp_output(tp));
1274 }
1275
1276 #endif /* INET6 */
1277
1278 /*
1279  * The new sockopt interface makes it possible for us to block in the
1280  * copyin/out step (if we take a page fault).  Taking a page fault while
1281  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1282  * both now use TSM, there probably isn't any need for this function to
1283  * run in a critical section any more.  This needs more examination.)
1284  */
1285 void
1286 tcp_ctloutput(netmsg_t msg)
1287 {
1288         struct socket *so = msg->base.nm_so;
1289         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1290         int     error, opt, optval, opthz;
1291         struct  inpcb *inp;
1292         struct  tcpcb *tp;
1293
1294         error = 0;
1295         inp = so->so_pcb;
1296         if (inp == NULL) {
1297                 error = ECONNRESET;
1298                 goto done;
1299         }
1300
1301         if (sopt->sopt_level != IPPROTO_TCP) {
1302 #ifdef INET6
1303                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1304                         ip6_ctloutput_dispatch(msg);
1305                 else
1306 #endif /* INET6 */
1307                 ip_ctloutput(msg);
1308                 /* msg invalid now */
1309                 return;
1310         }
1311         tp = intotcpcb(inp);
1312
1313         switch (sopt->sopt_dir) {
1314         case SOPT_SET:
1315                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1316                                       sizeof optval);
1317                 if (error)
1318                         break;
1319                 switch (sopt->sopt_name) {
1320                 case TCP_FASTKEEP:
1321                         if (optval > 0) {
1322                                 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1323                                         tp->t_flags |= TF_FASTKEEP;
1324                                         tcp_timer_keep_activity(tp, 0);
1325                                 }
1326                         } else {
1327                                 tp->t_flags &= ~TF_FASTKEEP;
1328                         }
1329                         break;
1330 #ifdef TCP_SIGNATURE
1331                 case TCP_SIGNATURE_ENABLE:
1332                         if (optval > 0)
1333                                 tp->t_flags |= TF_SIGNATURE;
1334                         else
1335                                 tp->t_flags &= ~TF_SIGNATURE;
1336                         break;
1337 #endif /* TCP_SIGNATURE */
1338                 case TCP_NODELAY:
1339                 case TCP_NOOPT:
1340                         switch (sopt->sopt_name) {
1341                         case TCP_NODELAY:
1342                                 opt = TF_NODELAY;
1343                                 break;
1344                         case TCP_NOOPT:
1345                                 opt = TF_NOOPT;
1346                                 break;
1347                         default:
1348                                 opt = 0; /* dead code to fool gcc */
1349                                 break;
1350                         }
1351
1352                         if (optval)
1353                                 tp->t_flags |= opt;
1354                         else
1355                                 tp->t_flags &= ~opt;
1356                         break;
1357
1358                 case TCP_NOPUSH:
1359                         if (optval)
1360                                 tp->t_flags |= TF_NOPUSH;
1361                         else {
1362                                 tp->t_flags &= ~TF_NOPUSH;
1363                                 error = tcp_output(tp);
1364                         }
1365                         break;
1366
1367                 case TCP_MAXSEG:
1368                         /*
1369                          * Must be between 0 and maxseg.  If the requested
1370                          * maxseg is too small to satisfy the desired minmss,
1371                          * pump it up (silently so sysctl modifications of
1372                          * minmss do not create unexpected program failures).
1373                          * Handle degenerate cases.
1374                          */
1375                         if (optval > 0 && optval <= tp->t_maxseg) {
1376                                 if (optval + 40 < tcp_minmss) {
1377                                         optval = tcp_minmss - 40;
1378                                         if (optval < 0)
1379                                                 optval = 1;
1380                                 }
1381                                 tp->t_maxseg = optval;
1382                         } else {
1383                                 error = EINVAL;
1384                         }
1385                         break;
1386
1387                 case TCP_KEEPINIT:
1388                         opthz = ((int64_t)optval * hz) / 1000;
1389                         if (opthz >= 1)
1390                                 tp->t_keepinit = opthz;
1391                         else
1392                                 error = EINVAL;
1393                         break;
1394
1395                 case TCP_KEEPIDLE:
1396                         opthz = ((int64_t)optval * hz) / 1000;
1397                         if (opthz >= 1)
1398                                 tp->t_keepidle = opthz;
1399                         else
1400                                 error = EINVAL;
1401                         break;
1402
1403                 case TCP_KEEPINTVL:
1404                         opthz = ((int64_t)optval * hz) / 1000;
1405                         if (opthz >= 1) {
1406                                 tp->t_keepintvl = opthz;
1407                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1408                         } else {
1409                                 error = EINVAL;
1410                         }
1411                         break;
1412
1413                 case TCP_KEEPCNT:
1414                         if (optval > 0) {
1415                                 tp->t_keepcnt = optval;
1416                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1417                         } else {
1418                                 error = EINVAL;
1419                         }
1420                         break;
1421
1422                 default:
1423                         error = ENOPROTOOPT;
1424                         break;
1425                 }
1426                 break;
1427
1428         case SOPT_GET:
1429                 switch (sopt->sopt_name) {
1430 #ifdef TCP_SIGNATURE
1431                 case TCP_SIGNATURE_ENABLE:
1432                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1433                         break;
1434 #endif /* TCP_SIGNATURE */
1435                 case TCP_NODELAY:
1436                         optval = tp->t_flags & TF_NODELAY;
1437                         break;
1438                 case TCP_MAXSEG:
1439                         optval = tp->t_maxseg;
1440                         break;
1441                 case TCP_NOOPT:
1442                         optval = tp->t_flags & TF_NOOPT;
1443                         break;
1444                 case TCP_NOPUSH:
1445                         optval = tp->t_flags & TF_NOPUSH;
1446                         break;
1447                 case TCP_KEEPINIT:
1448                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1449                         break;
1450                 case TCP_KEEPIDLE:
1451                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1452                         break;
1453                 case TCP_KEEPINTVL:
1454                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1455                         break;
1456                 case TCP_KEEPCNT:
1457                         optval = tp->t_keepcnt;
1458                         break;
1459                 default:
1460                         error = ENOPROTOOPT;
1461                         break;
1462                 }
1463                 if (error == 0)
1464                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1465                 break;
1466         }
1467 done:
1468         lwkt_replymsg(&msg->lmsg, error);
1469 }
1470
1471 /*
1472  * tcp_sendspace and tcp_recvspace are the default send and receive window
1473  * sizes, respectively.  These are obsolescent (this information should
1474  * be set by the route).
1475  *
1476  * Use a default that does not require tcp window scaling to be turned
1477  * on.  Individual programs or the administrator can increase the default.
1478  */
1479 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1480 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1481     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1482 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1483 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1484     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1485
1486 /*
1487  * Attach TCP protocol to socket, allocating internet protocol control
1488  * block, tcp control block, bufer space, and entering LISTEN state
1489  * if to accept connections.
1490  */
1491 static int
1492 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1493 {
1494         struct tcpcb *tp;
1495         struct inpcb *inp;
1496         int error;
1497         int cpu;
1498 #ifdef INET6
1499         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1500 #endif
1501
1502         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1503                 lwkt_gettoken(&so->so_rcv.ssb_token);
1504                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1505                                   ai->sb_rlimit);
1506                 lwkt_reltoken(&so->so_rcv.ssb_token);
1507                 if (error)
1508                         return (error);
1509         }
1510         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1511         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1512         cpu = mycpu->gd_cpuid;
1513
1514         /*
1515          * Set the default port for protocol processing. This will likely
1516          * change when we connect.
1517          */
1518         error = in_pcballoc(so, &tcbinfo[cpu]);
1519         if (error)
1520                 return (error);
1521         inp = so->so_pcb;
1522 #ifdef INET6
1523         if (isipv6) {
1524                 inp->inp_vflag |= INP_IPV6;
1525                 inp->in6p_hops = -1;    /* use kernel default */
1526         }
1527         else
1528 #endif
1529         inp->inp_vflag |= INP_IPV4;
1530         tp = tcp_newtcpcb(inp);
1531         if (tp == NULL) {
1532                 /*
1533                  * Make sure the socket is destroyed by the pcbdetach.
1534                  */
1535                 soreference(so);
1536 #ifdef INET6
1537                 if (isipv6)
1538                         in6_pcbdetach(inp);
1539                 else
1540 #endif
1541                 in_pcbdetach(inp);
1542                 sofree(so);     /* from ref above */
1543                 return (ENOBUFS);
1544         }
1545         tp->t_state = TCPS_CLOSED;
1546         return (0);
1547 }
1548
1549 /*
1550  * Initiate (or continue) disconnect.
1551  * If embryonic state, just send reset (once).
1552  * If in ``let data drain'' option and linger null, just drop.
1553  * Otherwise (hard), mark socket disconnecting and drop
1554  * current input data; switch states based on user close, and
1555  * send segment to peer (with FIN).
1556  */
1557 static struct tcpcb *
1558 tcp_disconnect(struct tcpcb *tp)
1559 {
1560         struct socket *so = tp->t_inpcb->inp_socket;
1561
1562         if (tp->t_state < TCPS_ESTABLISHED) {
1563                 tp = tcp_close(tp);
1564         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1565                 tp = tcp_drop(tp, 0);
1566         } else {
1567                 lwkt_gettoken(&so->so_rcv.ssb_token);
1568                 soisdisconnecting(so);
1569                 sbflush(&so->so_rcv.sb);
1570                 tp = tcp_usrclosed(tp);
1571                 if (tp)
1572                         tcp_output(tp);
1573                 lwkt_reltoken(&so->so_rcv.ssb_token);
1574         }
1575         return (tp);
1576 }
1577
1578 /*
1579  * User issued close, and wish to trail through shutdown states:
1580  * if never received SYN, just forget it.  If got a SYN from peer,
1581  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1582  * If already got a FIN from peer, then almost done; go to LAST_ACK
1583  * state.  In all other cases, have already sent FIN to peer (e.g.
1584  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1585  * for peer to send FIN or not respond to keep-alives, etc.
1586  * We can let the user exit from the close as soon as the FIN is acked.
1587  */
1588 static struct tcpcb *
1589 tcp_usrclosed(struct tcpcb *tp)
1590 {
1591
1592         switch (tp->t_state) {
1593
1594         case TCPS_CLOSED:
1595         case TCPS_LISTEN:
1596                 tp->t_state = TCPS_CLOSED;
1597                 tp = tcp_close(tp);
1598                 break;
1599
1600         case TCPS_SYN_SENT:
1601         case TCPS_SYN_RECEIVED:
1602                 tp->t_flags |= TF_NEEDFIN;
1603                 break;
1604
1605         case TCPS_ESTABLISHED:
1606                 tp->t_state = TCPS_FIN_WAIT_1;
1607                 break;
1608
1609         case TCPS_CLOSE_WAIT:
1610                 tp->t_state = TCPS_LAST_ACK;
1611                 break;
1612         }
1613         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1614                 soisdisconnected(tp->t_inpcb->inp_socket);
1615                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1616                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1617                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1618                             tcp_timer_2msl);
1619                 }
1620         }
1621         return (tp);
1622 }