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