sys/netinet/tcp_usrreq.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  64  */
  65
  66 #include "opt_ipsec.h"
  67 #include "opt_inet.h"
  68 #include "opt_inet6.h"
  69 #include "opt_tcpdebug.h"
  70
  71 #include <sys/param.h>
  72 #include <sys/systm.h>
  73 #include <sys/kernel.h>
  74 #include <sys/malloc.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/globaldata.h>
  77 #include <sys/thread.h>
  78
  79 #include <sys/mbuf.h>
  80 #ifdef INET6
  81 #include <sys/domain.h>
  82 #endif /* INET6 */
  83 #include <sys/socket.h>
  84 #include <sys/socketvar.h>
  85 #include <sys/socketops.h>
  86 #include <sys/protosw.h>
  87
  88 #include <sys/thread2.h>
  89 #include <sys/msgport2.h>
  90 #include <sys/socketvar2.h>
  91
  92 #include <net/if.h>
  93 #include <net/netisr.h>
  94 #include <net/route.h>
  95
  96 #include <net/netmsg2.h>
  97 #include <net/netisr2.h>
  98
  99 #include <netinet/in.h>
 100 #include <netinet/in_systm.h>
 101 #ifdef INET6
 102 #include <netinet/ip6.h>
 103 #endif
 104 #include <netinet/in_pcb.h>
 105 #ifdef INET6
 106 #include <netinet6/in6_pcb.h>
 107 #endif
 108 #include <netinet/in_var.h>
 109 #include <netinet/ip_var.h>
 110 #ifdef INET6
 111 #include <netinet6/ip6_var.h>
 112 #include <netinet6/tcp6_var.h>
 113 #endif
 114 #include <netinet/tcp.h>
 115 #include <netinet/tcp_fsm.h>
 116 #include <netinet/tcp_seq.h>
 117 #include <netinet/tcp_timer.h>
 118 #include <netinet/tcp_timer2.h>
 119 #include <netinet/tcp_var.h>
 120 #include <netinet/tcpip.h>
 121 #ifdef TCPDEBUG
 122 #include <netinet/tcp_debug.h>
 123 #endif
 124
 125 #ifdef IPSEC
 126 #include <netinet6/ipsec.h>
 127 #endif /*IPSEC*/
 128
 129 /*
 130  * TCP protocol interface to socket abstraction.
 131  */
 132 extern  char *tcpstates[];      /* XXX ??? */
 133
 134 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 135 static void     tcp_connect (netmsg_t msg);
 136 #ifdef INET6
 137 static void     tcp6_connect (netmsg_t msg);
 138 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 139                                 struct mbuf **mp,
 140                                 struct sockaddr_in6 *sin6,
 141                                 struct in6_addr *addr6);
 142 #endif /* INET6 */
 143 static struct tcpcb *
 144                 tcp_disconnect (struct tcpcb *);
 145 static struct tcpcb *
 146                 tcp_usrclosed (struct tcpcb *);
 147
 148 #ifdef TCPDEBUG
 149 #define TCPDEBUG0       int ostate = 0
 150 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 151 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 152                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 153 #else
 154 #define TCPDEBUG0
 155 #define TCPDEBUG1()
 156 #define TCPDEBUG2(req)
 157 #endif
 158
 159 static int      tcp_lport_extension = 1;
 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 161     &tcp_lport_extension, 0, "");
 162
 163 /*
 164  * For some ill optimized programs, which try to use TCP_NOPUSH
 165  * to improve performance, will have small amount of data sits
 166  * in the sending buffer.  These small amount of data will _not_
 167  * be pushed into the network until more data are written into
 168  * the socket or the socket write side is shutdown.
 169  */
 170 static int      tcp_disable_nopush = 1;
 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
 172     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
 173
 174 /*
 175  * TCP attaches to socket via pru_attach(), reserving space,
 176  * and an internet control block.  This is likely occuring on
 177  * cpu0 and may have to move later when we bind/connect.
 178  */
 179 static void
 180 tcp_usr_attach(netmsg_t msg)
 181 {
 182         struct socket *so = msg->base.nm_so;
 183         struct pru_attach_info *ai = msg->attach.nm_ai;
 184         int error;
 185         struct inpcb *inp;
 186         struct tcpcb *tp = NULL;
 187         TCPDEBUG0;
 188
 189         soreference(so);
 190         inp = so->so_pcb;
 191         TCPDEBUG1();
 192         if (inp) {
 193                 error = EISCONN;
 194                 goto out;
 195         }
 196
 197         error = tcp_attach(so, ai);
 198         if (error)
 199                 goto out;
 200
 201         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 202                 so->so_linger = TCP_LINGERTIME;
 203         tp = sototcpcb(so);
 204 out:
 205         sofree(so);             /* from ref above */
 206         TCPDEBUG2(PRU_ATTACH);
 207         lwkt_replymsg(&msg->lmsg, error);
 208 }
 209
 210 /*
 211  * pru_detach() detaches the TCP protocol from the socket.
 212  * If the protocol state is non-embryonic, then can't
 213  * do this directly: have to initiate a pru_disconnect(),
 214  * which may finish later; embryonic TCB's can just
 215  * be discarded here.
 216  */
 217 static void
 218 tcp_usr_detach(netmsg_t msg)
 219 {
 220         struct socket *so = msg->base.nm_so;
 221         int error = 0;
 222         struct inpcb *inp;
 223         struct tcpcb *tp;
 224         TCPDEBUG0;
 225
 226         inp = so->so_pcb;
 227
 228         /*
 229          * If the inp is already detached it may have been due to an async
 230          * close.  Just return as if no error occured.
 231          *
 232          * It's possible for the tcpcb (tp) to disconnect from the inp due
 233          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 234          * the detach message has been queued so we may find a NULL tp here.
 235          */
 236         if (inp) {
 237                 if ((tp = intotcpcb(inp)) != NULL) {
 238                         TCPDEBUG1();
 239                         tp = tcp_disconnect(tp);
 240                         TCPDEBUG2(PRU_DETACH);
 241                 }
 242         }
 243         lwkt_replymsg(&msg->lmsg, error);
 244 }
 245
 246 /*
 247  * NOTE: ignore_error is non-zero for certain disconnection races
 248  * which we want to silently allow, otherwise close() may return
 249  * an unexpected error.
 250  *
 251  * NOTE: The variables (msg) and (tp) are assumed.
 252  */
 253 #define COMMON_START(so, inp, ignore_error)                     \
 254         TCPDEBUG0;                                              \
 255                                                                 \
 256         inp = so->so_pcb;                                       \
 257         do {                                                    \
 258                  if (inp == NULL) {                             \
 259                         error = ignore_error ? 0 : EINVAL;      \
 260                         tp = NULL;                              \
 261                         goto out;                               \
 262                  }                                              \
 263                  tp = intotcpcb(inp);                           \
 264                  TCPDEBUG1();                                   \
 265         } while(0)
 266
 267 #define COMMON_END1(req, noreply)                               \
 268         out: do {                                               \
 269                 TCPDEBUG2(req);                                 \
 270                 if (!(noreply))                                 \
 271                         lwkt_replymsg(&msg->lmsg, error);       \
 272                 return;                                         \
 273         } while(0)
 274
 275 #define COMMON_END(req)         COMMON_END1((req), 0)
 276
 277 /*
 278  * Give the socket an address.
 279  */
 280 static void
 281 tcp_usr_bind(netmsg_t msg)
 282 {
 283         struct socket *so = msg->bind.base.nm_so;
 284         struct sockaddr *nam = msg->bind.nm_nam;
 285         struct thread *td = msg->bind.nm_td;
 286         int error = 0;
 287         struct inpcb *inp;
 288         struct tcpcb *tp;
 289         struct sockaddr_in *sinp;
 290
 291         COMMON_START(so, inp, 0);
 292
 293         /*
 294          * Must check for multicast addresses and disallow binding
 295          * to them.
 296          */
 297         sinp = (struct sockaddr_in *)nam;
 298         if (sinp->sin_family == AF_INET &&
 299             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 300                 error = EAFNOSUPPORT;
 301                 goto out;
 302         }
 303         error = in_pcbbind(inp, nam, td);
 304         if (error)
 305                 goto out;
 306         COMMON_END(PRU_BIND);
 307
 308 }
 309
 310 #ifdef INET6
 311
 312 static void
 313 tcp6_usr_bind(netmsg_t msg)
 314 {
 315         struct socket *so = msg->bind.base.nm_so;
 316         struct sockaddr *nam = msg->bind.nm_nam;
 317         struct thread *td = msg->bind.nm_td;
 318         int error = 0;
 319         struct inpcb *inp;
 320         struct tcpcb *tp;
 321         struct sockaddr_in6 *sin6p;
 322
 323         COMMON_START(so, inp, 0);
 324
 325         /*
 326          * Must check for multicast addresses and disallow binding
 327          * to them.
 328          */
 329         sin6p = (struct sockaddr_in6 *)nam;
 330         if (sin6p->sin6_family == AF_INET6 &&
 331             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 332                 error = EAFNOSUPPORT;
 333                 goto out;
 334         }
 335         inp->inp_vflag &= ~INP_IPV4;
 336         inp->inp_vflag |= INP_IPV6;
 337         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 338                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 339                         inp->inp_vflag |= INP_IPV4;
 340                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 341                         struct sockaddr_in sin;
 342
 343                         in6_sin6_2_sin(&sin, sin6p);
 344                         inp->inp_vflag |= INP_IPV4;
 345                         inp->inp_vflag &= ~INP_IPV6;
 346                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 347                         goto out;
 348                 }
 349         }
 350         error = in6_pcbbind(inp, nam, td);
 351         if (error)
 352                 goto out;
 353         COMMON_END(PRU_BIND);
 354 }
 355 #endif /* INET6 */
 356
 357 struct netmsg_inswildcard {
 358         struct netmsg_base      base;
 359         struct inpcb            *nm_inp;
 360 };
 361
 362 static void
 363 in_pcbinswildcardhash_handler(netmsg_t msg)
 364 {
 365         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 366         int cpu = mycpuid, nextcpu;
 367
 368         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 369
 370         nextcpu = cpu + 1;
 371         if (nextcpu < ncpus2)
 372                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 373         else
 374                 lwkt_replymsg(&nm->base.lmsg, 0);
 375 }
 376
 377 /*
 378  * Prepare to accept connections.
 379  */
 380 static void
 381 tcp_usr_listen(netmsg_t msg)
 382 {
 383         struct socket *so = msg->listen.base.nm_so;
 384         struct thread *td = msg->listen.nm_td;
 385         int error = 0;
 386         struct inpcb *inp;
 387         struct tcpcb *tp;
 388         struct netmsg_inswildcard nm;
 389         lwkt_port_t port0 = netisr_cpuport(0);
 390
 391         COMMON_START(so, inp, 0);
 392
 393         if (&curthread->td_msgport != port0) {
 394                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 395                     ("already asked to relink"));
 396
 397                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 398                 sosetport(so, port0);
 399                 msg->listen.nm_flags |= PRUL_RELINK;
 400
 401                 lwkt_forwardmsg(port0, &msg->listen.base.lmsg);
 402                 /* msg invalid now */
 403                 return;
 404         }
 405         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 406
 407         if (msg->listen.nm_flags & PRUL_RELINK) {
 408                 msg->listen.nm_flags &= ~PRUL_RELINK;
 409                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 410         }
 411         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 412         KASSERT(inp->inp_cpcbinfo == &tcbinfo[0], ("cpcbinfo is not tcbinfo0"));
 413
 414         if (tp->t_flags & TF_LISTEN)
 415                 goto out;
 416
 417         if (inp->inp_lport == 0) {
 418                 error = in_pcbbind(inp, NULL, td);
 419                 if (error)
 420                         goto out;
 421         }
 422
 423         tp->t_state = TCPS_LISTEN;
 424         tp->t_flags |= TF_LISTEN;
 425         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 426
 427         if (ncpus > 1) {
 428                 /*
 429                  * We have to set the flag because we can't have other cpus
 430                  * messing with our inp's flags.
 431                  */
 432                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 433                         ("already on connhash"));
 434                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 435                         ("already on wildcardhash"));
 436                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 437                         ("already on MP wildcardhash"));
 438                 inp->inp_flags |= INP_WILDCARD_MP;
 439
 440                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 441                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 442                 nm.nm_inp = inp;
 443                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 444         }
 445         in_pcbinswildcardhash(inp);
 446         COMMON_END(PRU_LISTEN);
 447 }
 448
 449 #ifdef INET6
 450
 451 static void
 452 tcp6_usr_listen(netmsg_t msg)
 453 {
 454         struct socket *so = msg->listen.base.nm_so;
 455         struct thread *td = msg->listen.nm_td;
 456         int error = 0;
 457         struct inpcb *inp;
 458         struct tcpcb *tp;
 459         struct netmsg_inswildcard nm;
 460
 461         COMMON_START(so, inp, 0);
 462
 463         if (tp->t_flags & TF_LISTEN)
 464                 goto out;
 465
 466         if (inp->inp_lport == 0) {
 467                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 468                         inp->inp_vflag |= INP_IPV4;
 469                 else
 470                         inp->inp_vflag &= ~INP_IPV4;
 471                 error = in6_pcbbind(inp, NULL, td);
 472                 if (error)
 473                         goto out;
 474         }
 475
 476         tp->t_state = TCPS_LISTEN;
 477         tp->t_flags |= TF_LISTEN;
 478         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 479
 480         if (ncpus > 1) {
 481                 /*
 482                  * We have to set the flag because we can't have other cpus
 483                  * messing with our inp's flags.
 484                  */
 485                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 486                         ("already on connhash"));
 487                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 488                         ("already on wildcardhash"));
 489                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 490                         ("already on MP wildcardhash"));
 491                 inp->inp_flags |= INP_WILDCARD_MP;
 492
 493                 KKASSERT(so->so_port == netisr_cpuport(0));
 494                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 495                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 496
 497                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 498                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 499                 nm.nm_inp = inp;
 500                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 501         }
 502         in_pcbinswildcardhash(inp);
 503         COMMON_END(PRU_LISTEN);
 504 }
 505 #endif /* INET6 */
 506
 507 /*
 508  * Initiate connection to peer.
 509  * Create a template for use in transmissions on this connection.
 510  * Enter SYN_SENT state, and mark socket as connecting.
 511  * Start keep-alive timer, and seed output sequence space.
 512  * Send initial segment on connection.
 513  */
 514 static void
 515 tcp_usr_connect(netmsg_t msg)
 516 {
 517         struct socket *so = msg->connect.base.nm_so;
 518         struct sockaddr *nam = msg->connect.nm_nam;
 519         struct thread *td = msg->connect.nm_td;
 520         int error = 0;
 521         struct inpcb *inp;
 522         struct tcpcb *tp;
 523         struct sockaddr_in *sinp;
 524
 525         COMMON_START(so, inp, 0);
 526
 527         /*
 528          * Must disallow TCP ``connections'' to multicast addresses.
 529          */
 530         sinp = (struct sockaddr_in *)nam;
 531         if (sinp->sin_family == AF_INET
 532             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 533                 error = EAFNOSUPPORT;
 534                 goto out;
 535         }
 536
 537         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 538                 error = EAFNOSUPPORT; /* IPv6 only jail */
 539                 goto out;
 540         }
 541
 542         tcp_connect(msg);
 543         /* msg is invalid now */
 544         return;
 545 out:
 546         if (msg->connect.nm_m) {
 547                 m_freem(msg->connect.nm_m);
 548                 msg->connect.nm_m = NULL;
 549         }
 550         if (msg->connect.nm_flags & PRUC_HELDTD)
 551                 lwkt_rele(td);
 552         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 553                 so->so_error = error;
 554                 soisdisconnected(so);
 555         }
 556         lwkt_replymsg(&msg->lmsg, error);
 557 }
 558
 559 #ifdef INET6
 560
 561 static void
 562 tcp6_usr_connect(netmsg_t msg)
 563 {
 564         struct socket *so = msg->connect.base.nm_so;
 565         struct sockaddr *nam = msg->connect.nm_nam;
 566         struct thread *td = msg->connect.nm_td;
 567         int error = 0;
 568         struct inpcb *inp;
 569         struct tcpcb *tp;
 570         struct sockaddr_in6 *sin6p;
 571
 572         COMMON_START(so, inp, 0);
 573
 574         /*
 575          * Must disallow TCP ``connections'' to multicast addresses.
 576          */
 577         sin6p = (struct sockaddr_in6 *)nam;
 578         if (sin6p->sin6_family == AF_INET6
 579             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 580                 error = EAFNOSUPPORT;
 581                 goto out;
 582         }
 583
 584         if (!prison_remote_ip(td, nam)) {
 585                 error = EAFNOSUPPORT; /* IPv4 only jail */
 586                 goto out;
 587         }
 588
 589         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 590                 struct sockaddr_in *sinp;
 591
 592                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 593                         error = EINVAL;
 594                         goto out;
 595                 }
 596                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 597                 in6_sin6_2_sin(sinp, sin6p);
 598                 inp->inp_vflag |= INP_IPV4;
 599                 inp->inp_vflag &= ~INP_IPV6;
 600                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 601                 msg->connect.nm_flags |= PRUC_NAMALLOC;
 602                 tcp_connect(msg);
 603                 /* msg is invalid now */
 604                 return;
 605         }
 606         inp->inp_vflag &= ~INP_IPV4;
 607         inp->inp_vflag |= INP_IPV6;
 608         inp->inp_inc.inc_isipv6 = 1;
 609
 610         msg->connect.nm_flags |= PRUC_FALLBACK;
 611         tcp6_connect(msg);
 612         /* msg is invalid now */
 613         return;
 614 out:
 615         if (msg->connect.nm_m) {
 616                 m_freem(msg->connect.nm_m);
 617                 msg->connect.nm_m = NULL;
 618         }
 619         lwkt_replymsg(&msg->lmsg, error);
 620 }
 621
 622 #endif /* INET6 */
 623
 624 /*
 625  * Initiate disconnect from peer.
 626  * If connection never passed embryonic stage, just drop;
 627  * else if don't need to let data drain, then can just drop anyways,
 628  * else have to begin TCP shutdown process: mark socket disconnecting,
 629  * drain unread data, state switch to reflect user close, and
 630  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 631  * when peer sends FIN and acks ours.
 632  *
 633  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 634  */
 635 static void
 636 tcp_usr_disconnect(netmsg_t msg)
 637 {
 638         struct socket *so = msg->disconnect.base.nm_so;
 639         int error = 0;
 640         struct inpcb *inp;
 641         struct tcpcb *tp;
 642
 643         COMMON_START(so, inp, 1);
 644         tp = tcp_disconnect(tp);
 645         COMMON_END(PRU_DISCONNECT);
 646 }
 647
 648 /*
 649  * Accept a connection.  Essentially all the work is
 650  * done at higher levels; just return the address
 651  * of the peer, storing through addr.
 652  */
 653 static void
 654 tcp_usr_accept(netmsg_t msg)
 655 {
 656         struct socket *so = msg->accept.base.nm_so;
 657         struct sockaddr **nam = msg->accept.nm_nam;
 658         int error = 0;
 659         struct inpcb *inp;
 660         struct tcpcb *tp = NULL;
 661         TCPDEBUG0;
 662
 663         inp = so->so_pcb;
 664         if (so->so_state & SS_ISDISCONNECTED) {
 665                 error = ECONNABORTED;
 666                 goto out;
 667         }
 668         if (inp == 0) {
 669                 error = EINVAL;
 670                 goto out;
 671         }
 672
 673         tp = intotcpcb(inp);
 674         TCPDEBUG1();
 675         in_setpeeraddr(so, nam);
 676         COMMON_END(PRU_ACCEPT);
 677 }
 678
 679 #ifdef INET6
 680 static void
 681 tcp6_usr_accept(netmsg_t msg)
 682 {
 683         struct socket *so = msg->accept.base.nm_so;
 684         struct sockaddr **nam = msg->accept.nm_nam;
 685         int error = 0;
 686         struct inpcb *inp;
 687         struct tcpcb *tp = NULL;
 688         TCPDEBUG0;
 689
 690         inp = so->so_pcb;
 691
 692         if (so->so_state & SS_ISDISCONNECTED) {
 693                 error = ECONNABORTED;
 694                 goto out;
 695         }
 696         if (inp == 0) {
 697                 error = EINVAL;
 698                 goto out;
 699         }
 700         tp = intotcpcb(inp);
 701         TCPDEBUG1();
 702         in6_mapped_peeraddr(so, nam);
 703         COMMON_END(PRU_ACCEPT);
 704 }
 705 #endif /* INET6 */
 706 /*
 707  * Mark the connection as being incapable of further output.
 708  */
 709 static void
 710 tcp_usr_shutdown(netmsg_t msg)
 711 {
 712         struct socket *so = msg->shutdown.base.nm_so;
 713         int error = 0;
 714         struct inpcb *inp;
 715         struct tcpcb *tp;
 716
 717         COMMON_START(so, inp, 0);
 718         socantsendmore(so);
 719         tp = tcp_usrclosed(tp);
 720         if (tp)
 721                 error = tcp_output(tp);
 722         COMMON_END(PRU_SHUTDOWN);
 723 }
 724
 725 /*
 726  * After a receive, possibly send window update to peer.
 727  */
 728 static void
 729 tcp_usr_rcvd(netmsg_t msg)
 730 {
 731         struct socket *so = msg->rcvd.base.nm_so;
 732         int error = 0, noreply = 0;
 733         struct inpcb *inp;
 734         struct tcpcb *tp;
 735
 736         COMMON_START(so, inp, 0);
 737
 738         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 739                 noreply = 1;
 740                 so_async_rcvd_reply(so);
 741         }
 742         tcp_output(tp);
 743
 744         COMMON_END1(PRU_RCVD, noreply);
 745 }
 746
 747 /*
 748  * Do a send by putting data in output queue and updating urgent
 749  * marker if URG set.  Possibly send more data.  Unlike the other
 750  * pru_*() routines, the mbuf chains are our responsibility.  We
 751  * must either enqueue them or free them.  The other pru_* routines
 752  * generally are caller-frees.
 753  */
 754 static void
 755 tcp_usr_send(netmsg_t msg)
 756 {
 757         struct socket *so = msg->send.base.nm_so;
 758         int flags = msg->send.nm_flags;
 759         struct mbuf *m = msg->send.nm_m;
 760         int error = 0;
 761         struct inpcb *inp;
 762         struct tcpcb *tp;
 763         TCPDEBUG0;
 764
 765         KKASSERT(msg->send.nm_control == NULL);
 766         KKASSERT(msg->send.nm_addr == NULL);
 767         KKASSERT((flags & PRUS_FREEADDR) == 0);
 768
 769         inp = so->so_pcb;
 770
 771         if (inp == NULL) {
 772                 /*
 773                  * OOPS! we lost a race, the TCP session got reset after
 774                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 775                  * network interrupt in the non-critical section of sosend().
 776                  */
 777                 m_freem(m);
 778                 error = ECONNRESET;     /* XXX EPIPE? */
 779                 tp = NULL;
 780                 TCPDEBUG1();
 781                 goto out;
 782         }
 783         tp = intotcpcb(inp);
 784         TCPDEBUG1();
 785
 786 #ifdef foo
 787         /*
 788          * This is no longer necessary, since:
 789          * - sosendtcp() has already checked it for us
 790          * - It does not work with asynchronized send
 791          */
 792
 793         /*
 794          * Don't let too much OOB data build up
 795          */
 796         if (flags & PRUS_OOB) {
 797                 if (ssb_space(&so->so_snd) < -512) {
 798                         m_freem(m);
 799                         error = ENOBUFS;
 800                         goto out;
 801                 }
 802         }
 803 #endif
 804
 805         /*
 806          * Pump the data into the socket.
 807          */
 808         if (m)
 809                 ssb_appendstream(&so->so_snd, m);
 810         if (flags & PRUS_OOB) {
 811                 /*
 812                  * According to RFC961 (Assigned Protocols),
 813                  * the urgent pointer points to the last octet
 814                  * of urgent data.  We continue, however,
 815                  * to consider it to indicate the first octet
 816                  * of data past the urgent section.
 817                  * Otherwise, snd_up should be one lower.
 818                  */
 819                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 820                 tp->t_flags |= TF_FORCE;
 821                 error = tcp_output(tp);
 822                 tp->t_flags &= ~TF_FORCE;
 823         } else {
 824                 if (flags & PRUS_EOF) {
 825                         /*
 826                          * Close the send side of the connection after
 827                          * the data is sent.
 828                          */
 829                         socantsendmore(so);
 830                         tp = tcp_usrclosed(tp);
 831                 }
 832                 if (tp != NULL && !tcp_output_pending(tp)) {
 833                         if (flags & PRUS_MORETOCOME)
 834                                 tp->t_flags |= TF_MORETOCOME;
 835                         error = tcp_output_fair(tp);
 836                         if (flags & PRUS_MORETOCOME)
 837                                 tp->t_flags &= ~TF_MORETOCOME;
 838                 }
 839         }
 840         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 841                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 842                    (flags & PRUS_NOREPLY));
 843 }
 844
 845 /*
 846  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 847  *       will sofree() it when we return.
 848  */
 849 static void
 850 tcp_usr_abort(netmsg_t msg)
 851 {
 852         struct socket *so = msg->abort.base.nm_so;
 853         int error = 0;
 854         struct inpcb *inp;
 855         struct tcpcb *tp;
 856
 857         COMMON_START(so, inp, 1);
 858         tp = tcp_drop(tp, ECONNABORTED);
 859         COMMON_END(PRU_ABORT);
 860 }
 861
 862 /*
 863  * Receive out-of-band data.
 864  */
 865 static void
 866 tcp_usr_rcvoob(netmsg_t msg)
 867 {
 868         struct socket *so = msg->rcvoob.base.nm_so;
 869         struct mbuf *m = msg->rcvoob.nm_m;
 870         int flags = msg->rcvoob.nm_flags;
 871         int error = 0;
 872         struct inpcb *inp;
 873         struct tcpcb *tp;
 874
 875         COMMON_START(so, inp, 0);
 876         if ((so->so_oobmark == 0 &&
 877              (so->so_state & SS_RCVATMARK) == 0) ||
 878             so->so_options & SO_OOBINLINE ||
 879             tp->t_oobflags & TCPOOB_HADDATA) {
 880                 error = EINVAL;
 881                 goto out;
 882         }
 883         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 884                 error = EWOULDBLOCK;
 885                 goto out;
 886         }
 887         m->m_len = 1;
 888         *mtod(m, caddr_t) = tp->t_iobc;
 889         if ((flags & MSG_PEEK) == 0)
 890                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 891         COMMON_END(PRU_RCVOOB);
 892 }
 893
 894 static void
 895 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 896 {
 897         in_savefaddr(so, faddr);
 898 }
 899
 900 #ifdef INET6
 901 static void
 902 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 903 {
 904         in6_mapped_savefaddr(so, faddr);
 905 }
 906 #endif
 907
 908 static int
 909 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 910     struct thread *td __unused)
 911 {
 912         const struct sockaddr_in *sinp;
 913
 914         sinp = (const struct sockaddr_in *)nam;
 915         if (sinp->sin_family == AF_INET &&
 916             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 917                 return EAFNOSUPPORT;
 918
 919         soisconnecting(so);
 920         return 0;
 921 }
 922
 923 /* xxx - should be const */
 924 struct pr_usrreqs tcp_usrreqs = {
 925         .pru_abort = tcp_usr_abort,
 926         .pru_accept = tcp_usr_accept,
 927         .pru_attach = tcp_usr_attach,
 928         .pru_bind = tcp_usr_bind,
 929         .pru_connect = tcp_usr_connect,
 930         .pru_connect2 = pr_generic_notsupp,
 931         .pru_control = in_control_dispatch,
 932         .pru_detach = tcp_usr_detach,
 933         .pru_disconnect = tcp_usr_disconnect,
 934         .pru_listen = tcp_usr_listen,
 935         .pru_peeraddr = in_setpeeraddr_dispatch,
 936         .pru_rcvd = tcp_usr_rcvd,
 937         .pru_rcvoob = tcp_usr_rcvoob,
 938         .pru_send = tcp_usr_send,
 939         .pru_sense = pru_sense_null,
 940         .pru_shutdown = tcp_usr_shutdown,
 941         .pru_sockaddr = in_setsockaddr_dispatch,
 942         .pru_sosend = sosendtcp,
 943         .pru_soreceive = sorecvtcp,
 944         .pru_savefaddr = tcp_usr_savefaddr,
 945         .pru_preconnect = tcp_usr_preconnect
 946 };
 947
 948 #ifdef INET6
 949 struct pr_usrreqs tcp6_usrreqs = {
 950         .pru_abort = tcp_usr_abort,
 951         .pru_accept = tcp6_usr_accept,
 952         .pru_attach = tcp_usr_attach,
 953         .pru_bind = tcp6_usr_bind,
 954         .pru_connect = tcp6_usr_connect,
 955         .pru_connect2 = pr_generic_notsupp,
 956         .pru_control = in6_control_dispatch,
 957         .pru_detach = tcp_usr_detach,
 958         .pru_disconnect = tcp_usr_disconnect,
 959         .pru_listen = tcp6_usr_listen,
 960         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 961         .pru_rcvd = tcp_usr_rcvd,
 962         .pru_rcvoob = tcp_usr_rcvoob,
 963         .pru_send = tcp_usr_send,
 964         .pru_sense = pru_sense_null,
 965         .pru_shutdown = tcp_usr_shutdown,
 966         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 967         .pru_sosend = sosendtcp,
 968         .pru_soreceive = sorecvtcp,
 969         .pru_savefaddr = tcp6_usr_savefaddr
 970 };
 971 #endif /* INET6 */
 972
 973 static int
 974 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 975                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 976 {
 977         struct inpcb *inp = tp->t_inpcb, *oinp;
 978         struct socket *so = inp->inp_socket;
 979         struct route *ro = &inp->inp_route;
 980
 981         KASSERT(inp->inp_cpcbinfo == &tcbinfo[mycpu->gd_cpuid],
 982             ("cpcbinfo mismatch"));
 983
 984         oinp = in_pcblookup_hash(inp->inp_cpcbinfo,
 985                                  sin->sin_addr, sin->sin_port,
 986                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 987                                   inp->inp_laddr : if_sin->sin_addr),
 988                                 inp->inp_lport, 0, NULL);
 989         if (oinp != NULL) {
 990                 m_freem(m);
 991                 return (EADDRINUSE);
 992         }
 993         if (inp->inp_laddr.s_addr == INADDR_ANY)
 994                 inp->inp_laddr = if_sin->sin_addr;
 995         inp->inp_faddr = sin->sin_addr;
 996         inp->inp_fport = sin->sin_port;
 997         in_pcbinsconnhash(inp);
 998
 999         /*
1000          * We are now on the inpcb's owner CPU, if the cached route was
1001          * freed because the rtentry's owner CPU is not the current CPU
1002          * (e.g. in tcp_connect()), then we try to reallocate it here with
1003          * the hope that a rtentry may be cloned from a RTF_PRCLONING
1004          * rtentry.
1005          */
1006         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1007             ro->ro_rt == NULL) {
1008                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1009                 ro->ro_dst.sa_family = AF_INET;
1010                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1011                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1012                         sin->sin_addr;
1013                 rtalloc(ro);
1014         }
1015
1016         /*
1017          * Now that no more errors can occur, change the protocol processing
1018          * port to the current thread (which is the correct thread).
1019          *
1020          * Create TCP timer message now; we are on the tcpcb's owner
1021          * CPU/thread.
1022          */
1023         tcp_create_timermsg(tp, &curthread->td_msgport);
1024
1025         /*
1026          * Compute window scaling to request.  Use a larger scaling then
1027          * needed for the initial receive buffer in case the receive buffer
1028          * gets expanded.
1029          */
1030         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1031                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1032         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1033                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1034         ) {
1035                 tp->request_r_scale++;
1036         }
1037
1038         soisconnecting(so);
1039         tcpstat.tcps_connattempt++;
1040         tp->t_state = TCPS_SYN_SENT;
1041         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1042         tp->iss = tcp_new_isn(tp);
1043         tcp_sendseqinit(tp);
1044         if (m) {
1045                 ssb_appendstream(&so->so_snd, m);
1046                 m = NULL;
1047                 if (flags & PRUS_OOB)
1048                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1049         }
1050
1051         /*
1052          * Close the send side of the connection after
1053          * the data is sent if flagged.
1054          */
1055         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1056                 socantsendmore(so);
1057                 tp = tcp_usrclosed(tp);
1058         }
1059         return (tcp_output(tp));
1060 }
1061
1062 /*
1063  * Common subroutine to open a TCP connection to remote host specified
1064  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1065  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1066  * a local host address (interface).
1067  * Initialize connection parameters and enter SYN-SENT state.
1068  */
1069 static void
1070 tcp_connect(netmsg_t msg)
1071 {
1072         struct socket *so = msg->connect.base.nm_so;
1073         struct sockaddr *nam = msg->connect.nm_nam;
1074         struct thread *td = msg->connect.nm_td;
1075         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1076         struct sockaddr_in *if_sin;
1077         struct inpcb *inp;
1078         struct tcpcb *tp;
1079         int error, calc_laddr = 1;
1080         lwkt_port_t port;
1081
1082         COMMON_START(so, inp, 0);
1083
1084         /*
1085          * Reconnect our pcb if we have to
1086          */
1087         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1088                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1089                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1090         }
1091
1092         /*
1093          * Bind if we have to
1094          */
1095         if (inp->inp_lport == 0) {
1096                 if (tcp_lport_extension) {
1097                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1098
1099                         error = in_pcbladdr(inp, nam, &if_sin, td);
1100                         if (error)
1101                                 goto out;
1102                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1103
1104                         error = in_pcbconn_bind(inp, nam, td);
1105                         if (error)
1106                                 goto out;
1107
1108                         calc_laddr = 0;
1109                 } else {
1110                         error = in_pcbbind(inp, NULL, td);
1111                         if (error)
1112                                 goto out;
1113                 }
1114         }
1115
1116         if (calc_laddr) {
1117                 /*
1118                  * Calculate the correct protocol processing thread.  The
1119                  * connect operation must run there.  Set the forwarding
1120                  * port before we forward the message or it will get bounced
1121                  * right back to us.
1122                  */
1123                 error = in_pcbladdr(inp, nam, &if_sin, td);
1124                 if (error)
1125                         goto out;
1126         }
1127         KKASSERT(inp->inp_socket == so);
1128
1129         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1130                             (inp->inp_laddr.s_addr ?
1131                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1132                             inp->inp_lport);
1133
1134         if (port != &curthread->td_msgport) {
1135                 struct route *ro = &inp->inp_route;
1136
1137                 /*
1138                  * in_pcbladdr() may have allocated a route entry for us
1139                  * on the current CPU, but we need a route entry on the
1140                  * inpcb's owner CPU, so free it here.
1141                  */
1142                 if (ro->ro_rt != NULL)
1143                         RTFREE(ro->ro_rt);
1144                 bzero(ro, sizeof(*ro));
1145
1146                 /*
1147                  * We are moving the protocol processing port the socket
1148                  * is on, we have to unlink here and re-link on the
1149                  * target cpu.
1150                  */
1151                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1152                 sosetport(so, port);
1153                 msg->connect.nm_flags |= PRUC_RECONNECT;
1154                 msg->connect.base.nm_dispatch = tcp_connect;
1155
1156                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1157                 /* msg invalid now */
1158                 return;
1159         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1160                 /*
1161                  * The original thread is no longer needed; release it.
1162                  */
1163                 lwkt_rele(td);
1164                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1165         }
1166         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1167                                   msg->connect.nm_m, sin, if_sin);
1168         msg->connect.nm_m = NULL;
1169 out:
1170         if (msg->connect.nm_m) {
1171                 m_freem(msg->connect.nm_m);
1172                 msg->connect.nm_m = NULL;
1173         }
1174         if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1175                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1176                 msg->connect.nm_nam = NULL;
1177         }
1178         if (msg->connect.nm_flags & PRUC_HELDTD)
1179                 lwkt_rele(td);
1180         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1181                 so->so_error = error;
1182                 soisdisconnected(so);
1183         }
1184         lwkt_replymsg(&msg->connect.base.lmsg, error);
1185         /* msg invalid now */
1186 }
1187
1188 #ifdef INET6
1189
1190 static void
1191 tcp6_connect(netmsg_t msg)
1192 {
1193         struct tcpcb *tp;
1194         struct socket *so = msg->connect.base.nm_so;
1195         struct sockaddr *nam = msg->connect.nm_nam;
1196         struct thread *td = msg->connect.nm_td;
1197         struct inpcb *inp;
1198         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1199         struct in6_addr *addr6;
1200         lwkt_port_t port;
1201         int error;
1202
1203         COMMON_START(so, inp, 0);
1204
1205         /*
1206          * Reconnect our pcb if we have to
1207          */
1208         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1209                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1210                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1211         }
1212
1213         /*
1214          * Bind if we have to
1215          */
1216         if (inp->inp_lport == 0) {
1217                 error = in6_pcbbind(inp, NULL, td);
1218                 if (error)
1219                         goto out;
1220         }
1221
1222         /*
1223          * Cannot simply call in_pcbconnect, because there might be an
1224          * earlier incarnation of this same connection still in
1225          * TIME_WAIT state, creating an ADDRINUSE error.
1226          */
1227         error = in6_pcbladdr(inp, nam, &addr6, td);
1228         if (error)
1229                 goto out;
1230
1231         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1232
1233         if (port != &curthread->td_msgport) {
1234                 struct route *ro = &inp->inp_route;
1235
1236                 /*
1237                  * in_pcbladdr() may have allocated a route entry for us
1238                  * on the current CPU, but we need a route entry on the
1239                  * inpcb's owner CPU, so free it here.
1240                  */
1241                 if (ro->ro_rt != NULL)
1242                         RTFREE(ro->ro_rt);
1243                 bzero(ro, sizeof(*ro));
1244
1245                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1246                 sosetport(so, port);
1247                 msg->connect.nm_flags |= PRUC_RECONNECT;
1248                 msg->connect.base.nm_dispatch = tcp6_connect;
1249
1250                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1251                 /* msg invalid now */
1252                 return;
1253         }
1254         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1255                                    &msg->connect.nm_m, sin6, addr6);
1256         /* nm_m may still be intact */
1257 out:
1258         if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1259                 tcp_connect(msg);
1260                 /* msg invalid now */
1261         } else {
1262                 if (msg->connect.nm_m) {
1263                         m_freem(msg->connect.nm_m);
1264                         msg->connect.nm_m = NULL;
1265                 }
1266                 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1267                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1268                         msg->connect.nm_nam = NULL;
1269                 }
1270                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1271                 /* msg invalid now */
1272         }
1273 }
1274
1275 static int
1276 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1277                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1278 {
1279         struct mbuf *m = *mp;
1280         struct inpcb *inp = tp->t_inpcb;
1281         struct socket *so = inp->inp_socket;
1282         struct inpcb *oinp;
1283
1284         /*
1285          * Cannot simply call in_pcbconnect, because there might be an
1286          * earlier incarnation of this same connection still in
1287          * TIME_WAIT state, creating an ADDRINUSE error.
1288          */
1289         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1290                                   &sin6->sin6_addr, sin6->sin6_port,
1291                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1292                                       addr6 : &inp->in6p_laddr),
1293                                   inp->inp_lport,  0, NULL);
1294         if (oinp)
1295                 return (EADDRINUSE);
1296
1297         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1298                 inp->in6p_laddr = *addr6;
1299         inp->in6p_faddr = sin6->sin6_addr;
1300         inp->inp_fport = sin6->sin6_port;
1301         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1302                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1303         in_pcbinsconnhash(inp);
1304
1305         /*
1306          * Now that no more errors can occur, change the protocol processing
1307          * port to the current thread (which is the correct thread).
1308          *
1309          * Create TCP timer message now; we are on the tcpcb's owner
1310          * CPU/thread.
1311          */
1312         tcp_create_timermsg(tp, &curthread->td_msgport);
1313
1314         /* Compute window scaling to request.  */
1315         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1316                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1317         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1318             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1319                 tp->request_r_scale++;
1320         }
1321
1322         soisconnecting(so);
1323         tcpstat.tcps_connattempt++;
1324         tp->t_state = TCPS_SYN_SENT;
1325         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1326         tp->iss = tcp_new_isn(tp);
1327         tcp_sendseqinit(tp);
1328         if (m) {
1329                 ssb_appendstream(&so->so_snd, m);
1330                 *mp = NULL;
1331                 if (flags & PRUS_OOB)
1332                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1333         }
1334
1335         /*
1336          * Close the send side of the connection after
1337          * the data is sent if flagged.
1338          */
1339         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1340                 socantsendmore(so);
1341                 tp = tcp_usrclosed(tp);
1342         }
1343         return (tcp_output(tp));
1344 }
1345
1346 #endif /* INET6 */
1347
1348 /*
1349  * The new sockopt interface makes it possible for us to block in the
1350  * copyin/out step (if we take a page fault).  Taking a page fault while
1351  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1352  * both now use TSM, there probably isn't any need for this function to
1353  * run in a critical section any more.  This needs more examination.)
1354  */
1355 void
1356 tcp_ctloutput(netmsg_t msg)
1357 {
1358         struct socket *so = msg->base.nm_so;
1359         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1360         int     error, opt, optval, opthz;
1361         struct  inpcb *inp;
1362         struct  tcpcb *tp;
1363
1364         error = 0;
1365         inp = so->so_pcb;
1366         if (inp == NULL) {
1367                 error = ECONNRESET;
1368                 goto done;
1369         }
1370
1371         if (sopt->sopt_level != IPPROTO_TCP) {
1372 #ifdef INET6
1373                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1374                         ip6_ctloutput_dispatch(msg);
1375                 else
1376 #endif /* INET6 */
1377                 ip_ctloutput(msg);
1378                 /* msg invalid now */
1379                 return;
1380         }
1381         tp = intotcpcb(inp);
1382
1383         switch (sopt->sopt_dir) {
1384         case SOPT_SET:
1385                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1386                                       sizeof optval);
1387                 if (error)
1388                         break;
1389                 switch (sopt->sopt_name) {
1390                 case TCP_FASTKEEP:
1391                         if (optval > 0)
1392                                 tp->t_keepidle = tp->t_keepintvl;
1393                         else
1394                                 tp->t_keepidle = tcp_keepidle;
1395                         tcp_timer_keep_activity(tp, 0);
1396                         break;
1397 #ifdef TCP_SIGNATURE
1398                 case TCP_SIGNATURE_ENABLE:
1399                         if (tp->t_state == TCPS_CLOSED) {
1400                                 /*
1401                                  * This is the only safe state that this
1402                                  * option could be changed.  Some segments
1403                                  * could already have been sent in other
1404                                  * states.
1405                                  */
1406                                 if (optval > 0)
1407                                         tp->t_flags |= TF_SIGNATURE;
1408                                 else
1409                                         tp->t_flags &= ~TF_SIGNATURE;
1410                         } else {
1411                                 error = EOPNOTSUPP;
1412                         }
1413                         break;
1414 #endif /* TCP_SIGNATURE */
1415                 case TCP_NODELAY:
1416                 case TCP_NOOPT:
1417                         switch (sopt->sopt_name) {
1418                         case TCP_NODELAY:
1419                                 opt = TF_NODELAY;
1420                                 break;
1421                         case TCP_NOOPT:
1422                                 opt = TF_NOOPT;
1423                                 break;
1424                         default:
1425                                 opt = 0; /* dead code to fool gcc */
1426                                 break;
1427                         }
1428
1429                         if (optval)
1430                                 tp->t_flags |= opt;
1431                         else
1432                                 tp->t_flags &= ~opt;
1433                         break;
1434
1435                 case TCP_NOPUSH:
1436                         if (tcp_disable_nopush)
1437                                 break;
1438                         if (optval)
1439                                 tp->t_flags |= TF_NOPUSH;
1440                         else {
1441                                 tp->t_flags &= ~TF_NOPUSH;
1442                                 error = tcp_output(tp);
1443                         }
1444                         break;
1445
1446                 case TCP_MAXSEG:
1447                         /*
1448                          * Must be between 0 and maxseg.  If the requested
1449                          * maxseg is too small to satisfy the desired minmss,
1450                          * pump it up (silently so sysctl modifications of
1451                          * minmss do not create unexpected program failures).
1452                          * Handle degenerate cases.
1453                          */
1454                         if (optval > 0 && optval <= tp->t_maxseg) {
1455                                 if (optval + 40 < tcp_minmss) {
1456                                         optval = tcp_minmss - 40;
1457                                         if (optval < 0)
1458                                                 optval = 1;
1459                                 }
1460                                 tp->t_maxseg = optval;
1461                         } else {
1462                                 error = EINVAL;
1463                         }
1464                         break;
1465
1466                 case TCP_KEEPINIT:
1467                         opthz = ((int64_t)optval * hz) / 1000;
1468                         if (opthz >= 1)
1469                                 tp->t_keepinit = opthz;
1470                         else
1471                                 error = EINVAL;
1472                         break;
1473
1474                 case TCP_KEEPIDLE:
1475                         opthz = ((int64_t)optval * hz) / 1000;
1476                         if (opthz >= 1) {
1477                                 tp->t_keepidle = opthz;
1478                                 tcp_timer_keep_activity(tp, 0);
1479                         } else {
1480                                 error = EINVAL;
1481                         }
1482                         break;
1483
1484                 case TCP_KEEPINTVL:
1485                         opthz = ((int64_t)optval * hz) / 1000;
1486                         if (opthz >= 1) {
1487                                 tp->t_keepintvl = opthz;
1488                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1489                         } else {
1490                                 error = EINVAL;
1491                         }
1492                         break;
1493
1494                 case TCP_KEEPCNT:
1495                         if (optval > 0) {
1496                                 tp->t_keepcnt = optval;
1497                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1498                         } else {
1499                                 error = EINVAL;
1500                         }
1501                         break;
1502
1503                 default:
1504                         error = ENOPROTOOPT;
1505                         break;
1506                 }
1507                 break;
1508
1509         case SOPT_GET:
1510                 switch (sopt->sopt_name) {
1511 #ifdef TCP_SIGNATURE
1512                 case TCP_SIGNATURE_ENABLE:
1513                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1514                         break;
1515 #endif /* TCP_SIGNATURE */
1516                 case TCP_NODELAY:
1517                         optval = tp->t_flags & TF_NODELAY;
1518                         break;
1519                 case TCP_MAXSEG:
1520                         optval = tp->t_maxseg;
1521                         break;
1522                 case TCP_NOOPT:
1523                         optval = tp->t_flags & TF_NOOPT;
1524                         break;
1525                 case TCP_NOPUSH:
1526                         optval = tp->t_flags & TF_NOPUSH;
1527                         break;
1528                 case TCP_KEEPINIT:
1529                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1530                         break;
1531                 case TCP_KEEPIDLE:
1532                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1533                         break;
1534                 case TCP_KEEPINTVL:
1535                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1536                         break;
1537                 case TCP_KEEPCNT:
1538                         optval = tp->t_keepcnt;
1539                         break;
1540                 default:
1541                         error = ENOPROTOOPT;
1542                         break;
1543                 }
1544                 if (error == 0)
1545                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1546                 break;
1547         }
1548 done:
1549         lwkt_replymsg(&msg->lmsg, error);
1550 }
1551
1552 /*
1553  * tcp_sendspace and tcp_recvspace are the default send and receive window
1554  * sizes, respectively.  These are obsolescent (this information should
1555  * be set by the route).
1556  *
1557  * Use a default that does not require tcp window scaling to be turned
1558  * on.  Individual programs or the administrator can increase the default.
1559  */
1560 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1561 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1562     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1563 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1564 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1565     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1566
1567 /*
1568  * Attach TCP protocol to socket, allocating internet protocol control
1569  * block, tcp control block, buffer space, and entering CLOSED state.
1570  */
1571 static int
1572 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1573 {
1574         struct tcpcb *tp;
1575         struct inpcb *inp;
1576         int error;
1577         int cpu;
1578 #ifdef INET6
1579         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1580 #endif
1581
1582         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1583                 lwkt_gettoken(&so->so_rcv.ssb_token);
1584                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1585                                   ai->sb_rlimit);
1586                 lwkt_reltoken(&so->so_rcv.ssb_token);
1587                 if (error)
1588                         return (error);
1589         }
1590         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1591         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1592         cpu = mycpu->gd_cpuid;
1593
1594         /*
1595          * Set the default port for protocol processing. This will likely
1596          * change when we connect.
1597          */
1598         error = in_pcballoc(so, &tcbinfo[cpu]);
1599         if (error)
1600                 return (error);
1601         inp = so->so_pcb;
1602 #ifdef INET6
1603         if (isipv6) {
1604                 inp->inp_vflag |= INP_IPV6;
1605                 inp->in6p_hops = -1;    /* use kernel default */
1606         }
1607         else
1608 #endif
1609         inp->inp_vflag |= INP_IPV4;
1610         tp = tcp_newtcpcb(inp);
1611         if (tp == NULL) {
1612                 /*
1613                  * Make sure the socket is destroyed by the pcbdetach.
1614                  */
1615                 soreference(so);
1616 #ifdef INET6
1617                 if (isipv6)
1618                         in6_pcbdetach(inp);
1619                 else
1620 #endif
1621                 in_pcbdetach(inp);
1622                 sofree(so);     /* from ref above */
1623                 return (ENOBUFS);
1624         }
1625         tp->t_state = TCPS_CLOSED;
1626         /* Keep a reference for asynchronized pru_rcvd */
1627         soreference(so);
1628         return (0);
1629 }
1630
1631 /*
1632  * Initiate (or continue) disconnect.
1633  * If embryonic state, just send reset (once).
1634  * If in ``let data drain'' option and linger null, just drop.
1635  * Otherwise (hard), mark socket disconnecting and drop
1636  * current input data; switch states based on user close, and
1637  * send segment to peer (with FIN).
1638  */
1639 static struct tcpcb *
1640 tcp_disconnect(struct tcpcb *tp)
1641 {
1642         struct socket *so = tp->t_inpcb->inp_socket;
1643
1644         if (tp->t_state < TCPS_ESTABLISHED) {
1645                 tp = tcp_close(tp);
1646         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1647                 tp = tcp_drop(tp, 0);
1648         } else {
1649                 lwkt_gettoken(&so->so_rcv.ssb_token);
1650                 soisdisconnecting(so);
1651                 sbflush(&so->so_rcv.sb);
1652                 tp = tcp_usrclosed(tp);
1653                 if (tp)
1654                         tcp_output(tp);
1655                 lwkt_reltoken(&so->so_rcv.ssb_token);
1656         }
1657         return (tp);
1658 }
1659
1660 /*
1661  * User issued close, and wish to trail through shutdown states:
1662  * if never received SYN, just forget it.  If got a SYN from peer,
1663  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1664  * If already got a FIN from peer, then almost done; go to LAST_ACK
1665  * state.  In all other cases, have already sent FIN to peer (e.g.
1666  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1667  * for peer to send FIN or not respond to keep-alives, etc.
1668  * We can let the user exit from the close as soon as the FIN is acked.
1669  */
1670 static struct tcpcb *
1671 tcp_usrclosed(struct tcpcb *tp)
1672 {
1673
1674         switch (tp->t_state) {
1675
1676         case TCPS_CLOSED:
1677         case TCPS_LISTEN:
1678                 tp->t_state = TCPS_CLOSED;
1679                 tp = tcp_close(tp);
1680                 break;
1681
1682         case TCPS_SYN_SENT:
1683         case TCPS_SYN_RECEIVED:
1684                 tp->t_flags |= TF_NEEDFIN;
1685                 break;
1686
1687         case TCPS_ESTABLISHED:
1688                 tp->t_state = TCPS_FIN_WAIT_1;
1689                 break;
1690
1691         case TCPS_CLOSE_WAIT:
1692                 tp->t_state = TCPS_LAST_ACK;
1693                 break;
1694         }
1695         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1696                 soisdisconnected(tp->t_inpcb->inp_socket);
1697                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1698                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1699                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1700                             tcp_timer_2msl);
1701                 }
1702         }
1703         return (tp);
1704 }