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 static void
 378 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 379 {
 380         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 381 }
 382
 383 /*
 384  * Prepare to accept connections.
 385  */
 386 static void
 387 tcp_usr_listen(netmsg_t msg)
 388 {
 389         struct socket *so = msg->listen.base.nm_so;
 390         struct thread *td = msg->listen.nm_td;
 391         int error = 0;
 392         struct inpcb *inp;
 393         struct tcpcb *tp;
 394         struct netmsg_inswildcard nm;
 395         lwkt_port_t port0 = netisr_cpuport(0);
 396
 397         COMMON_START(so, inp, 0);
 398
 399         if (&curthread->td_msgport != port0) {
 400                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 401
 402                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 403                     ("already asked to relink"));
 404
 405                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 406                 msg->listen.nm_flags |= PRUL_RELINK;
 407
 408                 /* See the related comment in tcp_connect() */
 409                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 410                 lwkt_forwardmsg(port0, lmsg);
 411                 /* msg invalid now */
 412                 return;
 413         }
 414         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 415
 416         if (msg->listen.nm_flags & PRUL_RELINK) {
 417                 msg->listen.nm_flags &= ~PRUL_RELINK;
 418                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 419         }
 420         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 421
 422         if (tp->t_flags & TF_LISTEN)
 423                 goto out;
 424
 425         if (inp->inp_lport == 0) {
 426                 error = in_pcbbind(inp, NULL, td);
 427                 if (error)
 428                         goto out;
 429         }
 430
 431         tp->t_state = TCPS_LISTEN;
 432         tp->t_flags |= TF_LISTEN;
 433         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 434
 435         if (ncpus2 > 1) {
 436                 /*
 437                  * Put this inpcb into wildcard hash on other cpus.
 438                  */
 439                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 440                         ("already on connhash"));
 441                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 442                         ("already on wildcardhash"));
 443
 444                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 445                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 446                 nm.nm_inp = inp;
 447                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 448         }
 449         in_pcbinswildcardhash(inp);
 450         COMMON_END(PRU_LISTEN);
 451 }
 452
 453 #ifdef INET6
 454
 455 static void
 456 tcp6_usr_listen(netmsg_t msg)
 457 {
 458         struct socket *so = msg->listen.base.nm_so;
 459         struct thread *td = msg->listen.nm_td;
 460         int error = 0;
 461         struct inpcb *inp;
 462         struct tcpcb *tp;
 463         struct netmsg_inswildcard nm;
 464
 465         COMMON_START(so, inp, 0);
 466
 467         if (tp->t_flags & TF_LISTEN)
 468                 goto out;
 469
 470         if (inp->inp_lport == 0) {
 471                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 472                         inp->inp_vflag |= INP_IPV4;
 473                 else
 474                         inp->inp_vflag &= ~INP_IPV4;
 475                 error = in6_pcbbind(inp, NULL, td);
 476                 if (error)
 477                         goto out;
 478         }
 479
 480         tp->t_state = TCPS_LISTEN;
 481         tp->t_flags |= TF_LISTEN;
 482         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 483
 484         if (ncpus2 > 1) {
 485                 /*
 486                  * Put this inpcb into wildcard hash on other cpus.
 487                  */
 488                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 489                         ("already on connhash"));
 490                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 491                         ("already on wildcardhash"));
 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                 sowwakeup(so);
 811         }
 812         if (flags & PRUS_OOB) {
 813                 /*
 814                  * According to RFC961 (Assigned Protocols),
 815                  * the urgent pointer points to the last octet
 816                  * of urgent data.  We continue, however,
 817                  * to consider it to indicate the first octet
 818                  * of data past the urgent section.
 819                  * Otherwise, snd_up should be one lower.
 820                  */
 821                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 822                 tp->t_flags |= TF_FORCE;
 823                 error = tcp_output(tp);
 824                 tp->t_flags &= ~TF_FORCE;
 825         } else {
 826                 if (flags & PRUS_EOF) {
 827                         /*
 828                          * Close the send side of the connection after
 829                          * the data is sent.
 830                          */
 831                         socantsendmore(so);
 832                         tp = tcp_usrclosed(tp);
 833                 }
 834                 if (tp != NULL && !tcp_output_pending(tp)) {
 835                         if (flags & PRUS_MORETOCOME)
 836                                 tp->t_flags |= TF_MORETOCOME;
 837                         error = tcp_output_fair(tp);
 838                         if (flags & PRUS_MORETOCOME)
 839                                 tp->t_flags &= ~TF_MORETOCOME;
 840                 }
 841         }
 842         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 843                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 844                    (flags & PRUS_NOREPLY));
 845 }
 846
 847 /*
 848  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 849  *       will sofree() it when we return.
 850  */
 851 static void
 852 tcp_usr_abort(netmsg_t msg)
 853 {
 854         struct socket *so = msg->abort.base.nm_so;
 855         int error = 0;
 856         struct inpcb *inp;
 857         struct tcpcb *tp;
 858
 859         COMMON_START(so, inp, 1);
 860         tp = tcp_drop(tp, ECONNABORTED);
 861         COMMON_END(PRU_ABORT);
 862 }
 863
 864 /*
 865  * Receive out-of-band data.
 866  */
 867 static void
 868 tcp_usr_rcvoob(netmsg_t msg)
 869 {
 870         struct socket *so = msg->rcvoob.base.nm_so;
 871         struct mbuf *m = msg->rcvoob.nm_m;
 872         int flags = msg->rcvoob.nm_flags;
 873         int error = 0;
 874         struct inpcb *inp;
 875         struct tcpcb *tp;
 876
 877         COMMON_START(so, inp, 0);
 878         if ((so->so_oobmark == 0 &&
 879              (so->so_state & SS_RCVATMARK) == 0) ||
 880             so->so_options & SO_OOBINLINE ||
 881             tp->t_oobflags & TCPOOB_HADDATA) {
 882                 error = EINVAL;
 883                 goto out;
 884         }
 885         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 886                 error = EWOULDBLOCK;
 887                 goto out;
 888         }
 889         m->m_len = 1;
 890         *mtod(m, caddr_t) = tp->t_iobc;
 891         if ((flags & MSG_PEEK) == 0)
 892                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 893         COMMON_END(PRU_RCVOOB);
 894 }
 895
 896 static void
 897 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 898 {
 899         in_savefaddr(so, faddr);
 900 }
 901
 902 #ifdef INET6
 903 static void
 904 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 905 {
 906         in6_mapped_savefaddr(so, faddr);
 907 }
 908 #endif
 909
 910 static int
 911 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 912     struct thread *td __unused)
 913 {
 914         const struct sockaddr_in *sinp;
 915
 916         sinp = (const struct sockaddr_in *)nam;
 917         if (sinp->sin_family == AF_INET &&
 918             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 919                 return EAFNOSUPPORT;
 920
 921         soisconnecting(so);
 922         return 0;
 923 }
 924
 925 /* xxx - should be const */
 926 struct pr_usrreqs tcp_usrreqs = {
 927         .pru_abort = tcp_usr_abort,
 928         .pru_accept = tcp_usr_accept,
 929         .pru_attach = tcp_usr_attach,
 930         .pru_bind = tcp_usr_bind,
 931         .pru_connect = tcp_usr_connect,
 932         .pru_connect2 = pr_generic_notsupp,
 933         .pru_control = in_control_dispatch,
 934         .pru_detach = tcp_usr_detach,
 935         .pru_disconnect = tcp_usr_disconnect,
 936         .pru_listen = tcp_usr_listen,
 937         .pru_peeraddr = in_setpeeraddr_dispatch,
 938         .pru_rcvd = tcp_usr_rcvd,
 939         .pru_rcvoob = tcp_usr_rcvoob,
 940         .pru_send = tcp_usr_send,
 941         .pru_sense = pru_sense_null,
 942         .pru_shutdown = tcp_usr_shutdown,
 943         .pru_sockaddr = in_setsockaddr_dispatch,
 944         .pru_sosend = sosendtcp,
 945         .pru_soreceive = sorecvtcp,
 946         .pru_savefaddr = tcp_usr_savefaddr,
 947         .pru_preconnect = tcp_usr_preconnect
 948 };
 949
 950 #ifdef INET6
 951 struct pr_usrreqs tcp6_usrreqs = {
 952         .pru_abort = tcp_usr_abort,
 953         .pru_accept = tcp6_usr_accept,
 954         .pru_attach = tcp_usr_attach,
 955         .pru_bind = tcp6_usr_bind,
 956         .pru_connect = tcp6_usr_connect,
 957         .pru_connect2 = pr_generic_notsupp,
 958         .pru_control = in6_control_dispatch,
 959         .pru_detach = tcp_usr_detach,
 960         .pru_disconnect = tcp_usr_disconnect,
 961         .pru_listen = tcp6_usr_listen,
 962         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 963         .pru_rcvd = tcp_usr_rcvd,
 964         .pru_rcvoob = tcp_usr_rcvoob,
 965         .pru_send = tcp_usr_send,
 966         .pru_sense = pru_sense_null,
 967         .pru_shutdown = tcp_usr_shutdown,
 968         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 969         .pru_sosend = sosendtcp,
 970         .pru_soreceive = sorecvtcp,
 971         .pru_savefaddr = tcp6_usr_savefaddr
 972 };
 973 #endif /* INET6 */
 974
 975 static int
 976 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 977                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 978 {
 979         struct inpcb *inp = tp->t_inpcb, *oinp;
 980         struct socket *so = inp->inp_socket;
 981         struct route *ro = &inp->inp_route;
 982
 983         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
 984             ("pcbinfo mismatch"));
 985
 986         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
 987                                  sin->sin_addr, sin->sin_port,
 988                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 989                                   inp->inp_laddr : if_sin->sin_addr),
 990                                 inp->inp_lport, 0, NULL);
 991         if (oinp != NULL) {
 992                 m_freem(m);
 993                 return (EADDRINUSE);
 994         }
 995         if (inp->inp_laddr.s_addr == INADDR_ANY)
 996                 inp->inp_laddr = if_sin->sin_addr;
 997         inp->inp_faddr = sin->sin_addr;
 998         inp->inp_fport = sin->sin_port;
 999         in_pcbinsconnhash(inp);
1000
1001         /*
1002          * We are now on the inpcb's owner CPU, if the cached route was
1003          * freed because the rtentry's owner CPU is not the current CPU
1004          * (e.g. in tcp_connect()), then we try to reallocate it here with
1005          * the hope that a rtentry may be cloned from a RTF_PRCLONING
1006          * rtentry.
1007          */
1008         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1009             ro->ro_rt == NULL) {
1010                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1011                 ro->ro_dst.sa_family = AF_INET;
1012                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1013                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1014                         sin->sin_addr;
1015                 rtalloc(ro);
1016         }
1017
1018         /*
1019          * Now that no more errors can occur, change the protocol processing
1020          * port to the current thread (which is the correct thread).
1021          *
1022          * Create TCP timer message now; we are on the tcpcb's owner
1023          * CPU/thread.
1024          */
1025         tcp_create_timermsg(tp, &curthread->td_msgport);
1026
1027         /*
1028          * Compute window scaling to request.  Use a larger scaling then
1029          * needed for the initial receive buffer in case the receive buffer
1030          * gets expanded.
1031          */
1032         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1033                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1034         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1035                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1036         ) {
1037                 tp->request_r_scale++;
1038         }
1039
1040         soisconnecting(so);
1041         tcpstat.tcps_connattempt++;
1042         tp->t_state = TCPS_SYN_SENT;
1043         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1044         tp->iss = tcp_new_isn(tp);
1045         tcp_sendseqinit(tp);
1046         if (m) {
1047                 ssb_appendstream(&so->so_snd, m);
1048                 m = NULL;
1049                 if (flags & PRUS_OOB)
1050                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1051         }
1052
1053         /*
1054          * Close the send side of the connection after
1055          * the data is sent if flagged.
1056          */
1057         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1058                 socantsendmore(so);
1059                 tp = tcp_usrclosed(tp);
1060         }
1061         return (tcp_output(tp));
1062 }
1063
1064 /*
1065  * Common subroutine to open a TCP connection to remote host specified
1066  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1067  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1068  * a local host address (interface).
1069  * Initialize connection parameters and enter SYN-SENT state.
1070  */
1071 static void
1072 tcp_connect(netmsg_t msg)
1073 {
1074         struct socket *so = msg->connect.base.nm_so;
1075         struct sockaddr *nam = msg->connect.nm_nam;
1076         struct thread *td = msg->connect.nm_td;
1077         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1078         struct sockaddr_in *if_sin = NULL;
1079         struct inpcb *inp;
1080         struct tcpcb *tp;
1081         int error;
1082         lwkt_port_t port;
1083
1084         COMMON_START(so, inp, 0);
1085
1086         /*
1087          * Reconnect our pcb if we have to
1088          */
1089         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1090                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1091                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1092         }
1093
1094         /*
1095          * Bind if we have to
1096          */
1097         if (inp->inp_lport == 0) {
1098                 if (tcp_lport_extension) {
1099                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1100
1101                         error = in_pcbladdr(inp, nam, &if_sin, td);
1102                         if (error)
1103                                 goto out;
1104                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1105
1106                         error = in_pcbbind_remote(inp, nam, td);
1107                         if (error)
1108                                 goto out;
1109
1110                         msg->connect.nm_flags |= PRUC_HASLADDR;
1111                 } else {
1112                         error = in_pcbbind(inp, NULL, td);
1113                         if (error)
1114                                 goto out;
1115                 }
1116         }
1117
1118         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1119                 /*
1120                  * Calculate the correct protocol processing thread.  The
1121                  * connect operation must run there.  Set the forwarding
1122                  * port before we forward the message or it will get bounced
1123                  * right back to us.
1124                  */
1125                 error = in_pcbladdr(inp, nam, &if_sin, td);
1126                 if (error)
1127                         goto out;
1128         }
1129         KKASSERT(inp->inp_socket == so);
1130
1131         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1132                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1133                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1134                             inp->inp_lport);
1135
1136         if (port != &curthread->td_msgport) {
1137                 struct route *ro = &inp->inp_route;
1138                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1139
1140                 /*
1141                  * in_pcbladdr() may have allocated a route entry for us
1142                  * on the current CPU, but we need a route entry on the
1143                  * inpcb's owner CPU, so free it here.
1144                  */
1145                 if (ro->ro_rt != NULL)
1146                         RTFREE(ro->ro_rt);
1147                 bzero(ro, sizeof(*ro));
1148
1149                 /*
1150                  * We are moving the protocol processing port the socket
1151                  * is on, we have to unlink here and re-link on the
1152                  * target cpu.
1153                  */
1154                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1155                 msg->connect.nm_flags |= PRUC_RECONNECT;
1156                 msg->connect.base.nm_dispatch = tcp_connect;
1157
1158                 /*
1159                  * Use message put done receipt to change this socket's
1160                  * so_port, i.e. _after_ this message was put onto the
1161                  * target netisr's msgport but _before_ the message could
1162                  * be pulled from the target netisr's msgport, so that:
1163                  * - The upper half (socket code) will not see the new
1164                  *   msgport before this message reaches the new msgport
1165                  *   and messages for this socket will be ordered.
1166                  * - This message will see the new msgport, when its
1167                  *   handler is called in the target netisr.
1168                  *
1169                  * NOTE:
1170                  * We MUST use messege put done receipt to change this
1171                  * socket's so_port:
1172                  * If we changed the so_port in this netisr after the
1173                  * lwkt_forwardmsg (so messages for this socket will be
1174                  * ordered) and changed the so_port in the target netisr
1175                  * at the very beginning of this message's handler, we
1176                  * would suffer so_port overwritten race, given this
1177                  * message might be forwarded again.
1178                  *
1179                  * NOTE:
1180                  * This mechanism depends on that the netisr's msgport
1181                  * is spin msgport (currently it is :).
1182                  *
1183                  * If the upper half saw the new msgport before this
1184                  * message reached the target netisr's msgport, the
1185                  * messages sent from the upper half could reach the new
1186                  * msgport before this message, thus there would be
1187                  * message reordering.  The worst case could be soclose()
1188                  * saw the new msgport and the detach message could reach
1189                  * the new msgport before this message, i.e. the inpcb
1190                  * could have been destroyed when this message was still
1191                  * pending on or on its way to the new msgport.  Other
1192                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1193                  * since we have unlinked this inpcb from the current
1194                  * pcbinfo first.
1195                  */
1196                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1197                 lwkt_forwardmsg(port, lmsg);
1198                 /* msg invalid now */
1199                 return;
1200         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1201                 /*
1202                  * The original thread is no longer needed; release it.
1203                  */
1204                 lwkt_rele(td);
1205                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1206         }
1207         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1208                                   msg->connect.nm_m, sin, if_sin);
1209         msg->connect.nm_m = NULL;
1210 out:
1211         if (msg->connect.nm_m) {
1212                 m_freem(msg->connect.nm_m);
1213                 msg->connect.nm_m = NULL;
1214         }
1215         if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1216                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1217                 msg->connect.nm_nam = NULL;
1218         }
1219         if (msg->connect.nm_flags & PRUC_HELDTD)
1220                 lwkt_rele(td);
1221         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1222                 so->so_error = error;
1223                 soisdisconnected(so);
1224         }
1225         lwkt_replymsg(&msg->connect.base.lmsg, error);
1226         /* msg invalid now */
1227 }
1228
1229 #ifdef INET6
1230
1231 static void
1232 tcp6_connect(netmsg_t msg)
1233 {
1234         struct tcpcb *tp;
1235         struct socket *so = msg->connect.base.nm_so;
1236         struct sockaddr *nam = msg->connect.nm_nam;
1237         struct thread *td = msg->connect.nm_td;
1238         struct inpcb *inp;
1239         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1240         struct in6_addr *addr6;
1241         lwkt_port_t port;
1242         int error;
1243
1244         COMMON_START(so, inp, 0);
1245
1246         /*
1247          * Reconnect our pcb if we have to
1248          */
1249         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1250                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1251                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1252         }
1253
1254         /*
1255          * Bind if we have to
1256          */
1257         if (inp->inp_lport == 0) {
1258                 error = in6_pcbbind(inp, NULL, td);
1259                 if (error)
1260                         goto out;
1261         }
1262
1263         /*
1264          * Cannot simply call in_pcbconnect, because there might be an
1265          * earlier incarnation of this same connection still in
1266          * TIME_WAIT state, creating an ADDRINUSE error.
1267          */
1268         error = in6_pcbladdr(inp, nam, &addr6, td);
1269         if (error)
1270                 goto out;
1271
1272         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1273
1274         if (port != &curthread->td_msgport) {
1275                 struct route *ro = &inp->inp_route;
1276                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1277
1278                 /*
1279                  * in_pcbladdr() may have allocated a route entry for us
1280                  * on the current CPU, but we need a route entry on the
1281                  * inpcb's owner CPU, so free it here.
1282                  */
1283                 if (ro->ro_rt != NULL)
1284                         RTFREE(ro->ro_rt);
1285                 bzero(ro, sizeof(*ro));
1286
1287                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1288                 msg->connect.nm_flags |= PRUC_RECONNECT;
1289                 msg->connect.base.nm_dispatch = tcp6_connect;
1290
1291                 /* See the related comment in tcp_connect() */
1292                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1293                 lwkt_forwardmsg(port, lmsg);
1294                 /* msg invalid now */
1295                 return;
1296         }
1297         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1298                                    &msg->connect.nm_m, sin6, addr6);
1299         /* nm_m may still be intact */
1300 out:
1301         if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1302                 tcp_connect(msg);
1303                 /* msg invalid now */
1304         } else {
1305                 if (msg->connect.nm_m) {
1306                         m_freem(msg->connect.nm_m);
1307                         msg->connect.nm_m = NULL;
1308                 }
1309                 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1310                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1311                         msg->connect.nm_nam = NULL;
1312                 }
1313                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1314                 /* msg invalid now */
1315         }
1316 }
1317
1318 static int
1319 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1320                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1321 {
1322         struct mbuf *m = *mp;
1323         struct inpcb *inp = tp->t_inpcb;
1324         struct socket *so = inp->inp_socket;
1325         struct inpcb *oinp;
1326
1327         /*
1328          * Cannot simply call in_pcbconnect, because there might be an
1329          * earlier incarnation of this same connection still in
1330          * TIME_WAIT state, creating an ADDRINUSE error.
1331          */
1332         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1333                                   &sin6->sin6_addr, sin6->sin6_port,
1334                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1335                                       addr6 : &inp->in6p_laddr),
1336                                   inp->inp_lport,  0, NULL);
1337         if (oinp)
1338                 return (EADDRINUSE);
1339
1340         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1341                 inp->in6p_laddr = *addr6;
1342         inp->in6p_faddr = sin6->sin6_addr;
1343         inp->inp_fport = sin6->sin6_port;
1344         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1345                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1346         in_pcbinsconnhash(inp);
1347
1348         /*
1349          * Now that no more errors can occur, change the protocol processing
1350          * port to the current thread (which is the correct thread).
1351          *
1352          * Create TCP timer message now; we are on the tcpcb's owner
1353          * CPU/thread.
1354          */
1355         tcp_create_timermsg(tp, &curthread->td_msgport);
1356
1357         /* Compute window scaling to request.  */
1358         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1359                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1360         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1361             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1362                 tp->request_r_scale++;
1363         }
1364
1365         soisconnecting(so);
1366         tcpstat.tcps_connattempt++;
1367         tp->t_state = TCPS_SYN_SENT;
1368         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1369         tp->iss = tcp_new_isn(tp);
1370         tcp_sendseqinit(tp);
1371         if (m) {
1372                 ssb_appendstream(&so->so_snd, m);
1373                 *mp = NULL;
1374                 if (flags & PRUS_OOB)
1375                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1376         }
1377
1378         /*
1379          * Close the send side of the connection after
1380          * the data is sent if flagged.
1381          */
1382         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1383                 socantsendmore(so);
1384                 tp = tcp_usrclosed(tp);
1385         }
1386         return (tcp_output(tp));
1387 }
1388
1389 #endif /* INET6 */
1390
1391 /*
1392  * The new sockopt interface makes it possible for us to block in the
1393  * copyin/out step (if we take a page fault).  Taking a page fault while
1394  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1395  * both now use TSM, there probably isn't any need for this function to
1396  * run in a critical section any more.  This needs more examination.)
1397  */
1398 void
1399 tcp_ctloutput(netmsg_t msg)
1400 {
1401         struct socket *so = msg->base.nm_so;
1402         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1403         int     error, opt, optval, opthz;
1404         struct  inpcb *inp;
1405         struct  tcpcb *tp;
1406
1407         error = 0;
1408         inp = so->so_pcb;
1409         if (inp == NULL) {
1410                 error = ECONNRESET;
1411                 goto done;
1412         }
1413         tp = intotcpcb(inp);
1414
1415         /* Get socket's owner cpuid hint */
1416         if (sopt->sopt_level == SOL_SOCKET &&
1417             sopt->sopt_dir == SOPT_GET &&
1418             sopt->sopt_name == SO_CPUHINT) {
1419                 if (tp->t_flags & TF_LISTEN) {
1420                         /*
1421                          * Listen sockets owner cpuid is always 0,
1422                          * which does not make sense if SO_REUSEPORT
1423                          * is not set.
1424                          */
1425                         if (so->so_options & SO_REUSEPORT)
1426                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1427                         else
1428                                 optval = -1; /* no hint */
1429                 } else {
1430                         optval = mycpuid;
1431                 }
1432                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1433                 goto done;
1434         }
1435
1436         if (sopt->sopt_level != IPPROTO_TCP) {
1437                 if (sopt->sopt_level == IPPROTO_IP) {
1438                         switch (sopt->sopt_name) {
1439                         case IP_MULTICAST_IF:
1440                         case IP_MULTICAST_VIF:
1441                         case IP_MULTICAST_TTL:
1442                         case IP_MULTICAST_LOOP:
1443                         case IP_ADD_MEMBERSHIP:
1444                         case IP_DROP_MEMBERSHIP:
1445                                 /*
1446                                  * Multicast does not make sense on
1447                                  * TCP sockets.
1448                                  */
1449                                 error = EOPNOTSUPP;
1450                                 goto done;
1451                         }
1452                 }
1453 #ifdef INET6
1454                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1455                         ip6_ctloutput_dispatch(msg);
1456                 else
1457 #endif /* INET6 */
1458                 ip_ctloutput(msg);
1459                 /* msg invalid now */
1460                 return;
1461         }
1462
1463         switch (sopt->sopt_dir) {
1464         case SOPT_SET:
1465                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1466                                       sizeof optval);
1467                 if (error)
1468                         break;
1469                 switch (sopt->sopt_name) {
1470                 case TCP_FASTKEEP:
1471                         if (optval > 0)
1472                                 tp->t_keepidle = tp->t_keepintvl;
1473                         else
1474                                 tp->t_keepidle = tcp_keepidle;
1475                         tcp_timer_keep_activity(tp, 0);
1476                         break;
1477 #ifdef TCP_SIGNATURE
1478                 case TCP_SIGNATURE_ENABLE:
1479                         if (tp->t_state == TCPS_CLOSED) {
1480                                 /*
1481                                  * This is the only safe state that this
1482                                  * option could be changed.  Some segments
1483                                  * could already have been sent in other
1484                                  * states.
1485                                  */
1486                                 if (optval > 0)
1487                                         tp->t_flags |= TF_SIGNATURE;
1488                                 else
1489                                         tp->t_flags &= ~TF_SIGNATURE;
1490                         } else {
1491                                 error = EOPNOTSUPP;
1492                         }
1493                         break;
1494 #endif /* TCP_SIGNATURE */
1495                 case TCP_NODELAY:
1496                 case TCP_NOOPT:
1497                         switch (sopt->sopt_name) {
1498                         case TCP_NODELAY:
1499                                 opt = TF_NODELAY;
1500                                 break;
1501                         case TCP_NOOPT:
1502                                 opt = TF_NOOPT;
1503                                 break;
1504                         default:
1505                                 opt = 0; /* dead code to fool gcc */
1506                                 break;
1507                         }
1508
1509                         if (optval)
1510                                 tp->t_flags |= opt;
1511                         else
1512                                 tp->t_flags &= ~opt;
1513                         break;
1514
1515                 case TCP_NOPUSH:
1516                         if (tcp_disable_nopush)
1517                                 break;
1518                         if (optval)
1519                                 tp->t_flags |= TF_NOPUSH;
1520                         else {
1521                                 tp->t_flags &= ~TF_NOPUSH;
1522                                 error = tcp_output(tp);
1523                         }
1524                         break;
1525
1526                 case TCP_MAXSEG:
1527                         /*
1528                          * Must be between 0 and maxseg.  If the requested
1529                          * maxseg is too small to satisfy the desired minmss,
1530                          * pump it up (silently so sysctl modifications of
1531                          * minmss do not create unexpected program failures).
1532                          * Handle degenerate cases.
1533                          */
1534                         if (optval > 0 && optval <= tp->t_maxseg) {
1535                                 if (optval + 40 < tcp_minmss) {
1536                                         optval = tcp_minmss - 40;
1537                                         if (optval < 0)
1538                                                 optval = 1;
1539                                 }
1540                                 tp->t_maxseg = optval;
1541                         } else {
1542                                 error = EINVAL;
1543                         }
1544                         break;
1545
1546                 case TCP_KEEPINIT:
1547                         opthz = ((int64_t)optval * hz) / 1000;
1548                         if (opthz >= 1)
1549                                 tp->t_keepinit = opthz;
1550                         else
1551                                 error = EINVAL;
1552                         break;
1553
1554                 case TCP_KEEPIDLE:
1555                         opthz = ((int64_t)optval * hz) / 1000;
1556                         if (opthz >= 1) {
1557                                 tp->t_keepidle = opthz;
1558                                 tcp_timer_keep_activity(tp, 0);
1559                         } else {
1560                                 error = EINVAL;
1561                         }
1562                         break;
1563
1564                 case TCP_KEEPINTVL:
1565                         opthz = ((int64_t)optval * hz) / 1000;
1566                         if (opthz >= 1) {
1567                                 tp->t_keepintvl = opthz;
1568                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1569                         } else {
1570                                 error = EINVAL;
1571                         }
1572                         break;
1573
1574                 case TCP_KEEPCNT:
1575                         if (optval > 0) {
1576                                 tp->t_keepcnt = optval;
1577                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1578                         } else {
1579                                 error = EINVAL;
1580                         }
1581                         break;
1582
1583                 default:
1584                         error = ENOPROTOOPT;
1585                         break;
1586                 }
1587                 break;
1588
1589         case SOPT_GET:
1590                 switch (sopt->sopt_name) {
1591 #ifdef TCP_SIGNATURE
1592                 case TCP_SIGNATURE_ENABLE:
1593                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1594                         break;
1595 #endif /* TCP_SIGNATURE */
1596                 case TCP_NODELAY:
1597                         optval = tp->t_flags & TF_NODELAY;
1598                         break;
1599                 case TCP_MAXSEG:
1600                         optval = tp->t_maxseg;
1601                         break;
1602                 case TCP_NOOPT:
1603                         optval = tp->t_flags & TF_NOOPT;
1604                         break;
1605                 case TCP_NOPUSH:
1606                         optval = tp->t_flags & TF_NOPUSH;
1607                         break;
1608                 case TCP_KEEPINIT:
1609                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1610                         break;
1611                 case TCP_KEEPIDLE:
1612                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1613                         break;
1614                 case TCP_KEEPINTVL:
1615                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1616                         break;
1617                 case TCP_KEEPCNT:
1618                         optval = tp->t_keepcnt;
1619                         break;
1620                 default:
1621                         error = ENOPROTOOPT;
1622                         break;
1623                 }
1624                 if (error == 0)
1625                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1626                 break;
1627         }
1628 done:
1629         lwkt_replymsg(&msg->lmsg, error);
1630 }
1631
1632 /*
1633  * tcp_sendspace and tcp_recvspace are the default send and receive window
1634  * sizes, respectively.  These are obsolescent (this information should
1635  * be set by the route).
1636  *
1637  * Use a default that does not require tcp window scaling to be turned
1638  * on.  Individual programs or the administrator can increase the default.
1639  */
1640 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1641 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1642     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1643 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1644 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1645     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1646
1647 /*
1648  * Attach TCP protocol to socket, allocating internet protocol control
1649  * block, tcp control block, buffer space, and entering CLOSED state.
1650  */
1651 static int
1652 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1653 {
1654         struct tcpcb *tp;
1655         struct inpcb *inp;
1656         int error;
1657         int cpu;
1658 #ifdef INET6
1659         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1660 #endif
1661
1662         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1663                 lwkt_gettoken(&so->so_rcv.ssb_token);
1664                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1665                                   ai->sb_rlimit);
1666                 lwkt_reltoken(&so->so_rcv.ssb_token);
1667                 if (error)
1668                         return (error);
1669         }
1670         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1671         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1672         cpu = mycpu->gd_cpuid;
1673
1674         /*
1675          * Set the default port for protocol processing. This will likely
1676          * change when we connect.
1677          */
1678         error = in_pcballoc(so, &tcbinfo[cpu]);
1679         if (error)
1680                 return (error);
1681         inp = so->so_pcb;
1682 #ifdef INET6
1683         if (isipv6) {
1684                 inp->inp_vflag |= INP_IPV6;
1685                 inp->in6p_hops = -1;    /* use kernel default */
1686         }
1687         else
1688 #endif
1689         inp->inp_vflag |= INP_IPV4;
1690         tp = tcp_newtcpcb(inp);
1691         if (tp == NULL) {
1692                 /*
1693                  * Make sure the socket is destroyed by the pcbdetach.
1694                  */
1695                 soreference(so);
1696 #ifdef INET6
1697                 if (isipv6)
1698                         in6_pcbdetach(inp);
1699                 else
1700 #endif
1701                 in_pcbdetach(inp);
1702                 sofree(so);     /* from ref above */
1703                 return (ENOBUFS);
1704         }
1705         tp->t_state = TCPS_CLOSED;
1706         /* Keep a reference for asynchronized pru_rcvd */
1707         soreference(so);
1708         return (0);
1709 }
1710
1711 /*
1712  * Initiate (or continue) disconnect.
1713  * If embryonic state, just send reset (once).
1714  * If in ``let data drain'' option and linger null, just drop.
1715  * Otherwise (hard), mark socket disconnecting and drop
1716  * current input data; switch states based on user close, and
1717  * send segment to peer (with FIN).
1718  */
1719 static struct tcpcb *
1720 tcp_disconnect(struct tcpcb *tp)
1721 {
1722         struct socket *so = tp->t_inpcb->inp_socket;
1723
1724         if (tp->t_state < TCPS_ESTABLISHED) {
1725                 tp = tcp_close(tp);
1726         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1727                 tp = tcp_drop(tp, 0);
1728         } else {
1729                 lwkt_gettoken(&so->so_rcv.ssb_token);
1730                 soisdisconnecting(so);
1731                 sbflush(&so->so_rcv.sb);
1732                 tp = tcp_usrclosed(tp);
1733                 if (tp)
1734                         tcp_output(tp);
1735                 lwkt_reltoken(&so->so_rcv.ssb_token);
1736         }
1737         return (tp);
1738 }
1739
1740 /*
1741  * User issued close, and wish to trail through shutdown states:
1742  * if never received SYN, just forget it.  If got a SYN from peer,
1743  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1744  * If already got a FIN from peer, then almost done; go to LAST_ACK
1745  * state.  In all other cases, have already sent FIN to peer (e.g.
1746  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1747  * for peer to send FIN or not respond to keep-alives, etc.
1748  * We can let the user exit from the close as soon as the FIN is acked.
1749  */
1750 static struct tcpcb *
1751 tcp_usrclosed(struct tcpcb *tp)
1752 {
1753
1754         switch (tp->t_state) {
1755
1756         case TCPS_CLOSED:
1757         case TCPS_LISTEN:
1758                 tp->t_state = TCPS_CLOSED;
1759                 tp = tcp_close(tp);
1760                 break;
1761
1762         case TCPS_SYN_SENT:
1763         case TCPS_SYN_RECEIVED:
1764                 tp->t_flags |= TF_NEEDFIN;
1765                 break;
1766
1767         case TCPS_ESTABLISHED:
1768                 tp->t_state = TCPS_FIN_WAIT_1;
1769                 break;
1770
1771         case TCPS_CLOSE_WAIT:
1772                 tp->t_state = TCPS_LAST_ACK;
1773                 break;
1774         }
1775         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1776                 soisdisconnected(tp->t_inpcb->inp_socket);
1777                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1778                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1779                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1780                             tcp_timer_2msl);
1781                 }
1782         }
1783         return (tp);
1784 }