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  * Allocate socket buffer space.
 176  */
 177 static int
 178 tcp_usr_preattach(struct socket *so, int proto __unused,
 179     struct pru_attach_info *ai)
 180 {
 181         int error;
 182
 183         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
 184                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
 185                                   ai->sb_rlimit);
 186                 if (error)
 187                         return (error);
 188         }
 189         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
 190         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
 191
 192         return 0;
 193 }
 194
 195 /*
 196  * TCP attaches to socket via pru_attach(), reserving space,
 197  * and an internet control block.  This socket may move to
 198  * other CPU later when we bind/connect.
 199  */
 200 static void
 201 tcp_usr_attach(netmsg_t msg)
 202 {
 203         struct socket *so = msg->base.nm_so;
 204         struct pru_attach_info *ai = msg->attach.nm_ai;
 205         int error;
 206         struct inpcb *inp;
 207         struct tcpcb *tp = NULL;
 208         TCPDEBUG0;
 209
 210         inp = so->so_pcb;
 211         KASSERT(inp == NULL, ("tcp socket attached"));
 212         TCPDEBUG1();
 213
 214         error = tcp_attach(so, ai);
 215         if (error)
 216                 goto out;
 217
 218         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 219                 so->so_linger = TCP_LINGERTIME;
 220         tp = sototcpcb(so);
 221 out:
 222         TCPDEBUG2(PRU_ATTACH);
 223         lwkt_replymsg(&msg->lmsg, error);
 224 }
 225
 226 /*
 227  * pru_detach() detaches the TCP protocol from the socket.
 228  * If the protocol state is non-embryonic, then can't
 229  * do this directly: have to initiate a pru_disconnect(),
 230  * which may finish later; embryonic TCB's can just
 231  * be discarded here.
 232  */
 233 static void
 234 tcp_usr_detach(netmsg_t msg)
 235 {
 236         struct socket *so = msg->base.nm_so;
 237         int error = 0;
 238         struct inpcb *inp;
 239         struct tcpcb *tp;
 240         TCPDEBUG0;
 241
 242         inp = so->so_pcb;
 243
 244         /*
 245          * If the inp is already detached or never attached, it may have
 246          * been due to an async close or async attach failure.  Just return
 247          * as if no error occured.
 248          */
 249         if (inp) {
 250                 tp = intotcpcb(inp);
 251                 KASSERT(tp != NULL, ("tcp_usr_detach: tp is NULL"));
 252                 TCPDEBUG1();
 253                 tp = tcp_disconnect(tp);
 254                 TCPDEBUG2(PRU_DETACH);
 255         }
 256         lwkt_replymsg(&msg->lmsg, error);
 257 }
 258
 259 /*
 260  * NOTE: ignore_error is non-zero for certain disconnection races
 261  * which we want to silently allow, otherwise close() may return
 262  * an unexpected error.
 263  *
 264  * NOTE: The variables (msg) and (tp) are assumed.
 265  */
 266 #define COMMON_START(so, inp, ignore_error)                     \
 267         TCPDEBUG0;                                              \
 268                                                                 \
 269         inp = so->so_pcb;                                       \
 270         do {                                                    \
 271                 if (inp == NULL) {                              \
 272                         error = ignore_error ? 0 : EINVAL;      \
 273                         tp = NULL;                              \
 274                         goto out;                               \
 275                 }                                               \
 276                 tp = intotcpcb(inp);                            \
 277                 TCPDEBUG1();                                    \
 278         } while(0)
 279
 280 #define COMMON_END1(req, noreply)                               \
 281         out: do {                                               \
 282                 TCPDEBUG2(req);                                 \
 283                 if (!(noreply))                                 \
 284                         lwkt_replymsg(&msg->lmsg, error);       \
 285                 return;                                         \
 286         } while(0)
 287
 288 #define COMMON_END(req)         COMMON_END1((req), 0)
 289
 290 /*
 291  * Give the socket an address.
 292  */
 293 static void
 294 tcp_usr_bind(netmsg_t msg)
 295 {
 296         struct socket *so = msg->bind.base.nm_so;
 297         struct sockaddr *nam = msg->bind.nm_nam;
 298         struct thread *td = msg->bind.nm_td;
 299         int error = 0;
 300         struct inpcb *inp;
 301         struct tcpcb *tp;
 302         struct sockaddr_in *sinp;
 303
 304         COMMON_START(so, inp, 0);
 305
 306         /*
 307          * Must check for multicast addresses and disallow binding
 308          * to them.
 309          */
 310         sinp = (struct sockaddr_in *)nam;
 311         if (sinp->sin_family == AF_INET &&
 312             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 313                 error = EAFNOSUPPORT;
 314                 goto out;
 315         }
 316         error = in_pcbbind(inp, nam, td);
 317         if (error)
 318                 goto out;
 319
 320         COMMON_END(PRU_BIND);
 321 }
 322
 323 #ifdef INET6
 324
 325 static void
 326 tcp6_usr_bind(netmsg_t msg)
 327 {
 328         struct socket *so = msg->bind.base.nm_so;
 329         struct sockaddr *nam = msg->bind.nm_nam;
 330         struct thread *td = msg->bind.nm_td;
 331         int error = 0;
 332         struct inpcb *inp;
 333         struct tcpcb *tp;
 334         struct sockaddr_in6 *sin6p;
 335
 336         COMMON_START(so, inp, 0);
 337
 338         /*
 339          * Must check for multicast addresses and disallow binding
 340          * to them.
 341          */
 342         sin6p = (struct sockaddr_in6 *)nam;
 343         if (sin6p->sin6_family == AF_INET6 &&
 344             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 345                 error = EAFNOSUPPORT;
 346                 goto out;
 347         }
 348         error = in6_pcbbind(inp, nam, td);
 349         if (error)
 350                 goto out;
 351         COMMON_END(PRU_BIND);
 352 }
 353 #endif /* INET6 */
 354
 355 struct netmsg_inswildcard {
 356         struct netmsg_base      base;
 357         struct inpcb            *nm_inp;
 358 };
 359
 360 static void
 361 in_pcbinswildcardhash_handler(netmsg_t msg)
 362 {
 363         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 364         int cpu = mycpuid, nextcpu;
 365
 366         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 367
 368         nextcpu = cpu + 1;
 369         if (nextcpu < ncpus2)
 370                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 371         else
 372                 lwkt_replymsg(&nm->base.lmsg, 0);
 373 }
 374
 375 static void
 376 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 377 {
 378         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 379 }
 380
 381 /*
 382  * Prepare to accept connections.
 383  */
 384 static void
 385 tcp_usr_listen(netmsg_t msg)
 386 {
 387         struct socket *so = msg->listen.base.nm_so;
 388         struct thread *td = msg->listen.nm_td;
 389         int error = 0;
 390         struct inpcb *inp;
 391         struct tcpcb *tp;
 392         struct netmsg_inswildcard nm;
 393         lwkt_port_t port0 = netisr_cpuport(0);
 394
 395         COMMON_START(so, inp, 0);
 396
 397         if (&curthread->td_msgport != port0) {
 398                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 399
 400                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 401                     ("already asked to relink"));
 402
 403                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 404                 msg->listen.nm_flags |= PRUL_RELINK;
 405
 406                 /* See the related comment in tcp_connect() */
 407                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 408                 lwkt_forwardmsg(port0, lmsg);
 409                 /* msg invalid now */
 410                 return;
 411         }
 412         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 413
 414         if (msg->listen.nm_flags & PRUL_RELINK) {
 415                 msg->listen.nm_flags &= ~PRUL_RELINK;
 416                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 417         }
 418         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 419
 420         if (tp->t_flags & TF_LISTEN)
 421                 goto out;
 422
 423         if (inp->inp_lport == 0) {
 424                 error = in_pcbbind(inp, NULL, td);
 425                 if (error)
 426                         goto out;
 427         }
 428
 429         tp->t_state = TCPS_LISTEN;
 430         tp->t_flags |= TF_LISTEN;
 431         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 432
 433         if (ncpus2 > 1) {
 434                 /*
 435                  * Put this inpcb into wildcard hash on other cpus.
 436                  */
 437                 ASSERT_INP_NOTINHASH(inp);
 438                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 439                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 440                 nm.nm_inp = inp;
 441                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 442         }
 443         in_pcbinswildcardhash(inp);
 444         COMMON_END(PRU_LISTEN);
 445 }
 446
 447 #ifdef INET6
 448
 449 static void
 450 tcp6_usr_listen(netmsg_t msg)
 451 {
 452         struct socket *so = msg->listen.base.nm_so;
 453         struct thread *td = msg->listen.nm_td;
 454         int error = 0;
 455         struct inpcb *inp;
 456         struct tcpcb *tp;
 457         struct netmsg_inswildcard nm;
 458
 459         COMMON_START(so, inp, 0);
 460
 461         if (tp->t_flags & TF_LISTEN)
 462                 goto out;
 463
 464         if (inp->inp_lport == 0) {
 465                 error = in6_pcbbind(inp, NULL, td);
 466                 if (error)
 467                         goto out;
 468         }
 469
 470         tp->t_state = TCPS_LISTEN;
 471         tp->t_flags |= TF_LISTEN;
 472         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 473
 474         if (ncpus2 > 1) {
 475                 /*
 476                  * Put this inpcb into wildcard hash on other cpus.
 477                  */
 478                 KKASSERT(so->so_port == netisr_cpuport(0));
 479                 ASSERT_IN_NETISR(0);
 480                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 481                 ASSERT_INP_NOTINHASH(inp);
 482
 483                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 484                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 485                 nm.nm_inp = inp;
 486                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 487         }
 488         in_pcbinswildcardhash(inp);
 489         COMMON_END(PRU_LISTEN);
 490 }
 491 #endif /* INET6 */
 492
 493 /*
 494  * Initiate connection to peer.
 495  * Create a template for use in transmissions on this connection.
 496  * Enter SYN_SENT state, and mark socket as connecting.
 497  * Start keep-alive timer, and seed output sequence space.
 498  * Send initial segment on connection.
 499  */
 500 static void
 501 tcp_usr_connect(netmsg_t msg)
 502 {
 503         struct socket *so = msg->connect.base.nm_so;
 504         struct sockaddr *nam = msg->connect.nm_nam;
 505         struct thread *td = msg->connect.nm_td;
 506         int error = 0;
 507         struct inpcb *inp;
 508         struct tcpcb *tp;
 509         struct sockaddr_in *sinp;
 510
 511         COMMON_START(so, inp, 0);
 512
 513         /*
 514          * Must disallow TCP ``connections'' to multicast addresses.
 515          */
 516         sinp = (struct sockaddr_in *)nam;
 517         if (sinp->sin_family == AF_INET
 518             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 519                 error = EAFNOSUPPORT;
 520                 goto out;
 521         }
 522
 523         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 524                 error = EAFNOSUPPORT; /* IPv6 only jail */
 525                 goto out;
 526         }
 527
 528         tcp_connect(msg);
 529         /* msg is invalid now */
 530         return;
 531 out:
 532         if (msg->connect.nm_m) {
 533                 m_freem(msg->connect.nm_m);
 534                 msg->connect.nm_m = NULL;
 535         }
 536         if (msg->connect.nm_flags & PRUC_HELDTD)
 537                 lwkt_rele(td);
 538         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 539                 so->so_error = error;
 540                 soisdisconnected(so);
 541         }
 542         lwkt_replymsg(&msg->lmsg, error);
 543 }
 544
 545 #ifdef INET6
 546
 547 static void
 548 tcp6_usr_connect(netmsg_t msg)
 549 {
 550         struct socket *so = msg->connect.base.nm_so;
 551         struct sockaddr *nam = msg->connect.nm_nam;
 552         struct thread *td = msg->connect.nm_td;
 553         int error = 0;
 554         struct inpcb *inp;
 555         struct tcpcb *tp;
 556         struct sockaddr_in6 *sin6p;
 557
 558         COMMON_START(so, inp, 0);
 559
 560         /*
 561          * Must disallow TCP ``connections'' to multicast addresses.
 562          */
 563         sin6p = (struct sockaddr_in6 *)nam;
 564         if (sin6p->sin6_family == AF_INET6
 565             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 566                 error = EAFNOSUPPORT;
 567                 goto out;
 568         }
 569
 570         if (!prison_remote_ip(td, nam)) {
 571                 error = EAFNOSUPPORT; /* IPv4 only jail */
 572                 goto out;
 573         }
 574
 575         /* Reject v4-mapped address */
 576         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 577                 error = EADDRNOTAVAIL;
 578                 goto out;
 579         }
 580
 581         inp->inp_inc.inc_isipv6 = 1;
 582         tcp6_connect(msg);
 583         /* msg is invalid now */
 584         return;
 585 out:
 586         if (msg->connect.nm_m) {
 587                 m_freem(msg->connect.nm_m);
 588                 msg->connect.nm_m = NULL;
 589         }
 590         lwkt_replymsg(&msg->lmsg, error);
 591 }
 592
 593 #endif /* INET6 */
 594
 595 /*
 596  * Initiate disconnect from peer.
 597  * If connection never passed embryonic stage, just drop;
 598  * else if don't need to let data drain, then can just drop anyways,
 599  * else have to begin TCP shutdown process: mark socket disconnecting,
 600  * drain unread data, state switch to reflect user close, and
 601  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 602  * when peer sends FIN and acks ours.
 603  *
 604  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 605  */
 606 static void
 607 tcp_usr_disconnect(netmsg_t msg)
 608 {
 609         struct socket *so = msg->disconnect.base.nm_so;
 610         int error = 0;
 611         struct inpcb *inp;
 612         struct tcpcb *tp;
 613
 614         COMMON_START(so, inp, 1);
 615         tp = tcp_disconnect(tp);
 616         COMMON_END(PRU_DISCONNECT);
 617 }
 618
 619 /*
 620  * Accept a connection.  Essentially all the work is
 621  * done at higher levels; just return the address
 622  * of the peer, storing through addr.
 623  */
 624 static void
 625 tcp_usr_accept(netmsg_t msg)
 626 {
 627         struct socket *so = msg->accept.base.nm_so;
 628         struct sockaddr **nam = msg->accept.nm_nam;
 629         int error = 0;
 630         struct inpcb *inp;
 631         struct tcpcb *tp = NULL;
 632         TCPDEBUG0;
 633
 634         inp = so->so_pcb;
 635         if (so->so_state & SS_ISDISCONNECTED) {
 636                 error = ECONNABORTED;
 637                 goto out;
 638         }
 639         if (inp == NULL) {
 640                 error = EINVAL;
 641                 goto out;
 642         }
 643
 644         tp = intotcpcb(inp);
 645         TCPDEBUG1();
 646         in_setpeeraddr(so, nam);
 647         COMMON_END(PRU_ACCEPT);
 648 }
 649
 650 #ifdef INET6
 651 static void
 652 tcp6_usr_accept(netmsg_t msg)
 653 {
 654         struct socket *so = msg->accept.base.nm_so;
 655         struct sockaddr **nam = msg->accept.nm_nam;
 656         int error = 0;
 657         struct inpcb *inp;
 658         struct tcpcb *tp = NULL;
 659         TCPDEBUG0;
 660
 661         inp = so->so_pcb;
 662
 663         if (so->so_state & SS_ISDISCONNECTED) {
 664                 error = ECONNABORTED;
 665                 goto out;
 666         }
 667         if (inp == NULL) {
 668                 error = EINVAL;
 669                 goto out;
 670         }
 671         tp = intotcpcb(inp);
 672         TCPDEBUG1();
 673         in6_setpeeraddr(so, nam);
 674         COMMON_END(PRU_ACCEPT);
 675 }
 676 #endif /* INET6 */
 677
 678 /*
 679  * Mark the connection as being incapable of further output.
 680  */
 681 static void
 682 tcp_usr_shutdown(netmsg_t msg)
 683 {
 684         struct socket *so = msg->shutdown.base.nm_so;
 685         int error = 0;
 686         struct inpcb *inp;
 687         struct tcpcb *tp;
 688
 689         COMMON_START(so, inp, 0);
 690         socantsendmore(so);
 691         tp = tcp_usrclosed(tp);
 692         if (tp)
 693                 error = tcp_output(tp);
 694         COMMON_END(PRU_SHUTDOWN);
 695 }
 696
 697 /*
 698  * After a receive, possibly send window update to peer.
 699  */
 700 static void
 701 tcp_usr_rcvd(netmsg_t msg)
 702 {
 703         struct socket *so = msg->rcvd.base.nm_so;
 704         int error = 0, noreply = 0;
 705         struct inpcb *inp;
 706         struct tcpcb *tp;
 707
 708         COMMON_START(so, inp, 0);
 709
 710         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 711                 noreply = 1;
 712                 so_async_rcvd_reply(so);
 713         }
 714         tcp_output(tp);
 715
 716         COMMON_END1(PRU_RCVD, noreply);
 717 }
 718
 719 /*
 720  * Do a send by putting data in output queue and updating urgent
 721  * marker if URG set.  Possibly send more data.  Unlike the other
 722  * pru_*() routines, the mbuf chains are our responsibility.  We
 723  * must either enqueue them or free them.  The other pru_* routines
 724  * generally are caller-frees.
 725  */
 726 static void
 727 tcp_usr_send(netmsg_t msg)
 728 {
 729         struct socket *so = msg->send.base.nm_so;
 730         int flags = msg->send.nm_flags;
 731         struct mbuf *m = msg->send.nm_m;
 732         int error = 0;
 733         struct inpcb *inp;
 734         struct tcpcb *tp;
 735         TCPDEBUG0;
 736
 737         KKASSERT(msg->send.nm_control == NULL);
 738         KKASSERT(msg->send.nm_addr == NULL);
 739         KKASSERT((flags & PRUS_FREEADDR) == 0);
 740
 741         inp = so->so_pcb;
 742
 743         if (inp == NULL) {
 744                 /*
 745                  * OOPS! we lost a race, the TCP session got reset after
 746                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 747                  * network interrupt in the non-critical section of sosend().
 748                  */
 749                 m_freem(m);
 750                 error = ECONNRESET;     /* XXX EPIPE? */
 751                 tp = NULL;
 752                 TCPDEBUG1();
 753                 goto out;
 754         }
 755         tp = intotcpcb(inp);
 756         TCPDEBUG1();
 757
 758 #ifdef foo
 759         /*
 760          * This is no longer necessary, since:
 761          * - sosendtcp() has already checked it for us
 762          * - It does not work with asynchronized send
 763          */
 764
 765         /*
 766          * Don't let too much OOB data build up
 767          */
 768         if (flags & PRUS_OOB) {
 769                 if (ssb_space(&so->so_snd) < -512) {
 770                         m_freem(m);
 771                         error = ENOBUFS;
 772                         goto out;
 773                 }
 774         }
 775 #endif
 776
 777         /*
 778          * Pump the data into the socket.
 779          */
 780         if (m) {
 781                 ssb_appendstream(&so->so_snd, m);
 782                 sowwakeup(so);
 783         }
 784         if (flags & PRUS_OOB) {
 785                 /*
 786                  * According to RFC961 (Assigned Protocols),
 787                  * the urgent pointer points to the last octet
 788                  * of urgent data.  We continue, however,
 789                  * to consider it to indicate the first octet
 790                  * of data past the urgent section.
 791                  * Otherwise, snd_up should be one lower.
 792                  */
 793                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 794                 tp->t_flags |= TF_FORCE;
 795                 error = tcp_output(tp);
 796                 tp->t_flags &= ~TF_FORCE;
 797         } else {
 798                 if (flags & PRUS_EOF) {
 799                         /*
 800                          * Close the send side of the connection after
 801                          * the data is sent.
 802                          */
 803                         socantsendmore(so);
 804                         tp = tcp_usrclosed(tp);
 805                 }
 806                 if (tp != NULL && !tcp_output_pending(tp)) {
 807                         if (flags & PRUS_MORETOCOME)
 808                                 tp->t_flags |= TF_MORETOCOME;
 809                         error = tcp_output_fair(tp);
 810                         if (flags & PRUS_MORETOCOME)
 811                                 tp->t_flags &= ~TF_MORETOCOME;
 812                 }
 813         }
 814         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 815                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 816                    (flags & PRUS_NOREPLY));
 817 }
 818
 819 /*
 820  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 821  *       will sofree() it when we return.
 822  */
 823 static void
 824 tcp_usr_abort(netmsg_t msg)
 825 {
 826         struct socket *so = msg->abort.base.nm_so;
 827         int error = 0;
 828         struct inpcb *inp;
 829         struct tcpcb *tp;
 830
 831         COMMON_START(so, inp, 1);
 832         tp = tcp_drop(tp, ECONNABORTED);
 833         COMMON_END(PRU_ABORT);
 834 }
 835
 836 /*
 837  * Receive out-of-band data.
 838  */
 839 static void
 840 tcp_usr_rcvoob(netmsg_t msg)
 841 {
 842         struct socket *so = msg->rcvoob.base.nm_so;
 843         struct mbuf *m = msg->rcvoob.nm_m;
 844         int flags = msg->rcvoob.nm_flags;
 845         int error = 0;
 846         struct inpcb *inp;
 847         struct tcpcb *tp;
 848
 849         COMMON_START(so, inp, 0);
 850         if ((so->so_oobmark == 0 &&
 851              (so->so_state & SS_RCVATMARK) == 0) ||
 852             so->so_options & SO_OOBINLINE ||
 853             tp->t_oobflags & TCPOOB_HADDATA) {
 854                 error = EINVAL;
 855                 goto out;
 856         }
 857         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 858                 error = EWOULDBLOCK;
 859                 goto out;
 860         }
 861         m->m_len = 1;
 862         *mtod(m, caddr_t) = tp->t_iobc;
 863         if ((flags & MSG_PEEK) == 0)
 864                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 865         COMMON_END(PRU_RCVOOB);
 866 }
 867
 868 static void
 869 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 870 {
 871         in_savefaddr(so, faddr);
 872 }
 873
 874 #ifdef INET6
 875 static void
 876 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 877 {
 878         in6_savefaddr(so, faddr);
 879 }
 880 #endif
 881
 882 static int
 883 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 884     struct thread *td __unused)
 885 {
 886         const struct sockaddr_in *sinp;
 887
 888         sinp = (const struct sockaddr_in *)nam;
 889         if (sinp->sin_family == AF_INET &&
 890             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 891                 return EAFNOSUPPORT;
 892
 893         soisconnecting(so);
 894         return 0;
 895 }
 896
 897 /* xxx - should be const */
 898 struct pr_usrreqs tcp_usrreqs = {
 899         .pru_abort = tcp_usr_abort,
 900         .pru_accept = tcp_usr_accept,
 901         .pru_attach = tcp_usr_attach,
 902         .pru_bind = tcp_usr_bind,
 903         .pru_connect = tcp_usr_connect,
 904         .pru_connect2 = pr_generic_notsupp,
 905         .pru_control = in_control_dispatch,
 906         .pru_detach = tcp_usr_detach,
 907         .pru_disconnect = tcp_usr_disconnect,
 908         .pru_listen = tcp_usr_listen,
 909         .pru_peeraddr = in_setpeeraddr_dispatch,
 910         .pru_rcvd = tcp_usr_rcvd,
 911         .pru_rcvoob = tcp_usr_rcvoob,
 912         .pru_send = tcp_usr_send,
 913         .pru_sense = pru_sense_null,
 914         .pru_shutdown = tcp_usr_shutdown,
 915         .pru_sockaddr = in_setsockaddr_dispatch,
 916         .pru_sosend = sosendtcp,
 917         .pru_soreceive = sorecvtcp,
 918         .pru_savefaddr = tcp_usr_savefaddr,
 919         .pru_preconnect = tcp_usr_preconnect,
 920         .pru_preattach = tcp_usr_preattach
 921 };
 922
 923 #ifdef INET6
 924 struct pr_usrreqs tcp6_usrreqs = {
 925         .pru_abort = tcp_usr_abort,
 926         .pru_accept = tcp6_usr_accept,
 927         .pru_attach = tcp_usr_attach,
 928         .pru_bind = tcp6_usr_bind,
 929         .pru_connect = tcp6_usr_connect,
 930         .pru_connect2 = pr_generic_notsupp,
 931         .pru_control = in6_control_dispatch,
 932         .pru_detach = tcp_usr_detach,
 933         .pru_disconnect = tcp_usr_disconnect,
 934         .pru_listen = tcp6_usr_listen,
 935         .pru_peeraddr = in6_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 = in6_setsockaddr_dispatch,
 942         .pru_sosend = sosendtcp,
 943         .pru_soreceive = sorecvtcp,
 944         .pru_savefaddr = tcp6_usr_savefaddr
 945 };
 946 #endif /* INET6 */
 947
 948 static int
 949 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 950                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 951 {
 952         struct inpcb *inp = tp->t_inpcb, *oinp;
 953         struct socket *so = inp->inp_socket;
 954         struct route *ro = &inp->inp_route;
 955
 956         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
 957             ("pcbinfo mismatch"));
 958
 959         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
 960                                  sin->sin_addr, sin->sin_port,
 961                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 962                                   inp->inp_laddr : if_sin->sin_addr),
 963                                 inp->inp_lport, 0, NULL);
 964         if (oinp != NULL) {
 965                 m_freem(m);
 966                 return (EADDRINUSE);
 967         }
 968         if (inp->inp_laddr.s_addr == INADDR_ANY)
 969                 inp->inp_laddr = if_sin->sin_addr;
 970         inp->inp_faddr = sin->sin_addr;
 971         inp->inp_fport = sin->sin_port;
 972         in_pcbinsconnhash(inp);
 973
 974         /*
 975          * We are now on the inpcb's owner CPU, if the cached route was
 976          * freed because the rtentry's owner CPU is not the current CPU
 977          * (e.g. in tcp_connect()), then we try to reallocate it here with
 978          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 979          * rtentry.
 980          */
 981         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 982             ro->ro_rt == NULL) {
 983                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 984                 ro->ro_dst.sa_family = AF_INET;
 985                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 986                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 987                         sin->sin_addr;
 988                 rtalloc(ro);
 989         }
 990
 991         /*
 992          * Now that no more errors can occur, change the protocol processing
 993          * port to the current thread (which is the correct thread).
 994          *
 995          * Create TCP timer message now; we are on the tcpcb's owner
 996          * CPU/thread.
 997          */
 998         tcp_create_timermsg(tp, &curthread->td_msgport);
 999
1000         /*
1001          * Compute window scaling to request.  Use a larger scaling then
1002          * needed for the initial receive buffer in case the receive buffer
1003          * gets expanded.
1004          */
1005         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1006                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1007         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1008                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1009         ) {
1010                 tp->request_r_scale++;
1011         }
1012
1013         soisconnecting(so);
1014         tcpstat.tcps_connattempt++;
1015         tp->t_state = TCPS_SYN_SENT;
1016         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1017         tp->iss = tcp_new_isn(tp);
1018         tcp_sendseqinit(tp);
1019         if (m) {
1020                 ssb_appendstream(&so->so_snd, m);
1021                 m = NULL;
1022                 if (flags & PRUS_OOB)
1023                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1024         }
1025
1026         /*
1027          * Close the send side of the connection after
1028          * the data is sent if flagged.
1029          */
1030         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1031                 socantsendmore(so);
1032                 tp = tcp_usrclosed(tp);
1033         }
1034         return (tcp_output(tp));
1035 }
1036
1037 /*
1038  * Common subroutine to open a TCP connection to remote host specified
1039  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1040  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1041  * a local host address (interface).
1042  * Initialize connection parameters and enter SYN-SENT state.
1043  */
1044 static void
1045 tcp_connect(netmsg_t msg)
1046 {
1047         struct socket *so = msg->connect.base.nm_so;
1048         struct sockaddr *nam = msg->connect.nm_nam;
1049         struct thread *td = msg->connect.nm_td;
1050         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1051         struct sockaddr_in *if_sin = NULL;
1052         struct inpcb *inp;
1053         struct tcpcb *tp;
1054         int error;
1055         lwkt_port_t port;
1056
1057         COMMON_START(so, inp, 0);
1058
1059         /*
1060          * Reconnect our pcb if we have to
1061          */
1062         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1063                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1064                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1065         }
1066
1067         /*
1068          * Bind if we have to
1069          */
1070         if (inp->inp_lport == 0) {
1071                 if (tcp_lport_extension) {
1072                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1073
1074                         error = in_pcbladdr(inp, nam, &if_sin, td);
1075                         if (error)
1076                                 goto out;
1077                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1078
1079                         error = in_pcbbind_remote(inp, nam, td);
1080                         if (error)
1081                                 goto out;
1082
1083                         msg->connect.nm_flags |= PRUC_HASLADDR;
1084                 } else {
1085                         error = in_pcbbind(inp, NULL, td);
1086                         if (error)
1087                                 goto out;
1088                 }
1089         }
1090
1091         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1092                 /*
1093                  * Calculate the correct protocol processing thread.  The
1094                  * connect operation must run there.  Set the forwarding
1095                  * port before we forward the message or it will get bounced
1096                  * right back to us.
1097                  */
1098                 error = in_pcbladdr(inp, nam, &if_sin, td);
1099                 if (error)
1100                         goto out;
1101         }
1102         KKASSERT(inp->inp_socket == so);
1103
1104         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1105                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1106                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1107                             inp->inp_lport);
1108
1109         if (port != &curthread->td_msgport) {
1110                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1111
1112                 /*
1113                  * in_pcbladdr() may have allocated a route entry for us
1114                  * on the current CPU, but we need a route entry on the
1115                  * inpcb's owner CPU, so free it here.
1116                  */
1117                 in_pcbresetroute(inp);
1118
1119                 /*
1120                  * We are moving the protocol processing port the socket
1121                  * is on, we have to unlink here and re-link on the
1122                  * target cpu.
1123                  */
1124                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1125                 msg->connect.nm_flags |= PRUC_RECONNECT;
1126                 msg->connect.base.nm_dispatch = tcp_connect;
1127
1128                 /*
1129                  * Use message put done receipt to change this socket's
1130                  * so_port, i.e. _after_ this message was put onto the
1131                  * target netisr's msgport but _before_ the message could
1132                  * be pulled from the target netisr's msgport, so that:
1133                  * - The upper half (socket code) will not see the new
1134                  *   msgport before this message reaches the new msgport
1135                  *   and messages for this socket will be ordered.
1136                  * - This message will see the new msgport, when its
1137                  *   handler is called in the target netisr.
1138                  *
1139                  * NOTE:
1140                  * We MUST use messege put done receipt to change this
1141                  * socket's so_port:
1142                  * If we changed the so_port in this netisr after the
1143                  * lwkt_forwardmsg (so messages for this socket will be
1144                  * ordered) and changed the so_port in the target netisr
1145                  * at the very beginning of this message's handler, we
1146                  * would suffer so_port overwritten race, given this
1147                  * message might be forwarded again.
1148                  *
1149                  * NOTE:
1150                  * This mechanism depends on that the netisr's msgport
1151                  * is spin msgport (currently it is :).
1152                  *
1153                  * If the upper half saw the new msgport before this
1154                  * message reached the target netisr's msgport, the
1155                  * messages sent from the upper half could reach the new
1156                  * msgport before this message, thus there would be
1157                  * message reordering.  The worst case could be soclose()
1158                  * saw the new msgport and the detach message could reach
1159                  * the new msgport before this message, i.e. the inpcb
1160                  * could have been destroyed when this message was still
1161                  * pending on or on its way to the new msgport.  Other
1162                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1163                  * since we have unlinked this inpcb from the current
1164                  * pcbinfo first.
1165                  */
1166                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1167                 lwkt_forwardmsg(port, lmsg);
1168                 /* msg invalid now */
1169                 return;
1170         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1171                 /*
1172                  * The original thread is no longer needed; release it.
1173                  */
1174                 lwkt_rele(td);
1175                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1176         }
1177         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1178                                   msg->connect.nm_m, sin, if_sin);
1179         msg->connect.nm_m = NULL;
1180 out:
1181         if (msg->connect.nm_m) {
1182                 m_freem(msg->connect.nm_m);
1183                 msg->connect.nm_m = NULL;
1184         }
1185         if (msg->connect.nm_flags & PRUC_HELDTD)
1186                 lwkt_rele(td);
1187         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1188                 so->so_error = error;
1189                 soisdisconnected(so);
1190         }
1191         lwkt_replymsg(&msg->connect.base.lmsg, error);
1192         /* msg invalid now */
1193 }
1194
1195 #ifdef INET6
1196
1197 static void
1198 tcp6_connect(netmsg_t msg)
1199 {
1200         struct tcpcb *tp;
1201         struct socket *so = msg->connect.base.nm_so;
1202         struct sockaddr *nam = msg->connect.nm_nam;
1203         struct thread *td = msg->connect.nm_td;
1204         struct inpcb *inp;
1205         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1206         struct in6_addr *addr6;
1207         lwkt_port_t port;
1208         int error;
1209
1210         COMMON_START(so, inp, 0);
1211
1212         /*
1213          * Reconnect our pcb if we have to
1214          */
1215         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1216                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1217                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1218         }
1219
1220         /*
1221          * Bind if we have to
1222          */
1223         if (inp->inp_lport == 0) {
1224                 error = in6_pcbbind(inp, NULL, td);
1225                 if (error)
1226                         goto out;
1227         }
1228
1229         /*
1230          * Cannot simply call in_pcbconnect, because there might be an
1231          * earlier incarnation of this same connection still in
1232          * TIME_WAIT state, creating an ADDRINUSE error.
1233          */
1234         error = in6_pcbladdr(inp, nam, &addr6, td);
1235         if (error)
1236                 goto out;
1237
1238         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1239
1240         if (port != &curthread->td_msgport) {
1241                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1242
1243                 /*
1244                  * in_pcbladdr() may have allocated a route entry for us
1245                  * on the current CPU, but we need a route entry on the
1246                  * inpcb's owner CPU, so free it here.
1247                  */
1248                 in_pcbresetroute(inp);
1249
1250                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1251                 msg->connect.nm_flags |= PRUC_RECONNECT;
1252                 msg->connect.base.nm_dispatch = tcp6_connect;
1253
1254                 /* See the related comment in tcp_connect() */
1255                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1256                 lwkt_forwardmsg(port, lmsg);
1257                 /* msg invalid now */
1258                 return;
1259         }
1260         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1261                                    &msg->connect.nm_m, sin6, addr6);
1262         /* nm_m may still be intact */
1263 out:
1264         if (msg->connect.nm_m) {
1265                 m_freem(msg->connect.nm_m);
1266                 msg->connect.nm_m = NULL;
1267         }
1268         lwkt_replymsg(&msg->connect.base.lmsg, error);
1269         /* msg invalid now */
1270 }
1271
1272 static int
1273 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1274                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1275 {
1276         struct mbuf *m = *mp;
1277         struct inpcb *inp = tp->t_inpcb;
1278         struct socket *so = inp->inp_socket;
1279         struct inpcb *oinp;
1280
1281         /*
1282          * Cannot simply call in_pcbconnect, because there might be an
1283          * earlier incarnation of this same connection still in
1284          * TIME_WAIT state, creating an ADDRINUSE error.
1285          */
1286         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1287                                   &sin6->sin6_addr, sin6->sin6_port,
1288                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1289                                       addr6 : &inp->in6p_laddr),
1290                                   inp->inp_lport,  0, NULL);
1291         if (oinp)
1292                 return (EADDRINUSE);
1293
1294         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1295                 inp->in6p_laddr = *addr6;
1296         inp->in6p_faddr = sin6->sin6_addr;
1297         inp->inp_fport = sin6->sin6_port;
1298         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1299                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1300         in_pcbinsconnhash(inp);
1301
1302         /*
1303          * Now that no more errors can occur, change the protocol processing
1304          * port to the current thread (which is the correct thread).
1305          *
1306          * Create TCP timer message now; we are on the tcpcb's owner
1307          * CPU/thread.
1308          */
1309         tcp_create_timermsg(tp, &curthread->td_msgport);
1310
1311         /* Compute window scaling to request.  */
1312         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1313                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1314         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1315             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1316                 tp->request_r_scale++;
1317         }
1318
1319         soisconnecting(so);
1320         tcpstat.tcps_connattempt++;
1321         tp->t_state = TCPS_SYN_SENT;
1322         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1323         tp->iss = tcp_new_isn(tp);
1324         tcp_sendseqinit(tp);
1325         if (m) {
1326                 ssb_appendstream(&so->so_snd, m);
1327                 *mp = NULL;
1328                 if (flags & PRUS_OOB)
1329                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1330         }
1331
1332         /*
1333          * Close the send side of the connection after
1334          * the data is sent if flagged.
1335          */
1336         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1337                 socantsendmore(so);
1338                 tp = tcp_usrclosed(tp);
1339         }
1340         return (tcp_output(tp));
1341 }
1342
1343 #endif /* INET6 */
1344
1345 /*
1346  * The new sockopt interface makes it possible for us to block in the
1347  * copyin/out step (if we take a page fault).  Taking a page fault while
1348  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1349  * both now use TSM, there probably isn't any need for this function to
1350  * run in a critical section any more.  This needs more examination.)
1351  */
1352 void
1353 tcp_ctloutput(netmsg_t msg)
1354 {
1355         struct socket *so = msg->base.nm_so;
1356         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1357         int     error, opt, optval, opthz;
1358         struct  inpcb *inp;
1359         struct  tcpcb *tp;
1360
1361         error = 0;
1362         inp = so->so_pcb;
1363         if (inp == NULL) {
1364                 error = ECONNRESET;
1365                 goto done;
1366         }
1367         tp = intotcpcb(inp);
1368
1369         /* Get socket's owner cpuid hint */
1370         if (sopt->sopt_level == SOL_SOCKET &&
1371             sopt->sopt_dir == SOPT_GET &&
1372             sopt->sopt_name == SO_CPUHINT) {
1373                 if (tp->t_flags & TF_LISTEN) {
1374                         /*
1375                          * Listen sockets owner cpuid is always 0,
1376                          * which does not make sense if SO_REUSEPORT
1377                          * is not set.
1378                          */
1379                         if (so->so_options & SO_REUSEPORT)
1380                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1381                         else
1382                                 optval = -1; /* no hint */
1383                 } else {
1384                         optval = mycpuid;
1385                 }
1386                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1387                 goto done;
1388         }
1389
1390         if (sopt->sopt_level != IPPROTO_TCP) {
1391                 if (sopt->sopt_level == IPPROTO_IP) {
1392                         switch (sopt->sopt_name) {
1393                         case IP_MULTICAST_IF:
1394                         case IP_MULTICAST_VIF:
1395                         case IP_MULTICAST_TTL:
1396                         case IP_MULTICAST_LOOP:
1397                         case IP_ADD_MEMBERSHIP:
1398                         case IP_DROP_MEMBERSHIP:
1399                                 /*
1400                                  * Multicast does not make sense on
1401                                  * TCP sockets.
1402                                  */
1403                                 error = EOPNOTSUPP;
1404                                 goto done;
1405                         }
1406                 }
1407 #ifdef INET6
1408                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1409                         ip6_ctloutput_dispatch(msg);
1410                 else
1411 #endif /* INET6 */
1412                 ip_ctloutput(msg);
1413                 /* msg invalid now */
1414                 return;
1415         }
1416
1417         switch (sopt->sopt_dir) {
1418         case SOPT_SET:
1419                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1420                                       sizeof optval);
1421                 if (error)
1422                         break;
1423                 switch (sopt->sopt_name) {
1424                 case TCP_FASTKEEP:
1425                         if (optval > 0)
1426                                 tp->t_keepidle = tp->t_keepintvl;
1427                         else
1428                                 tp->t_keepidle = tcp_keepidle;
1429                         tcp_timer_keep_activity(tp, 0);
1430                         break;
1431 #ifdef TCP_SIGNATURE
1432                 case TCP_SIGNATURE_ENABLE:
1433                         if (tp->t_state == TCPS_CLOSED) {
1434                                 /*
1435                                  * This is the only safe state that this
1436                                  * option could be changed.  Some segments
1437                                  * could already have been sent in other
1438                                  * states.
1439                                  */
1440                                 if (optval > 0)
1441                                         tp->t_flags |= TF_SIGNATURE;
1442                                 else
1443                                         tp->t_flags &= ~TF_SIGNATURE;
1444                         } else {
1445                                 error = EOPNOTSUPP;
1446                         }
1447                         break;
1448 #endif /* TCP_SIGNATURE */
1449                 case TCP_NODELAY:
1450                 case TCP_NOOPT:
1451                         switch (sopt->sopt_name) {
1452                         case TCP_NODELAY:
1453                                 opt = TF_NODELAY;
1454                                 break;
1455                         case TCP_NOOPT:
1456                                 opt = TF_NOOPT;
1457                                 break;
1458                         default:
1459                                 opt = 0; /* dead code to fool gcc */
1460                                 break;
1461                         }
1462
1463                         if (optval)
1464                                 tp->t_flags |= opt;
1465                         else
1466                                 tp->t_flags &= ~opt;
1467                         break;
1468
1469                 case TCP_NOPUSH:
1470                         if (tcp_disable_nopush)
1471                                 break;
1472                         if (optval)
1473                                 tp->t_flags |= TF_NOPUSH;
1474                         else {
1475                                 tp->t_flags &= ~TF_NOPUSH;
1476                                 error = tcp_output(tp);
1477                         }
1478                         break;
1479
1480                 case TCP_MAXSEG:
1481                         /*
1482                          * Must be between 0 and maxseg.  If the requested
1483                          * maxseg is too small to satisfy the desired minmss,
1484                          * pump it up (silently so sysctl modifications of
1485                          * minmss do not create unexpected program failures).
1486                          * Handle degenerate cases.
1487                          */
1488                         if (optval > 0 && optval <= tp->t_maxseg) {
1489                                 if (optval + 40 < tcp_minmss) {
1490                                         optval = tcp_minmss - 40;
1491                                         if (optval < 0)
1492                                                 optval = 1;
1493                                 }
1494                                 tp->t_maxseg = optval;
1495                         } else {
1496                                 error = EINVAL;
1497                         }
1498                         break;
1499
1500                 case TCP_KEEPINIT:
1501                         opthz = ((int64_t)optval * hz) / 1000;
1502                         if (opthz >= 1)
1503                                 tp->t_keepinit = opthz;
1504                         else
1505                                 error = EINVAL;
1506                         break;
1507
1508                 case TCP_KEEPIDLE:
1509                         opthz = ((int64_t)optval * hz) / 1000;
1510                         if (opthz >= 1) {
1511                                 tp->t_keepidle = opthz;
1512                                 tcp_timer_keep_activity(tp, 0);
1513                         } else {
1514                                 error = EINVAL;
1515                         }
1516                         break;
1517
1518                 case TCP_KEEPINTVL:
1519                         opthz = ((int64_t)optval * hz) / 1000;
1520                         if (opthz >= 1) {
1521                                 tp->t_keepintvl = opthz;
1522                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1523                         } else {
1524                                 error = EINVAL;
1525                         }
1526                         break;
1527
1528                 case TCP_KEEPCNT:
1529                         if (optval > 0) {
1530                                 tp->t_keepcnt = optval;
1531                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1532                         } else {
1533                                 error = EINVAL;
1534                         }
1535                         break;
1536
1537                 default:
1538                         error = ENOPROTOOPT;
1539                         break;
1540                 }
1541                 break;
1542
1543         case SOPT_GET:
1544                 switch (sopt->sopt_name) {
1545 #ifdef TCP_SIGNATURE
1546                 case TCP_SIGNATURE_ENABLE:
1547                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1548                         break;
1549 #endif /* TCP_SIGNATURE */
1550                 case TCP_NODELAY:
1551                         optval = tp->t_flags & TF_NODELAY;
1552                         break;
1553                 case TCP_MAXSEG:
1554                         optval = tp->t_maxseg;
1555                         break;
1556                 case TCP_NOOPT:
1557                         optval = tp->t_flags & TF_NOOPT;
1558                         break;
1559                 case TCP_NOPUSH:
1560                         optval = tp->t_flags & TF_NOPUSH;
1561                         break;
1562                 case TCP_KEEPINIT:
1563                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1564                         break;
1565                 case TCP_KEEPIDLE:
1566                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1567                         break;
1568                 case TCP_KEEPINTVL:
1569                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1570                         break;
1571                 case TCP_KEEPCNT:
1572                         optval = tp->t_keepcnt;
1573                         break;
1574                 default:
1575                         error = ENOPROTOOPT;
1576                         break;
1577                 }
1578                 if (error == 0)
1579                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1580                 break;
1581         }
1582 done:
1583         lwkt_replymsg(&msg->lmsg, error);
1584 }
1585
1586 /*
1587  * tcp_sendspace and tcp_recvspace are the default send and receive window
1588  * sizes, respectively.  These are obsolescent (this information should
1589  * be set by the route).
1590  *
1591  * Use a default that does not require tcp window scaling to be turned
1592  * on.  Individual programs or the administrator can increase the default.
1593  */
1594 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1595 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1596     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1597 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1598 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1599     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1600
1601 /*
1602  * Attach TCP protocol to socket, allocating internet protocol control
1603  * block, tcp control block, buffer space, and entering CLOSED state.
1604  */
1605 static int
1606 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1607 {
1608         struct tcpcb *tp;
1609         struct inpcb *inp;
1610         int error;
1611         int cpu;
1612 #ifdef INET6
1613         boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1614 #endif
1615
1616         if (ai != NULL) {
1617                 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1618                 if (error)
1619                         return (error);
1620         } else {
1621                 /* Post attach; do nothing */
1622         }
1623
1624         cpu = mycpu->gd_cpuid;
1625
1626         /*
1627          * Set the default pcbinfo.  This will likely change when we
1628          * bind/connect.
1629          */
1630         error = in_pcballoc(so, &tcbinfo[cpu]);
1631         if (error)
1632                 return (error);
1633         inp = so->so_pcb;
1634 #ifdef INET6
1635         if (isipv6)
1636                 inp->in6p_hops = -1;    /* use kernel default */
1637 #endif
1638         tp = tcp_newtcpcb(inp);
1639         KASSERT(tp != NULL, ("tcp_newtcpcb failed"));
1640         tp->t_state = TCPS_CLOSED;
1641         /* Keep a reference for asynchronized pru_rcvd */
1642         soreference(so);
1643         return (0);
1644 }
1645
1646 /*
1647  * Initiate (or continue) disconnect.
1648  * If embryonic state, just send reset (once).
1649  * If in ``let data drain'' option and linger null, just drop.
1650  * Otherwise (hard), mark socket disconnecting and drop
1651  * current input data; switch states based on user close, and
1652  * send segment to peer (with FIN).
1653  */
1654 static struct tcpcb *
1655 tcp_disconnect(struct tcpcb *tp)
1656 {
1657         struct socket *so = tp->t_inpcb->inp_socket;
1658
1659         if (tp->t_state < TCPS_ESTABLISHED) {
1660                 tp = tcp_close(tp);
1661         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1662                 tp = tcp_drop(tp, 0);
1663         } else {
1664                 lwkt_gettoken(&so->so_rcv.ssb_token);
1665                 soisdisconnecting(so);
1666                 sbflush(&so->so_rcv.sb);
1667                 tp = tcp_usrclosed(tp);
1668                 if (tp)
1669                         tcp_output(tp);
1670                 lwkt_reltoken(&so->so_rcv.ssb_token);
1671         }
1672         return (tp);
1673 }
1674
1675 /*
1676  * User issued close, and wish to trail through shutdown states:
1677  * if never received SYN, just forget it.  If got a SYN from peer,
1678  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1679  * If already got a FIN from peer, then almost done; go to LAST_ACK
1680  * state.  In all other cases, have already sent FIN to peer (e.g.
1681  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1682  * for peer to send FIN or not respond to keep-alives, etc.
1683  * We can let the user exit from the close as soon as the FIN is acked.
1684  */
1685 static struct tcpcb *
1686 tcp_usrclosed(struct tcpcb *tp)
1687 {
1688
1689         switch (tp->t_state) {
1690
1691         case TCPS_CLOSED:
1692         case TCPS_LISTEN:
1693                 tp->t_state = TCPS_CLOSED;
1694                 tp = tcp_close(tp);
1695                 break;
1696
1697         case TCPS_SYN_SENT:
1698         case TCPS_SYN_RECEIVED:
1699                 tp->t_flags |= TF_NEEDFIN;
1700                 break;
1701
1702         case TCPS_ESTABLISHED:
1703                 tp->t_state = TCPS_FIN_WAIT_1;
1704                 break;
1705
1706         case TCPS_CLOSE_WAIT:
1707                 tp->t_state = TCPS_LAST_ACK;
1708                 break;
1709         }
1710         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1711                 soisdisconnected(tp->t_inpcb->inp_socket);
1712                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1713                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1714                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1715                             tcp_timer_2msl);
1716                 }
1717         }
1718         return (tp);
1719 }