sys/netinet/tcp_usrreq.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. All advertising materials mentioning features or use of this software
  47  *    must display the following acknowledgement:
  48  *      This product includes software developed by the University of
  49  *      California, Berkeley and its contributors.
  50  * 4. Neither the name of the University nor the names of its contributors
  51  *    may be used to endorse or promote products derived from this software
  52  *    without specific prior written permission.
  53  *
  54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  64  * SUCH DAMAGE.
  65  *
  66  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  67  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  68  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
  69  */
  70
  71 #include "opt_ipsec.h"
  72 #include "opt_inet.h"
  73 #include "opt_inet6.h"
  74 #include "opt_tcpdebug.h"
  75
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/kernel.h>
  79 #include <sys/malloc.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/globaldata.h>
  82 #include <sys/thread.h>
  83
  84 #include <sys/mbuf.h>
  85 #ifdef INET6
  86 #include <sys/domain.h>
  87 #endif /* INET6 */
  88 #include <sys/socket.h>
  89 #include <sys/socketvar.h>
  90 #include <sys/protosw.h>
  91
  92 #include <sys/thread2.h>
  93 #include <sys/msgport2.h>
  94 #include <sys/socketvar2.h>
  95
  96 #include <net/if.h>
  97 #include <net/netisr.h>
  98 #include <net/route.h>
  99
 100 #include <net/netmsg2.h>
 101
 102 #include <netinet/in.h>
 103 #include <netinet/in_systm.h>
 104 #ifdef INET6
 105 #include <netinet/ip6.h>
 106 #endif
 107 #include <netinet/in_pcb.h>
 108 #ifdef INET6
 109 #include <netinet6/in6_pcb.h>
 110 #endif
 111 #include <netinet/in_var.h>
 112 #include <netinet/ip_var.h>
 113 #ifdef INET6
 114 #include <netinet6/ip6_var.h>
 115 #include <netinet6/tcp6_var.h>
 116 #endif
 117 #include <netinet/tcp.h>
 118 #include <netinet/tcp_fsm.h>
 119 #include <netinet/tcp_seq.h>
 120 #include <netinet/tcp_timer.h>
 121 #include <netinet/tcp_timer2.h>
 122 #include <netinet/tcp_var.h>
 123 #include <netinet/tcpip.h>
 124 #ifdef TCPDEBUG
 125 #include <netinet/tcp_debug.h>
 126 #endif
 127
 128 #ifdef IPSEC
 129 #include <netinet6/ipsec.h>
 130 #endif /*IPSEC*/
 131
 132 /*
 133  * TCP protocol interface to socket abstraction.
 134  */
 135 extern  char *tcpstates[];      /* XXX ??? */
 136
 137 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 138 static void     tcp_connect (netmsg_t msg);
 139 #ifdef INET6
 140 static void     tcp6_connect (netmsg_t msg);
 141 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 142                                 struct mbuf **mp,
 143                                 struct sockaddr_in6 *sin6,
 144                                 struct in6_addr *addr6);
 145 #endif /* INET6 */
 146 static struct tcpcb *
 147                 tcp_disconnect (struct tcpcb *);
 148 static struct tcpcb *
 149                 tcp_usrclosed (struct tcpcb *);
 150
 151 #ifdef TCPDEBUG
 152 #define TCPDEBUG0       int ostate = 0
 153 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 154 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 155                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 156 #else
 157 #define TCPDEBUG0
 158 #define TCPDEBUG1()
 159 #define TCPDEBUG2(req)
 160 #endif
 161
 162 static int      tcp_lport_extension = 1;
 163
 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 165     &tcp_lport_extension, 0, "");
 166
 167 /*
 168  * TCP attaches to socket via pru_attach(), reserving space,
 169  * and an internet control block.  This is likely occuring on
 170  * cpu0 and may have to move later when we bind/connect.
 171  */
 172 static void
 173 tcp_usr_attach(netmsg_t msg)
 174 {
 175         struct socket *so = msg->base.nm_so;
 176         struct pru_attach_info *ai = msg->attach.nm_ai;
 177         int error;
 178         struct inpcb *inp;
 179         struct tcpcb *tp = 0;
 180         TCPDEBUG0;
 181
 182         soreference(so);
 183         inp = so->so_pcb;
 184         TCPDEBUG1();
 185         if (inp) {
 186                 error = EISCONN;
 187                 goto out;
 188         }
 189
 190         error = tcp_attach(so, ai);
 191         if (error)
 192                 goto out;
 193
 194         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 195                 so->so_linger = TCP_LINGERTIME;
 196         tp = sototcpcb(so);
 197 out:
 198         sofree(so);             /* from ref above */
 199         TCPDEBUG2(PRU_ATTACH);
 200         lwkt_replymsg(&msg->lmsg, error);
 201 }
 202
 203 /*
 204  * pru_detach() detaches the TCP protocol from the socket.
 205  * If the protocol state is non-embryonic, then can't
 206  * do this directly: have to initiate a pru_disconnect(),
 207  * which may finish later; embryonic TCB's can just
 208  * be discarded here.
 209  */
 210 static void
 211 tcp_usr_detach(netmsg_t msg)
 212 {
 213         struct socket *so = msg->base.nm_so;
 214         int error = 0;
 215         struct inpcb *inp;
 216         struct tcpcb *tp;
 217         TCPDEBUG0;
 218
 219         inp = so->so_pcb;
 220
 221         /*
 222          * If the inp is already detached it may have been due to an async
 223          * close.  Just return as if no error occured.
 224          *
 225          * It's possible for the tcpcb (tp) to disconnect from the inp due
 226          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 227          * the detach message has been queued so we may find a NULL tp here.
 228          */
 229         if (inp) {
 230                 if ((tp = intotcpcb(inp)) != NULL) {
 231                         TCPDEBUG1();
 232                         tp = tcp_disconnect(tp);
 233                         TCPDEBUG2(PRU_DETACH);
 234                 }
 235         }
 236         lwkt_replymsg(&msg->lmsg, error);
 237 }
 238
 239 /*
 240  * NOTE: ignore_error is non-zero for certain disconnection races
 241  * which we want to silently allow, otherwise close() may return
 242  * an unexpected error.
 243  *
 244  * NOTE: The variables (msg) and (tp) are assumed.
 245  */
 246 #define COMMON_START(so, inp, ignore_error)                     \
 247         TCPDEBUG0;                                              \
 248                                                                 \
 249         inp = so->so_pcb;                                       \
 250         do {                                                    \
 251                  if (inp == NULL) {                             \
 252                         error = ignore_error ? 0 : EINVAL;      \
 253                         tp = NULL;                              \
 254                         goto out;                               \
 255                  }                                              \
 256                  tp = intotcpcb(inp);                           \
 257                  TCPDEBUG1();                                   \
 258         } while(0)
 259
 260 #define COMMON_END(req)                                         \
 261         out: do {                                               \
 262                 TCPDEBUG2(req);                                 \
 263                 lwkt_replymsg(&msg->lmsg, error);               \
 264                 return;                                         \
 265         } while(0)
 266
 267 /*
 268  * Give the socket an address.
 269  */
 270 static void
 271 tcp_usr_bind(netmsg_t msg)
 272 {
 273         struct socket *so = msg->bind.base.nm_so;
 274         struct sockaddr *nam = msg->bind.nm_nam;
 275         struct thread *td = msg->bind.nm_td;
 276         int error = 0;
 277         struct inpcb *inp;
 278         struct tcpcb *tp;
 279         struct sockaddr_in *sinp;
 280
 281         COMMON_START(so, inp, 0);
 282
 283         /*
 284          * Must check for multicast addresses and disallow binding
 285          * to them.
 286          */
 287         sinp = (struct sockaddr_in *)nam;
 288         if (sinp->sin_family == AF_INET &&
 289             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 290                 error = EAFNOSUPPORT;
 291                 goto out;
 292         }
 293         error = in_pcbbind(inp, nam, td);
 294         if (error)
 295                 goto out;
 296         COMMON_END(PRU_BIND);
 297
 298 }
 299
 300 #ifdef INET6
 301
 302 static void
 303 tcp6_usr_bind(netmsg_t msg)
 304 {
 305         struct socket *so = msg->bind.base.nm_so;
 306         struct sockaddr *nam = msg->bind.nm_nam;
 307         struct thread *td = msg->bind.nm_td;
 308         int error = 0;
 309         struct inpcb *inp;
 310         struct tcpcb *tp;
 311         struct sockaddr_in6 *sin6p;
 312
 313         COMMON_START(so, inp, 0);
 314
 315         /*
 316          * Must check for multicast addresses and disallow binding
 317          * to them.
 318          */
 319         sin6p = (struct sockaddr_in6 *)nam;
 320         if (sin6p->sin6_family == AF_INET6 &&
 321             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 322                 error = EAFNOSUPPORT;
 323                 goto out;
 324         }
 325         inp->inp_vflag &= ~INP_IPV4;
 326         inp->inp_vflag |= INP_IPV6;
 327         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 328                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 329                         inp->inp_vflag |= INP_IPV4;
 330                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 331                         struct sockaddr_in sin;
 332
 333                         in6_sin6_2_sin(&sin, sin6p);
 334                         inp->inp_vflag |= INP_IPV4;
 335                         inp->inp_vflag &= ~INP_IPV6;
 336                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 337                         goto out;
 338                 }
 339         }
 340         error = in6_pcbbind(inp, nam, td);
 341         if (error)
 342                 goto out;
 343         COMMON_END(PRU_BIND);
 344 }
 345 #endif /* INET6 */
 346
 347 #ifdef SMP
 348
 349 struct netmsg_inswildcard {
 350         struct netmsg_base      base;
 351         struct inpcb            *nm_inp;
 352 };
 353
 354 static void
 355 in_pcbinswildcardhash_handler(netmsg_t msg)
 356 {
 357         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 358         int cpu = mycpuid, nextcpu;
 359
 360         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 361
 362         nextcpu = cpu + 1;
 363         if (nextcpu < ncpus2)
 364                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
 365         else
 366                 lwkt_replymsg(&nm->base.lmsg, 0);
 367 }
 368
 369 #endif
 370
 371 /*
 372  * Prepare to accept connections.
 373  */
 374 static void
 375 tcp_usr_listen(netmsg_t msg)
 376 {
 377         struct socket *so = msg->listen.base.nm_so;
 378         struct thread *td = msg->listen.nm_td;
 379         int error = 0;
 380         struct inpcb *inp;
 381         struct tcpcb *tp;
 382 #ifdef SMP
 383         struct netmsg_inswildcard nm;
 384 #endif
 385
 386         COMMON_START(so, inp, 0);
 387
 388         if (tp->t_flags & TF_LISTEN)
 389                 goto out;
 390
 391         if (inp->inp_lport == 0) {
 392                 error = in_pcbbind(inp, NULL, td);
 393                 if (error)
 394                         goto out;
 395         }
 396
 397         tp->t_state = TCPS_LISTEN;
 398         tp->t_flags |= TF_LISTEN;
 399         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 400
 401 #ifdef SMP
 402         if (ncpus > 1) {
 403                 /*
 404                  * We have to set the flag because we can't have other cpus
 405                  * messing with our inp's flags.
 406                  */
 407                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 408                         ("already on connhash\n"));
 409                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 410                         ("already on wildcardhash\n"));
 411                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 412                         ("already on MP wildcardhash\n"));
 413                 inp->inp_flags |= INP_WILDCARD_MP;
 414
 415                 KKASSERT(so->so_port == cpu_portfn(0));
 416                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 417                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 418
 419                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 420                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 421                 nm.nm_inp = inp;
 422                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 423         }
 424 #endif
 425         in_pcbinswildcardhash(inp);
 426         COMMON_END(PRU_LISTEN);
 427 }
 428
 429 #ifdef INET6
 430
 431 static void
 432 tcp6_usr_listen(netmsg_t msg)
 433 {
 434         struct socket *so = msg->listen.base.nm_so;
 435         struct thread *td = msg->listen.nm_td;
 436         int error = 0;
 437         struct inpcb *inp;
 438         struct tcpcb *tp;
 439 #ifdef SMP
 440         struct netmsg_inswildcard nm;
 441 #endif
 442
 443         COMMON_START(so, inp, 0);
 444
 445         if (tp->t_flags & TF_LISTEN)
 446                 goto out;
 447
 448         if (inp->inp_lport == 0) {
 449                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 450                         inp->inp_vflag |= INP_IPV4;
 451                 else
 452                         inp->inp_vflag &= ~INP_IPV4;
 453                 error = in6_pcbbind(inp, NULL, td);
 454                 if (error)
 455                         goto out;
 456         }
 457
 458         tp->t_state = TCPS_LISTEN;
 459         tp->t_flags |= TF_LISTEN;
 460         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 461
 462 #ifdef SMP
 463         if (ncpus > 1) {
 464                 /*
 465                  * We have to set the flag because we can't have other cpus
 466                  * messing with our inp's flags.
 467                  */
 468                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 469                         ("already on connhash\n"));
 470                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 471                         ("already on wildcardhash\n"));
 472                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 473                         ("already on MP wildcardhash\n"));
 474                 inp->inp_flags |= INP_WILDCARD_MP;
 475
 476                 KKASSERT(so->so_port == cpu_portfn(0));
 477                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 478                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 479
 480                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 481                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 482                 nm.nm_inp = inp;
 483                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 484         }
 485 #endif
 486         in_pcbinswildcardhash(inp);
 487         COMMON_END(PRU_LISTEN);
 488 }
 489 #endif /* INET6 */
 490
 491 /*
 492  * Initiate connection to peer.
 493  * Create a template for use in transmissions on this connection.
 494  * Enter SYN_SENT state, and mark socket as connecting.
 495  * Start keep-alive timer, and seed output sequence space.
 496  * Send initial segment on connection.
 497  */
 498 static void
 499 tcp_usr_connect(netmsg_t msg)
 500 {
 501         struct socket *so = msg->connect.base.nm_so;
 502         struct sockaddr *nam = msg->connect.nm_nam;
 503         struct thread *td = msg->connect.nm_td;
 504         int error = 0;
 505         struct inpcb *inp;
 506         struct tcpcb *tp;
 507         struct sockaddr_in *sinp;
 508
 509         COMMON_START(so, inp, 0);
 510
 511         /*
 512          * Must disallow TCP ``connections'' to multicast addresses.
 513          */
 514         sinp = (struct sockaddr_in *)nam;
 515         if (sinp->sin_family == AF_INET
 516             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 517                 error = EAFNOSUPPORT;
 518                 goto out;
 519         }
 520
 521         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 522                 error = EAFNOSUPPORT; /* IPv6 only jail */
 523                 goto out;
 524         }
 525
 526         tcp_connect(msg);
 527         /* msg is invalid now */
 528         return;
 529 out:
 530         if (msg->connect.nm_m) {
 531                 m_freem(msg->connect.nm_m);
 532                 msg->connect.nm_m = NULL;
 533         }
 534         lwkt_replymsg(&msg->lmsg, error);
 535 }
 536
 537 #ifdef INET6
 538
 539 static void
 540 tcp6_usr_connect(netmsg_t msg)
 541 {
 542         struct socket *so = msg->connect.base.nm_so;
 543         struct sockaddr *nam = msg->connect.nm_nam;
 544         struct thread *td = msg->connect.nm_td;
 545         int error = 0;
 546         struct inpcb *inp;
 547         struct tcpcb *tp;
 548         struct sockaddr_in6 *sin6p;
 549
 550         COMMON_START(so, inp, 0);
 551
 552         /*
 553          * Must disallow TCP ``connections'' to multicast addresses.
 554          */
 555         sin6p = (struct sockaddr_in6 *)nam;
 556         if (sin6p->sin6_family == AF_INET6
 557             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 558                 error = EAFNOSUPPORT;
 559                 goto out;
 560         }
 561
 562         if (!prison_remote_ip(td, nam)) {
 563                 error = EAFNOSUPPORT; /* IPv4 only jail */
 564                 goto out;
 565         }
 566
 567         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 568                 struct sockaddr_in *sinp;
 569
 570                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 571                         error = EINVAL;
 572                         goto out;
 573                 }
 574                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 575                 in6_sin6_2_sin(sinp, sin6p);
 576                 inp->inp_vflag |= INP_IPV4;
 577                 inp->inp_vflag &= ~INP_IPV6;
 578                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 579                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
 580                 tcp_connect(msg);
 581                 /* msg is invalid now */
 582                 return;
 583         }
 584         inp->inp_vflag &= ~INP_IPV4;
 585         inp->inp_vflag |= INP_IPV6;
 586         inp->inp_inc.inc_isipv6 = 1;
 587
 588         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
 589         tcp6_connect(msg);
 590         /* msg is invalid now */
 591         return;
 592 out:
 593         if (msg->connect.nm_m) {
 594                 m_freem(msg->connect.nm_m);
 595                 msg->connect.nm_m = NULL;
 596         }
 597         lwkt_replymsg(&msg->lmsg, error);
 598 }
 599
 600 #endif /* INET6 */
 601
 602 /*
 603  * Initiate disconnect from peer.
 604  * If connection never passed embryonic stage, just drop;
 605  * else if don't need to let data drain, then can just drop anyways,
 606  * else have to begin TCP shutdown process: mark socket disconnecting,
 607  * drain unread data, state switch to reflect user close, and
 608  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 609  * when peer sends FIN and acks ours.
 610  *
 611  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 612  */
 613 static void
 614 tcp_usr_disconnect(netmsg_t msg)
 615 {
 616         struct socket *so = msg->disconnect.base.nm_so;
 617         int error = 0;
 618         struct inpcb *inp;
 619         struct tcpcb *tp;
 620
 621         COMMON_START(so, inp, 1);
 622         tp = tcp_disconnect(tp);
 623         COMMON_END(PRU_DISCONNECT);
 624 }
 625
 626 /*
 627  * Accept a connection.  Essentially all the work is
 628  * done at higher levels; just return the address
 629  * of the peer, storing through addr.
 630  */
 631 static void
 632 tcp_usr_accept(netmsg_t msg)
 633 {
 634         struct socket *so = msg->accept.base.nm_so;
 635         struct sockaddr **nam = msg->accept.nm_nam;
 636         int error = 0;
 637         struct inpcb *inp;
 638         struct tcpcb *tp = NULL;
 639         TCPDEBUG0;
 640
 641         inp = so->so_pcb;
 642         if (so->so_state & SS_ISDISCONNECTED) {
 643                 error = ECONNABORTED;
 644                 goto out;
 645         }
 646         if (inp == 0) {
 647                 error = EINVAL;
 648                 goto out;
 649         }
 650
 651         tp = intotcpcb(inp);
 652         TCPDEBUG1();
 653         in_setpeeraddr(so, nam);
 654         COMMON_END(PRU_ACCEPT);
 655 }
 656
 657 #ifdef INET6
 658 static void
 659 tcp6_usr_accept(netmsg_t msg)
 660 {
 661         struct socket *so = msg->accept.base.nm_so;
 662         struct sockaddr **nam = msg->accept.nm_nam;
 663         int error = 0;
 664         struct inpcb *inp;
 665         struct tcpcb *tp = NULL;
 666         TCPDEBUG0;
 667
 668         inp = so->so_pcb;
 669
 670         if (so->so_state & SS_ISDISCONNECTED) {
 671                 error = ECONNABORTED;
 672                 goto out;
 673         }
 674         if (inp == 0) {
 675                 error = EINVAL;
 676                 goto out;
 677         }
 678         tp = intotcpcb(inp);
 679         TCPDEBUG1();
 680         in6_mapped_peeraddr(so, nam);
 681         COMMON_END(PRU_ACCEPT);
 682 }
 683 #endif /* INET6 */
 684 /*
 685  * Mark the connection as being incapable of further output.
 686  */
 687 static void
 688 tcp_usr_shutdown(netmsg_t msg)
 689 {
 690         struct socket *so = msg->shutdown.base.nm_so;
 691         int error = 0;
 692         struct inpcb *inp;
 693         struct tcpcb *tp;
 694
 695         COMMON_START(so, inp, 0);
 696         socantsendmore(so);
 697         tp = tcp_usrclosed(tp);
 698         if (tp)
 699                 error = tcp_output(tp);
 700         COMMON_END(PRU_SHUTDOWN);
 701 }
 702
 703 /*
 704  * After a receive, possibly send window update to peer.
 705  */
 706 static void
 707 tcp_usr_rcvd(netmsg_t msg)
 708 {
 709         struct socket *so = msg->rcvd.base.nm_so;
 710         int error = 0;
 711         struct inpcb *inp;
 712         struct tcpcb *tp;
 713
 714         COMMON_START(so, inp, 0);
 715         tcp_output(tp);
 716         COMMON_END(PRU_RCVD);
 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         struct mbuf *control = msg->send.nm_control;
 733         int error = 0;
 734         struct inpcb *inp;
 735         struct tcpcb *tp;
 736         TCPDEBUG0;
 737
 738         inp = so->so_pcb;
 739
 740         if (inp == NULL) {
 741                 /*
 742                  * OOPS! we lost a race, the TCP session got reset after
 743                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 744                  * network interrupt in the non-critical section of sosend().
 745                  */
 746                 m_freem(m);
 747                 if (control)
 748                         m_freem(control);
 749                 error = ECONNRESET;     /* XXX EPIPE? */
 750                 tp = NULL;
 751                 TCPDEBUG1();
 752                 goto out;
 753         }
 754         tp = intotcpcb(inp);
 755         TCPDEBUG1();
 756         if (control) {
 757                 /* TCP doesn't do control messages (rights, creds, etc) */
 758                 if (control->m_len) {
 759                         m_freem(control);
 760                         m_freem(m);
 761                         error = EINVAL;
 762                         goto out;
 763                 }
 764                 m_freem(control);       /* empty control, just free it */
 765         }
 766
 767         /*
 768          * Don't let too much OOB data build up
 769          */
 770         if (flags & PRUS_OOB) {
 771                 if (ssb_space(&so->so_snd) < -512) {
 772                         m_freem(m);
 773                         error = ENOBUFS;
 774                         goto out;
 775                 }
 776         }
 777
 778         /*
 779          * Pump the data into the socket.
 780          */
 781         if (m)
 782                 ssb_appendstream(&so->so_snd, m);
 783         if (flags & PRUS_OOB) {
 784                 /*
 785                  * According to RFC961 (Assigned Protocols),
 786                  * the urgent pointer points to the last octet
 787                  * of urgent data.  We continue, however,
 788                  * to consider it to indicate the first octet
 789                  * of data past the urgent section.
 790                  * Otherwise, snd_up should be one lower.
 791                  */
 792                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 793                 tp->t_flags |= TF_FORCE;
 794                 error = tcp_output(tp);
 795                 tp->t_flags &= ~TF_FORCE;
 796         } else {
 797                 if (flags & PRUS_EOF) {
 798                         /*
 799                          * Close the send side of the connection after
 800                          * the data is sent.
 801                          */
 802                         socantsendmore(so);
 803                         tp = tcp_usrclosed(tp);
 804                 }
 805                 if (tp != NULL) {
 806                         if (flags & PRUS_MORETOCOME)
 807                                 tp->t_flags |= TF_MORETOCOME;
 808                         error = tcp_output(tp);
 809                         if (flags & PRUS_MORETOCOME)
 810                                 tp->t_flags &= ~TF_MORETOCOME;
 811                 }
 812         }
 813         COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
 814                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 815 }
 816
 817 /*
 818  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 819  *       will sofree() it when we return.
 820  */
 821 static void
 822 tcp_usr_abort(netmsg_t msg)
 823 {
 824         struct socket *so = msg->abort.base.nm_so;
 825         int error = 0;
 826         struct inpcb *inp;
 827         struct tcpcb *tp;
 828
 829         COMMON_START(so, inp, 1);
 830         tp = tcp_drop(tp, ECONNABORTED);
 831         COMMON_END(PRU_ABORT);
 832 }
 833
 834 /*
 835  * Receive out-of-band data.
 836  */
 837 static void
 838 tcp_usr_rcvoob(netmsg_t msg)
 839 {
 840         struct socket *so = msg->rcvoob.base.nm_so;
 841         struct mbuf *m = msg->rcvoob.nm_m;
 842         int flags = msg->rcvoob.nm_flags;
 843         int error = 0;
 844         struct inpcb *inp;
 845         struct tcpcb *tp;
 846
 847         COMMON_START(so, inp, 0);
 848         if ((so->so_oobmark == 0 &&
 849              (so->so_state & SS_RCVATMARK) == 0) ||
 850             so->so_options & SO_OOBINLINE ||
 851             tp->t_oobflags & TCPOOB_HADDATA) {
 852                 error = EINVAL;
 853                 goto out;
 854         }
 855         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 856                 error = EWOULDBLOCK;
 857                 goto out;
 858         }
 859         m->m_len = 1;
 860         *mtod(m, caddr_t) = tp->t_iobc;
 861         if ((flags & MSG_PEEK) == 0)
 862                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 863         COMMON_END(PRU_RCVOOB);
 864 }
 865
 866 /* xxx - should be const */
 867 struct pr_usrreqs tcp_usrreqs = {
 868         .pru_abort = tcp_usr_abort,
 869         .pru_accept = tcp_usr_accept,
 870         .pru_attach = tcp_usr_attach,
 871         .pru_bind = tcp_usr_bind,
 872         .pru_connect = tcp_usr_connect,
 873         .pru_connect2 = pr_generic_notsupp,
 874         .pru_control = in_control_dispatch,
 875         .pru_detach = tcp_usr_detach,
 876         .pru_disconnect = tcp_usr_disconnect,
 877         .pru_listen = tcp_usr_listen,
 878         .pru_peeraddr = in_setpeeraddr_dispatch,
 879         .pru_rcvd = tcp_usr_rcvd,
 880         .pru_rcvoob = tcp_usr_rcvoob,
 881         .pru_send = tcp_usr_send,
 882         .pru_sense = pru_sense_null,
 883         .pru_shutdown = tcp_usr_shutdown,
 884         .pru_sockaddr = in_setsockaddr_dispatch,
 885         .pru_sosend = sosend,
 886         .pru_soreceive = soreceive
 887 };
 888
 889 #ifdef INET6
 890 struct pr_usrreqs tcp6_usrreqs = {
 891         .pru_abort = tcp_usr_abort,
 892         .pru_accept = tcp6_usr_accept,
 893         .pru_attach = tcp_usr_attach,
 894         .pru_bind = tcp6_usr_bind,
 895         .pru_connect = tcp6_usr_connect,
 896         .pru_connect2 = pr_generic_notsupp,
 897         .pru_control = in6_control_dispatch,
 898         .pru_detach = tcp_usr_detach,
 899         .pru_disconnect = tcp_usr_disconnect,
 900         .pru_listen = tcp6_usr_listen,
 901         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 902         .pru_rcvd = tcp_usr_rcvd,
 903         .pru_rcvoob = tcp_usr_rcvoob,
 904         .pru_send = tcp_usr_send,
 905         .pru_sense = pru_sense_null,
 906         .pru_shutdown = tcp_usr_shutdown,
 907         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 908         .pru_sosend = sosend,
 909         .pru_soreceive = soreceive
 910 };
 911 #endif /* INET6 */
 912
 913 static int
 914 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 915                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 916 {
 917         struct inpcb *inp = tp->t_inpcb, *oinp;
 918         struct socket *so = inp->inp_socket;
 919         struct route *ro = &inp->inp_route;
 920
 921         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
 922                                  sin->sin_addr, sin->sin_port,
 923                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 924                                   inp->inp_laddr : if_sin->sin_addr),
 925                                 inp->inp_lport, 0, NULL);
 926         if (oinp != NULL) {
 927                 m_freem(m);
 928                 return (EADDRINUSE);
 929         }
 930         if (inp->inp_laddr.s_addr == INADDR_ANY)
 931                 inp->inp_laddr = if_sin->sin_addr;
 932         inp->inp_faddr = sin->sin_addr;
 933         inp->inp_fport = sin->sin_port;
 934         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
 935         in_pcbinsconnhash(inp);
 936
 937         /*
 938          * We are now on the inpcb's owner CPU, if the cached route was
 939          * freed because the rtentry's owner CPU is not the current CPU
 940          * (e.g. in tcp_connect()), then we try to reallocate it here with
 941          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 942          * rtentry.
 943          */
 944         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 945             ro->ro_rt == NULL) {
 946                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 947                 ro->ro_dst.sa_family = AF_INET;
 948                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 949                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 950                         sin->sin_addr;
 951                 rtalloc(ro);
 952         }
 953
 954         /*
 955          * Now that no more errors can occur, change the protocol processing
 956          * port to the current thread (which is the correct thread).
 957          *
 958          * Create TCP timer message now; we are on the tcpcb's owner
 959          * CPU/thread.
 960          */
 961         tcp_create_timermsg(tp, &curthread->td_msgport);
 962
 963         /*
 964          * Compute window scaling to request.  Use a larger scaling then
 965          * needed for the initial receive buffer in case the receive buffer
 966          * gets expanded.
 967          */
 968         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
 969                 tp->request_r_scale = TCP_MIN_WINSHIFT;
 970         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
 971                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
 972         ) {
 973                 tp->request_r_scale++;
 974         }
 975
 976         soisconnecting(so);
 977         tcpstat.tcps_connattempt++;
 978         tp->t_state = TCPS_SYN_SENT;
 979         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
 980         tp->iss = tcp_new_isn(tp);
 981         tcp_sendseqinit(tp);
 982         if (m) {
 983                 ssb_appendstream(&so->so_snd, m);
 984                 m = NULL;
 985                 if (flags & PRUS_OOB)
 986                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 987         }
 988
 989         /*
 990          * Close the send side of the connection after
 991          * the data is sent if flagged.
 992          */
 993         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
 994                 socantsendmore(so);
 995                 tp = tcp_usrclosed(tp);
 996         }
 997         return (tcp_output(tp));
 998 }
 999
1000 /*
1001  * Common subroutine to open a TCP connection to remote host specified
1002  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1003  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1004  * a local host address (interface).
1005  * Initialize connection parameters and enter SYN-SENT state.
1006  */
1007 static void
1008 tcp_connect(netmsg_t msg)
1009 {
1010         struct socket *so = msg->connect.base.nm_so;
1011         struct sockaddr *nam = msg->connect.nm_nam;
1012         struct thread *td = msg->connect.nm_td;
1013         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1014         struct sockaddr_in *if_sin;
1015         struct inpcb *inp;
1016         struct tcpcb *tp;
1017         int error, calc_laddr = 1;
1018 #ifdef SMP
1019         lwkt_port_t port;
1020 #endif
1021
1022         COMMON_START(so, inp, 0);
1023
1024         /*
1025          * Reconnect our pcb if we have to
1026          */
1027         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1028                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1029                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1030         }
1031
1032         /*
1033          * Bind if we have to
1034          */
1035         if (inp->inp_lport == 0) {
1036                 if (tcp_lport_extension) {
1037                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1038
1039                         error = in_pcbladdr(inp, nam, &if_sin, td);
1040                         if (error)
1041                                 goto out;
1042                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1043
1044                         error = in_pcbconn_bind(inp, nam, td);
1045                         if (error)
1046                                 goto out;
1047
1048                         calc_laddr = 0;
1049                 } else {
1050                         error = in_pcbbind(inp, NULL, td);
1051                         if (error)
1052                                 goto out;
1053                 }
1054         }
1055
1056         if (calc_laddr) {
1057                 /*
1058                  * Calculate the correct protocol processing thread.  The
1059                  * connect operation must run there.  Set the forwarding
1060                  * port before we forward the message or it will get bounced
1061                  * right back to us.
1062                  */
1063                 error = in_pcbladdr(inp, nam, &if_sin, td);
1064                 if (error)
1065                         goto out;
1066         }
1067         KKASSERT(inp->inp_socket == so);
1068
1069 #ifdef SMP
1070         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1071                             (inp->inp_laddr.s_addr ?
1072                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1073                             inp->inp_lport);
1074
1075         if (port != &curthread->td_msgport) {
1076                 struct route *ro = &inp->inp_route;
1077
1078                 /*
1079                  * in_pcbladdr() may have allocated a route entry for us
1080                  * on the current CPU, but we need a route entry on the
1081                  * inpcb's owner CPU, so free it here.
1082                  */
1083                 if (ro->ro_rt != NULL)
1084                         RTFREE(ro->ro_rt);
1085                 bzero(ro, sizeof(*ro));
1086
1087                 /*
1088                  * We are moving the protocol processing port the socket
1089                  * is on, we have to unlink here and re-link on the
1090                  * target cpu.
1091                  */
1092                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1093                 sosetport(so, port);
1094                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1095                 msg->connect.base.nm_dispatch = tcp_connect;
1096
1097                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1098                 /* msg invalid now */
1099                 return;
1100         }
1101 #else
1102         KKASSERT(so->so_port == &curthread->td_msgport);
1103 #endif
1104         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1105                                   msg->connect.nm_m, sin, if_sin);
1106         msg->connect.nm_m = NULL;
1107 out:
1108         if (msg->connect.nm_m) {
1109                 m_freem(msg->connect.nm_m);
1110                 msg->connect.nm_m = NULL;
1111         }
1112         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1113                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1114                 msg->connect.nm_nam = NULL;
1115         }
1116         lwkt_replymsg(&msg->connect.base.lmsg, error);
1117         /* msg invalid now */
1118 }
1119
1120 #ifdef INET6
1121
1122 static void
1123 tcp6_connect(netmsg_t msg)
1124 {
1125         struct tcpcb *tp;
1126         struct socket *so = msg->connect.base.nm_so;
1127         struct sockaddr *nam = msg->connect.nm_nam;
1128         struct thread *td = msg->connect.nm_td;
1129         struct inpcb *inp;
1130         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1131         struct in6_addr *addr6;
1132 #ifdef SMP
1133         lwkt_port_t port;
1134 #endif
1135         int error;
1136
1137         COMMON_START(so, inp, 0);
1138
1139         /*
1140          * Reconnect our pcb if we have to
1141          */
1142         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1143                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1144                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1145         }
1146
1147         /*
1148          * Bind if we have to
1149          */
1150         if (inp->inp_lport == 0) {
1151                 error = in6_pcbbind(inp, NULL, td);
1152                 if (error)
1153                         goto out;
1154         }
1155
1156         /*
1157          * Cannot simply call in_pcbconnect, because there might be an
1158          * earlier incarnation of this same connection still in
1159          * TIME_WAIT state, creating an ADDRINUSE error.
1160          */
1161         error = in6_pcbladdr(inp, nam, &addr6, td);
1162         if (error)
1163                 goto out;
1164
1165 #ifdef SMP
1166         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1167
1168         if (port != &curthread->td_msgport) {
1169                 struct route *ro = &inp->inp_route;
1170
1171                 /*
1172                  * in_pcbladdr() may have allocated a route entry for us
1173                  * on the current CPU, but we need a route entry on the
1174                  * inpcb's owner CPU, so free it here.
1175                  */
1176                 if (ro->ro_rt != NULL)
1177                         RTFREE(ro->ro_rt);
1178                 bzero(ro, sizeof(*ro));
1179
1180                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1181                 sosetport(so, port);
1182                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1183                 msg->connect.base.nm_dispatch = tcp6_connect;
1184
1185                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1186                 /* msg invalid now */
1187                 return;
1188         }
1189 #endif
1190         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1191                                    &msg->connect.nm_m, sin6, addr6);
1192         /* nm_m may still be intact */
1193 out:
1194         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1195                 tcp_connect(msg);
1196                 /* msg invalid now */
1197         } else {
1198                 if (msg->connect.nm_m) {
1199                         m_freem(msg->connect.nm_m);
1200                         msg->connect.nm_m = NULL;
1201                 }
1202                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1203                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1204                         msg->connect.nm_nam = NULL;
1205                 }
1206                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1207                 /* msg invalid now */
1208         }
1209 }
1210
1211 static int
1212 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1213                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1214 {
1215         struct mbuf *m = *mp;
1216         struct inpcb *inp = tp->t_inpcb;
1217         struct socket *so = inp->inp_socket;
1218         struct inpcb *oinp;
1219
1220         /*
1221          * Cannot simply call in_pcbconnect, because there might be an
1222          * earlier incarnation of this same connection still in
1223          * TIME_WAIT state, creating an ADDRINUSE error.
1224          */
1225         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1226                                   &sin6->sin6_addr, sin6->sin6_port,
1227                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1228                                       addr6 : &inp->in6p_laddr),
1229                                   inp->inp_lport,  0, NULL);
1230         if (oinp)
1231                 return (EADDRINUSE);
1232
1233         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1234                 inp->in6p_laddr = *addr6;
1235         inp->in6p_faddr = sin6->sin6_addr;
1236         inp->inp_fport = sin6->sin6_port;
1237         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1238                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1239         in_pcbinsconnhash(inp);
1240
1241         /*
1242          * Now that no more errors can occur, change the protocol processing
1243          * port to the current thread (which is the correct thread).
1244          *
1245          * Create TCP timer message now; we are on the tcpcb's owner
1246          * CPU/thread.
1247          */
1248         tcp_create_timermsg(tp, &curthread->td_msgport);
1249
1250         /* Compute window scaling to request.  */
1251         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1252                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1253         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1254             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1255                 tp->request_r_scale++;
1256         }
1257
1258         soisconnecting(so);
1259         tcpstat.tcps_connattempt++;
1260         tp->t_state = TCPS_SYN_SENT;
1261         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1262         tp->iss = tcp_new_isn(tp);
1263         tcp_sendseqinit(tp);
1264         if (m) {
1265                 ssb_appendstream(&so->so_snd, m);
1266                 *mp = NULL;
1267                 if (flags & PRUS_OOB)
1268                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1269         }
1270
1271         /*
1272          * Close the send side of the connection after
1273          * the data is sent if flagged.
1274          */
1275         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1276                 socantsendmore(so);
1277                 tp = tcp_usrclosed(tp);
1278         }
1279         return (tcp_output(tp));
1280 }
1281
1282 #endif /* INET6 */
1283
1284 /*
1285  * The new sockopt interface makes it possible for us to block in the
1286  * copyin/out step (if we take a page fault).  Taking a page fault while
1287  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1288  * both now use TSM, there probably isn't any need for this function to
1289  * run in a critical section any more.  This needs more examination.)
1290  */
1291 void
1292 tcp_ctloutput(netmsg_t msg)
1293 {
1294         struct socket *so = msg->base.nm_so;
1295         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1296         int     error, opt, optval, opthz;
1297         struct  inpcb *inp;
1298         struct  tcpcb *tp;
1299
1300         error = 0;
1301         inp = so->so_pcb;
1302         if (inp == NULL) {
1303                 error = ECONNRESET;
1304                 goto done;
1305         }
1306
1307         if (sopt->sopt_level != IPPROTO_TCP) {
1308 #ifdef INET6
1309                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1310                         ip6_ctloutput_dispatch(msg);
1311                 else
1312 #endif /* INET6 */
1313                 ip_ctloutput(msg);
1314                 /* msg invalid now */
1315                 return;
1316         }
1317         tp = intotcpcb(inp);
1318
1319         switch (sopt->sopt_dir) {
1320         case SOPT_SET:
1321                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1322                                       sizeof optval);
1323                 if (error)
1324                         break;
1325                 switch (sopt->sopt_name) {
1326                 case TCP_FASTKEEP:
1327                         if (optval > 0) {
1328                                 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1329                                         tp->t_flags |= TF_FASTKEEP;
1330                                         tcp_timer_keep_activity(tp, 0);
1331                                 }
1332                         } else {
1333                                 tp->t_flags &= ~TF_FASTKEEP;
1334                         }
1335                         break;
1336 #ifdef TCP_SIGNATURE
1337                 case TCP_SIGNATURE_ENABLE:
1338                         if (optval > 0)
1339                                 tp->t_flags |= TF_SIGNATURE;
1340                         else
1341                                 tp->t_flags &= ~TF_SIGNATURE;
1342                         break;
1343 #endif /* TCP_SIGNATURE */
1344                 case TCP_NODELAY:
1345                 case TCP_NOOPT:
1346                         switch (sopt->sopt_name) {
1347                         case TCP_NODELAY:
1348                                 opt = TF_NODELAY;
1349                                 break;
1350                         case TCP_NOOPT:
1351                                 opt = TF_NOOPT;
1352                                 break;
1353                         default:
1354                                 opt = 0; /* dead code to fool gcc */
1355                                 break;
1356                         }
1357
1358                         if (optval)
1359                                 tp->t_flags |= opt;
1360                         else
1361                                 tp->t_flags &= ~opt;
1362                         break;
1363
1364                 case TCP_NOPUSH:
1365                         if (optval)
1366                                 tp->t_flags |= TF_NOPUSH;
1367                         else {
1368                                 tp->t_flags &= ~TF_NOPUSH;
1369                                 error = tcp_output(tp);
1370                         }
1371                         break;
1372
1373                 case TCP_MAXSEG:
1374                         /*
1375                          * Must be between 0 and maxseg.  If the requested
1376                          * maxseg is too small to satisfy the desired minmss,
1377                          * pump it up (silently so sysctl modifications of
1378                          * minmss do not create unexpected program failures).
1379                          * Handle degenerate cases.
1380                          */
1381                         if (optval > 0 && optval <= tp->t_maxseg) {
1382                                 if (optval + 40 < tcp_minmss) {
1383                                         optval = tcp_minmss - 40;
1384                                         if (optval < 0)
1385                                                 optval = 1;
1386                                 }
1387                                 tp->t_maxseg = optval;
1388                         } else {
1389                                 error = EINVAL;
1390                         }
1391                         break;
1392
1393                 case TCP_KEEPINIT:
1394                         opthz = ((int64_t)optval * hz) / 1000;
1395                         if (opthz >= 1)
1396                                 tp->t_keepinit = opthz;
1397                         else
1398                                 error = EINVAL;
1399                         break;
1400
1401                 case TCP_KEEPIDLE:
1402                         opthz = ((int64_t)optval * hz) / 1000;
1403                         if (opthz >= 1)
1404                                 tp->t_keepidle = opthz;
1405                         else
1406                                 error = EINVAL;
1407                         break;
1408
1409                 case TCP_KEEPINTVL:
1410                         opthz = ((int64_t)optval * hz) / 1000;
1411                         if (opthz >= 1) {
1412                                 tp->t_keepintvl = opthz;
1413                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1414                         } else {
1415                                 error = EINVAL;
1416                         }
1417                         break;
1418
1419                 case TCP_KEEPCNT:
1420                         if (optval > 0) {
1421                                 tp->t_keepcnt = optval;
1422                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1423                         } else {
1424                                 error = EINVAL;
1425                         }
1426                         break;
1427
1428                 default:
1429                         error = ENOPROTOOPT;
1430                         break;
1431                 }
1432                 break;
1433
1434         case SOPT_GET:
1435                 switch (sopt->sopt_name) {
1436 #ifdef TCP_SIGNATURE
1437                 case TCP_SIGNATURE_ENABLE:
1438                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1439                         break;
1440 #endif /* TCP_SIGNATURE */
1441                 case TCP_NODELAY:
1442                         optval = tp->t_flags & TF_NODELAY;
1443                         break;
1444                 case TCP_MAXSEG:
1445                         optval = tp->t_maxseg;
1446                         break;
1447                 case TCP_NOOPT:
1448                         optval = tp->t_flags & TF_NOOPT;
1449                         break;
1450                 case TCP_NOPUSH:
1451                         optval = tp->t_flags & TF_NOPUSH;
1452                         break;
1453                 case TCP_KEEPINIT:
1454                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1455                         break;
1456                 case TCP_KEEPIDLE:
1457                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1458                         break;
1459                 case TCP_KEEPINTVL:
1460                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1461                         break;
1462                 case TCP_KEEPCNT:
1463                         optval = tp->t_keepcnt;
1464                         break;
1465                 default:
1466                         error = ENOPROTOOPT;
1467                         break;
1468                 }
1469                 if (error == 0)
1470                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1471                 break;
1472         }
1473 done:
1474         lwkt_replymsg(&msg->lmsg, error);
1475 }
1476
1477 /*
1478  * tcp_sendspace and tcp_recvspace are the default send and receive window
1479  * sizes, respectively.  These are obsolescent (this information should
1480  * be set by the route).
1481  *
1482  * Use a default that does not require tcp window scaling to be turned
1483  * on.  Individual programs or the administrator can increase the default.
1484  */
1485 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1486 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1487     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1488 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1489 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1490     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1491
1492 /*
1493  * Attach TCP protocol to socket, allocating internet protocol control
1494  * block, tcp control block, bufer space, and entering LISTEN state
1495  * if to accept connections.
1496  */
1497 static int
1498 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1499 {
1500         struct tcpcb *tp;
1501         struct inpcb *inp;
1502         int error;
1503         int cpu;
1504 #ifdef INET6
1505         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1506 #endif
1507
1508         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1509                 lwkt_gettoken(&so->so_rcv.ssb_token);
1510                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1511                                   ai->sb_rlimit);
1512                 lwkt_reltoken(&so->so_rcv.ssb_token);
1513                 if (error)
1514                         return (error);
1515         }
1516         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1517         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1518         cpu = mycpu->gd_cpuid;
1519
1520         /*
1521          * Set the default port for protocol processing. This will likely
1522          * change when we connect.
1523          */
1524         error = in_pcballoc(so, &tcbinfo[cpu]);
1525         if (error)
1526                 return (error);
1527         inp = so->so_pcb;
1528 #ifdef INET6
1529         if (isipv6) {
1530                 inp->inp_vflag |= INP_IPV6;
1531                 inp->in6p_hops = -1;    /* use kernel default */
1532         }
1533         else
1534 #endif
1535         inp->inp_vflag |= INP_IPV4;
1536         tp = tcp_newtcpcb(inp);
1537         if (tp == NULL) {
1538                 /*
1539                  * Make sure the socket is destroyed by the pcbdetach.
1540                  */
1541                 soreference(so);
1542 #ifdef INET6
1543                 if (isipv6)
1544                         in6_pcbdetach(inp);
1545                 else
1546 #endif
1547                 in_pcbdetach(inp);
1548                 sofree(so);     /* from ref above */
1549                 return (ENOBUFS);
1550         }
1551         tp->t_state = TCPS_CLOSED;
1552         return (0);
1553 }
1554
1555 /*
1556  * Initiate (or continue) disconnect.
1557  * If embryonic state, just send reset (once).
1558  * If in ``let data drain'' option and linger null, just drop.
1559  * Otherwise (hard), mark socket disconnecting and drop
1560  * current input data; switch states based on user close, and
1561  * send segment to peer (with FIN).
1562  */
1563 static struct tcpcb *
1564 tcp_disconnect(struct tcpcb *tp)
1565 {
1566         struct socket *so = tp->t_inpcb->inp_socket;
1567
1568         if (tp->t_state < TCPS_ESTABLISHED) {
1569                 tp = tcp_close(tp);
1570         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1571                 tp = tcp_drop(tp, 0);
1572         } else {
1573                 lwkt_gettoken(&so->so_rcv.ssb_token);
1574                 soisdisconnecting(so);
1575                 sbflush(&so->so_rcv.sb);
1576                 tp = tcp_usrclosed(tp);
1577                 if (tp)
1578                         tcp_output(tp);
1579                 lwkt_reltoken(&so->so_rcv.ssb_token);
1580         }
1581         return (tp);
1582 }
1583
1584 /*
1585  * User issued close, and wish to trail through shutdown states:
1586  * if never received SYN, just forget it.  If got a SYN from peer,
1587  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1588  * If already got a FIN from peer, then almost done; go to LAST_ACK
1589  * state.  In all other cases, have already sent FIN to peer (e.g.
1590  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1591  * for peer to send FIN or not respond to keep-alives, etc.
1592  * We can let the user exit from the close as soon as the FIN is acked.
1593  */
1594 static struct tcpcb *
1595 tcp_usrclosed(struct tcpcb *tp)
1596 {
1597
1598         switch (tp->t_state) {
1599
1600         case TCPS_CLOSED:
1601         case TCPS_LISTEN:
1602                 tp->t_state = TCPS_CLOSED;
1603                 tp = tcp_close(tp);
1604                 break;
1605
1606         case TCPS_SYN_SENT:
1607         case TCPS_SYN_RECEIVED:
1608                 tp->t_flags |= TF_NEEDFIN;
1609                 break;
1610
1611         case TCPS_ESTABLISHED:
1612                 tp->t_state = TCPS_FIN_WAIT_1;
1613                 break;
1614
1615         case TCPS_CLOSE_WAIT:
1616                 tp->t_state = TCPS_LAST_ACK;
1617                 break;
1618         }
1619         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1620                 soisdisconnected(tp->t_inpcb->inp_socket);
1621                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1622                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1623                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1624                             tcp_timer_2msl);
1625                 }
1626         }
1627         return (tp);
1628 }