sys/netinet/tcp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2006 Robert N. M. Watson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  32  * $FreeBSD$
  33  */
  34
  35 #include "opt_inet.h"
  36 #include "opt_inet6.h"
  37 #include "opt_tcpdebug.h"
  38
  39 #include <sys/param.h>
  40 #include <sys/systm.h>
  41 #include <sys/malloc.h>
  42 #include <sys/kernel.h>
  43 #include <sys/sysctl.h>
  44 #include <sys/mbuf.h>
  45 #ifdef INET6
  46 #include <sys/domain.h>
  47 #endif /* INET6 */
  48 #include <sys/socket.h>
  49 #include <sys/socketvar.h>
  50 #include <sys/protosw.h>
  51 #include <sys/proc.h>
  52 #include <sys/jail.h>
  53
  54 #include <net/if.h>
  55 #include <net/route.h>
  56
  57 #include <netinet/in.h>
  58 #include <netinet/in_systm.h>
  59 #ifdef INET6
  60 #include <netinet/ip6.h>
  61 #endif
  62 #include <netinet/in_pcb.h>
  63 #ifdef INET6
  64 #include <netinet6/in6_pcb.h>
  65 #endif
  66 #include <netinet/in_var.h>
  67 #include <netinet/ip_var.h>
  68 #ifdef INET6
  69 #include <netinet6/ip6_var.h>
  70 #include <netinet6/scope6_var.h>
  71 #endif
  72 #include <netinet/tcp.h>
  73 #include <netinet/tcp_fsm.h>
  74 #include <netinet/tcp_seq.h>
  75 #include <netinet/tcp_timer.h>
  76 #include <netinet/tcp_var.h>
  77 #include <netinet/tcpip.h>
  78 #ifdef TCPDEBUG
  79 #include <netinet/tcp_debug.h>
  80 #endif
  81
  82 /*
  83  * TCP protocol interface to socket abstraction.
  84  */
  85 extern  char *tcpstates[];      /* XXX ??? */
  86
  87 static int      tcp_attach(struct socket *);
  88 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
  89                     struct thread *td);
  90 #ifdef INET6
  91 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
  92                     struct thread *td);
  93 #endif /* INET6 */
  94 static void     tcp_disconnect(struct tcpcb *);
  95 static void     tcp_usrclosed(struct tcpcb *);
  96 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
  97
  98 #ifdef TCPDEBUG
  99 #define TCPDEBUG0       int ostate = 0
 100 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 101 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 102                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 103 #else
 104 #define TCPDEBUG0
 105 #define TCPDEBUG1()
 106 #define TCPDEBUG2(req)
 107 #endif
 108
 109 /*
 110  * TCP attaches to socket via pru_attach(), reserving space,
 111  * and an internet control block.
 112  */
 113 static int
 114 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 115 {
 116         struct inpcb *inp;
 117         struct tcpcb *tp = NULL;
 118         int error;
 119         TCPDEBUG0;
 120
 121         inp = sotoinpcb(so);
 122         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 123         INP_INFO_WLOCK(&tcbinfo);
 124         TCPDEBUG1();
 125
 126         error = tcp_attach(so);
 127         if (error)
 128                 goto out;
 129
 130         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 131                 so->so_linger = TCP_LINGERTIME;
 132
 133         inp = sotoinpcb(so);
 134         tp = intotcpcb(inp);
 135 out:
 136         TCPDEBUG2(PRU_ATTACH);
 137         INP_INFO_WUNLOCK(&tcbinfo);
 138         return error;
 139 }
 140
 141 /*
 142  * pru_detach() detaches the TCP protocol from the socket.
 143  * If the protocol state is non-embryonic, then can't
 144  * do this directly: have to initiate a pru_disconnect(),
 145  * which may finish later; embryonic TCB's can just
 146  * be discarded here.
 147  */
 148 static void
 149 tcp_usr_detach(struct socket *so)
 150 {
 151         struct inpcb *inp;
 152         struct tcpcb *tp;
 153 #ifdef INET6
 154         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
 155 #endif
 156         TCPDEBUG0;
 157
 158         inp = sotoinpcb(so);
 159         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 160         INP_INFO_WLOCK(&tcbinfo);
 161         INP_LOCK(inp);
 162         KASSERT(inp->inp_socket != NULL,
 163             ("tcp_usr_detach: inp_socket == NULL"));
 164         TCPDEBUG1();
 165
 166         /*
 167          * First, if we still have full TCP state, and we're not dropped,
 168          * initiate a disconnect.
 169          */
 170         if (!(inp->inp_vflag & INP_TIMEWAIT) &&
 171             !(inp->inp_vflag & INP_DROPPED)) {
 172                 tp = intotcpcb(inp);
 173                 tcp_disconnect(tp);
 174         }
 175
 176         /*
 177          * Second, release any protocol state that we can reasonably release.
 178          * Note that the call to tcp_disconnect() may actually have changed
 179          * the TCP state, so we have to re-evaluate INP_TIMEWAIT and
 180          * INP_DROPPED.
 181          */
 182         if (inp->inp_vflag & INP_TIMEWAIT) {
 183                 if (inp->inp_vflag & INP_DROPPED) {
 184                         /*
 185                          * Connection was in time wait and has been dropped;
 186                          * the calling path is either via tcp_twclose(), or
 187                          * as a result of an eventual soclose() after
 188                          * tcp_twclose() has been called.  In either case,
 189                          * tcp_twclose() has detached the tcptw from the
 190                          * inpcb, so we just detach and free the inpcb.
 191                          *
 192                          * XXXRW: Would it be cleaner to free the tcptw
 193                          * here?
 194                          */
 195 #ifdef INET6
 196                         if (isipv6) {
 197                                 in6_pcbdetach(inp);
 198                                 in6_pcbfree(inp);
 199                         } else {
 200 #endif
 201                                 in_pcbdetach(inp);
 202                                 in_pcbfree(inp);
 203 #ifdef INET6
 204                         }
 205 #endif
 206                 } else {
 207                         /*
 208                          * Connection is in time wait and has not yet been
 209                          * dropped; allow the socket to be discarded, but
 210                          * need to keep inpcb until end of time wait.
 211                          */
 212 #ifdef INET6
 213                         if (isipv6)
 214                                 in6_pcbdetach(inp);
 215                         else
 216 #endif
 217                                 in_pcbdetach(inp);
 218                         INP_UNLOCK(inp);
 219                 }
 220         } else {
 221                 /*
 222                  * If not in timewait, there are two possible paths.  First,
 223                  * the TCP connection is either embryonic or done, in which
 224                  * case we tear down all state.  Second, it may still be
 225                  * active, in which case we acquire a reference to the socket
 226                  * and will free it later when TCP is done.
 227                  */
 228                 tp = intotcpcb(inp);
 229                 if (inp->inp_vflag & INP_DROPPED ||
 230                     tp->t_state < TCPS_SYN_SENT) {
 231                         tcp_discardcb(tp);
 232 #ifdef INET6
 233                         if (isipv6) {
 234                                 in_pcbdetach(inp);
 235                                 in_pcbfree(inp);
 236                         } else {
 237 #endif
 238                                 in_pcbdetach(inp);
 239                                 in_pcbfree(inp);
 240 #ifdef INET6
 241                         }
 242 #endif
 243                 } else {
 244                         SOCK_LOCK(so);
 245                         so->so_state |= SS_PROTOREF;
 246                         SOCK_UNLOCK(so);
 247                         inp->inp_vflag |= INP_SOCKREF;
 248                         INP_UNLOCK(inp);
 249                 }
 250         }
 251         tp = NULL;
 252         TCPDEBUG2(PRU_DETACH);
 253         INP_INFO_WUNLOCK(&tcbinfo);
 254 }
 255
 256 /*
 257  * Give the socket an address.
 258  */
 259 static int
 260 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 261 {
 262         int error = 0;
 263         struct inpcb *inp;
 264         struct tcpcb *tp = NULL;
 265         struct sockaddr_in *sinp;
 266
 267         sinp = (struct sockaddr_in *)nam;
 268         if (nam->sa_len != sizeof (*sinp))
 269                 return (EINVAL);
 270         /*
 271          * Must check for multicast addresses and disallow binding
 272          * to them.
 273          */
 274         if (sinp->sin_family == AF_INET &&
 275             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 276                 return (EAFNOSUPPORT);
 277
 278         TCPDEBUG0;
 279         INP_INFO_WLOCK(&tcbinfo);
 280         inp = sotoinpcb(so);
 281         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 282         INP_LOCK(inp);
 283         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 284                 error = EINVAL;
 285                 goto out;
 286         }
 287         tp = intotcpcb(inp);
 288         TCPDEBUG1();
 289         error = in_pcbbind(inp, nam, td->td_ucred);
 290 out:
 291         TCPDEBUG2(PRU_BIND);
 292         INP_UNLOCK(inp);
 293         INP_INFO_WUNLOCK(&tcbinfo);
 294
 295         return (error);
 296 }
 297
 298 #ifdef INET6
 299 static int
 300 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 301 {
 302         int error = 0;
 303         struct inpcb *inp;
 304         struct tcpcb *tp = NULL;
 305         struct sockaddr_in6 *sin6p;
 306
 307         sin6p = (struct sockaddr_in6 *)nam;
 308         if (nam->sa_len != sizeof (*sin6p))
 309                 return (EINVAL);
 310         /*
 311          * Must check for multicast addresses and disallow binding
 312          * to them.
 313          */
 314         if (sin6p->sin6_family == AF_INET6 &&
 315             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 316                 return (EAFNOSUPPORT);
 317
 318         TCPDEBUG0;
 319         INP_INFO_WLOCK(&tcbinfo);
 320         inp = sotoinpcb(so);
 321         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 322         INP_LOCK(inp);
 323         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 324                 error = EINVAL;
 325                 goto out;
 326         }
 327         tp = intotcpcb(inp);
 328         TCPDEBUG1();
 329         inp->inp_vflag &= ~INP_IPV4;
 330         inp->inp_vflag |= INP_IPV6;
 331         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 332                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 333                         inp->inp_vflag |= INP_IPV4;
 334                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 335                         struct sockaddr_in sin;
 336
 337                         in6_sin6_2_sin(&sin, sin6p);
 338                         inp->inp_vflag |= INP_IPV4;
 339                         inp->inp_vflag &= ~INP_IPV6;
 340                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 341                             td->td_ucred);
 342                         goto out;
 343                 }
 344         }
 345         error = in6_pcbbind(inp, nam, td->td_ucred);
 346 out:
 347         TCPDEBUG2(PRU_BIND);
 348         INP_UNLOCK(inp);
 349         INP_INFO_WUNLOCK(&tcbinfo);
 350         return (error);
 351 }
 352 #endif /* INET6 */
 353
 354 /*
 355  * Prepare to accept connections.
 356  */
 357 static int
 358 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 359 {
 360         int error = 0;
 361         struct inpcb *inp;
 362         struct tcpcb *tp = NULL;
 363
 364         TCPDEBUG0;
 365         INP_INFO_WLOCK(&tcbinfo);
 366         inp = sotoinpcb(so);
 367         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 368         INP_LOCK(inp);
 369         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 370                 error = EINVAL;
 371                 goto out;
 372         }
 373         tp = intotcpcb(inp);
 374         TCPDEBUG1();
 375         SOCK_LOCK(so);
 376         error = solisten_proto_check(so);
 377         if (error == 0 && inp->inp_lport == 0)
 378                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 379         if (error == 0) {
 380                 tp->t_state = TCPS_LISTEN;
 381                 solisten_proto(so, backlog);
 382         }
 383         SOCK_UNLOCK(so);
 384
 385 out:
 386         TCPDEBUG2(PRU_LISTEN);
 387         INP_UNLOCK(inp);
 388         INP_INFO_WUNLOCK(&tcbinfo);
 389         return (error);
 390 }
 391
 392 #ifdef INET6
 393 static int
 394 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 395 {
 396         int error = 0;
 397         struct inpcb *inp;
 398         struct tcpcb *tp = NULL;
 399
 400         TCPDEBUG0;
 401         INP_INFO_WLOCK(&tcbinfo);
 402         inp = sotoinpcb(so);
 403         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 404         INP_LOCK(inp);
 405         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 406                 error = EINVAL;
 407                 goto out;
 408         }
 409         tp = intotcpcb(inp);
 410         TCPDEBUG1();
 411         SOCK_LOCK(so);
 412         error = solisten_proto_check(so);
 413         if (error == 0 && inp->inp_lport == 0) {
 414                 inp->inp_vflag &= ~INP_IPV4;
 415                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 416                         inp->inp_vflag |= INP_IPV4;
 417                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 418         }
 419         if (error == 0) {
 420                 tp->t_state = TCPS_LISTEN;
 421                 solisten_proto(so, backlog);
 422         }
 423         SOCK_UNLOCK(so);
 424
 425 out:
 426         TCPDEBUG2(PRU_LISTEN);
 427         INP_UNLOCK(inp);
 428         INP_INFO_WUNLOCK(&tcbinfo);
 429         return (error);
 430 }
 431 #endif /* INET6 */
 432
 433 /*
 434  * Initiate connection to peer.
 435  * Create a template for use in transmissions on this connection.
 436  * Enter SYN_SENT state, and mark socket as connecting.
 437  * Start keep-alive timer, and seed output sequence space.
 438  * Send initial segment on connection.
 439  */
 440 static int
 441 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 442 {
 443         int error = 0;
 444         struct inpcb *inp;
 445         struct tcpcb *tp = NULL;
 446         struct sockaddr_in *sinp;
 447
 448         sinp = (struct sockaddr_in *)nam;
 449         if (nam->sa_len != sizeof (*sinp))
 450                 return (EINVAL);
 451         /*
 452          * Must disallow TCP ``connections'' to multicast addresses.
 453          */
 454         if (sinp->sin_family == AF_INET
 455             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 456                 return (EAFNOSUPPORT);
 457         if (jailed(td->td_ucred))
 458                 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr);
 459
 460         TCPDEBUG0;
 461         INP_INFO_WLOCK(&tcbinfo);
 462         inp = sotoinpcb(so);
 463         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 464         INP_LOCK(inp);
 465         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 466                 error = EINVAL;
 467                 goto out;
 468         }
 469         tp = intotcpcb(inp);
 470         TCPDEBUG1();
 471         if ((error = tcp_connect(tp, nam, td)) != 0)
 472                 goto out;
 473         error = tcp_output(tp);
 474 out:
 475         TCPDEBUG2(PRU_CONNECT);
 476         INP_UNLOCK(inp);
 477         INP_INFO_WUNLOCK(&tcbinfo);
 478         return (error);
 479 }
 480
 481 #ifdef INET6
 482 static int
 483 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 484 {
 485         int error = 0;
 486         struct inpcb *inp;
 487         struct tcpcb *tp = NULL;
 488         struct sockaddr_in6 *sin6p;
 489
 490         TCPDEBUG0;
 491
 492         sin6p = (struct sockaddr_in6 *)nam;
 493         if (nam->sa_len != sizeof (*sin6p))
 494                 return (EINVAL);
 495         /*
 496          * Must disallow TCP ``connections'' to multicast addresses.
 497          */
 498         if (sin6p->sin6_family == AF_INET6
 499             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 500                 return (EAFNOSUPPORT);
 501
 502         INP_INFO_WLOCK(&tcbinfo);
 503         inp = sotoinpcb(so);
 504         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 505         INP_LOCK(inp);
 506         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 507                 error = EINVAL;
 508                 goto out;
 509         }
 510         tp = intotcpcb(inp);
 511         TCPDEBUG1();
 512         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 513                 struct sockaddr_in sin;
 514
 515                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 516                         error = EINVAL;
 517                         goto out;
 518                 }
 519
 520                 in6_sin6_2_sin(&sin, sin6p);
 521                 inp->inp_vflag |= INP_IPV4;
 522                 inp->inp_vflag &= ~INP_IPV6;
 523                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 524                         goto out;
 525                 error = tcp_output(tp);
 526                 goto out;
 527         }
 528         inp->inp_vflag &= ~INP_IPV4;
 529         inp->inp_vflag |= INP_IPV6;
 530         inp->inp_inc.inc_isipv6 = 1;
 531         if ((error = tcp6_connect(tp, nam, td)) != 0)
 532                 goto out;
 533         error = tcp_output(tp);
 534
 535 out:
 536         TCPDEBUG2(PRU_CONNECT);
 537         INP_UNLOCK(inp);
 538         INP_INFO_WUNLOCK(&tcbinfo);
 539         return (error);
 540 }
 541 #endif /* INET6 */
 542
 543 /*
 544  * Initiate disconnect from peer.
 545  * If connection never passed embryonic stage, just drop;
 546  * else if don't need to let data drain, then can just drop anyways,
 547  * else have to begin TCP shutdown process: mark socket disconnecting,
 548  * drain unread data, state switch to reflect user close, and
 549  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 550  * when peer sends FIN and acks ours.
 551  *
 552  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 553  */
 554 static int
 555 tcp_usr_disconnect(struct socket *so)
 556 {
 557         struct inpcb *inp;
 558         struct tcpcb *tp = NULL;
 559         int error = 0;
 560
 561         TCPDEBUG0;
 562         INP_INFO_WLOCK(&tcbinfo);
 563         inp = sotoinpcb(so);
 564         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 565         INP_LOCK(inp);
 566         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 567                 error = EINVAL;
 568                 goto out;
 569         }
 570         tp = intotcpcb(inp);
 571         TCPDEBUG1();
 572         tcp_disconnect(tp);
 573 out:
 574         TCPDEBUG2(PRU_DISCONNECT);
 575         INP_UNLOCK(inp);
 576         INP_INFO_WUNLOCK(&tcbinfo);
 577         return (error);
 578 }
 579
 580 /*
 581  * Accept a connection.  Essentially all the work is
 582  * done at higher levels; just return the address
 583  * of the peer, storing through addr.
 584  */
 585 static int
 586 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 587 {
 588         int error = 0;
 589         struct inpcb *inp = NULL;
 590         struct tcpcb *tp = NULL;
 591         struct in_addr addr;
 592         in_port_t port = 0;
 593         TCPDEBUG0;
 594
 595         if (so->so_state & SS_ISDISCONNECTED)
 596                 return (ECONNABORTED);
 597
 598         inp = sotoinpcb(so);
 599         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 600         INP_LOCK(inp);
 601         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 602                 error = ECONNABORTED;
 603                 goto out;
 604         }
 605         tp = intotcpcb(inp);
 606         TCPDEBUG1();
 607
 608         /*
 609          * We inline in_setpeeraddr and COMMON_END here, so that we can
 610          * copy the data of interest and defer the malloc until after we
 611          * release the lock.
 612          */
 613         port = inp->inp_fport;
 614         addr = inp->inp_faddr;
 615
 616 out:
 617         TCPDEBUG2(PRU_ACCEPT);
 618         INP_UNLOCK(inp);
 619         if (error == 0)
 620                 *nam = in_sockaddr(port, &addr);
 621         return error;
 622 }
 623
 624 #ifdef INET6
 625 static int
 626 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 627 {
 628         struct inpcb *inp = NULL;
 629         int error = 0;
 630         struct tcpcb *tp = NULL;
 631         struct in_addr addr;
 632         struct in6_addr addr6;
 633         in_port_t port = 0;
 634         int v4 = 0;
 635         TCPDEBUG0;
 636
 637         if (so->so_state & SS_ISDISCONNECTED) {
 638                 error = ECONNABORTED;
 639                 goto out;
 640         }
 641
 642         inp = sotoinpcb(so);
 643         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 644         INP_LOCK(inp);
 645         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 646                 error = EINVAL;
 647                 goto out;
 648         }
 649         tp = intotcpcb(inp);
 650         TCPDEBUG1();
 651
 652         /*
 653          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 654          * copy the data of interest and defer the malloc until after we
 655          * release the lock.
 656          */
 657         if (inp->inp_vflag & INP_IPV4) {
 658                 v4 = 1;
 659                 port = inp->inp_fport;
 660                 addr = inp->inp_faddr;
 661         } else {
 662                 port = inp->inp_fport;
 663                 addr6 = inp->in6p_faddr;
 664         }
 665
 666 out:
 667         TCPDEBUG2(PRU_ACCEPT);
 668         INP_UNLOCK(inp);
 669         if (error == 0) {
 670                 if (v4)
 671                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 672                 else
 673                         *nam = in6_sockaddr(port, &addr6);
 674         }
 675         return error;
 676 }
 677 #endif /* INET6 */
 678
 679 /*
 680  * This is the wrapper function for in_setsockaddr. We just pass down
 681  * the pcbinfo for in_setsockaddr to lock. We don't want to do the locking
 682  * here because in_setsockaddr will call malloc and can block.
 683  */
 684 static int
 685 tcp_sockaddr(struct socket *so, struct sockaddr **nam)
 686 {
 687         return (in_setsockaddr(so, nam, &tcbinfo));
 688 }
 689
 690 /*
 691  * This is the wrapper function for in_setpeeraddr. We just pass down
 692  * the pcbinfo for in_setpeeraddr to lock.
 693  */
 694 static int
 695 tcp_peeraddr(struct socket *so, struct sockaddr **nam)
 696 {
 697         return (in_setpeeraddr(so, nam, &tcbinfo));
 698 }
 699
 700 /*
 701  * Mark the connection as being incapable of further output.
 702  */
 703 static int
 704 tcp_usr_shutdown(struct socket *so)
 705 {
 706         int error = 0;
 707         struct inpcb *inp;
 708         struct tcpcb *tp = NULL;
 709
 710         TCPDEBUG0;
 711         INP_INFO_WLOCK(&tcbinfo);
 712         inp = sotoinpcb(so);
 713         KASSERT(inp != NULL, ("inp == NULL"));
 714         INP_LOCK(inp);
 715         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 716                 error = EINVAL;
 717                 goto out;
 718         }
 719         tp = intotcpcb(inp);
 720         TCPDEBUG1();
 721         socantsendmore(so);
 722         tcp_usrclosed(tp);
 723         error = tcp_output(tp);
 724
 725 out:
 726         TCPDEBUG2(PRU_SHUTDOWN);
 727         INP_UNLOCK(inp);
 728         INP_INFO_WUNLOCK(&tcbinfo);
 729
 730         return (error);
 731 }
 732
 733 /*
 734  * After a receive, possibly send window update to peer.
 735  */
 736 static int
 737 tcp_usr_rcvd(struct socket *so, int flags)
 738 {
 739         struct inpcb *inp;
 740         struct tcpcb *tp = NULL;
 741         int error = 0;
 742
 743         TCPDEBUG0;
 744         inp = sotoinpcb(so);
 745         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 746         INP_LOCK(inp);
 747         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 748                 error = EINVAL;
 749                 goto out;
 750         }
 751         tp = intotcpcb(inp);
 752         TCPDEBUG1();
 753         tcp_output(tp);
 754
 755 out:
 756         TCPDEBUG2(PRU_RCVD);
 757         INP_UNLOCK(inp);
 758         return (error);
 759 }
 760
 761 /*
 762  * Do a send by putting data in output queue and updating urgent
 763  * marker if URG set.  Possibly send more data.  Unlike the other
 764  * pru_*() routines, the mbuf chains are our responsibility.  We
 765  * must either enqueue them or free them.  The other pru_* routines
 766  * generally are caller-frees.
 767  */
 768 static int
 769 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 770              struct sockaddr *nam, struct mbuf *control, struct thread *td)
 771 {
 772         int error = 0;
 773         struct inpcb *inp;
 774         struct tcpcb *tp = NULL;
 775         int headlocked = 0;
 776 #ifdef INET6
 777         int isipv6;
 778 #endif
 779         TCPDEBUG0;
 780
 781         /*
 782          * We require the pcbinfo lock in two cases:
 783          *
 784          * (1) An implied connect is taking place, which can result in
 785          *     binding IPs and ports and hence modification of the pcb hash
 786          *     chains.
 787          *
 788          * (2) PRUS_EOF is set, resulting in explicit close on the send.
 789          */
 790         if ((nam != NULL) || (flags & PRUS_EOF)) {
 791                 INP_INFO_WLOCK(&tcbinfo);
 792                 headlocked = 1;
 793         }
 794         inp = sotoinpcb(so);
 795         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 796         INP_LOCK(inp);
 797         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 798                 error = EINVAL;
 799                 goto out;
 800         }
 801 #ifdef INET6
 802         isipv6 = nam && nam->sa_family == AF_INET6;
 803 #endif /* INET6 */
 804         tp = intotcpcb(inp);
 805         TCPDEBUG1();
 806         if (control) {
 807                 /* TCP doesn't do control messages (rights, creds, etc) */
 808                 if (control->m_len) {
 809                         m_freem(control);
 810                         if (m)
 811                                 m_freem(m);
 812                         error = EINVAL;
 813                         goto out;
 814                 }
 815                 m_freem(control);       /* empty control, just free it */
 816         }
 817         if (!(flags & PRUS_OOB)) {
 818                 sbappendstream(&so->so_snd, m);
 819                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 820                         /*
 821                          * Do implied connect if not yet connected,
 822                          * initialize window to default value, and
 823                          * initialize maxseg/maxopd using peer's cached
 824                          * MSS.
 825                          */
 826                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 827 #ifdef INET6
 828                         if (isipv6)
 829                                 error = tcp6_connect(tp, nam, td);
 830                         else
 831 #endif /* INET6 */
 832                         error = tcp_connect(tp, nam, td);
 833                         if (error)
 834                                 goto out;
 835                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 836                         tcp_mss(tp, -1);
 837                 }
 838                 if (flags & PRUS_EOF) {
 839                         /*
 840                          * Close the send side of the connection after
 841                          * the data is sent.
 842                          */
 843                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 844                         socantsendmore(so);
 845                         tcp_usrclosed(tp);
 846                 }
 847                 if (headlocked) {
 848                         INP_INFO_WUNLOCK(&tcbinfo);
 849                         headlocked = 0;
 850                 }
 851                 if (tp != NULL) {
 852                         if (flags & PRUS_MORETOCOME)
 853                                 tp->t_flags |= TF_MORETOCOME;
 854                         error = tcp_output(tp);
 855                         if (flags & PRUS_MORETOCOME)
 856                                 tp->t_flags &= ~TF_MORETOCOME;
 857                 }
 858         } else {
 859                 /*
 860                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 861                  */
 862                 SOCKBUF_LOCK(&so->so_snd);
 863                 if (sbspace(&so->so_snd) < -512) {
 864                         SOCKBUF_UNLOCK(&so->so_snd);
 865                         m_freem(m);
 866                         error = ENOBUFS;
 867                         goto out;
 868                 }
 869                 /*
 870                  * According to RFC961 (Assigned Protocols),
 871                  * the urgent pointer points to the last octet
 872                  * of urgent data.  We continue, however,
 873                  * to consider it to indicate the first octet
 874                  * of data past the urgent section.
 875                  * Otherwise, snd_up should be one lower.
 876                  */
 877                 sbappendstream_locked(&so->so_snd, m);
 878                 SOCKBUF_UNLOCK(&so->so_snd);
 879                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 880                         /*
 881                          * Do implied connect if not yet connected,
 882                          * initialize window to default value, and
 883                          * initialize maxseg/maxopd using peer's cached
 884                          * MSS.
 885                          */
 886                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 887 #ifdef INET6
 888                         if (isipv6)
 889                                 error = tcp6_connect(tp, nam, td);
 890                         else
 891 #endif /* INET6 */
 892                         error = tcp_connect(tp, nam, td);
 893                         if (error)
 894                                 goto out;
 895                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 896                         tcp_mss(tp, -1);
 897                         INP_INFO_WUNLOCK(&tcbinfo);
 898                         headlocked = 0;
 899                 } else if (nam) {
 900                         INP_INFO_WUNLOCK(&tcbinfo);
 901                         headlocked = 0;
 902                 }
 903                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 904                 tp->t_flags |= TF_FORCEDATA;
 905                 error = tcp_output(tp);
 906                 tp->t_flags &= ~TF_FORCEDATA;
 907         }
 908 out:
 909         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 910                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 911         INP_UNLOCK(inp);
 912         if (headlocked)
 913                 INP_INFO_WUNLOCK(&tcbinfo);
 914         return (error);
 915 }
 916
 917 /*
 918  * Abort the TCP.
 919  */
 920 static void
 921 tcp_usr_abort(struct socket *so)
 922 {
 923 #if 0
 924         struct inpcb *inp;
 925         struct tcpcb *tp;
 926 #endif
 927
 928         /*
 929          * XXXRW: This is not really quite the same, as we want to tcp_drop()
 930          * rather than tcp_disconnect(), I think, but for now I'll avoid
 931          * replicating all the tear-down logic here.
 932          */
 933         tcp_usr_detach(so);
 934
 935 #if 0
 936         TCPDEBUG0;
 937         INP_INFO_WLOCK(&tcbinfo);
 938         inp = sotoinpcb(so);
 939         INP_LOCK(inp);
 940         /*
 941          * Do we need to handle timewait here?  Aborted connections should
 942          * never generate a FIN?
 943          */
 944         KASSERT((inp->inp_vflag & INP_TIMEWAIT) == 0,
 945             ("tcp_usr_abort: timewait"));
 946         tp = intotcpcb(inp);
 947         TCPDEBUG1();
 948         tp = tcp_drop(tp, ECONNABORTED);
 949                 TCPDEBUG2(PRU_ABORT);
 950         if (tp != NULL)
 951                 INP_UNLOCK(inp);
 952         INP_INFO_WUNLOCK(&tcbinfo);
 953 #endif
 954 }
 955
 956 /*
 957  * Receive out-of-band data.
 958  */
 959 static int
 960 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
 961 {
 962         int error = 0;
 963         struct inpcb *inp;
 964         struct tcpcb *tp = NULL;
 965
 966         TCPDEBUG0;
 967         inp = sotoinpcb(so);
 968         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
 969         INP_LOCK(inp);
 970         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 971                 error = EINVAL;
 972                 goto out;
 973         }
 974         tp = intotcpcb(inp);
 975         TCPDEBUG1();
 976         if ((so->so_oobmark == 0 &&
 977              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
 978             so->so_options & SO_OOBINLINE ||
 979             tp->t_oobflags & TCPOOB_HADDATA) {
 980                 error = EINVAL;
 981                 goto out;
 982         }
 983         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 984                 error = EWOULDBLOCK;
 985                 goto out;
 986         }
 987         m->m_len = 1;
 988         *mtod(m, caddr_t) = tp->t_iobc;
 989         if ((flags & MSG_PEEK) == 0)
 990                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 991
 992 out:
 993         TCPDEBUG2(PRU_RCVOOB);
 994         INP_UNLOCK(inp);
 995         return (error);
 996 }
 997
 998 struct pr_usrreqs tcp_usrreqs = {
 999         .pru_abort =            tcp_usr_abort,
1000         .pru_accept =           tcp_usr_accept,
1001         .pru_attach =           tcp_usr_attach,
1002         .pru_bind =             tcp_usr_bind,
1003         .pru_connect =          tcp_usr_connect,
1004         .pru_control =          in_control,
1005         .pru_detach =           tcp_usr_detach,
1006         .pru_disconnect =       tcp_usr_disconnect,
1007         .pru_listen =           tcp_usr_listen,
1008         .pru_peeraddr =         tcp_peeraddr,
1009         .pru_rcvd =             tcp_usr_rcvd,
1010         .pru_rcvoob =           tcp_usr_rcvoob,
1011         .pru_send =             tcp_usr_send,
1012         .pru_shutdown =         tcp_usr_shutdown,
1013         .pru_sockaddr =         tcp_sockaddr,
1014         .pru_sosetlabel =       in_pcbsosetlabel
1015 };
1016
1017 #ifdef INET6
1018 struct pr_usrreqs tcp6_usrreqs = {
1019         .pru_abort =            tcp_usr_abort,
1020         .pru_accept =           tcp6_usr_accept,
1021         .pru_attach =           tcp_usr_attach,
1022         .pru_bind =             tcp6_usr_bind,
1023         .pru_connect =          tcp6_usr_connect,
1024         .pru_control =          in6_control,
1025         .pru_detach =           tcp_usr_detach,
1026         .pru_disconnect =       tcp_usr_disconnect,
1027         .pru_listen =           tcp6_usr_listen,
1028         .pru_peeraddr =         in6_mapped_peeraddr,
1029         .pru_rcvd =             tcp_usr_rcvd,
1030         .pru_rcvoob =           tcp_usr_rcvoob,
1031         .pru_send =             tcp_usr_send,
1032         .pru_shutdown =         tcp_usr_shutdown,
1033         .pru_sockaddr =         in6_mapped_sockaddr,
1034         .pru_sosetlabel =       in_pcbsosetlabel
1035 };
1036 #endif /* INET6 */
1037
1038 /*
1039  * Common subroutine to open a TCP connection to remote host specified
1040  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1041  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1042  * to choose a local host address (interface).  If there is an existing
1043  * incarnation of the same connection in TIME-WAIT state and if the remote
1044  * host was sending CC options and if the connection duration was < MSL, then
1045  * truncate the previous TIME-WAIT state and proceed.
1046  * Initialize connection parameters and enter SYN-SENT state.
1047  */
1048 static int
1049 tcp_connect(tp, nam, td)
1050         register struct tcpcb *tp;
1051         struct sockaddr *nam;
1052         struct thread *td;
1053 {
1054         struct inpcb *inp = tp->t_inpcb, *oinp;
1055         struct socket *so = inp->inp_socket;
1056         struct in_addr laddr;
1057         u_short lport;
1058         int error;
1059
1060         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1061         INP_LOCK_ASSERT(inp);
1062
1063         if (inp->inp_lport == 0) {
1064                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1065                 if (error)
1066                         return error;
1067         }
1068
1069         /*
1070          * Cannot simply call in_pcbconnect, because there might be an
1071          * earlier incarnation of this same connection still in
1072          * TIME_WAIT state, creating an ADDRINUSE error.
1073          */
1074         laddr = inp->inp_laddr;
1075         lport = inp->inp_lport;
1076         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1077             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1078         if (error && oinp == NULL)
1079                 return error;
1080         if (oinp)
1081                 return EADDRINUSE;
1082         inp->inp_laddr = laddr;
1083         in_pcbrehash(inp);
1084
1085         /* Compute window scaling to request.  */
1086         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1087             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1088                 tp->request_r_scale++;
1089
1090         soisconnecting(so);
1091         tcpstat.tcps_connattempt++;
1092         tp->t_state = TCPS_SYN_SENT;
1093         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1094         tp->iss = tcp_new_isn(tp);
1095         tp->t_bw_rtseq = tp->iss;
1096         tcp_sendseqinit(tp);
1097
1098         return 0;
1099 }
1100
1101 #ifdef INET6
1102 static int
1103 tcp6_connect(tp, nam, td)
1104         register struct tcpcb *tp;
1105         struct sockaddr *nam;
1106         struct thread *td;
1107 {
1108         struct inpcb *inp = tp->t_inpcb, *oinp;
1109         struct socket *so = inp->inp_socket;
1110         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1111         struct in6_addr *addr6;
1112         int error;
1113
1114         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1115         INP_LOCK_ASSERT(inp);
1116
1117         if (inp->inp_lport == 0) {
1118                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1119                 if (error)
1120                         return error;
1121         }
1122
1123         /*
1124          * Cannot simply call in_pcbconnect, because there might be an
1125          * earlier incarnation of this same connection still in
1126          * TIME_WAIT state, creating an ADDRINUSE error.
1127          * in6_pcbladdr() also handles scope zone IDs.
1128          */
1129         error = in6_pcbladdr(inp, nam, &addr6);
1130         if (error)
1131                 return error;
1132         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1133                                   &sin6->sin6_addr, sin6->sin6_port,
1134                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1135                                   ? addr6
1136                                   : &inp->in6p_laddr,
1137                                   inp->inp_lport,  0, NULL);
1138         if (oinp)
1139                 return EADDRINUSE;
1140         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1141                 inp->in6p_laddr = *addr6;
1142         inp->in6p_faddr = sin6->sin6_addr;
1143         inp->inp_fport = sin6->sin6_port;
1144         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1145         inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
1146         if (inp->in6p_flags & IN6P_AUTOFLOWLABEL)
1147                 inp->in6p_flowinfo |=
1148                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1149         in_pcbrehash(inp);
1150
1151         /* Compute window scaling to request.  */
1152         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1153             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1154                 tp->request_r_scale++;
1155
1156         soisconnecting(so);
1157         tcpstat.tcps_connattempt++;
1158         tp->t_state = TCPS_SYN_SENT;
1159         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1160         tp->iss = tcp_new_isn(tp);
1161         tp->t_bw_rtseq = tp->iss;
1162         tcp_sendseqinit(tp);
1163
1164         return 0;
1165 }
1166 #endif /* INET6 */
1167
1168 /*
1169  * Export TCP internal state information via a struct tcp_info, based on the
1170  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1171  * (TCP state machine, etc).  We export all information using FreeBSD-native
1172  * constants -- for example, the numeric values for tcpi_state will differ
1173  * from Linux.
1174  */
1175 static void
1176 tcp_fill_info(tp, ti)
1177         struct tcpcb *tp;
1178         struct tcp_info *ti;
1179 {
1180
1181         INP_LOCK_ASSERT(tp->t_inpcb);
1182         bzero(ti, sizeof(*ti));
1183
1184         ti->tcpi_state = tp->t_state;
1185         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1186                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1187         if (tp->sack_enable)
1188                 ti->tcpi_options |= TCPI_OPT_SACK;
1189         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1190                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1191                 ti->tcpi_snd_wscale = tp->snd_scale;
1192                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1193         }
1194         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1195         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1196
1197         /*
1198          * FreeBSD-specific extension fields for tcp_info.
1199          */
1200         ti->tcpi_rcv_space = tp->rcv_wnd;
1201         ti->tcpi_snd_wnd = tp->snd_wnd;
1202         ti->tcpi_snd_bwnd = tp->snd_bwnd;
1203 }
1204
1205 /*
1206  * The new sockopt interface makes it possible for us to block in the
1207  * copyin/out step (if we take a page fault).  Taking a page fault at
1208  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
1209  * use TSM, there probably isn't any need for this function to run at
1210  * splnet() any more.  This needs more examination.)
1211  *
1212  * XXXRW: The locking here is wrong; we may take a page fault while holding
1213  * the inpcb lock.
1214  */
1215 int
1216 tcp_ctloutput(so, sopt)
1217         struct socket *so;
1218         struct sockopt *sopt;
1219 {
1220         int     error, opt, optval;
1221         struct  inpcb *inp;
1222         struct  tcpcb *tp;
1223         struct  tcp_info ti;
1224
1225         error = 0;
1226         inp = sotoinpcb(so);
1227         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1228         INP_LOCK(inp);
1229         if (sopt->sopt_level != IPPROTO_TCP) {
1230                 INP_UNLOCK(inp);
1231 #ifdef INET6
1232                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1233                         error = ip6_ctloutput(so, sopt);
1234                 else
1235 #endif /* INET6 */
1236                 error = ip_ctloutput(so, sopt);
1237                 return (error);
1238         }
1239         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
1240                 error = ECONNRESET;
1241                 goto out;
1242         }
1243         tp = intotcpcb(inp);
1244
1245         switch (sopt->sopt_dir) {
1246         case SOPT_SET:
1247                 switch (sopt->sopt_name) {
1248 #ifdef TCP_SIGNATURE
1249                 case TCP_MD5SIG:
1250                         error = sooptcopyin(sopt, &optval, sizeof optval,
1251                                             sizeof optval);
1252                         if (error)
1253                                 break;
1254
1255                         if (optval > 0)
1256                                 tp->t_flags |= TF_SIGNATURE;
1257                         else
1258                                 tp->t_flags &= ~TF_SIGNATURE;
1259                         break;
1260 #endif /* TCP_SIGNATURE */
1261                 case TCP_NODELAY:
1262                 case TCP_NOOPT:
1263                         error = sooptcopyin(sopt, &optval, sizeof optval,
1264                                             sizeof optval);
1265                         if (error)
1266                                 break;
1267
1268                         switch (sopt->sopt_name) {
1269                         case TCP_NODELAY:
1270                                 opt = TF_NODELAY;
1271                                 break;
1272                         case TCP_NOOPT:
1273                                 opt = TF_NOOPT;
1274                                 break;
1275                         default:
1276                                 opt = 0; /* dead code to fool gcc */
1277                                 break;
1278                         }
1279
1280                         if (optval)
1281                                 tp->t_flags |= opt;
1282                         else
1283                                 tp->t_flags &= ~opt;
1284                         break;
1285
1286                 case TCP_NOPUSH:
1287                         error = sooptcopyin(sopt, &optval, sizeof optval,
1288                                             sizeof optval);
1289                         if (error)
1290                                 break;
1291
1292                         if (optval)
1293                                 tp->t_flags |= TF_NOPUSH;
1294                         else {
1295                                 tp->t_flags &= ~TF_NOPUSH;
1296                                 error = tcp_output(tp);
1297                         }
1298                         break;
1299
1300                 case TCP_MAXSEG:
1301                         error = sooptcopyin(sopt, &optval, sizeof optval,
1302                                             sizeof optval);
1303                         if (error)
1304                                 break;
1305
1306                         if (optval > 0 && optval <= tp->t_maxseg &&
1307                             optval + 40 >= tcp_minmss)
1308                                 tp->t_maxseg = optval;
1309                         else
1310                                 error = EINVAL;
1311                         break;
1312
1313                 case TCP_INFO:
1314                         error = EINVAL;
1315                         break;
1316
1317                 default:
1318                         error = ENOPROTOOPT;
1319                         break;
1320                 }
1321                 break;
1322
1323         case SOPT_GET:
1324                 switch (sopt->sopt_name) {
1325 #ifdef TCP_SIGNATURE
1326                 case TCP_MD5SIG:
1327                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1328                         error = sooptcopyout(sopt, &optval, sizeof optval);
1329                         break;
1330 #endif
1331                 case TCP_NODELAY:
1332                         optval = tp->t_flags & TF_NODELAY;
1333                         error = sooptcopyout(sopt, &optval, sizeof optval);
1334                         break;
1335                 case TCP_MAXSEG:
1336                         optval = tp->t_maxseg;
1337                         error = sooptcopyout(sopt, &optval, sizeof optval);
1338                         break;
1339                 case TCP_NOOPT:
1340                         optval = tp->t_flags & TF_NOOPT;
1341                         error = sooptcopyout(sopt, &optval, sizeof optval);
1342                         break;
1343                 case TCP_NOPUSH:
1344                         optval = tp->t_flags & TF_NOPUSH;
1345                         error = sooptcopyout(sopt, &optval, sizeof optval);
1346                         break;
1347                 case TCP_INFO:
1348                         tcp_fill_info(tp, &ti);
1349                         error = sooptcopyout(sopt, &ti, sizeof ti);
1350                         break;
1351                 default:
1352                         error = ENOPROTOOPT;
1353                         break;
1354                 }
1355                 break;
1356         }
1357 out:
1358         INP_UNLOCK(inp);
1359         return (error);
1360 }
1361
1362 /*
1363  * tcp_sendspace and tcp_recvspace are the default send and receive window
1364  * sizes, respectively.  These are obsolescent (this information should
1365  * be set by the route).
1366  */
1367 u_long  tcp_sendspace = 1024*32;
1368 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1369     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1370 u_long  tcp_recvspace = 1024*64;
1371 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1372     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1373
1374 /*
1375  * Attach TCP protocol to socket, allocating
1376  * internet protocol control block, tcp control block,
1377  * bufer space, and entering LISTEN state if to accept connections.
1378  */
1379 static int
1380 tcp_attach(so)
1381         struct socket *so;
1382 {
1383         register struct tcpcb *tp;
1384         struct inpcb *inp;
1385         int error;
1386 #ifdef INET6
1387         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1388 #endif
1389
1390         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1391
1392         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1393                 error = soreserve(so, tcp_sendspace, tcp_recvspace);
1394                 if (error)
1395                         return (error);
1396         }
1397         error = in_pcballoc(so, &tcbinfo, "tcpinp");
1398         if (error)
1399                 return (error);
1400         inp = sotoinpcb(so);
1401 #ifdef INET6
1402         if (isipv6) {
1403                 inp->inp_vflag |= INP_IPV6;
1404                 inp->in6p_hops = -1;    /* use kernel default */
1405         }
1406         else
1407 #endif
1408         inp->inp_vflag |= INP_IPV4;
1409         tp = tcp_newtcpcb(inp);
1410         if (tp == NULL) {
1411                 INP_LOCK(inp);
1412 #ifdef INET6
1413                 if (isipv6) {
1414                         in6_pcbdetach(inp);
1415                         in6_pcbfree(inp);
1416                 } else {
1417 #endif
1418                         in_pcbdetach(inp);
1419                         in_pcbfree(inp);
1420 #ifdef INET6
1421                 }
1422 #endif
1423                 return (ENOBUFS);
1424         }
1425         tp->t_state = TCPS_CLOSED;
1426         return (0);
1427 }
1428
1429 /*
1430  * Initiate (or continue) disconnect.
1431  * If embryonic state, just send reset (once).
1432  * If in ``let data drain'' option and linger null, just drop.
1433  * Otherwise (hard), mark socket disconnecting and drop
1434  * current input data; switch states based on user close, and
1435  * send segment to peer (with FIN).
1436  */
1437 static void
1438 tcp_disconnect(tp)
1439         register struct tcpcb *tp;
1440 {
1441         struct inpcb *inp = tp->t_inpcb;
1442         struct socket *so = inp->inp_socket;
1443
1444         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1445         INP_LOCK_ASSERT(inp);
1446
1447         /*
1448          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1449          * socket is still open.
1450          */
1451         if (tp->t_state < TCPS_ESTABLISHED) {
1452                 tp = tcp_close(tp);
1453                 KASSERT(tp != NULL,
1454                     ("tcp_disconnect: tcp_close() returned NULL"));
1455         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1456                 tp = tcp_drop(tp, 0);
1457                 KASSERT(tp != NULL,
1458                     ("tcp_disconnect: tcp_drop() returned NULL"));
1459         } else {
1460                 soisdisconnecting(so);
1461                 sbflush(&so->so_rcv);
1462                 tcp_usrclosed(tp);
1463                 if (!(inp->inp_vflag & INP_DROPPED))
1464                         tcp_output(tp);
1465         }
1466 }
1467
1468 /*
1469  * User issued close, and wish to trail through shutdown states:
1470  * if never received SYN, just forget it.  If got a SYN from peer,
1471  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1472  * If already got a FIN from peer, then almost done; go to LAST_ACK
1473  * state.  In all other cases, have already sent FIN to peer (e.g.
1474  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1475  * for peer to send FIN or not respond to keep-alives, etc.
1476  * We can let the user exit from the close as soon as the FIN is acked.
1477  */
1478 static void
1479 tcp_usrclosed(tp)
1480         register struct tcpcb *tp;
1481 {
1482
1483         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1484         INP_LOCK_ASSERT(tp->t_inpcb);
1485
1486         switch (tp->t_state) {
1487
1488         case TCPS_CLOSED:
1489         case TCPS_LISTEN:
1490                 tp->t_state = TCPS_CLOSED;
1491                 tp = tcp_close(tp);
1492                 /*
1493                  * tcp_close() should never return NULL here as the socket is
1494                  * still open.
1495                  */
1496                 KASSERT(tp != NULL,
1497                     ("tcp_usrclosed: tcp_close() returned NULL"));
1498                 break;
1499
1500         case TCPS_SYN_SENT:
1501         case TCPS_SYN_RECEIVED:
1502                 tp->t_flags |= TF_NEEDFIN;
1503                 break;
1504
1505         case TCPS_ESTABLISHED:
1506                 tp->t_state = TCPS_FIN_WAIT_1;
1507                 break;
1508
1509         case TCPS_CLOSE_WAIT:
1510                 tp->t_state = TCPS_LAST_ACK;
1511                 break;
1512         }
1513         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1514                 soisdisconnected(tp->t_inpcb->inp_socket);
1515                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1516                 if (tp->t_state == TCPS_FIN_WAIT_2)
1517                         callout_reset(tp->tt_2msl, tcp_maxidle,
1518                                       tcp_timer_2msl, tp);
1519         }
1520 }