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         TCPDEBUG1();
 124
 125         error = tcp_attach(so);
 126         if (error)
 127                 goto out;
 128
 129         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 130                 so->so_linger = TCP_LINGERTIME;
 131
 132         inp = sotoinpcb(so);
 133         tp = intotcpcb(inp);
 134 out:
 135         TCPDEBUG2(PRU_ATTACH);
 136         return error;
 137 }
 138
 139 /*
 140  * tcp_detach() releases any protocol state that can be reasonably released
 141  * when a socket shutdown is requested, and is a shared code path for
 142  * tcp_usr_detach() and tcp_usr_abort(), the two socket close entry points.
 143  *
 144  * Accepts pcbinfo, inpcb locked, will unlock the inpcb (if needed) on
 145  * return.
 146  */
 147 static void
 148 tcp_detach(struct socket *so, struct inpcb *inp)
 149 {
 150         struct tcpcb *tp;
 151 #ifdef INET6
 152         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
 153 #endif
 154
 155         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 156         INP_LOCK_ASSERT(inp);
 157
 158         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 159         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 160
 161         if (inp->inp_vflag & INP_TIMEWAIT) {
 162                 if (inp->inp_vflag & INP_DROPPED) {
 163                         /*
 164                          * Connection was in time wait and has been dropped;
 165                          * the calling path is either via tcp_twclose(), or
 166                          * as a result of an eventual soclose() after
 167                          * tcp_twclose() has been called.  In either case,
 168                          * tcp_twclose() has detached the tcptw from the
 169                          * inpcb, so we just detach and free the inpcb.
 170                          *
 171                          * XXXRW: Would it be cleaner to free the tcptw
 172                          * here?
 173                          */
 174 #ifdef INET6
 175                         if (isipv6) {
 176                                 in6_pcbdetach(inp);
 177                                 in6_pcbfree(inp);
 178                         } else {
 179 #endif
 180                                 in_pcbdetach(inp);
 181                                 in_pcbfree(inp);
 182 #ifdef INET6
 183                         }
 184 #endif
 185                 } else {
 186                         /*
 187                          * Connection is in time wait and has not yet been
 188                          * dropped; allow the socket to be discarded, but
 189                          * need to keep inpcb until end of time wait.
 190                          */
 191 #ifdef INET6
 192                         if (isipv6)
 193                                 in6_pcbdetach(inp);
 194                         else
 195 #endif
 196                                 in_pcbdetach(inp);
 197                         INP_UNLOCK(inp);
 198                 }
 199         } else {
 200                 /*
 201                  * If not in timewait, there are two possible paths.  First,
 202                  * the TCP connection is either embryonic or done, in which
 203                  * case we tear down all state.  Second, it may still be
 204                  * active, in which case we acquire a reference to the socket
 205                  * and will free it later when TCP is done.
 206                  */
 207                 tp = intotcpcb(inp);
 208                 if (inp->inp_vflag & INP_DROPPED ||
 209                     tp->t_state < TCPS_SYN_SENT) {
 210                         tcp_discardcb(tp);
 211 #ifdef INET6
 212                         if (isipv6) {
 213                                 in_pcbdetach(inp);
 214                                 in_pcbfree(inp);
 215                         } else {
 216 #endif
 217                                 in_pcbdetach(inp);
 218                                 in_pcbfree(inp);
 219 #ifdef INET6
 220                         }
 221 #endif
 222                 } else {
 223                         SOCK_LOCK(so);
 224                         so->so_state |= SS_PROTOREF;
 225                         SOCK_UNLOCK(so);
 226                         inp->inp_vflag |= INP_SOCKREF;
 227                         INP_UNLOCK(inp);
 228                 }
 229         }
 230 }
 231
 232 /*
 233  * pru_detach() detaches the TCP protocol from the socket.
 234  * If the protocol state is non-embryonic, then can't
 235  * do this directly: have to initiate a pru_disconnect(),
 236  * which may finish later; embryonic TCB's can just
 237  * be discarded here.
 238  */
 239 static void
 240 tcp_usr_detach(struct socket *so)
 241 {
 242         struct inpcb *inp;
 243         struct tcpcb *tp;
 244         TCPDEBUG0;
 245
 246         inp = sotoinpcb(so);
 247         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 248         INP_INFO_WLOCK(&tcbinfo);
 249         INP_LOCK(inp);
 250         KASSERT(inp->inp_socket != NULL,
 251             ("tcp_usr_detach: inp_socket == NULL"));
 252         TCPDEBUG1();
 253
 254         /*
 255          * First, if we still have full TCP state, and we're not dropped,
 256          * initiate a disconnect.
 257          */
 258         if (!(inp->inp_vflag & INP_TIMEWAIT) &&
 259             !(inp->inp_vflag & INP_DROPPED)) {
 260                 tp = intotcpcb(inp);
 261                 tcp_disconnect(tp);
 262         }
 263         tcp_detach(so, inp);
 264         tp = NULL;
 265         TCPDEBUG2(PRU_DETACH);
 266         INP_INFO_WUNLOCK(&tcbinfo);
 267 }
 268
 269 /*
 270  * Give the socket an address.
 271  */
 272 static int
 273 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 274 {
 275         int error = 0;
 276         struct inpcb *inp;
 277         struct tcpcb *tp = NULL;
 278         struct sockaddr_in *sinp;
 279
 280         sinp = (struct sockaddr_in *)nam;
 281         if (nam->sa_len != sizeof (*sinp))
 282                 return (EINVAL);
 283         /*
 284          * Must check for multicast addresses and disallow binding
 285          * to them.
 286          */
 287         if (sinp->sin_family == AF_INET &&
 288             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 289                 return (EAFNOSUPPORT);
 290
 291         TCPDEBUG0;
 292         INP_INFO_WLOCK(&tcbinfo);
 293         inp = sotoinpcb(so);
 294         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 295         INP_LOCK(inp);
 296         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 297                 error = EINVAL;
 298                 goto out;
 299         }
 300         tp = intotcpcb(inp);
 301         TCPDEBUG1();
 302         error = in_pcbbind(inp, nam, td->td_ucred);
 303 out:
 304         TCPDEBUG2(PRU_BIND);
 305         INP_UNLOCK(inp);
 306         INP_INFO_WUNLOCK(&tcbinfo);
 307
 308         return (error);
 309 }
 310
 311 #ifdef INET6
 312 static int
 313 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 314 {
 315         int error = 0;
 316         struct inpcb *inp;
 317         struct tcpcb *tp = NULL;
 318         struct sockaddr_in6 *sin6p;
 319
 320         sin6p = (struct sockaddr_in6 *)nam;
 321         if (nam->sa_len != sizeof (*sin6p))
 322                 return (EINVAL);
 323         /*
 324          * Must check for multicast addresses and disallow binding
 325          * to them.
 326          */
 327         if (sin6p->sin6_family == AF_INET6 &&
 328             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 329                 return (EAFNOSUPPORT);
 330
 331         TCPDEBUG0;
 332         INP_INFO_WLOCK(&tcbinfo);
 333         inp = sotoinpcb(so);
 334         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 335         INP_LOCK(inp);
 336         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 337                 error = EINVAL;
 338                 goto out;
 339         }
 340         tp = intotcpcb(inp);
 341         TCPDEBUG1();
 342         inp->inp_vflag &= ~INP_IPV4;
 343         inp->inp_vflag |= INP_IPV6;
 344         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 345                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 346                         inp->inp_vflag |= INP_IPV4;
 347                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 348                         struct sockaddr_in sin;
 349
 350                         in6_sin6_2_sin(&sin, sin6p);
 351                         inp->inp_vflag |= INP_IPV4;
 352                         inp->inp_vflag &= ~INP_IPV6;
 353                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 354                             td->td_ucred);
 355                         goto out;
 356                 }
 357         }
 358         error = in6_pcbbind(inp, nam, td->td_ucred);
 359 out:
 360         TCPDEBUG2(PRU_BIND);
 361         INP_UNLOCK(inp);
 362         INP_INFO_WUNLOCK(&tcbinfo);
 363         return (error);
 364 }
 365 #endif /* INET6 */
 366
 367 /*
 368  * Prepare to accept connections.
 369  */
 370 static int
 371 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 372 {
 373         int error = 0;
 374         struct inpcb *inp;
 375         struct tcpcb *tp = NULL;
 376
 377         TCPDEBUG0;
 378         INP_INFO_WLOCK(&tcbinfo);
 379         inp = sotoinpcb(so);
 380         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 381         INP_LOCK(inp);
 382         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 383                 error = EINVAL;
 384                 goto out;
 385         }
 386         tp = intotcpcb(inp);
 387         TCPDEBUG1();
 388         SOCK_LOCK(so);
 389         error = solisten_proto_check(so);
 390         if (error == 0 && inp->inp_lport == 0)
 391                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 392         if (error == 0) {
 393                 tp->t_state = TCPS_LISTEN;
 394                 solisten_proto(so, backlog);
 395         }
 396         SOCK_UNLOCK(so);
 397
 398 out:
 399         TCPDEBUG2(PRU_LISTEN);
 400         INP_UNLOCK(inp);
 401         INP_INFO_WUNLOCK(&tcbinfo);
 402         return (error);
 403 }
 404
 405 #ifdef INET6
 406 static int
 407 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 408 {
 409         int error = 0;
 410         struct inpcb *inp;
 411         struct tcpcb *tp = NULL;
 412
 413         TCPDEBUG0;
 414         INP_INFO_WLOCK(&tcbinfo);
 415         inp = sotoinpcb(so);
 416         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 417         INP_LOCK(inp);
 418         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 419                 error = EINVAL;
 420                 goto out;
 421         }
 422         tp = intotcpcb(inp);
 423         TCPDEBUG1();
 424         SOCK_LOCK(so);
 425         error = solisten_proto_check(so);
 426         if (error == 0 && inp->inp_lport == 0) {
 427                 inp->inp_vflag &= ~INP_IPV4;
 428                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 429                         inp->inp_vflag |= INP_IPV4;
 430                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 431         }
 432         if (error == 0) {
 433                 tp->t_state = TCPS_LISTEN;
 434                 solisten_proto(so, backlog);
 435         }
 436         SOCK_UNLOCK(so);
 437
 438 out:
 439         TCPDEBUG2(PRU_LISTEN);
 440         INP_UNLOCK(inp);
 441         INP_INFO_WUNLOCK(&tcbinfo);
 442         return (error);
 443 }
 444 #endif /* INET6 */
 445
 446 /*
 447  * Initiate connection to peer.
 448  * Create a template for use in transmissions on this connection.
 449  * Enter SYN_SENT state, and mark socket as connecting.
 450  * Start keep-alive timer, and seed output sequence space.
 451  * Send initial segment on connection.
 452  */
 453 static int
 454 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 455 {
 456         int error = 0;
 457         struct inpcb *inp;
 458         struct tcpcb *tp = NULL;
 459         struct sockaddr_in *sinp;
 460
 461         sinp = (struct sockaddr_in *)nam;
 462         if (nam->sa_len != sizeof (*sinp))
 463                 return (EINVAL);
 464         /*
 465          * Must disallow TCP ``connections'' to multicast addresses.
 466          */
 467         if (sinp->sin_family == AF_INET
 468             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 469                 return (EAFNOSUPPORT);
 470         if (jailed(td->td_ucred))
 471                 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr);
 472
 473         TCPDEBUG0;
 474         INP_INFO_WLOCK(&tcbinfo);
 475         inp = sotoinpcb(so);
 476         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 477         INP_LOCK(inp);
 478         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 479                 error = EINVAL;
 480                 goto out;
 481         }
 482         tp = intotcpcb(inp);
 483         TCPDEBUG1();
 484         if ((error = tcp_connect(tp, nam, td)) != 0)
 485                 goto out;
 486         error = tcp_output(tp);
 487 out:
 488         TCPDEBUG2(PRU_CONNECT);
 489         INP_UNLOCK(inp);
 490         INP_INFO_WUNLOCK(&tcbinfo);
 491         return (error);
 492 }
 493
 494 #ifdef INET6
 495 static int
 496 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 497 {
 498         int error = 0;
 499         struct inpcb *inp;
 500         struct tcpcb *tp = NULL;
 501         struct sockaddr_in6 *sin6p;
 502
 503         TCPDEBUG0;
 504
 505         sin6p = (struct sockaddr_in6 *)nam;
 506         if (nam->sa_len != sizeof (*sin6p))
 507                 return (EINVAL);
 508         /*
 509          * Must disallow TCP ``connections'' to multicast addresses.
 510          */
 511         if (sin6p->sin6_family == AF_INET6
 512             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 513                 return (EAFNOSUPPORT);
 514
 515         INP_INFO_WLOCK(&tcbinfo);
 516         inp = sotoinpcb(so);
 517         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 518         INP_LOCK(inp);
 519         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 520                 error = EINVAL;
 521                 goto out;
 522         }
 523         tp = intotcpcb(inp);
 524         TCPDEBUG1();
 525         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 526                 struct sockaddr_in sin;
 527
 528                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 529                         error = EINVAL;
 530                         goto out;
 531                 }
 532
 533                 in6_sin6_2_sin(&sin, sin6p);
 534                 inp->inp_vflag |= INP_IPV4;
 535                 inp->inp_vflag &= ~INP_IPV6;
 536                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 537                         goto out;
 538                 error = tcp_output(tp);
 539                 goto out;
 540         }
 541         inp->inp_vflag &= ~INP_IPV4;
 542         inp->inp_vflag |= INP_IPV6;
 543         inp->inp_inc.inc_isipv6 = 1;
 544         if ((error = tcp6_connect(tp, nam, td)) != 0)
 545                 goto out;
 546         error = tcp_output(tp);
 547
 548 out:
 549         TCPDEBUG2(PRU_CONNECT);
 550         INP_UNLOCK(inp);
 551         INP_INFO_WUNLOCK(&tcbinfo);
 552         return (error);
 553 }
 554 #endif /* INET6 */
 555
 556 /*
 557  * Initiate disconnect from peer.
 558  * If connection never passed embryonic stage, just drop;
 559  * else if don't need to let data drain, then can just drop anyways,
 560  * else have to begin TCP shutdown process: mark socket disconnecting,
 561  * drain unread data, state switch to reflect user close, and
 562  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 563  * when peer sends FIN and acks ours.
 564  *
 565  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 566  */
 567 static int
 568 tcp_usr_disconnect(struct socket *so)
 569 {
 570         struct inpcb *inp;
 571         struct tcpcb *tp = NULL;
 572         int error = 0;
 573
 574         TCPDEBUG0;
 575         INP_INFO_WLOCK(&tcbinfo);
 576         inp = sotoinpcb(so);
 577         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 578         INP_LOCK(inp);
 579         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 580                 error = EINVAL;
 581                 goto out;
 582         }
 583         tp = intotcpcb(inp);
 584         TCPDEBUG1();
 585         tcp_disconnect(tp);
 586 out:
 587         TCPDEBUG2(PRU_DISCONNECT);
 588         INP_UNLOCK(inp);
 589         INP_INFO_WUNLOCK(&tcbinfo);
 590         return (error);
 591 }
 592
 593 /*
 594  * Accept a connection.  Essentially all the work is
 595  * done at higher levels; just return the address
 596  * of the peer, storing through addr.
 597  */
 598 static int
 599 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 600 {
 601         int error = 0;
 602         struct inpcb *inp = NULL;
 603         struct tcpcb *tp = NULL;
 604         struct in_addr addr;
 605         in_port_t port = 0;
 606         TCPDEBUG0;
 607
 608         if (so->so_state & SS_ISDISCONNECTED)
 609                 return (ECONNABORTED);
 610
 611         inp = sotoinpcb(so);
 612         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 613         INP_LOCK(inp);
 614         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 615                 error = ECONNABORTED;
 616                 goto out;
 617         }
 618         tp = intotcpcb(inp);
 619         TCPDEBUG1();
 620
 621         /*
 622          * We inline in_setpeeraddr and COMMON_END here, so that we can
 623          * copy the data of interest and defer the malloc until after we
 624          * release the lock.
 625          */
 626         port = inp->inp_fport;
 627         addr = inp->inp_faddr;
 628
 629 out:
 630         TCPDEBUG2(PRU_ACCEPT);
 631         INP_UNLOCK(inp);
 632         if (error == 0)
 633                 *nam = in_sockaddr(port, &addr);
 634         return error;
 635 }
 636
 637 #ifdef INET6
 638 static int
 639 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 640 {
 641         struct inpcb *inp = NULL;
 642         int error = 0;
 643         struct tcpcb *tp = NULL;
 644         struct in_addr addr;
 645         struct in6_addr addr6;
 646         in_port_t port = 0;
 647         int v4 = 0;
 648         TCPDEBUG0;
 649
 650         if (so->so_state & SS_ISDISCONNECTED)
 651                 return (ECONNABORTED);
 652
 653         inp = sotoinpcb(so);
 654         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 655         INP_LOCK(inp);
 656         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 657                 error = EINVAL;
 658                 goto out;
 659         }
 660         tp = intotcpcb(inp);
 661         TCPDEBUG1();
 662
 663         /*
 664          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 665          * copy the data of interest and defer the malloc until after we
 666          * release the lock.
 667          */
 668         if (inp->inp_vflag & INP_IPV4) {
 669                 v4 = 1;
 670                 port = inp->inp_fport;
 671                 addr = inp->inp_faddr;
 672         } else {
 673                 port = inp->inp_fport;
 674                 addr6 = inp->in6p_faddr;
 675         }
 676
 677 out:
 678         TCPDEBUG2(PRU_ACCEPT);
 679         INP_UNLOCK(inp);
 680         if (error == 0) {
 681                 if (v4)
 682                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 683                 else
 684                         *nam = in6_sockaddr(port, &addr6);
 685         }
 686         return error;
 687 }
 688 #endif /* INET6 */
 689
 690 /*
 691  * This is the wrapper function for in_setsockaddr. We just pass down
 692  * the pcbinfo for in_setsockaddr to lock. We don't want to do the locking
 693  * here because in_setsockaddr will call malloc and can block.
 694  */
 695 static int
 696 tcp_sockaddr(struct socket *so, struct sockaddr **nam)
 697 {
 698         return (in_setsockaddr(so, nam, &tcbinfo));
 699 }
 700
 701 /*
 702  * This is the wrapper function for in_setpeeraddr. We just pass down
 703  * the pcbinfo for in_setpeeraddr to lock.
 704  */
 705 static int
 706 tcp_peeraddr(struct socket *so, struct sockaddr **nam)
 707 {
 708         return (in_setpeeraddr(so, nam, &tcbinfo));
 709 }
 710
 711 /*
 712  * Mark the connection as being incapable of further output.
 713  */
 714 static int
 715 tcp_usr_shutdown(struct socket *so)
 716 {
 717         int error = 0;
 718         struct inpcb *inp;
 719         struct tcpcb *tp = NULL;
 720
 721         TCPDEBUG0;
 722         INP_INFO_WLOCK(&tcbinfo);
 723         inp = sotoinpcb(so);
 724         KASSERT(inp != NULL, ("inp == NULL"));
 725         INP_LOCK(inp);
 726         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 727                 error = EINVAL;
 728                 goto out;
 729         }
 730         tp = intotcpcb(inp);
 731         TCPDEBUG1();
 732         socantsendmore(so);
 733         tcp_usrclosed(tp);
 734         error = tcp_output(tp);
 735
 736 out:
 737         TCPDEBUG2(PRU_SHUTDOWN);
 738         INP_UNLOCK(inp);
 739         INP_INFO_WUNLOCK(&tcbinfo);
 740
 741         return (error);
 742 }
 743
 744 /*
 745  * After a receive, possibly send window update to peer.
 746  */
 747 static int
 748 tcp_usr_rcvd(struct socket *so, int flags)
 749 {
 750         struct inpcb *inp;
 751         struct tcpcb *tp = NULL;
 752         int error = 0;
 753
 754         TCPDEBUG0;
 755         inp = sotoinpcb(so);
 756         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 757         INP_LOCK(inp);
 758         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 759                 error = EINVAL;
 760                 goto out;
 761         }
 762         tp = intotcpcb(inp);
 763         TCPDEBUG1();
 764         tcp_output(tp);
 765
 766 out:
 767         TCPDEBUG2(PRU_RCVD);
 768         INP_UNLOCK(inp);
 769         return (error);
 770 }
 771
 772 /*
 773  * Do a send by putting data in output queue and updating urgent
 774  * marker if URG set.  Possibly send more data.  Unlike the other
 775  * pru_*() routines, the mbuf chains are our responsibility.  We
 776  * must either enqueue them or free them.  The other pru_* routines
 777  * generally are caller-frees.
 778  */
 779 static int
 780 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 781              struct sockaddr *nam, struct mbuf *control, struct thread *td)
 782 {
 783         int error = 0;
 784         struct inpcb *inp;
 785         struct tcpcb *tp = NULL;
 786         int headlocked = 0;
 787 #ifdef INET6
 788         int isipv6;
 789 #endif
 790         TCPDEBUG0;
 791
 792         /*
 793          * We require the pcbinfo lock in two cases:
 794          *
 795          * (1) An implied connect is taking place, which can result in
 796          *     binding IPs and ports and hence modification of the pcb hash
 797          *     chains.
 798          *
 799          * (2) PRUS_EOF is set, resulting in explicit close on the send.
 800          */
 801         if ((nam != NULL) || (flags & PRUS_EOF)) {
 802                 INP_INFO_WLOCK(&tcbinfo);
 803                 headlocked = 1;
 804         }
 805         inp = sotoinpcb(so);
 806         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 807         INP_LOCK(inp);
 808         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 809                 error = EINVAL;
 810                 goto out;
 811         }
 812 #ifdef INET6
 813         isipv6 = nam && nam->sa_family == AF_INET6;
 814 #endif /* INET6 */
 815         tp = intotcpcb(inp);
 816         TCPDEBUG1();
 817         if (control) {
 818                 /* TCP doesn't do control messages (rights, creds, etc) */
 819                 if (control->m_len) {
 820                         m_freem(control);
 821                         if (m)
 822                                 m_freem(m);
 823                         error = EINVAL;
 824                         goto out;
 825                 }
 826                 m_freem(control);       /* empty control, just free it */
 827         }
 828         if (!(flags & PRUS_OOB)) {
 829                 sbappendstream(&so->so_snd, m);
 830                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 831                         /*
 832                          * Do implied connect if not yet connected,
 833                          * initialize window to default value, and
 834                          * initialize maxseg/maxopd using peer's cached
 835                          * MSS.
 836                          */
 837                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 838 #ifdef INET6
 839                         if (isipv6)
 840                                 error = tcp6_connect(tp, nam, td);
 841                         else
 842 #endif /* INET6 */
 843                         error = tcp_connect(tp, nam, td);
 844                         if (error)
 845                                 goto out;
 846                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 847                         tcp_mss(tp, -1);
 848                 }
 849                 if (flags & PRUS_EOF) {
 850                         /*
 851                          * Close the send side of the connection after
 852                          * the data is sent.
 853                          */
 854                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 855                         socantsendmore(so);
 856                         tcp_usrclosed(tp);
 857                 }
 858                 if (headlocked) {
 859                         INP_INFO_WUNLOCK(&tcbinfo);
 860                         headlocked = 0;
 861                 }
 862                 if (tp != NULL) {
 863                         if (flags & PRUS_MORETOCOME)
 864                                 tp->t_flags |= TF_MORETOCOME;
 865                         error = tcp_output(tp);
 866                         if (flags & PRUS_MORETOCOME)
 867                                 tp->t_flags &= ~TF_MORETOCOME;
 868                 }
 869         } else {
 870                 /*
 871                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 872                  */
 873                 SOCKBUF_LOCK(&so->so_snd);
 874                 if (sbspace(&so->so_snd) < -512) {
 875                         SOCKBUF_UNLOCK(&so->so_snd);
 876                         m_freem(m);
 877                         error = ENOBUFS;
 878                         goto out;
 879                 }
 880                 /*
 881                  * According to RFC961 (Assigned Protocols),
 882                  * the urgent pointer points to the last octet
 883                  * of urgent data.  We continue, however,
 884                  * to consider it to indicate the first octet
 885                  * of data past the urgent section.
 886                  * Otherwise, snd_up should be one lower.
 887                  */
 888                 sbappendstream_locked(&so->so_snd, m);
 889                 SOCKBUF_UNLOCK(&so->so_snd);
 890                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 891                         /*
 892                          * Do implied connect if not yet connected,
 893                          * initialize window to default value, and
 894                          * initialize maxseg/maxopd using peer's cached
 895                          * MSS.
 896                          */
 897                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 898 #ifdef INET6
 899                         if (isipv6)
 900                                 error = tcp6_connect(tp, nam, td);
 901                         else
 902 #endif /* INET6 */
 903                         error = tcp_connect(tp, nam, td);
 904                         if (error)
 905                                 goto out;
 906                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 907                         tcp_mss(tp, -1);
 908                         INP_INFO_WUNLOCK(&tcbinfo);
 909                         headlocked = 0;
 910                 } else if (nam) {
 911                         INP_INFO_WUNLOCK(&tcbinfo);
 912                         headlocked = 0;
 913                 }
 914                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 915                 tp->t_flags |= TF_FORCEDATA;
 916                 error = tcp_output(tp);
 917                 tp->t_flags &= ~TF_FORCEDATA;
 918         }
 919 out:
 920         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 921                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 922         INP_UNLOCK(inp);
 923         if (headlocked)
 924                 INP_INFO_WUNLOCK(&tcbinfo);
 925         return (error);
 926 }
 927
 928 /*
 929  * Abort the TCP.
 930  *
 931  * First, drop the connection.  Then collect state if possible.
 932  */
 933 static void
 934 tcp_usr_abort(struct socket *so)
 935 {
 936         struct inpcb *inp;
 937         struct tcpcb *tp;
 938         TCPDEBUG0;
 939
 940         inp = sotoinpcb(so);
 941         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
 942
 943         INP_INFO_WLOCK(&tcbinfo);
 944         INP_LOCK(inp);
 945         KASSERT(inp->inp_socket != NULL,
 946             ("tcp_usr_abort: inp_socket == NULL"));
 947         TCPDEBUG1();
 948
 949         /*
 950          * First, if we still have full TCP state, and we're not dropped,
 951          * drop.
 952          */
 953         if (!(inp->inp_vflag & INP_TIMEWAIT) &&
 954             !(inp->inp_vflag & INP_DROPPED)) {
 955                 tp = intotcpcb(inp);
 956                 tcp_drop(tp, ECONNABORTED);
 957         }
 958         tcp_detach(so, inp);
 959         tp = NULL;
 960         TCPDEBUG2(PRU_DETACH);
 961         INP_INFO_WUNLOCK(&tcbinfo);
 962 }
 963
 964 /*
 965  * Receive out-of-band data.
 966  */
 967 static int
 968 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
 969 {
 970         int error = 0;
 971         struct inpcb *inp;
 972         struct tcpcb *tp = NULL;
 973
 974         TCPDEBUG0;
 975         inp = sotoinpcb(so);
 976         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
 977         INP_LOCK(inp);
 978         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 979                 error = EINVAL;
 980                 goto out;
 981         }
 982         tp = intotcpcb(inp);
 983         TCPDEBUG1();
 984         if ((so->so_oobmark == 0 &&
 985              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
 986             so->so_options & SO_OOBINLINE ||
 987             tp->t_oobflags & TCPOOB_HADDATA) {
 988                 error = EINVAL;
 989                 goto out;
 990         }
 991         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 992                 error = EWOULDBLOCK;
 993                 goto out;
 994         }
 995         m->m_len = 1;
 996         *mtod(m, caddr_t) = tp->t_iobc;
 997         if ((flags & MSG_PEEK) == 0)
 998                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 999
1000 out:
1001         TCPDEBUG2(PRU_RCVOOB);
1002         INP_UNLOCK(inp);
1003         return (error);
1004 }
1005
1006 struct pr_usrreqs tcp_usrreqs = {
1007         .pru_abort =            tcp_usr_abort,
1008         .pru_accept =           tcp_usr_accept,
1009         .pru_attach =           tcp_usr_attach,
1010         .pru_bind =             tcp_usr_bind,
1011         .pru_connect =          tcp_usr_connect,
1012         .pru_control =          in_control,
1013         .pru_detach =           tcp_usr_detach,
1014         .pru_disconnect =       tcp_usr_disconnect,
1015         .pru_listen =           tcp_usr_listen,
1016         .pru_peeraddr =         tcp_peeraddr,
1017         .pru_rcvd =             tcp_usr_rcvd,
1018         .pru_rcvoob =           tcp_usr_rcvoob,
1019         .pru_send =             tcp_usr_send,
1020         .pru_shutdown =         tcp_usr_shutdown,
1021         .pru_sockaddr =         tcp_sockaddr,
1022         .pru_sosetlabel =       in_pcbsosetlabel
1023 };
1024
1025 #ifdef INET6
1026 struct pr_usrreqs tcp6_usrreqs = {
1027         .pru_abort =            tcp_usr_abort,
1028         .pru_accept =           tcp6_usr_accept,
1029         .pru_attach =           tcp_usr_attach,
1030         .pru_bind =             tcp6_usr_bind,
1031         .pru_connect =          tcp6_usr_connect,
1032         .pru_control =          in6_control,
1033         .pru_detach =           tcp_usr_detach,
1034         .pru_disconnect =       tcp_usr_disconnect,
1035         .pru_listen =           tcp6_usr_listen,
1036         .pru_peeraddr =         in6_mapped_peeraddr,
1037         .pru_rcvd =             tcp_usr_rcvd,
1038         .pru_rcvoob =           tcp_usr_rcvoob,
1039         .pru_send =             tcp_usr_send,
1040         .pru_shutdown =         tcp_usr_shutdown,
1041         .pru_sockaddr =         in6_mapped_sockaddr,
1042         .pru_sosetlabel =       in_pcbsosetlabel
1043 };
1044 #endif /* INET6 */
1045
1046 /*
1047  * Common subroutine to open a TCP connection to remote host specified
1048  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1049  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1050  * to choose a local host address (interface).  If there is an existing
1051  * incarnation of the same connection in TIME-WAIT state and if the remote
1052  * host was sending CC options and if the connection duration was < MSL, then
1053  * truncate the previous TIME-WAIT state and proceed.
1054  * Initialize connection parameters and enter SYN-SENT state.
1055  */
1056 static int
1057 tcp_connect(tp, nam, td)
1058         register struct tcpcb *tp;
1059         struct sockaddr *nam;
1060         struct thread *td;
1061 {
1062         struct inpcb *inp = tp->t_inpcb, *oinp;
1063         struct socket *so = inp->inp_socket;
1064         struct in_addr laddr;
1065         u_short lport;
1066         int error;
1067
1068         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1069         INP_LOCK_ASSERT(inp);
1070
1071         if (inp->inp_lport == 0) {
1072                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1073                 if (error)
1074                         return error;
1075         }
1076
1077         /*
1078          * Cannot simply call in_pcbconnect, because there might be an
1079          * earlier incarnation of this same connection still in
1080          * TIME_WAIT state, creating an ADDRINUSE error.
1081          */
1082         laddr = inp->inp_laddr;
1083         lport = inp->inp_lport;
1084         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1085             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1086         if (error && oinp == NULL)
1087                 return error;
1088         if (oinp)
1089                 return EADDRINUSE;
1090         inp->inp_laddr = laddr;
1091         in_pcbrehash(inp);
1092
1093         /* Compute window scaling to request.  */
1094         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1095             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1096                 tp->request_r_scale++;
1097
1098         soisconnecting(so);
1099         tcpstat.tcps_connattempt++;
1100         tp->t_state = TCPS_SYN_SENT;
1101         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1102         tp->iss = tcp_new_isn(tp);
1103         tp->t_bw_rtseq = tp->iss;
1104         tcp_sendseqinit(tp);
1105
1106         return 0;
1107 }
1108
1109 #ifdef INET6
1110 static int
1111 tcp6_connect(tp, nam, td)
1112         register struct tcpcb *tp;
1113         struct sockaddr *nam;
1114         struct thread *td;
1115 {
1116         struct inpcb *inp = tp->t_inpcb, *oinp;
1117         struct socket *so = inp->inp_socket;
1118         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1119         struct in6_addr *addr6;
1120         int error;
1121
1122         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1123         INP_LOCK_ASSERT(inp);
1124
1125         if (inp->inp_lport == 0) {
1126                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1127                 if (error)
1128                         return error;
1129         }
1130
1131         /*
1132          * Cannot simply call in_pcbconnect, because there might be an
1133          * earlier incarnation of this same connection still in
1134          * TIME_WAIT state, creating an ADDRINUSE error.
1135          * in6_pcbladdr() also handles scope zone IDs.
1136          */
1137         error = in6_pcbladdr(inp, nam, &addr6);
1138         if (error)
1139                 return error;
1140         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1141                                   &sin6->sin6_addr, sin6->sin6_port,
1142                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1143                                   ? addr6
1144                                   : &inp->in6p_laddr,
1145                                   inp->inp_lport,  0, NULL);
1146         if (oinp)
1147                 return EADDRINUSE;
1148         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1149                 inp->in6p_laddr = *addr6;
1150         inp->in6p_faddr = sin6->sin6_addr;
1151         inp->inp_fport = sin6->sin6_port;
1152         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1153         inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
1154         if (inp->in6p_flags & IN6P_AUTOFLOWLABEL)
1155                 inp->in6p_flowinfo |=
1156                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1157         in_pcbrehash(inp);
1158
1159         /* Compute window scaling to request.  */
1160         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1161             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1162                 tp->request_r_scale++;
1163
1164         soisconnecting(so);
1165         tcpstat.tcps_connattempt++;
1166         tp->t_state = TCPS_SYN_SENT;
1167         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1168         tp->iss = tcp_new_isn(tp);
1169         tp->t_bw_rtseq = tp->iss;
1170         tcp_sendseqinit(tp);
1171
1172         return 0;
1173 }
1174 #endif /* INET6 */
1175
1176 /*
1177  * Export TCP internal state information via a struct tcp_info, based on the
1178  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1179  * (TCP state machine, etc).  We export all information using FreeBSD-native
1180  * constants -- for example, the numeric values for tcpi_state will differ
1181  * from Linux.
1182  */
1183 static void
1184 tcp_fill_info(tp, ti)
1185         struct tcpcb *tp;
1186         struct tcp_info *ti;
1187 {
1188
1189         INP_LOCK_ASSERT(tp->t_inpcb);
1190         bzero(ti, sizeof(*ti));
1191
1192         ti->tcpi_state = tp->t_state;
1193         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1194                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1195         if (tp->sack_enable)
1196                 ti->tcpi_options |= TCPI_OPT_SACK;
1197         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1198                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1199                 ti->tcpi_snd_wscale = tp->snd_scale;
1200                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1201         }
1202         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1203         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1204
1205         /*
1206          * FreeBSD-specific extension fields for tcp_info.
1207          */
1208         ti->tcpi_rcv_space = tp->rcv_wnd;
1209         ti->tcpi_snd_wnd = tp->snd_wnd;
1210         ti->tcpi_snd_bwnd = tp->snd_bwnd;
1211 }
1212
1213 /*
1214  * The new sockopt interface makes it possible for us to block in the
1215  * copyin/out step (if we take a page fault).  Taking a page fault at
1216  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
1217  * use TSM, there probably isn't any need for this function to run at
1218  * splnet() any more.  This needs more examination.)
1219  *
1220  * XXXRW: The locking here is wrong; we may take a page fault while holding
1221  * the inpcb lock.
1222  */
1223 int
1224 tcp_ctloutput(so, sopt)
1225         struct socket *so;
1226         struct sockopt *sopt;
1227 {
1228         int     error, opt, optval;
1229         struct  inpcb *inp;
1230         struct  tcpcb *tp;
1231         struct  tcp_info ti;
1232
1233         error = 0;
1234         inp = sotoinpcb(so);
1235         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1236         INP_LOCK(inp);
1237         if (sopt->sopt_level != IPPROTO_TCP) {
1238                 INP_UNLOCK(inp);
1239 #ifdef INET6
1240                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1241                         error = ip6_ctloutput(so, sopt);
1242                 else
1243 #endif /* INET6 */
1244                 error = ip_ctloutput(so, sopt);
1245                 return (error);
1246         }
1247         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
1248                 error = ECONNRESET;
1249                 goto out;
1250         }
1251         tp = intotcpcb(inp);
1252
1253         switch (sopt->sopt_dir) {
1254         case SOPT_SET:
1255                 switch (sopt->sopt_name) {
1256 #ifdef TCP_SIGNATURE
1257                 case TCP_MD5SIG:
1258                         error = sooptcopyin(sopt, &optval, sizeof optval,
1259                                             sizeof optval);
1260                         if (error)
1261                                 break;
1262
1263                         if (optval > 0)
1264                                 tp->t_flags |= TF_SIGNATURE;
1265                         else
1266                                 tp->t_flags &= ~TF_SIGNATURE;
1267                         break;
1268 #endif /* TCP_SIGNATURE */
1269                 case TCP_NODELAY:
1270                 case TCP_NOOPT:
1271                         error = sooptcopyin(sopt, &optval, sizeof optval,
1272                                             sizeof optval);
1273                         if (error)
1274                                 break;
1275
1276                         switch (sopt->sopt_name) {
1277                         case TCP_NODELAY:
1278                                 opt = TF_NODELAY;
1279                                 break;
1280                         case TCP_NOOPT:
1281                                 opt = TF_NOOPT;
1282                                 break;
1283                         default:
1284                                 opt = 0; /* dead code to fool gcc */
1285                                 break;
1286                         }
1287
1288                         if (optval)
1289                                 tp->t_flags |= opt;
1290                         else
1291                                 tp->t_flags &= ~opt;
1292                         break;
1293
1294                 case TCP_NOPUSH:
1295                         error = sooptcopyin(sopt, &optval, sizeof optval,
1296                                             sizeof optval);
1297                         if (error)
1298                                 break;
1299
1300                         if (optval)
1301                                 tp->t_flags |= TF_NOPUSH;
1302                         else {
1303                                 tp->t_flags &= ~TF_NOPUSH;
1304                                 error = tcp_output(tp);
1305                         }
1306                         break;
1307
1308                 case TCP_MAXSEG:
1309                         error = sooptcopyin(sopt, &optval, sizeof optval,
1310                                             sizeof optval);
1311                         if (error)
1312                                 break;
1313
1314                         if (optval > 0 && optval <= tp->t_maxseg &&
1315                             optval + 40 >= tcp_minmss)
1316                                 tp->t_maxseg = optval;
1317                         else
1318                                 error = EINVAL;
1319                         break;
1320
1321                 case TCP_INFO:
1322                         error = EINVAL;
1323                         break;
1324
1325                 default:
1326                         error = ENOPROTOOPT;
1327                         break;
1328                 }
1329                 break;
1330
1331         case SOPT_GET:
1332                 switch (sopt->sopt_name) {
1333 #ifdef TCP_SIGNATURE
1334                 case TCP_MD5SIG:
1335                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1336                         error = sooptcopyout(sopt, &optval, sizeof optval);
1337                         break;
1338 #endif
1339                 case TCP_NODELAY:
1340                         optval = tp->t_flags & TF_NODELAY;
1341                         error = sooptcopyout(sopt, &optval, sizeof optval);
1342                         break;
1343                 case TCP_MAXSEG:
1344                         optval = tp->t_maxseg;
1345                         error = sooptcopyout(sopt, &optval, sizeof optval);
1346                         break;
1347                 case TCP_NOOPT:
1348                         optval = tp->t_flags & TF_NOOPT;
1349                         error = sooptcopyout(sopt, &optval, sizeof optval);
1350                         break;
1351                 case TCP_NOPUSH:
1352                         optval = tp->t_flags & TF_NOPUSH;
1353                         error = sooptcopyout(sopt, &optval, sizeof optval);
1354                         break;
1355                 case TCP_INFO:
1356                         tcp_fill_info(tp, &ti);
1357                         error = sooptcopyout(sopt, &ti, sizeof ti);
1358                         break;
1359                 default:
1360                         error = ENOPROTOOPT;
1361                         break;
1362                 }
1363                 break;
1364         }
1365 out:
1366         INP_UNLOCK(inp);
1367         return (error);
1368 }
1369
1370 /*
1371  * tcp_sendspace and tcp_recvspace are the default send and receive window
1372  * sizes, respectively.  These are obsolescent (this information should
1373  * be set by the route).
1374  */
1375 u_long  tcp_sendspace = 1024*32;
1376 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1377     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1378 u_long  tcp_recvspace = 1024*64;
1379 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1380     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1381
1382 /*
1383  * Attach TCP protocol to socket, allocating
1384  * internet protocol control block, tcp control block,
1385  * bufer space, and entering LISTEN state if to accept connections.
1386  */
1387 static int
1388 tcp_attach(so)
1389         struct socket *so;
1390 {
1391         register struct tcpcb *tp;
1392         struct inpcb *inp;
1393         int error;
1394 #ifdef INET6
1395         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1396 #endif
1397
1398         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1399                 error = soreserve(so, tcp_sendspace, tcp_recvspace);
1400                 if (error)
1401                         return (error);
1402         }
1403         INP_INFO_WLOCK(&tcbinfo);
1404         error = in_pcballoc(so, &tcbinfo);
1405         if (error) {
1406                 INP_INFO_WUNLOCK(&tcbinfo);
1407                 return (error);
1408         }
1409         inp = sotoinpcb(so);
1410 #ifdef INET6
1411         if (isipv6) {
1412                 inp->inp_vflag |= INP_IPV6;
1413                 inp->in6p_hops = -1;    /* use kernel default */
1414         }
1415         else
1416 #endif
1417         inp->inp_vflag |= INP_IPV4;
1418         tp = tcp_newtcpcb(inp);
1419         if (tp == NULL) {
1420 #ifdef INET6
1421                 if (isipv6) {
1422                         in6_pcbdetach(inp);
1423                         in6_pcbfree(inp);
1424                 } else {
1425 #endif
1426                         in_pcbdetach(inp);
1427                         in_pcbfree(inp);
1428 #ifdef INET6
1429                 }
1430 #endif
1431                 INP_INFO_WUNLOCK(&tcbinfo);
1432                 return (ENOBUFS);
1433         }
1434         tp->t_state = TCPS_CLOSED;
1435         INP_UNLOCK(inp);
1436         INP_INFO_WUNLOCK(&tcbinfo);
1437         return (0);
1438 }
1439
1440 /*
1441  * Initiate (or continue) disconnect.
1442  * If embryonic state, just send reset (once).
1443  * If in ``let data drain'' option and linger null, just drop.
1444  * Otherwise (hard), mark socket disconnecting and drop
1445  * current input data; switch states based on user close, and
1446  * send segment to peer (with FIN).
1447  */
1448 static void
1449 tcp_disconnect(tp)
1450         register struct tcpcb *tp;
1451 {
1452         struct inpcb *inp = tp->t_inpcb;
1453         struct socket *so = inp->inp_socket;
1454
1455         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1456         INP_LOCK_ASSERT(inp);
1457
1458         /*
1459          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1460          * socket is still open.
1461          */
1462         if (tp->t_state < TCPS_ESTABLISHED) {
1463                 tp = tcp_close(tp);
1464                 KASSERT(tp != NULL,
1465                     ("tcp_disconnect: tcp_close() returned NULL"));
1466         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1467                 tp = tcp_drop(tp, 0);
1468                 KASSERT(tp != NULL,
1469                     ("tcp_disconnect: tcp_drop() returned NULL"));
1470         } else {
1471                 soisdisconnecting(so);
1472                 sbflush(&so->so_rcv);
1473                 tcp_usrclosed(tp);
1474                 if (!(inp->inp_vflag & INP_DROPPED))
1475                         tcp_output(tp);
1476         }
1477 }
1478
1479 /*
1480  * User issued close, and wish to trail through shutdown states:
1481  * if never received SYN, just forget it.  If got a SYN from peer,
1482  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1483  * If already got a FIN from peer, then almost done; go to LAST_ACK
1484  * state.  In all other cases, have already sent FIN to peer (e.g.
1485  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1486  * for peer to send FIN or not respond to keep-alives, etc.
1487  * We can let the user exit from the close as soon as the FIN is acked.
1488  */
1489 static void
1490 tcp_usrclosed(tp)
1491         register struct tcpcb *tp;
1492 {
1493
1494         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1495         INP_LOCK_ASSERT(tp->t_inpcb);
1496
1497         switch (tp->t_state) {
1498
1499         case TCPS_CLOSED:
1500         case TCPS_LISTEN:
1501                 tp->t_state = TCPS_CLOSED;
1502                 tp = tcp_close(tp);
1503                 /*
1504                  * tcp_close() should never return NULL here as the socket is
1505                  * still open.
1506                  */
1507                 KASSERT(tp != NULL,
1508                     ("tcp_usrclosed: tcp_close() returned NULL"));
1509                 break;
1510
1511         case TCPS_SYN_SENT:
1512         case TCPS_SYN_RECEIVED:
1513                 tp->t_flags |= TF_NEEDFIN;
1514                 break;
1515
1516         case TCPS_ESTABLISHED:
1517                 tp->t_state = TCPS_FIN_WAIT_1;
1518                 break;
1519
1520         case TCPS_CLOSE_WAIT:
1521                 tp->t_state = TCPS_LAST_ACK;
1522                 break;
1523         }
1524         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1525                 soisdisconnected(tp->t_inpcb->inp_socket);
1526                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1527                 if (tp->t_state == TCPS_FIN_WAIT_2)
1528                         callout_reset(tp->tt_2msl, tcp_maxidle,
1529                                       tcp_timer_2msl, tp);
1530         }
1531 }