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 is called when the socket layer loses its final reference
 141  * to the socket, be it a file descriptor reference, a reference from TCP,
 142  * etc.  At this point, there is only one case in which we will keep around
 143  * inpcb state: time wait.
 144  *
 145  * This function can probably be re-absorbed back into tcp_usr_detach() now
 146  * that there is a single detach path.
 147  */
 148 static void
 149 tcp_detach(struct socket *so, struct inpcb *inp)
 150 {
 151         struct tcpcb *tp;
 152 #ifdef INET6
 153         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
 154 #endif
 155
 156         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 157         INP_LOCK_ASSERT(inp);
 158
 159         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 160         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 161
 162         tp = intotcpcb(inp);
 163
 164         if (inp->inp_vflag & INP_TIMEWAIT) {
 165                 /*
 166                  * There are two cases to handle: one in which the time wait
 167                  * state is being discarded (INP_DROPPED), and one in which
 168                  * this connection will remain in timewait.  In the former,
 169                  * it is time to discard all state (except tcptw, which has
 170                  * already been discarded by the timewait close code, which
 171                  * should be further up the call stack somewhere).  In the
 172                  * latter case, we detach from the socket, but leave the pcb
 173                  * present until timewait ends.
 174                  *
 175                  * XXXRW: Would it be cleaner to free the tcptw here?
 176                  */
 177                 if (inp->inp_vflag & INP_DROPPED) {
 178                         KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
 179                             "INP_DROPPED && tp != NULL"));
 180 #ifdef INET6
 181                         if (isipv6) {
 182                                 in6_pcbdetach(inp);
 183                                 in6_pcbfree(inp);
 184                         } else {
 185 #endif
 186                                 in_pcbdetach(inp);
 187                                 in_pcbfree(inp);
 188 #ifdef INET6
 189                         }
 190 #endif
 191                 } else {
 192 #ifdef INET6
 193                         if (isipv6)
 194                                 in6_pcbdetach(inp);
 195                         else
 196 #endif
 197                                 in_pcbdetach(inp);
 198                         INP_UNLOCK(inp);
 199                 }
 200         } else {
 201                 /*
 202                  * If the connection is not in timewait, we consider two
 203                  * two conditions: one in which no further processing is
 204                  * necessary (dropped || embryonic), and one in which TCP is
 205                  * not yet done, but no longer requires the socket, so the
 206                  * pcb will persist for the time being.
 207                  *
 208                  * XXXRW: Does the second case still occur?
 209                  */
 210                 if (inp->inp_vflag & INP_DROPPED ||
 211                     tp->t_state < TCPS_SYN_SENT) {
 212                         tcp_discardcb(tp);
 213 #ifdef INET6
 214                         if (isipv6) {
 215                                 in6_pcbdetach(inp);
 216                                 in6_pcbfree(inp);
 217                         } else {
 218 #endif
 219                                 in_pcbdetach(inp);
 220                                 in_pcbfree(inp);
 221 #ifdef INET6
 222                         }
 223 #endif
 224                 } else {
 225 #ifdef INET6
 226                         if (isipv6)
 227                                 in6_pcbdetach(inp);
 228                         else
 229 #endif
 230                                 in_pcbdetach(inp);
 231                 }
 232         }
 233 }
 234
 235 /*
 236  * pru_detach() detaches the TCP protocol from the socket.
 237  * If the protocol state is non-embryonic, then can't
 238  * do this directly: have to initiate a pru_disconnect(),
 239  * which may finish later; embryonic TCB's can just
 240  * be discarded here.
 241  */
 242 static void
 243 tcp_usr_detach(struct socket *so)
 244 {
 245         struct inpcb *inp;
 246         struct tcpcb *tp;
 247         TCPDEBUG0;
 248
 249         inp = sotoinpcb(so);
 250         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 251         INP_INFO_WLOCK(&tcbinfo);
 252         INP_LOCK(inp);
 253         KASSERT(inp->inp_socket != NULL,
 254             ("tcp_usr_detach: inp_socket == NULL"));
 255         TCPDEBUG1();
 256
 257         tcp_detach(so, inp);
 258         tp = NULL;
 259         TCPDEBUG2(PRU_DETACH);
 260         INP_INFO_WUNLOCK(&tcbinfo);
 261 }
 262
 263 /*
 264  * Give the socket an address.
 265  */
 266 static int
 267 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 268 {
 269         int error = 0;
 270         struct inpcb *inp;
 271         struct tcpcb *tp = NULL;
 272         struct sockaddr_in *sinp;
 273
 274         sinp = (struct sockaddr_in *)nam;
 275         if (nam->sa_len != sizeof (*sinp))
 276                 return (EINVAL);
 277         /*
 278          * Must check for multicast addresses and disallow binding
 279          * to them.
 280          */
 281         if (sinp->sin_family == AF_INET &&
 282             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 283                 return (EAFNOSUPPORT);
 284
 285         TCPDEBUG0;
 286         INP_INFO_WLOCK(&tcbinfo);
 287         inp = sotoinpcb(so);
 288         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 289         INP_LOCK(inp);
 290         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 291                 error = EINVAL;
 292                 goto out;
 293         }
 294         tp = intotcpcb(inp);
 295         TCPDEBUG1();
 296         error = in_pcbbind(inp, nam, td->td_ucred);
 297 out:
 298         TCPDEBUG2(PRU_BIND);
 299         INP_UNLOCK(inp);
 300         INP_INFO_WUNLOCK(&tcbinfo);
 301
 302         return (error);
 303 }
 304
 305 #ifdef INET6
 306 static int
 307 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 308 {
 309         int error = 0;
 310         struct inpcb *inp;
 311         struct tcpcb *tp = NULL;
 312         struct sockaddr_in6 *sin6p;
 313
 314         sin6p = (struct sockaddr_in6 *)nam;
 315         if (nam->sa_len != sizeof (*sin6p))
 316                 return (EINVAL);
 317         /*
 318          * Must check for multicast addresses and disallow binding
 319          * to them.
 320          */
 321         if (sin6p->sin6_family == AF_INET6 &&
 322             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 323                 return (EAFNOSUPPORT);
 324
 325         TCPDEBUG0;
 326         INP_INFO_WLOCK(&tcbinfo);
 327         inp = sotoinpcb(so);
 328         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 329         INP_LOCK(inp);
 330         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 331                 error = EINVAL;
 332                 goto out;
 333         }
 334         tp = intotcpcb(inp);
 335         TCPDEBUG1();
 336         inp->inp_vflag &= ~INP_IPV4;
 337         inp->inp_vflag |= INP_IPV6;
 338         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 339                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 340                         inp->inp_vflag |= INP_IPV4;
 341                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 342                         struct sockaddr_in sin;
 343
 344                         in6_sin6_2_sin(&sin, sin6p);
 345                         inp->inp_vflag |= INP_IPV4;
 346                         inp->inp_vflag &= ~INP_IPV6;
 347                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 348                             td->td_ucred);
 349                         goto out;
 350                 }
 351         }
 352         error = in6_pcbbind(inp, nam, td->td_ucred);
 353 out:
 354         TCPDEBUG2(PRU_BIND);
 355         INP_UNLOCK(inp);
 356         INP_INFO_WUNLOCK(&tcbinfo);
 357         return (error);
 358 }
 359 #endif /* INET6 */
 360
 361 /*
 362  * Prepare to accept connections.
 363  */
 364 static int
 365 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 366 {
 367         int error = 0;
 368         struct inpcb *inp;
 369         struct tcpcb *tp = NULL;
 370
 371         TCPDEBUG0;
 372         INP_INFO_WLOCK(&tcbinfo);
 373         inp = sotoinpcb(so);
 374         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 375         INP_LOCK(inp);
 376         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 377                 error = EINVAL;
 378                 goto out;
 379         }
 380         tp = intotcpcb(inp);
 381         TCPDEBUG1();
 382         SOCK_LOCK(so);
 383         error = solisten_proto_check(so);
 384         if (error == 0 && inp->inp_lport == 0)
 385                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 386         if (error == 0) {
 387                 tp->t_state = TCPS_LISTEN;
 388                 solisten_proto(so, backlog);
 389         }
 390         SOCK_UNLOCK(so);
 391
 392 out:
 393         TCPDEBUG2(PRU_LISTEN);
 394         INP_UNLOCK(inp);
 395         INP_INFO_WUNLOCK(&tcbinfo);
 396         return (error);
 397 }
 398
 399 #ifdef INET6
 400 static int
 401 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 402 {
 403         int error = 0;
 404         struct inpcb *inp;
 405         struct tcpcb *tp = NULL;
 406
 407         TCPDEBUG0;
 408         INP_INFO_WLOCK(&tcbinfo);
 409         inp = sotoinpcb(so);
 410         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 411         INP_LOCK(inp);
 412         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 413                 error = EINVAL;
 414                 goto out;
 415         }
 416         tp = intotcpcb(inp);
 417         TCPDEBUG1();
 418         SOCK_LOCK(so);
 419         error = solisten_proto_check(so);
 420         if (error == 0 && inp->inp_lport == 0) {
 421                 inp->inp_vflag &= ~INP_IPV4;
 422                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 423                         inp->inp_vflag |= INP_IPV4;
 424                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 425         }
 426         if (error == 0) {
 427                 tp->t_state = TCPS_LISTEN;
 428                 solisten_proto(so, backlog);
 429         }
 430         SOCK_UNLOCK(so);
 431
 432 out:
 433         TCPDEBUG2(PRU_LISTEN);
 434         INP_UNLOCK(inp);
 435         INP_INFO_WUNLOCK(&tcbinfo);
 436         return (error);
 437 }
 438 #endif /* INET6 */
 439
 440 /*
 441  * Initiate connection to peer.
 442  * Create a template for use in transmissions on this connection.
 443  * Enter SYN_SENT state, and mark socket as connecting.
 444  * Start keep-alive timer, and seed output sequence space.
 445  * Send initial segment on connection.
 446  */
 447 static int
 448 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 449 {
 450         int error = 0;
 451         struct inpcb *inp;
 452         struct tcpcb *tp = NULL;
 453         struct sockaddr_in *sinp;
 454
 455         sinp = (struct sockaddr_in *)nam;
 456         if (nam->sa_len != sizeof (*sinp))
 457                 return (EINVAL);
 458         /*
 459          * Must disallow TCP ``connections'' to multicast addresses.
 460          */
 461         if (sinp->sin_family == AF_INET
 462             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 463                 return (EAFNOSUPPORT);
 464         if (jailed(td->td_ucred))
 465                 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr);
 466
 467         TCPDEBUG0;
 468         INP_INFO_WLOCK(&tcbinfo);
 469         inp = sotoinpcb(so);
 470         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 471         INP_LOCK(inp);
 472         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 473                 error = EINVAL;
 474                 goto out;
 475         }
 476         tp = intotcpcb(inp);
 477         TCPDEBUG1();
 478         if ((error = tcp_connect(tp, nam, td)) != 0)
 479                 goto out;
 480         error = tcp_output(tp);
 481 out:
 482         TCPDEBUG2(PRU_CONNECT);
 483         INP_UNLOCK(inp);
 484         INP_INFO_WUNLOCK(&tcbinfo);
 485         return (error);
 486 }
 487
 488 #ifdef INET6
 489 static int
 490 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 491 {
 492         int error = 0;
 493         struct inpcb *inp;
 494         struct tcpcb *tp = NULL;
 495         struct sockaddr_in6 *sin6p;
 496
 497         TCPDEBUG0;
 498
 499         sin6p = (struct sockaddr_in6 *)nam;
 500         if (nam->sa_len != sizeof (*sin6p))
 501                 return (EINVAL);
 502         /*
 503          * Must disallow TCP ``connections'' to multicast addresses.
 504          */
 505         if (sin6p->sin6_family == AF_INET6
 506             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 507                 return (EAFNOSUPPORT);
 508
 509         INP_INFO_WLOCK(&tcbinfo);
 510         inp = sotoinpcb(so);
 511         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 512         INP_LOCK(inp);
 513         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 514                 error = EINVAL;
 515                 goto out;
 516         }
 517         tp = intotcpcb(inp);
 518         TCPDEBUG1();
 519         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 520                 struct sockaddr_in sin;
 521
 522                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 523                         error = EINVAL;
 524                         goto out;
 525                 }
 526
 527                 in6_sin6_2_sin(&sin, sin6p);
 528                 inp->inp_vflag |= INP_IPV4;
 529                 inp->inp_vflag &= ~INP_IPV6;
 530                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 531                         goto out;
 532                 error = tcp_output(tp);
 533                 goto out;
 534         }
 535         inp->inp_vflag &= ~INP_IPV4;
 536         inp->inp_vflag |= INP_IPV6;
 537         inp->inp_inc.inc_isipv6 = 1;
 538         if ((error = tcp6_connect(tp, nam, td)) != 0)
 539                 goto out;
 540         error = tcp_output(tp);
 541
 542 out:
 543         TCPDEBUG2(PRU_CONNECT);
 544         INP_UNLOCK(inp);
 545         INP_INFO_WUNLOCK(&tcbinfo);
 546         return (error);
 547 }
 548 #endif /* INET6 */
 549
 550 /*
 551  * Initiate disconnect from peer.
 552  * If connection never passed embryonic stage, just drop;
 553  * else if don't need to let data drain, then can just drop anyways,
 554  * else have to begin TCP shutdown process: mark socket disconnecting,
 555  * drain unread data, state switch to reflect user close, and
 556  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 557  * when peer sends FIN and acks ours.
 558  *
 559  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 560  */
 561 static int
 562 tcp_usr_disconnect(struct socket *so)
 563 {
 564         struct inpcb *inp;
 565         struct tcpcb *tp = NULL;
 566         int error = 0;
 567
 568         TCPDEBUG0;
 569         INP_INFO_WLOCK(&tcbinfo);
 570         inp = sotoinpcb(so);
 571         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 572         INP_LOCK(inp);
 573         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 574                 error = EINVAL;
 575                 goto out;
 576         }
 577         tp = intotcpcb(inp);
 578         TCPDEBUG1();
 579         tcp_disconnect(tp);
 580 out:
 581         TCPDEBUG2(PRU_DISCONNECT);
 582         INP_UNLOCK(inp);
 583         INP_INFO_WUNLOCK(&tcbinfo);
 584         return (error);
 585 }
 586
 587 /*
 588  * Accept a connection.  Essentially all the work is
 589  * done at higher levels; just return the address
 590  * of the peer, storing through addr.
 591  */
 592 static int
 593 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 594 {
 595         int error = 0;
 596         struct inpcb *inp = NULL;
 597         struct tcpcb *tp = NULL;
 598         struct in_addr addr;
 599         in_port_t port = 0;
 600         TCPDEBUG0;
 601
 602         if (so->so_state & SS_ISDISCONNECTED)
 603                 return (ECONNABORTED);
 604
 605         inp = sotoinpcb(so);
 606         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 607         INP_LOCK(inp);
 608         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 609                 error = ECONNABORTED;
 610                 goto out;
 611         }
 612         tp = intotcpcb(inp);
 613         TCPDEBUG1();
 614
 615         /*
 616          * We inline in_setpeeraddr and COMMON_END here, so that we can
 617          * copy the data of interest and defer the malloc until after we
 618          * release the lock.
 619          */
 620         port = inp->inp_fport;
 621         addr = inp->inp_faddr;
 622
 623 out:
 624         TCPDEBUG2(PRU_ACCEPT);
 625         INP_UNLOCK(inp);
 626         if (error == 0)
 627                 *nam = in_sockaddr(port, &addr);
 628         return error;
 629 }
 630
 631 #ifdef INET6
 632 static int
 633 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 634 {
 635         struct inpcb *inp = NULL;
 636         int error = 0;
 637         struct tcpcb *tp = NULL;
 638         struct in_addr addr;
 639         struct in6_addr addr6;
 640         in_port_t port = 0;
 641         int v4 = 0;
 642         TCPDEBUG0;
 643
 644         if (so->so_state & SS_ISDISCONNECTED)
 645                 return (ECONNABORTED);
 646
 647         inp = sotoinpcb(so);
 648         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 649         INP_LOCK(inp);
 650         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 651                 error = EINVAL;
 652                 goto out;
 653         }
 654         tp = intotcpcb(inp);
 655         TCPDEBUG1();
 656
 657         /*
 658          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 659          * copy the data of interest and defer the malloc until after we
 660          * release the lock.
 661          */
 662         if (inp->inp_vflag & INP_IPV4) {
 663                 v4 = 1;
 664                 port = inp->inp_fport;
 665                 addr = inp->inp_faddr;
 666         } else {
 667                 port = inp->inp_fport;
 668                 addr6 = inp->in6p_faddr;
 669         }
 670
 671 out:
 672         TCPDEBUG2(PRU_ACCEPT);
 673         INP_UNLOCK(inp);
 674         if (error == 0) {
 675                 if (v4)
 676                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 677                 else
 678                         *nam = in6_sockaddr(port, &addr6);
 679         }
 680         return error;
 681 }
 682 #endif /* INET6 */
 683
 684 /*
 685  * This is the wrapper function for in_setsockaddr. We just pass down
 686  * the pcbinfo for in_setsockaddr to lock. We don't want to do the locking
 687  * here because in_setsockaddr will call malloc and can block.
 688  */
 689 static int
 690 tcp_sockaddr(struct socket *so, struct sockaddr **nam)
 691 {
 692         return (in_setsockaddr(so, nam, &tcbinfo));
 693 }
 694
 695 /*
 696  * This is the wrapper function for in_setpeeraddr. We just pass down
 697  * the pcbinfo for in_setpeeraddr to lock.
 698  */
 699 static int
 700 tcp_peeraddr(struct socket *so, struct sockaddr **nam)
 701 {
 702         return (in_setpeeraddr(so, nam, &tcbinfo));
 703 }
 704
 705 /*
 706  * Mark the connection as being incapable of further output.
 707  */
 708 static int
 709 tcp_usr_shutdown(struct socket *so)
 710 {
 711         int error = 0;
 712         struct inpcb *inp;
 713         struct tcpcb *tp = NULL;
 714
 715         TCPDEBUG0;
 716         INP_INFO_WLOCK(&tcbinfo);
 717         inp = sotoinpcb(so);
 718         KASSERT(inp != NULL, ("inp == NULL"));
 719         INP_LOCK(inp);
 720         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 721                 error = EINVAL;
 722                 goto out;
 723         }
 724         tp = intotcpcb(inp);
 725         TCPDEBUG1();
 726         socantsendmore(so);
 727         tcp_usrclosed(tp);
 728         error = tcp_output(tp);
 729
 730 out:
 731         TCPDEBUG2(PRU_SHUTDOWN);
 732         INP_UNLOCK(inp);
 733         INP_INFO_WUNLOCK(&tcbinfo);
 734
 735         return (error);
 736 }
 737
 738 /*
 739  * After a receive, possibly send window update to peer.
 740  */
 741 static int
 742 tcp_usr_rcvd(struct socket *so, int flags)
 743 {
 744         struct inpcb *inp;
 745         struct tcpcb *tp = NULL;
 746         int error = 0;
 747
 748         TCPDEBUG0;
 749         inp = sotoinpcb(so);
 750         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 751         INP_LOCK(inp);
 752         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 753                 error = EINVAL;
 754                 goto out;
 755         }
 756         tp = intotcpcb(inp);
 757         TCPDEBUG1();
 758         tcp_output(tp);
 759
 760 out:
 761         TCPDEBUG2(PRU_RCVD);
 762         INP_UNLOCK(inp);
 763         return (error);
 764 }
 765
 766 /*
 767  * Do a send by putting data in output queue and updating urgent
 768  * marker if URG set.  Possibly send more data.  Unlike the other
 769  * pru_*() routines, the mbuf chains are our responsibility.  We
 770  * must either enqueue them or free them.  The other pru_* routines
 771  * generally are caller-frees.
 772  */
 773 static int
 774 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 775              struct sockaddr *nam, struct mbuf *control, struct thread *td)
 776 {
 777         int error = 0;
 778         struct inpcb *inp;
 779         struct tcpcb *tp = NULL;
 780         int headlocked = 0;
 781 #ifdef INET6
 782         int isipv6;
 783 #endif
 784         TCPDEBUG0;
 785
 786         /*
 787          * We require the pcbinfo lock in two cases:
 788          *
 789          * (1) An implied connect is taking place, which can result in
 790          *     binding IPs and ports and hence modification of the pcb hash
 791          *     chains.
 792          *
 793          * (2) PRUS_EOF is set, resulting in explicit close on the send.
 794          */
 795         if ((nam != NULL) || (flags & PRUS_EOF)) {
 796                 INP_INFO_WLOCK(&tcbinfo);
 797                 headlocked = 1;
 798         }
 799         inp = sotoinpcb(so);
 800         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 801         INP_LOCK(inp);
 802         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
 803                 error = EINVAL;
 804                 goto out;
 805         }
 806 #ifdef INET6
 807         isipv6 = nam && nam->sa_family == AF_INET6;
 808 #endif /* INET6 */
 809         tp = intotcpcb(inp);
 810         TCPDEBUG1();
 811         if (control) {
 812                 /* TCP doesn't do control messages (rights, creds, etc) */
 813                 if (control->m_len) {
 814                         m_freem(control);
 815                         if (m)
 816                                 m_freem(m);
 817                         error = EINVAL;
 818                         goto out;
 819                 }
 820                 m_freem(control);       /* empty control, just free it */
 821         }
 822         if (!(flags & PRUS_OOB)) {
 823                 sbappendstream(&so->so_snd, m);
 824                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 825                         /*
 826                          * Do implied connect if not yet connected,
 827                          * initialize window to default value, and
 828                          * initialize maxseg/maxopd using peer's cached
 829                          * MSS.
 830                          */
 831                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 832 #ifdef INET6
 833                         if (isipv6)
 834                                 error = tcp6_connect(tp, nam, td);
 835                         else
 836 #endif /* INET6 */
 837                         error = tcp_connect(tp, nam, td);
 838                         if (error)
 839                                 goto out;
 840                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 841                         tcp_mss(tp, -1);
 842                 }
 843                 if (flags & PRUS_EOF) {
 844                         /*
 845                          * Close the send side of the connection after
 846                          * the data is sent.
 847                          */
 848                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 849                         socantsendmore(so);
 850                         tcp_usrclosed(tp);
 851                 }
 852                 if (headlocked) {
 853                         INP_INFO_WUNLOCK(&tcbinfo);
 854                         headlocked = 0;
 855                 }
 856                 if (tp != NULL) {
 857                         if (flags & PRUS_MORETOCOME)
 858                                 tp->t_flags |= TF_MORETOCOME;
 859                         error = tcp_output(tp);
 860                         if (flags & PRUS_MORETOCOME)
 861                                 tp->t_flags &= ~TF_MORETOCOME;
 862                 }
 863         } else {
 864                 /*
 865                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 866                  */
 867                 SOCKBUF_LOCK(&so->so_snd);
 868                 if (sbspace(&so->so_snd) < -512) {
 869                         SOCKBUF_UNLOCK(&so->so_snd);
 870                         m_freem(m);
 871                         error = ENOBUFS;
 872                         goto out;
 873                 }
 874                 /*
 875                  * According to RFC961 (Assigned Protocols),
 876                  * the urgent pointer points to the last octet
 877                  * of urgent data.  We continue, however,
 878                  * to consider it to indicate the first octet
 879                  * of data past the urgent section.
 880                  * Otherwise, snd_up should be one lower.
 881                  */
 882                 sbappendstream_locked(&so->so_snd, m);
 883                 SOCKBUF_UNLOCK(&so->so_snd);
 884                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 885                         /*
 886                          * Do implied connect if not yet connected,
 887                          * initialize window to default value, and
 888                          * initialize maxseg/maxopd using peer's cached
 889                          * MSS.
 890                          */
 891                         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 892 #ifdef INET6
 893                         if (isipv6)
 894                                 error = tcp6_connect(tp, nam, td);
 895                         else
 896 #endif /* INET6 */
 897                         error = tcp_connect(tp, nam, td);
 898                         if (error)
 899                                 goto out;
 900                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 901                         tcp_mss(tp, -1);
 902                         INP_INFO_WUNLOCK(&tcbinfo);
 903                         headlocked = 0;
 904                 } else if (nam) {
 905                         INP_INFO_WUNLOCK(&tcbinfo);
 906                         headlocked = 0;
 907                 }
 908                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 909                 tp->t_flags |= TF_FORCEDATA;
 910                 error = tcp_output(tp);
 911                 tp->t_flags &= ~TF_FORCEDATA;
 912         }
 913 out:
 914         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 915                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 916         INP_UNLOCK(inp);
 917         if (headlocked)
 918                 INP_INFO_WUNLOCK(&tcbinfo);
 919         return (error);
 920 }
 921
 922 /*
 923  * Abort the TCP.  Drop the connection abruptly.
 924  */
 925 static void
 926 tcp_usr_abort(struct socket *so)
 927 {
 928         struct inpcb *inp;
 929         struct tcpcb *tp = NULL;
 930         TCPDEBUG0;
 931
 932         inp = sotoinpcb(so);
 933         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
 934
 935         INP_INFO_WLOCK(&tcbinfo);
 936         INP_LOCK(inp);
 937         KASSERT(inp->inp_socket != NULL,
 938             ("tcp_usr_abort: inp_socket == NULL"));
 939
 940         /*
 941          * If we still have full TCP state, and we're not dropped, drop.
 942          */
 943         if (!(inp->inp_vflag & INP_TIMEWAIT) &&
 944             !(inp->inp_vflag & INP_DROPPED)) {
 945                 tp = intotcpcb(inp);
 946                 TCPDEBUG1();
 947                 tcp_drop(tp, ECONNABORTED);
 948                 TCPDEBUG2(PRU_ABORT);
 949         }
 950         if (!(inp->inp_vflag & INP_DROPPED)) {
 951                 SOCK_LOCK(so);
 952                 so->so_state |= SS_PROTOREF;
 953                 SOCK_UNLOCK(so);
 954                 inp->inp_vflag |= INP_SOCKREF;
 955         }
 956         INP_UNLOCK(inp);
 957         INP_INFO_WUNLOCK(&tcbinfo);
 958 }
 959
 960 /*
 961  * TCP socket is closed.  Start friendly disconnect.
 962  */
 963 static void
 964 tcp_usr_close(struct socket *so)
 965 {
 966         struct inpcb *inp;
 967         struct tcpcb *tp = NULL;
 968         TCPDEBUG0;
 969
 970         inp = sotoinpcb(so);
 971         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
 972
 973         INP_INFO_WLOCK(&tcbinfo);
 974         INP_LOCK(inp);
 975         KASSERT(inp->inp_socket != NULL,
 976             ("tcp_usr_close: inp_socket == NULL"));
 977
 978         /*
 979          * If we still have full TCP state, and we're not dropped, initiate
 980          * a disconnect.
 981          */
 982         if (!(inp->inp_vflag & INP_TIMEWAIT) &&
 983             !(inp->inp_vflag & INP_DROPPED)) {
 984                 tp = intotcpcb(inp);
 985                 TCPDEBUG1();
 986                 tcp_disconnect(tp);
 987                 TCPDEBUG2(PRU_CLOSE);
 988         }
 989         if (!(inp->inp_vflag & INP_DROPPED)) {
 990                 SOCK_LOCK(so);
 991                 so->so_state |= SS_PROTOREF;
 992                 SOCK_UNLOCK(so);
 993                 inp->inp_vflag |= INP_SOCKREF;
 994         }
 995         INP_UNLOCK(inp);
 996         INP_INFO_WUNLOCK(&tcbinfo);
 997 }
 998
 999 /*
1000  * Receive out-of-band data.
1001  */
1002 static int
1003 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1004 {
1005         int error = 0;
1006         struct inpcb *inp;
1007         struct tcpcb *tp = NULL;
1008
1009         TCPDEBUG0;
1010         inp = sotoinpcb(so);
1011         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1012         INP_LOCK(inp);
1013         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
1014                 error = EINVAL;
1015                 goto out;
1016         }
1017         tp = intotcpcb(inp);
1018         TCPDEBUG1();
1019         if ((so->so_oobmark == 0 &&
1020              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1021             so->so_options & SO_OOBINLINE ||
1022             tp->t_oobflags & TCPOOB_HADDATA) {
1023                 error = EINVAL;
1024                 goto out;
1025         }
1026         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1027                 error = EWOULDBLOCK;
1028                 goto out;
1029         }
1030         m->m_len = 1;
1031         *mtod(m, caddr_t) = tp->t_iobc;
1032         if ((flags & MSG_PEEK) == 0)
1033                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1034
1035 out:
1036         TCPDEBUG2(PRU_RCVOOB);
1037         INP_UNLOCK(inp);
1038         return (error);
1039 }
1040
1041 struct pr_usrreqs tcp_usrreqs = {
1042         .pru_abort =            tcp_usr_abort,
1043         .pru_accept =           tcp_usr_accept,
1044         .pru_attach =           tcp_usr_attach,
1045         .pru_bind =             tcp_usr_bind,
1046         .pru_connect =          tcp_usr_connect,
1047         .pru_control =          in_control,
1048         .pru_detach =           tcp_usr_detach,
1049         .pru_disconnect =       tcp_usr_disconnect,
1050         .pru_listen =           tcp_usr_listen,
1051         .pru_peeraddr =         tcp_peeraddr,
1052         .pru_rcvd =             tcp_usr_rcvd,
1053         .pru_rcvoob =           tcp_usr_rcvoob,
1054         .pru_send =             tcp_usr_send,
1055         .pru_shutdown =         tcp_usr_shutdown,
1056         .pru_sockaddr =         tcp_sockaddr,
1057         .pru_sosetlabel =       in_pcbsosetlabel,
1058         .pru_close =            tcp_usr_close,
1059 };
1060
1061 #ifdef INET6
1062 struct pr_usrreqs tcp6_usrreqs = {
1063         .pru_abort =            tcp_usr_abort,
1064         .pru_accept =           tcp6_usr_accept,
1065         .pru_attach =           tcp_usr_attach,
1066         .pru_bind =             tcp6_usr_bind,
1067         .pru_connect =          tcp6_usr_connect,
1068         .pru_control =          in6_control,
1069         .pru_detach =           tcp_usr_detach,
1070         .pru_disconnect =       tcp_usr_disconnect,
1071         .pru_listen =           tcp6_usr_listen,
1072         .pru_peeraddr =         in6_mapped_peeraddr,
1073         .pru_rcvd =             tcp_usr_rcvd,
1074         .pru_rcvoob =           tcp_usr_rcvoob,
1075         .pru_send =             tcp_usr_send,
1076         .pru_shutdown =         tcp_usr_shutdown,
1077         .pru_sockaddr =         in6_mapped_sockaddr,
1078         .pru_sosetlabel =       in_pcbsosetlabel,
1079         .pru_close =            tcp_usr_close,
1080 };
1081 #endif /* INET6 */
1082
1083 /*
1084  * Common subroutine to open a TCP connection to remote host specified
1085  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1086  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1087  * to choose a local host address (interface).  If there is an existing
1088  * incarnation of the same connection in TIME-WAIT state and if the remote
1089  * host was sending CC options and if the connection duration was < MSL, then
1090  * truncate the previous TIME-WAIT state and proceed.
1091  * Initialize connection parameters and enter SYN-SENT state.
1092  */
1093 static int
1094 tcp_connect(tp, nam, td)
1095         register struct tcpcb *tp;
1096         struct sockaddr *nam;
1097         struct thread *td;
1098 {
1099         struct inpcb *inp = tp->t_inpcb, *oinp;
1100         struct socket *so = inp->inp_socket;
1101         struct in_addr laddr;
1102         u_short lport;
1103         int error;
1104
1105         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1106         INP_LOCK_ASSERT(inp);
1107
1108         if (inp->inp_lport == 0) {
1109                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1110                 if (error)
1111                         return error;
1112         }
1113
1114         /*
1115          * Cannot simply call in_pcbconnect, because there might be an
1116          * earlier incarnation of this same connection still in
1117          * TIME_WAIT state, creating an ADDRINUSE error.
1118          */
1119         laddr = inp->inp_laddr;
1120         lport = inp->inp_lport;
1121         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1122             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1123         if (error && oinp == NULL)
1124                 return error;
1125         if (oinp)
1126                 return EADDRINUSE;
1127         inp->inp_laddr = laddr;
1128         in_pcbrehash(inp);
1129
1130         /* Compute window scaling to request.  */
1131         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1132             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1133                 tp->request_r_scale++;
1134
1135         soisconnecting(so);
1136         tcpstat.tcps_connattempt++;
1137         tp->t_state = TCPS_SYN_SENT;
1138         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1139         tp->iss = tcp_new_isn(tp);
1140         tp->t_bw_rtseq = tp->iss;
1141         tcp_sendseqinit(tp);
1142
1143         return 0;
1144 }
1145
1146 #ifdef INET6
1147 static int
1148 tcp6_connect(tp, nam, td)
1149         register struct tcpcb *tp;
1150         struct sockaddr *nam;
1151         struct thread *td;
1152 {
1153         struct inpcb *inp = tp->t_inpcb, *oinp;
1154         struct socket *so = inp->inp_socket;
1155         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1156         struct in6_addr *addr6;
1157         int error;
1158
1159         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1160         INP_LOCK_ASSERT(inp);
1161
1162         if (inp->inp_lport == 0) {
1163                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1164                 if (error)
1165                         return error;
1166         }
1167
1168         /*
1169          * Cannot simply call in_pcbconnect, because there might be an
1170          * earlier incarnation of this same connection still in
1171          * TIME_WAIT state, creating an ADDRINUSE error.
1172          * in6_pcbladdr() also handles scope zone IDs.
1173          */
1174         error = in6_pcbladdr(inp, nam, &addr6);
1175         if (error)
1176                 return error;
1177         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1178                                   &sin6->sin6_addr, sin6->sin6_port,
1179                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1180                                   ? addr6
1181                                   : &inp->in6p_laddr,
1182                                   inp->inp_lport,  0, NULL);
1183         if (oinp)
1184                 return EADDRINUSE;
1185         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1186                 inp->in6p_laddr = *addr6;
1187         inp->in6p_faddr = sin6->sin6_addr;
1188         inp->inp_fport = sin6->sin6_port;
1189         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1190         inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
1191         if (inp->in6p_flags & IN6P_AUTOFLOWLABEL)
1192                 inp->in6p_flowinfo |=
1193                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1194         in_pcbrehash(inp);
1195
1196         /* Compute window scaling to request.  */
1197         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1198             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
1199                 tp->request_r_scale++;
1200
1201         soisconnecting(so);
1202         tcpstat.tcps_connattempt++;
1203         tp->t_state = TCPS_SYN_SENT;
1204         callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1205         tp->iss = tcp_new_isn(tp);
1206         tp->t_bw_rtseq = tp->iss;
1207         tcp_sendseqinit(tp);
1208
1209         return 0;
1210 }
1211 #endif /* INET6 */
1212
1213 /*
1214  * Export TCP internal state information via a struct tcp_info, based on the
1215  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1216  * (TCP state machine, etc).  We export all information using FreeBSD-native
1217  * constants -- for example, the numeric values for tcpi_state will differ
1218  * from Linux.
1219  */
1220 static void
1221 tcp_fill_info(tp, ti)
1222         struct tcpcb *tp;
1223         struct tcp_info *ti;
1224 {
1225
1226         INP_LOCK_ASSERT(tp->t_inpcb);
1227         bzero(ti, sizeof(*ti));
1228
1229         ti->tcpi_state = tp->t_state;
1230         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1231                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1232         if (tp->sack_enable)
1233                 ti->tcpi_options |= TCPI_OPT_SACK;
1234         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1235                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1236                 ti->tcpi_snd_wscale = tp->snd_scale;
1237                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1238         }
1239         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1240         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1241
1242         /*
1243          * FreeBSD-specific extension fields for tcp_info.
1244          */
1245         ti->tcpi_rcv_space = tp->rcv_wnd;
1246         ti->tcpi_snd_wnd = tp->snd_wnd;
1247         ti->tcpi_snd_bwnd = tp->snd_bwnd;
1248 }
1249
1250 /*
1251  * The new sockopt interface makes it possible for us to block in the
1252  * copyin/out step (if we take a page fault).  Taking a page fault at
1253  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
1254  * use TSM, there probably isn't any need for this function to run at
1255  * splnet() any more.  This needs more examination.)
1256  *
1257  * XXXRW: The locking here is wrong; we may take a page fault while holding
1258  * the inpcb lock.
1259  */
1260 int
1261 tcp_ctloutput(so, sopt)
1262         struct socket *so;
1263         struct sockopt *sopt;
1264 {
1265         int     error, opt, optval;
1266         struct  inpcb *inp;
1267         struct  tcpcb *tp;
1268         struct  tcp_info ti;
1269
1270         error = 0;
1271         inp = sotoinpcb(so);
1272         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1273         INP_LOCK(inp);
1274         if (sopt->sopt_level != IPPROTO_TCP) {
1275                 INP_UNLOCK(inp);
1276 #ifdef INET6
1277                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1278                         error = ip6_ctloutput(so, sopt);
1279                 else
1280 #endif /* INET6 */
1281                 error = ip_ctloutput(so, sopt);
1282                 return (error);
1283         }
1284         if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) {
1285                 error = ECONNRESET;
1286                 goto out;
1287         }
1288         tp = intotcpcb(inp);
1289
1290         switch (sopt->sopt_dir) {
1291         case SOPT_SET:
1292                 switch (sopt->sopt_name) {
1293 #ifdef TCP_SIGNATURE
1294                 case TCP_MD5SIG:
1295                         error = sooptcopyin(sopt, &optval, sizeof optval,
1296                                             sizeof optval);
1297                         if (error)
1298                                 break;
1299
1300                         if (optval > 0)
1301                                 tp->t_flags |= TF_SIGNATURE;
1302                         else
1303                                 tp->t_flags &= ~TF_SIGNATURE;
1304                         break;
1305 #endif /* TCP_SIGNATURE */
1306                 case TCP_NODELAY:
1307                 case TCP_NOOPT:
1308                         error = sooptcopyin(sopt, &optval, sizeof optval,
1309                                             sizeof optval);
1310                         if (error)
1311                                 break;
1312
1313                         switch (sopt->sopt_name) {
1314                         case TCP_NODELAY:
1315                                 opt = TF_NODELAY;
1316                                 break;
1317                         case TCP_NOOPT:
1318                                 opt = TF_NOOPT;
1319                                 break;
1320                         default:
1321                                 opt = 0; /* dead code to fool gcc */
1322                                 break;
1323                         }
1324
1325                         if (optval)
1326                                 tp->t_flags |= opt;
1327                         else
1328                                 tp->t_flags &= ~opt;
1329                         break;
1330
1331                 case TCP_NOPUSH:
1332                         error = sooptcopyin(sopt, &optval, sizeof optval,
1333                                             sizeof optval);
1334                         if (error)
1335                                 break;
1336
1337                         if (optval)
1338                                 tp->t_flags |= TF_NOPUSH;
1339                         else {
1340                                 tp->t_flags &= ~TF_NOPUSH;
1341                                 error = tcp_output(tp);
1342                         }
1343                         break;
1344
1345                 case TCP_MAXSEG:
1346                         error = sooptcopyin(sopt, &optval, sizeof optval,
1347                                             sizeof optval);
1348                         if (error)
1349                                 break;
1350
1351                         if (optval > 0 && optval <= tp->t_maxseg &&
1352                             optval + 40 >= tcp_minmss)
1353                                 tp->t_maxseg = optval;
1354                         else
1355                                 error = EINVAL;
1356                         break;
1357
1358                 case TCP_INFO:
1359                         error = EINVAL;
1360                         break;
1361
1362                 default:
1363                         error = ENOPROTOOPT;
1364                         break;
1365                 }
1366                 break;
1367
1368         case SOPT_GET:
1369                 switch (sopt->sopt_name) {
1370 #ifdef TCP_SIGNATURE
1371                 case TCP_MD5SIG:
1372                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1373                         error = sooptcopyout(sopt, &optval, sizeof optval);
1374                         break;
1375 #endif
1376                 case TCP_NODELAY:
1377                         optval = tp->t_flags & TF_NODELAY;
1378                         error = sooptcopyout(sopt, &optval, sizeof optval);
1379                         break;
1380                 case TCP_MAXSEG:
1381                         optval = tp->t_maxseg;
1382                         error = sooptcopyout(sopt, &optval, sizeof optval);
1383                         break;
1384                 case TCP_NOOPT:
1385                         optval = tp->t_flags & TF_NOOPT;
1386                         error = sooptcopyout(sopt, &optval, sizeof optval);
1387                         break;
1388                 case TCP_NOPUSH:
1389                         optval = tp->t_flags & TF_NOPUSH;
1390                         error = sooptcopyout(sopt, &optval, sizeof optval);
1391                         break;
1392                 case TCP_INFO:
1393                         tcp_fill_info(tp, &ti);
1394                         error = sooptcopyout(sopt, &ti, sizeof ti);
1395                         break;
1396                 default:
1397                         error = ENOPROTOOPT;
1398                         break;
1399                 }
1400                 break;
1401         }
1402 out:
1403         INP_UNLOCK(inp);
1404         return (error);
1405 }
1406
1407 /*
1408  * tcp_sendspace and tcp_recvspace are the default send and receive window
1409  * sizes, respectively.  These are obsolescent (this information should
1410  * be set by the route).
1411  */
1412 u_long  tcp_sendspace = 1024*32;
1413 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1414     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1415 u_long  tcp_recvspace = 1024*64;
1416 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1417     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1418
1419 /*
1420  * Attach TCP protocol to socket, allocating
1421  * internet protocol control block, tcp control block,
1422  * bufer space, and entering LISTEN state if to accept connections.
1423  */
1424 static int
1425 tcp_attach(so)
1426         struct socket *so;
1427 {
1428         register struct tcpcb *tp;
1429         struct inpcb *inp;
1430         int error;
1431 #ifdef INET6
1432         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1433 #endif
1434
1435         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1436                 error = soreserve(so, tcp_sendspace, tcp_recvspace);
1437                 if (error)
1438                         return (error);
1439         }
1440         INP_INFO_WLOCK(&tcbinfo);
1441         error = in_pcballoc(so, &tcbinfo);
1442         if (error) {
1443                 INP_INFO_WUNLOCK(&tcbinfo);
1444                 return (error);
1445         }
1446         inp = sotoinpcb(so);
1447 #ifdef INET6
1448         if (isipv6) {
1449                 inp->inp_vflag |= INP_IPV6;
1450                 inp->in6p_hops = -1;    /* use kernel default */
1451         }
1452         else
1453 #endif
1454         inp->inp_vflag |= INP_IPV4;
1455         tp = tcp_newtcpcb(inp);
1456         if (tp == NULL) {
1457 #ifdef INET6
1458                 if (isipv6) {
1459                         in6_pcbdetach(inp);
1460                         in6_pcbfree(inp);
1461                 } else {
1462 #endif
1463                         in_pcbdetach(inp);
1464                         in_pcbfree(inp);
1465 #ifdef INET6
1466                 }
1467 #endif
1468                 INP_INFO_WUNLOCK(&tcbinfo);
1469                 return (ENOBUFS);
1470         }
1471         tp->t_state = TCPS_CLOSED;
1472         INP_UNLOCK(inp);
1473         INP_INFO_WUNLOCK(&tcbinfo);
1474         return (0);
1475 }
1476
1477 /*
1478  * Initiate (or continue) disconnect.
1479  * If embryonic state, just send reset (once).
1480  * If in ``let data drain'' option and linger null, just drop.
1481  * Otherwise (hard), mark socket disconnecting and drop
1482  * current input data; switch states based on user close, and
1483  * send segment to peer (with FIN).
1484  */
1485 static void
1486 tcp_disconnect(tp)
1487         register struct tcpcb *tp;
1488 {
1489         struct inpcb *inp = tp->t_inpcb;
1490         struct socket *so = inp->inp_socket;
1491
1492         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1493         INP_LOCK_ASSERT(inp);
1494
1495         /*
1496          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1497          * socket is still open.
1498          */
1499         if (tp->t_state < TCPS_ESTABLISHED) {
1500                 tp = tcp_close(tp);
1501                 KASSERT(tp != NULL,
1502                     ("tcp_disconnect: tcp_close() returned NULL"));
1503         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1504                 tp = tcp_drop(tp, 0);
1505                 KASSERT(tp != NULL,
1506                     ("tcp_disconnect: tcp_drop() returned NULL"));
1507         } else {
1508                 soisdisconnecting(so);
1509                 sbflush(&so->so_rcv);
1510                 tcp_usrclosed(tp);
1511                 if (!(inp->inp_vflag & INP_DROPPED))
1512                         tcp_output(tp);
1513         }
1514 }
1515
1516 /*
1517  * User issued close, and wish to trail through shutdown states:
1518  * if never received SYN, just forget it.  If got a SYN from peer,
1519  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1520  * If already got a FIN from peer, then almost done; go to LAST_ACK
1521  * state.  In all other cases, have already sent FIN to peer (e.g.
1522  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1523  * for peer to send FIN or not respond to keep-alives, etc.
1524  * We can let the user exit from the close as soon as the FIN is acked.
1525  */
1526 static void
1527 tcp_usrclosed(tp)
1528         register struct tcpcb *tp;
1529 {
1530
1531         INP_INFO_WLOCK_ASSERT(&tcbinfo);
1532         INP_LOCK_ASSERT(tp->t_inpcb);
1533
1534         switch (tp->t_state) {
1535
1536         case TCPS_CLOSED:
1537         case TCPS_LISTEN:
1538                 tp->t_state = TCPS_CLOSED;
1539                 tp = tcp_close(tp);
1540                 /*
1541                  * tcp_close() should never return NULL here as the socket is
1542                  * still open.
1543                  */
1544                 KASSERT(tp != NULL,
1545                     ("tcp_usrclosed: tcp_close() returned NULL"));
1546                 break;
1547
1548         case TCPS_SYN_SENT:
1549         case TCPS_SYN_RECEIVED:
1550                 tp->t_flags |= TF_NEEDFIN;
1551                 break;
1552
1553         case TCPS_ESTABLISHED:
1554                 tp->t_state = TCPS_FIN_WAIT_1;
1555                 break;
1556
1557         case TCPS_CLOSE_WAIT:
1558                 tp->t_state = TCPS_LAST_ACK;
1559                 break;
1560         }
1561         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1562                 soisdisconnected(tp->t_inpcb->inp_socket);
1563                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1564                 if (tp->t_state == TCPS_FIN_WAIT_2)
1565                         callout_reset(tp->tt_2msl, tcp_maxidle,
1566                                       tcp_timer_2msl, tp);
1567         }
1568 }