sys/netinet/tcp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2006-2007 Robert N. M. Watson
   5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   6  * All rights reserved.
   7  *
   8  * Portions of this software were developed by Robert N. M. Watson under
   9  * contract to Juniper Networks, Inc.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 4. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_ddb.h"
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44 #include "opt_tcpdebug.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/limits.h>
  49 #include <sys/malloc.h>
  50 #include <sys/kernel.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/mbuf.h>
  53 #ifdef INET6
  54 #include <sys/domain.h>
  55 #endif /* INET6 */
  56 #include <sys/socket.h>
  57 #include <sys/socketvar.h>
  58 #include <sys/protosw.h>
  59 #include <sys/proc.h>
  60 #include <sys/jail.h>
  61 #include <sys/syslog.h>
  62
  63 #ifdef DDB
  64 #include <ddb/ddb.h>
  65 #endif
  66
  67 #include <net/if.h>
  68 #include <net/route.h>
  69 #include <net/vnet.h>
  70
  71 #include <netinet/cc.h>
  72 #include <netinet/in.h>
  73 #include <netinet/in_pcb.h>
  74 #include <netinet/in_systm.h>
  75 #include <netinet/in_var.h>
  76 #include <netinet/ip_var.h>
  77 #ifdef INET6
  78 #include <netinet/ip6.h>
  79 #include <netinet6/in6_pcb.h>
  80 #include <netinet6/ip6_var.h>
  81 #include <netinet6/scope6_var.h>
  82 #endif
  83 #ifdef TCP_RFC7413
  84 #include <netinet/tcp_fastopen.h>
  85 #endif
  86 #include <netinet/tcp_fsm.h>
  87 #include <netinet/tcp_seq.h>
  88 #include <netinet/tcp_timer.h>
  89 #include <netinet/tcp_var.h>
  90 #include <netinet/tcpip.h>
  91 #ifdef TCPDEBUG
  92 #include <netinet/tcp_debug.h>
  93 #endif
  94 #ifdef TCP_OFFLOAD
  95 #include <netinet/tcp_offload.h>
  96 #endif
  97
  98 /*
  99  * TCP protocol interface to socket abstraction.
 100  */
 101 static int      tcp_attach(struct socket *);
 102 #ifdef INET
 103 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 104                     struct thread *td);
 105 #endif /* INET */
 106 #ifdef INET6
 107 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 108                     struct thread *td);
 109 #endif /* INET6 */
 110 static void     tcp_disconnect(struct tcpcb *);
 111 static void     tcp_usrclosed(struct tcpcb *);
 112 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 113
 114 #ifdef TCPDEBUG
 115 #define TCPDEBUG0       int ostate = 0
 116 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 117 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 118                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 119 #else
 120 #define TCPDEBUG0
 121 #define TCPDEBUG1()
 122 #define TCPDEBUG2(req)
 123 #endif
 124
 125 /*
 126  * TCP attaches to socket via pru_attach(), reserving space,
 127  * and an internet control block.
 128  */
 129 static int
 130 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 131 {
 132         struct inpcb *inp;
 133         struct tcpcb *tp = NULL;
 134         int error;
 135         TCPDEBUG0;
 136
 137         inp = sotoinpcb(so);
 138         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 139         TCPDEBUG1();
 140
 141         error = tcp_attach(so);
 142         if (error)
 143                 goto out;
 144
 145         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 146                 so->so_linger = TCP_LINGERTIME;
 147
 148         inp = sotoinpcb(so);
 149         tp = intotcpcb(inp);
 150 out:
 151         TCPDEBUG2(PRU_ATTACH);
 152         return error;
 153 }
 154
 155 /*
 156  * tcp_detach is called when the socket layer loses its final reference
 157  * to the socket, be it a file descriptor reference, a reference from TCP,
 158  * etc.  At this point, there is only one case in which we will keep around
 159  * inpcb state: time wait.
 160  *
 161  * This function can probably be re-absorbed back into tcp_usr_detach() now
 162  * that there is a single detach path.
 163  */
 164 static void
 165 tcp_detach(struct socket *so, struct inpcb *inp)
 166 {
 167         struct tcpcb *tp;
 168
 169         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
 170         INP_WLOCK_ASSERT(inp);
 171
 172         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 173         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 174
 175         tp = intotcpcb(inp);
 176
 177         if (inp->inp_flags & INP_TIMEWAIT) {
 178                 /*
 179                  * There are two cases to handle: one in which the time wait
 180                  * state is being discarded (INP_DROPPED), and one in which
 181                  * this connection will remain in timewait.  In the former,
 182                  * it is time to discard all state (except tcptw, which has
 183                  * already been discarded by the timewait close code, which
 184                  * should be further up the call stack somewhere).  In the
 185                  * latter case, we detach from the socket, but leave the pcb
 186                  * present until timewait ends.
 187                  *
 188                  * XXXRW: Would it be cleaner to free the tcptw here?
 189                  *
 190                  * Astute question indeed, from twtcp perspective there are
 191                  * three cases to consider:
 192                  *
 193                  * #1 tcp_detach is called at tcptw creation time by
 194                  *  tcp_twstart, then do not discard the newly created tcptw
 195                  *  and leave inpcb present until timewait ends
 196                  * #2 tcp_detach is called at timewait end (or reuse) by
 197                  *  tcp_twclose, then the tcptw has already been discarded
 198                  *  (or reused) and inpcb is freed here
 199                  * #3 tcp_detach is called() after timewait ends (or reuse)
 200                  *  (e.g. by soclose), then tcptw has already been discarded
 201                  *  (or reused) and inpcb is freed here
 202                  *
 203                  *  In all three cases the tcptw should not be freed here.
 204                  */
 205                 if (inp->inp_flags & INP_DROPPED) {
 206                         in_pcbdetach(inp);
 207                         if (__predict_true(tp == NULL)) {
 208                                 in_pcbfree(inp);
 209                         } else {
 210                                 /*
 211                                  * This case should not happen as in TIMEWAIT
 212                                  * state the inp should not be destroyed before
 213                                  * its tcptw.  If INVARIANTS is defined, panic.
 214                                  */
 215 #ifdef INVARIANTS
 216                                 panic("%s: Panic before an inp double-free: "
 217                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 218                                     , __func__);
 219 #else
 220                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 221                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 222                                     , __func__);
 223 #endif
 224                                 INP_WUNLOCK(inp);
 225                         }
 226                 } else {
 227                         in_pcbdetach(inp);
 228                         INP_WUNLOCK(inp);
 229                 }
 230         } else {
 231                 /*
 232                  * If the connection is not in timewait, we consider two
 233                  * two conditions: one in which no further processing is
 234                  * necessary (dropped || embryonic), and one in which TCP is
 235                  * not yet done, but no longer requires the socket, so the
 236                  * pcb will persist for the time being.
 237                  *
 238                  * XXXRW: Does the second case still occur?
 239                  */
 240                 if (inp->inp_flags & INP_DROPPED ||
 241                     tp->t_state < TCPS_SYN_SENT) {
 242                         tcp_discardcb(tp);
 243                         in_pcbdetach(inp);
 244                         in_pcbfree(inp);
 245                 } else {
 246                         in_pcbdetach(inp);
 247                         INP_WUNLOCK(inp);
 248                 }
 249         }
 250 }
 251
 252 /*
 253  * pru_detach() detaches the TCP protocol from the socket.
 254  * If the protocol state is non-embryonic, then can't
 255  * do this directly: have to initiate a pru_disconnect(),
 256  * which may finish later; embryonic TCB's can just
 257  * be discarded here.
 258  */
 259 static void
 260 tcp_usr_detach(struct socket *so)
 261 {
 262         struct inpcb *inp;
 263         int rlock = 0;
 264
 265         inp = sotoinpcb(so);
 266         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 267         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 268                 INP_INFO_RLOCK(&V_tcbinfo);
 269                 rlock = 1;
 270         }
 271         INP_WLOCK(inp);
 272         KASSERT(inp->inp_socket != NULL,
 273             ("tcp_usr_detach: inp_socket == NULL"));
 274         tcp_detach(so, inp);
 275         if (rlock)
 276                 INP_INFO_RUNLOCK(&V_tcbinfo);
 277 }
 278
 279 #ifdef INET
 280 /*
 281  * Give the socket an address.
 282  */
 283 static int
 284 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 285 {
 286         int error = 0;
 287         struct inpcb *inp;
 288         struct tcpcb *tp = NULL;
 289         struct sockaddr_in *sinp;
 290
 291         sinp = (struct sockaddr_in *)nam;
 292         if (nam->sa_len != sizeof (*sinp))
 293                 return (EINVAL);
 294         /*
 295          * Must check for multicast addresses and disallow binding
 296          * to them.
 297          */
 298         if (sinp->sin_family == AF_INET &&
 299             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 300                 return (EAFNOSUPPORT);
 301
 302         TCPDEBUG0;
 303         inp = sotoinpcb(so);
 304         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 305         INP_WLOCK(inp);
 306         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 307                 error = EINVAL;
 308                 goto out;
 309         }
 310         tp = intotcpcb(inp);
 311         TCPDEBUG1();
 312         INP_HASH_WLOCK(&V_tcbinfo);
 313         error = in_pcbbind(inp, nam, td->td_ucred);
 314         INP_HASH_WUNLOCK(&V_tcbinfo);
 315 out:
 316         TCPDEBUG2(PRU_BIND);
 317         INP_WUNLOCK(inp);
 318
 319         return (error);
 320 }
 321 #endif /* INET */
 322
 323 #ifdef INET6
 324 static int
 325 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 326 {
 327         int error = 0;
 328         struct inpcb *inp;
 329         struct tcpcb *tp = NULL;
 330         struct sockaddr_in6 *sin6p;
 331
 332         sin6p = (struct sockaddr_in6 *)nam;
 333         if (nam->sa_len != sizeof (*sin6p))
 334                 return (EINVAL);
 335         /*
 336          * Must check for multicast addresses and disallow binding
 337          * to them.
 338          */
 339         if (sin6p->sin6_family == AF_INET6 &&
 340             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 341                 return (EAFNOSUPPORT);
 342
 343         TCPDEBUG0;
 344         inp = sotoinpcb(so);
 345         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 346         INP_WLOCK(inp);
 347         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 348                 error = EINVAL;
 349                 goto out;
 350         }
 351         tp = intotcpcb(inp);
 352         TCPDEBUG1();
 353         INP_HASH_WLOCK(&V_tcbinfo);
 354         inp->inp_vflag &= ~INP_IPV4;
 355         inp->inp_vflag |= INP_IPV6;
 356 #ifdef INET
 357         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 358                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 359                         inp->inp_vflag |= INP_IPV4;
 360                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 361                         struct sockaddr_in sin;
 362
 363                         in6_sin6_2_sin(&sin, sin6p);
 364                         inp->inp_vflag |= INP_IPV4;
 365                         inp->inp_vflag &= ~INP_IPV6;
 366                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 367                             td->td_ucred);
 368                         INP_HASH_WUNLOCK(&V_tcbinfo);
 369                         goto out;
 370                 }
 371         }
 372 #endif
 373         error = in6_pcbbind(inp, nam, td->td_ucred);
 374         INP_HASH_WUNLOCK(&V_tcbinfo);
 375 out:
 376         TCPDEBUG2(PRU_BIND);
 377         INP_WUNLOCK(inp);
 378         return (error);
 379 }
 380 #endif /* INET6 */
 381
 382 #ifdef INET
 383 /*
 384  * Prepare to accept connections.
 385  */
 386 static int
 387 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 388 {
 389         int error = 0;
 390         struct inpcb *inp;
 391         struct tcpcb *tp = NULL;
 392
 393         TCPDEBUG0;
 394         inp = sotoinpcb(so);
 395         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 396         INP_WLOCK(inp);
 397         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 398                 error = EINVAL;
 399                 goto out;
 400         }
 401         tp = intotcpcb(inp);
 402         TCPDEBUG1();
 403         SOCK_LOCK(so);
 404         error = solisten_proto_check(so);
 405         INP_HASH_WLOCK(&V_tcbinfo);
 406         if (error == 0 && inp->inp_lport == 0)
 407                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 408         INP_HASH_WUNLOCK(&V_tcbinfo);
 409         if (error == 0) {
 410                 tcp_state_change(tp, TCPS_LISTEN);
 411                 solisten_proto(so, backlog);
 412 #ifdef TCP_OFFLOAD
 413                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 414                         tcp_offload_listen_start(tp);
 415 #endif
 416         }
 417         SOCK_UNLOCK(so);
 418
 419 #ifdef TCP_RFC7413
 420         if (tp->t_flags & TF_FASTOPEN)
 421                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 422 #endif
 423 out:
 424         TCPDEBUG2(PRU_LISTEN);
 425         INP_WUNLOCK(inp);
 426         return (error);
 427 }
 428 #endif /* INET */
 429
 430 #ifdef INET6
 431 static int
 432 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 433 {
 434         int error = 0;
 435         struct inpcb *inp;
 436         struct tcpcb *tp = NULL;
 437
 438         TCPDEBUG0;
 439         inp = sotoinpcb(so);
 440         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 441         INP_WLOCK(inp);
 442         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 443                 error = EINVAL;
 444                 goto out;
 445         }
 446         tp = intotcpcb(inp);
 447         TCPDEBUG1();
 448         SOCK_LOCK(so);
 449         error = solisten_proto_check(so);
 450         INP_HASH_WLOCK(&V_tcbinfo);
 451         if (error == 0 && inp->inp_lport == 0) {
 452                 inp->inp_vflag &= ~INP_IPV4;
 453                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 454                         inp->inp_vflag |= INP_IPV4;
 455                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 456         }
 457         INP_HASH_WUNLOCK(&V_tcbinfo);
 458         if (error == 0) {
 459                 tcp_state_change(tp, TCPS_LISTEN);
 460                 solisten_proto(so, backlog);
 461 #ifdef TCP_OFFLOAD
 462                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 463                         tcp_offload_listen_start(tp);
 464 #endif
 465         }
 466         SOCK_UNLOCK(so);
 467
 468 #ifdef TCP_RFC7413
 469         if (tp->t_flags & TF_FASTOPEN)
 470                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 471 #endif
 472 out:
 473         TCPDEBUG2(PRU_LISTEN);
 474         INP_WUNLOCK(inp);
 475         return (error);
 476 }
 477 #endif /* INET6 */
 478
 479 #ifdef INET
 480 /*
 481  * Initiate connection to peer.
 482  * Create a template for use in transmissions on this connection.
 483  * Enter SYN_SENT state, and mark socket as connecting.
 484  * Start keep-alive timer, and seed output sequence space.
 485  * Send initial segment on connection.
 486  */
 487 static int
 488 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 489 {
 490         int error = 0;
 491         struct inpcb *inp;
 492         struct tcpcb *tp = NULL;
 493         struct sockaddr_in *sinp;
 494
 495         sinp = (struct sockaddr_in *)nam;
 496         if (nam->sa_len != sizeof (*sinp))
 497                 return (EINVAL);
 498         /*
 499          * Must disallow TCP ``connections'' to multicast addresses.
 500          */
 501         if (sinp->sin_family == AF_INET
 502             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 503                 return (EAFNOSUPPORT);
 504         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 505                 return (error);
 506
 507         TCPDEBUG0;
 508         inp = sotoinpcb(so);
 509         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 510         INP_WLOCK(inp);
 511         if (inp->inp_flags & INP_TIMEWAIT) {
 512                 error = EADDRINUSE;
 513                 goto out;
 514         }
 515         if (inp->inp_flags & INP_DROPPED) {
 516                 error = ECONNREFUSED;
 517                 goto out;
 518         }
 519         tp = intotcpcb(inp);
 520         TCPDEBUG1();
 521         if ((error = tcp_connect(tp, nam, td)) != 0)
 522                 goto out;
 523 #ifdef TCP_OFFLOAD
 524         if (registered_toedevs > 0 &&
 525             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 526             (error = tcp_offload_connect(so, nam)) == 0)
 527                 goto out;
 528 #endif
 529         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 530         error = tcp_output(tp);
 531 out:
 532         TCPDEBUG2(PRU_CONNECT);
 533         INP_WUNLOCK(inp);
 534         return (error);
 535 }
 536 #endif /* INET */
 537
 538 #ifdef INET6
 539 static int
 540 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 541 {
 542         int error = 0;
 543         struct inpcb *inp;
 544         struct tcpcb *tp = NULL;
 545         struct sockaddr_in6 *sin6p;
 546
 547         TCPDEBUG0;
 548
 549         sin6p = (struct sockaddr_in6 *)nam;
 550         if (nam->sa_len != sizeof (*sin6p))
 551                 return (EINVAL);
 552         /*
 553          * Must disallow TCP ``connections'' to multicast addresses.
 554          */
 555         if (sin6p->sin6_family == AF_INET6
 556             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 557                 return (EAFNOSUPPORT);
 558
 559         inp = sotoinpcb(so);
 560         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 561         INP_WLOCK(inp);
 562         if (inp->inp_flags & INP_TIMEWAIT) {
 563                 error = EADDRINUSE;
 564                 goto out;
 565         }
 566         if (inp->inp_flags & INP_DROPPED) {
 567                 error = ECONNREFUSED;
 568                 goto out;
 569         }
 570         tp = intotcpcb(inp);
 571         TCPDEBUG1();
 572 #ifdef INET
 573         /*
 574          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 575          * therefore probably require the hash lock, which isn't held here.
 576          * Is this a significant problem?
 577          */
 578         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 579                 struct sockaddr_in sin;
 580
 581                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 582                         error = EINVAL;
 583                         goto out;
 584                 }
 585
 586                 in6_sin6_2_sin(&sin, sin6p);
 587                 inp->inp_vflag |= INP_IPV4;
 588                 inp->inp_vflag &= ~INP_IPV6;
 589                 if ((error = prison_remote_ip4(td->td_ucred,
 590                     &sin.sin_addr)) != 0)
 591                         goto out;
 592                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 593                         goto out;
 594 #ifdef TCP_OFFLOAD
 595                 if (registered_toedevs > 0 &&
 596                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 597                     (error = tcp_offload_connect(so, nam)) == 0)
 598                         goto out;
 599 #endif
 600                 error = tcp_output(tp);
 601                 goto out;
 602         }
 603 #endif
 604         inp->inp_vflag &= ~INP_IPV4;
 605         inp->inp_vflag |= INP_IPV6;
 606         inp->inp_inc.inc_flags |= INC_ISIPV6;
 607         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 608                 goto out;
 609         if ((error = tcp6_connect(tp, nam, td)) != 0)
 610                 goto out;
 611 #ifdef TCP_OFFLOAD
 612         if (registered_toedevs > 0 &&
 613             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 614             (error = tcp_offload_connect(so, nam)) == 0)
 615                 goto out;
 616 #endif
 617         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 618         error = tcp_output(tp);
 619
 620 out:
 621         TCPDEBUG2(PRU_CONNECT);
 622         INP_WUNLOCK(inp);
 623         return (error);
 624 }
 625 #endif /* INET6 */
 626
 627 /*
 628  * Initiate disconnect from peer.
 629  * If connection never passed embryonic stage, just drop;
 630  * else if don't need to let data drain, then can just drop anyways,
 631  * else have to begin TCP shutdown process: mark socket disconnecting,
 632  * drain unread data, state switch to reflect user close, and
 633  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 634  * when peer sends FIN and acks ours.
 635  *
 636  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 637  */
 638 static int
 639 tcp_usr_disconnect(struct socket *so)
 640 {
 641         struct inpcb *inp;
 642         struct tcpcb *tp = NULL;
 643         int error = 0;
 644
 645         TCPDEBUG0;
 646         INP_INFO_RLOCK(&V_tcbinfo);
 647         inp = sotoinpcb(so);
 648         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 649         INP_WLOCK(inp);
 650         if (inp->inp_flags & INP_TIMEWAIT)
 651                 goto out;
 652         if (inp->inp_flags & INP_DROPPED) {
 653                 error = ECONNRESET;
 654                 goto out;
 655         }
 656         tp = intotcpcb(inp);
 657         TCPDEBUG1();
 658         tcp_disconnect(tp);
 659 out:
 660         TCPDEBUG2(PRU_DISCONNECT);
 661         INP_WUNLOCK(inp);
 662         INP_INFO_RUNLOCK(&V_tcbinfo);
 663         return (error);
 664 }
 665
 666 #ifdef INET
 667 /*
 668  * Accept a connection.  Essentially all the work is done at higher levels;
 669  * just return the address of the peer, storing through addr.
 670  */
 671 static int
 672 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 673 {
 674         int error = 0;
 675         struct inpcb *inp = NULL;
 676         struct tcpcb *tp = NULL;
 677         struct in_addr addr;
 678         in_port_t port = 0;
 679         TCPDEBUG0;
 680
 681         if (so->so_state & SS_ISDISCONNECTED)
 682                 return (ECONNABORTED);
 683
 684         inp = sotoinpcb(so);
 685         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 686         INP_WLOCK(inp);
 687         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 688                 error = ECONNABORTED;
 689                 goto out;
 690         }
 691         tp = intotcpcb(inp);
 692         TCPDEBUG1();
 693
 694         /*
 695          * We inline in_getpeeraddr and COMMON_END here, so that we can
 696          * copy the data of interest and defer the malloc until after we
 697          * release the lock.
 698          */
 699         port = inp->inp_fport;
 700         addr = inp->inp_faddr;
 701
 702 out:
 703         TCPDEBUG2(PRU_ACCEPT);
 704         INP_WUNLOCK(inp);
 705         if (error == 0)
 706                 *nam = in_sockaddr(port, &addr);
 707         return error;
 708 }
 709 #endif /* INET */
 710
 711 #ifdef INET6
 712 static int
 713 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 714 {
 715         struct inpcb *inp = NULL;
 716         int error = 0;
 717         struct tcpcb *tp = NULL;
 718         struct in_addr addr;
 719         struct in6_addr addr6;
 720         in_port_t port = 0;
 721         int v4 = 0;
 722         TCPDEBUG0;
 723
 724         if (so->so_state & SS_ISDISCONNECTED)
 725                 return (ECONNABORTED);
 726
 727         inp = sotoinpcb(so);
 728         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 729         INP_INFO_RLOCK(&V_tcbinfo);
 730         INP_WLOCK(inp);
 731         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 732                 error = ECONNABORTED;
 733                 goto out;
 734         }
 735         tp = intotcpcb(inp);
 736         TCPDEBUG1();
 737
 738         /*
 739          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 740          * copy the data of interest and defer the malloc until after we
 741          * release the lock.
 742          */
 743         if (inp->inp_vflag & INP_IPV4) {
 744                 v4 = 1;
 745                 port = inp->inp_fport;
 746                 addr = inp->inp_faddr;
 747         } else {
 748                 port = inp->inp_fport;
 749                 addr6 = inp->in6p_faddr;
 750         }
 751
 752 out:
 753         TCPDEBUG2(PRU_ACCEPT);
 754         INP_WUNLOCK(inp);
 755         INP_INFO_RUNLOCK(&V_tcbinfo);
 756         if (error == 0) {
 757                 if (v4)
 758                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 759                 else
 760                         *nam = in6_sockaddr(port, &addr6);
 761         }
 762         return error;
 763 }
 764 #endif /* INET6 */
 765
 766 /*
 767  * Mark the connection as being incapable of further output.
 768  */
 769 static int
 770 tcp_usr_shutdown(struct socket *so)
 771 {
 772         int error = 0;
 773         struct inpcb *inp;
 774         struct tcpcb *tp = NULL;
 775
 776         TCPDEBUG0;
 777         INP_INFO_RLOCK(&V_tcbinfo);
 778         inp = sotoinpcb(so);
 779         KASSERT(inp != NULL, ("inp == NULL"));
 780         INP_WLOCK(inp);
 781         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 782                 error = ECONNRESET;
 783                 goto out;
 784         }
 785         tp = intotcpcb(inp);
 786         TCPDEBUG1();
 787         socantsendmore(so);
 788         tcp_usrclosed(tp);
 789         if (!(inp->inp_flags & INP_DROPPED))
 790                 error = tcp_output(tp);
 791
 792 out:
 793         TCPDEBUG2(PRU_SHUTDOWN);
 794         INP_WUNLOCK(inp);
 795         INP_INFO_RUNLOCK(&V_tcbinfo);
 796
 797         return (error);
 798 }
 799
 800 /*
 801  * After a receive, possibly send window update to peer.
 802  */
 803 static int
 804 tcp_usr_rcvd(struct socket *so, int flags)
 805 {
 806         struct inpcb *inp;
 807         struct tcpcb *tp = NULL;
 808         int error = 0;
 809
 810         TCPDEBUG0;
 811         inp = sotoinpcb(so);
 812         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 813         INP_WLOCK(inp);
 814         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 815                 error = ECONNRESET;
 816                 goto out;
 817         }
 818         tp = intotcpcb(inp);
 819         TCPDEBUG1();
 820 #ifdef TCP_RFC7413
 821         /*
 822          * For passively-created TFO connections, don't attempt a window
 823          * update while still in SYN_RECEIVED as this may trigger an early
 824          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 825          * application response data, or failing that, when the DELACK timer
 826          * expires.
 827          */
 828         if ((tp->t_flags & TF_FASTOPEN) &&
 829             (tp->t_state == TCPS_SYN_RECEIVED))
 830                 goto out;
 831 #endif
 832 #ifdef TCP_OFFLOAD
 833         if (tp->t_flags & TF_TOE)
 834                 tcp_offload_rcvd(tp);
 835         else
 836 #endif
 837         tcp_output(tp);
 838
 839 out:
 840         TCPDEBUG2(PRU_RCVD);
 841         INP_WUNLOCK(inp);
 842         return (error);
 843 }
 844
 845 /*
 846  * Do a send by putting data in output queue and updating urgent
 847  * marker if URG set.  Possibly send more data.  Unlike the other
 848  * pru_*() routines, the mbuf chains are our responsibility.  We
 849  * must either enqueue them or free them.  The other pru_* routines
 850  * generally are caller-frees.
 851  */
 852 static int
 853 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 854     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 855 {
 856         int error = 0;
 857         struct inpcb *inp;
 858         struct tcpcb *tp = NULL;
 859 #ifdef INET6
 860         int isipv6;
 861 #endif
 862         TCPDEBUG0;
 863
 864         /*
 865          * We require the pcbinfo lock if we will close the socket as part of
 866          * this call.
 867          */
 868         if (flags & PRUS_EOF)
 869                 INP_INFO_RLOCK(&V_tcbinfo);
 870         inp = sotoinpcb(so);
 871         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 872         INP_WLOCK(inp);
 873         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 874                 if (control)
 875                         m_freem(control);
 876                 if (m)
 877                         m_freem(m);
 878                 error = ECONNRESET;
 879                 goto out;
 880         }
 881 #ifdef INET6
 882         isipv6 = nam && nam->sa_family == AF_INET6;
 883 #endif /* INET6 */
 884         tp = intotcpcb(inp);
 885         TCPDEBUG1();
 886         if (control) {
 887                 /* TCP doesn't do control messages (rights, creds, etc) */
 888                 if (control->m_len) {
 889                         m_freem(control);
 890                         if (m)
 891                                 m_freem(m);
 892                         error = EINVAL;
 893                         goto out;
 894                 }
 895                 m_freem(control);       /* empty control, just free it */
 896         }
 897         if (!(flags & PRUS_OOB)) {
 898                 sbappendstream(&so->so_snd, m);
 899                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 900                         /*
 901                          * Do implied connect if not yet connected,
 902                          * initialize window to default value, and
 903                          * initialize maxseg/maxopd using peer's cached
 904                          * MSS.
 905                          */
 906 #ifdef INET6
 907                         if (isipv6)
 908                                 error = tcp6_connect(tp, nam, td);
 909 #endif /* INET6 */
 910 #if defined(INET6) && defined(INET)
 911                         else
 912 #endif
 913 #ifdef INET
 914                                 error = tcp_connect(tp, nam, td);
 915 #endif
 916                         if (error)
 917                                 goto out;
 918                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 919                         tcp_mss(tp, -1);
 920                 }
 921                 if (flags & PRUS_EOF) {
 922                         /*
 923                          * Close the send side of the connection after
 924                          * the data is sent.
 925                          */
 926                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 927                         socantsendmore(so);
 928                         tcp_usrclosed(tp);
 929                 }
 930                 if (!(inp->inp_flags & INP_DROPPED)) {
 931                         if (flags & PRUS_MORETOCOME)
 932                                 tp->t_flags |= TF_MORETOCOME;
 933                         error = tcp_output(tp);
 934                         if (flags & PRUS_MORETOCOME)
 935                                 tp->t_flags &= ~TF_MORETOCOME;
 936                 }
 937         } else {
 938                 /*
 939                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 940                  */
 941                 SOCKBUF_LOCK(&so->so_snd);
 942                 if (sbspace(&so->so_snd) < -512) {
 943                         SOCKBUF_UNLOCK(&so->so_snd);
 944                         m_freem(m);
 945                         error = ENOBUFS;
 946                         goto out;
 947                 }
 948                 /*
 949                  * According to RFC961 (Assigned Protocols),
 950                  * the urgent pointer points to the last octet
 951                  * of urgent data.  We continue, however,
 952                  * to consider it to indicate the first octet
 953                  * of data past the urgent section.
 954                  * Otherwise, snd_up should be one lower.
 955                  */
 956                 sbappendstream_locked(&so->so_snd, m);
 957                 SOCKBUF_UNLOCK(&so->so_snd);
 958                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 959                         /*
 960                          * Do implied connect if not yet connected,
 961                          * initialize window to default value, and
 962                          * initialize maxseg/maxopd using peer's cached
 963                          * MSS.
 964                          */
 965 #ifdef INET6
 966                         if (isipv6)
 967                                 error = tcp6_connect(tp, nam, td);
 968 #endif /* INET6 */
 969 #if defined(INET6) && defined(INET)
 970                         else
 971 #endif
 972 #ifdef INET
 973                                 error = tcp_connect(tp, nam, td);
 974 #endif
 975                         if (error)
 976                                 goto out;
 977                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 978                         tcp_mss(tp, -1);
 979                 }
 980                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 981                 tp->t_flags |= TF_FORCEDATA;
 982                 error = tcp_output(tp);
 983                 tp->t_flags &= ~TF_FORCEDATA;
 984         }
 985 out:
 986         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 987                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 988         INP_WUNLOCK(inp);
 989         if (flags & PRUS_EOF)
 990                 INP_INFO_RUNLOCK(&V_tcbinfo);
 991         return (error);
 992 }
 993
 994 /*
 995  * Abort the TCP.  Drop the connection abruptly.
 996  */
 997 static void
 998 tcp_usr_abort(struct socket *so)
 999 {
1000         struct inpcb *inp;
1001         struct tcpcb *tp = NULL;
1002         TCPDEBUG0;
1003
1004         inp = sotoinpcb(so);
1005         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1006
1007         INP_INFO_RLOCK(&V_tcbinfo);
1008         INP_WLOCK(inp);
1009         KASSERT(inp->inp_socket != NULL,
1010             ("tcp_usr_abort: inp_socket == NULL"));
1011
1012         /*
1013          * If we still have full TCP state, and we're not dropped, drop.
1014          */
1015         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1016             !(inp->inp_flags & INP_DROPPED)) {
1017                 tp = intotcpcb(inp);
1018                 TCPDEBUG1();
1019                 tcp_drop(tp, ECONNABORTED);
1020                 TCPDEBUG2(PRU_ABORT);
1021         }
1022         if (!(inp->inp_flags & INP_DROPPED)) {
1023                 SOCK_LOCK(so);
1024                 so->so_state |= SS_PROTOREF;
1025                 SOCK_UNLOCK(so);
1026                 inp->inp_flags |= INP_SOCKREF;
1027         }
1028         INP_WUNLOCK(inp);
1029         INP_INFO_RUNLOCK(&V_tcbinfo);
1030 }
1031
1032 /*
1033  * TCP socket is closed.  Start friendly disconnect.
1034  */
1035 static void
1036 tcp_usr_close(struct socket *so)
1037 {
1038         struct inpcb *inp;
1039         struct tcpcb *tp = NULL;
1040         TCPDEBUG0;
1041
1042         inp = sotoinpcb(so);
1043         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1044
1045         INP_INFO_RLOCK(&V_tcbinfo);
1046         INP_WLOCK(inp);
1047         KASSERT(inp->inp_socket != NULL,
1048             ("tcp_usr_close: inp_socket == NULL"));
1049
1050         /*
1051          * If we still have full TCP state, and we're not dropped, initiate
1052          * a disconnect.
1053          */
1054         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1055             !(inp->inp_flags & INP_DROPPED)) {
1056                 tp = intotcpcb(inp);
1057                 TCPDEBUG1();
1058                 tcp_disconnect(tp);
1059                 TCPDEBUG2(PRU_CLOSE);
1060         }
1061         if (!(inp->inp_flags & INP_DROPPED)) {
1062                 SOCK_LOCK(so);
1063                 so->so_state |= SS_PROTOREF;
1064                 SOCK_UNLOCK(so);
1065                 inp->inp_flags |= INP_SOCKREF;
1066         }
1067         INP_WUNLOCK(inp);
1068         INP_INFO_RUNLOCK(&V_tcbinfo);
1069 }
1070
1071 /*
1072  * Receive out-of-band data.
1073  */
1074 static int
1075 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1076 {
1077         int error = 0;
1078         struct inpcb *inp;
1079         struct tcpcb *tp = NULL;
1080
1081         TCPDEBUG0;
1082         inp = sotoinpcb(so);
1083         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1084         INP_WLOCK(inp);
1085         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1086                 error = ECONNRESET;
1087                 goto out;
1088         }
1089         tp = intotcpcb(inp);
1090         TCPDEBUG1();
1091         if ((so->so_oobmark == 0 &&
1092              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1093             so->so_options & SO_OOBINLINE ||
1094             tp->t_oobflags & TCPOOB_HADDATA) {
1095                 error = EINVAL;
1096                 goto out;
1097         }
1098         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1099                 error = EWOULDBLOCK;
1100                 goto out;
1101         }
1102         m->m_len = 1;
1103         *mtod(m, caddr_t) = tp->t_iobc;
1104         if ((flags & MSG_PEEK) == 0)
1105                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1106
1107 out:
1108         TCPDEBUG2(PRU_RCVOOB);
1109         INP_WUNLOCK(inp);
1110         return (error);
1111 }
1112
1113 #ifdef INET
1114 struct pr_usrreqs tcp_usrreqs = {
1115         .pru_abort =            tcp_usr_abort,
1116         .pru_accept =           tcp_usr_accept,
1117         .pru_attach =           tcp_usr_attach,
1118         .pru_bind =             tcp_usr_bind,
1119         .pru_connect =          tcp_usr_connect,
1120         .pru_control =          in_control,
1121         .pru_detach =           tcp_usr_detach,
1122         .pru_disconnect =       tcp_usr_disconnect,
1123         .pru_listen =           tcp_usr_listen,
1124         .pru_peeraddr =         in_getpeeraddr,
1125         .pru_rcvd =             tcp_usr_rcvd,
1126         .pru_rcvoob =           tcp_usr_rcvoob,
1127         .pru_send =             tcp_usr_send,
1128         .pru_shutdown =         tcp_usr_shutdown,
1129         .pru_sockaddr =         in_getsockaddr,
1130         .pru_sosetlabel =       in_pcbsosetlabel,
1131         .pru_close =            tcp_usr_close,
1132 };
1133 #endif /* INET */
1134
1135 #ifdef INET6
1136 struct pr_usrreqs tcp6_usrreqs = {
1137         .pru_abort =            tcp_usr_abort,
1138         .pru_accept =           tcp6_usr_accept,
1139         .pru_attach =           tcp_usr_attach,
1140         .pru_bind =             tcp6_usr_bind,
1141         .pru_connect =          tcp6_usr_connect,
1142         .pru_control =          in6_control,
1143         .pru_detach =           tcp_usr_detach,
1144         .pru_disconnect =       tcp_usr_disconnect,
1145         .pru_listen =           tcp6_usr_listen,
1146         .pru_peeraddr =         in6_mapped_peeraddr,
1147         .pru_rcvd =             tcp_usr_rcvd,
1148         .pru_rcvoob =           tcp_usr_rcvoob,
1149         .pru_send =             tcp_usr_send,
1150         .pru_shutdown =         tcp_usr_shutdown,
1151         .pru_sockaddr =         in6_mapped_sockaddr,
1152         .pru_sosetlabel =       in_pcbsosetlabel,
1153         .pru_close =            tcp_usr_close,
1154 };
1155 #endif /* INET6 */
1156
1157 #ifdef INET
1158 /*
1159  * Common subroutine to open a TCP connection to remote host specified
1160  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1161  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1162  * to choose a local host address (interface).  If there is an existing
1163  * incarnation of the same connection in TIME-WAIT state and if the remote
1164  * host was sending CC options and if the connection duration was < MSL, then
1165  * truncate the previous TIME-WAIT state and proceed.
1166  * Initialize connection parameters and enter SYN-SENT state.
1167  */
1168 static int
1169 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1170 {
1171         struct inpcb *inp = tp->t_inpcb, *oinp;
1172         struct socket *so = inp->inp_socket;
1173         struct in_addr laddr;
1174         u_short lport;
1175         int error;
1176
1177         INP_WLOCK_ASSERT(inp);
1178         INP_HASH_WLOCK(&V_tcbinfo);
1179
1180         if (inp->inp_lport == 0) {
1181                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1182                 if (error)
1183                         goto out;
1184         }
1185
1186         /*
1187          * Cannot simply call in_pcbconnect, because there might be an
1188          * earlier incarnation of this same connection still in
1189          * TIME_WAIT state, creating an ADDRINUSE error.
1190          */
1191         laddr = inp->inp_laddr;
1192         lport = inp->inp_lport;
1193         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1194             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1195         if (error && oinp == NULL)
1196                 goto out;
1197         if (oinp) {
1198                 error = EADDRINUSE;
1199                 goto out;
1200         }
1201         inp->inp_laddr = laddr;
1202         in_pcbrehash(inp);
1203         INP_HASH_WUNLOCK(&V_tcbinfo);
1204
1205         /*
1206          * Compute window scaling to request:
1207          * Scale to fit into sweet spot.  See tcp_syncache.c.
1208          * XXX: This should move to tcp_output().
1209          */
1210         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1211             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1212                 tp->request_r_scale++;
1213
1214         soisconnecting(so);
1215         TCPSTAT_INC(tcps_connattempt);
1216         tcp_state_change(tp, TCPS_SYN_SENT);
1217         tp->iss = tcp_new_isn(tp);
1218         tcp_sendseqinit(tp);
1219
1220         return 0;
1221
1222 out:
1223         INP_HASH_WUNLOCK(&V_tcbinfo);
1224         return (error);
1225 }
1226 #endif /* INET */
1227
1228 #ifdef INET6
1229 static int
1230 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1231 {
1232         struct inpcb *inp = tp->t_inpcb, *oinp;
1233         struct socket *so = inp->inp_socket;
1234         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1235         struct in6_addr addr6;
1236         int error;
1237
1238         INP_WLOCK_ASSERT(inp);
1239         INP_HASH_WLOCK(&V_tcbinfo);
1240
1241         if (inp->inp_lport == 0) {
1242                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1243                 if (error)
1244                         goto out;
1245         }
1246
1247         /*
1248          * Cannot simply call in_pcbconnect, because there might be an
1249          * earlier incarnation of this same connection still in
1250          * TIME_WAIT state, creating an ADDRINUSE error.
1251          * in6_pcbladdr() also handles scope zone IDs.
1252          *
1253          * XXXRW: We wouldn't need to expose in6_pcblookup_hash_locked()
1254          * outside of in6_pcb.c if there were an in6_pcbconnect_setup().
1255          */
1256         error = in6_pcbladdr(inp, nam, &addr6);
1257         if (error)
1258                 goto out;
1259         oinp = in6_pcblookup_hash_locked(inp->inp_pcbinfo,
1260                                   &sin6->sin6_addr, sin6->sin6_port,
1261                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1262                                   ? &addr6
1263                                   : &inp->in6p_laddr,
1264                                   inp->inp_lport,  0, NULL);
1265         if (oinp) {
1266                 error = EADDRINUSE;
1267                 goto out;
1268         }
1269         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1270                 inp->in6p_laddr = addr6;
1271         inp->in6p_faddr = sin6->sin6_addr;
1272         inp->inp_fport = sin6->sin6_port;
1273         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1274         inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
1275         if (inp->inp_flags & IN6P_AUTOFLOWLABEL)
1276                 inp->inp_flow |=
1277                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1278         in_pcbrehash(inp);
1279         INP_HASH_WUNLOCK(&V_tcbinfo);
1280
1281         /* Compute window scaling to request.  */
1282         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1283             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1284                 tp->request_r_scale++;
1285
1286         soisconnecting(so);
1287         TCPSTAT_INC(tcps_connattempt);
1288         tcp_state_change(tp, TCPS_SYN_SENT);
1289         tp->iss = tcp_new_isn(tp);
1290         tcp_sendseqinit(tp);
1291
1292         return 0;
1293
1294 out:
1295         INP_HASH_WUNLOCK(&V_tcbinfo);
1296         return error;
1297 }
1298 #endif /* INET6 */
1299
1300 /*
1301  * Export TCP internal state information via a struct tcp_info, based on the
1302  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1303  * (TCP state machine, etc).  We export all information using FreeBSD-native
1304  * constants -- for example, the numeric values for tcpi_state will differ
1305  * from Linux.
1306  */
1307 static void
1308 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1309 {
1310
1311         INP_WLOCK_ASSERT(tp->t_inpcb);
1312         bzero(ti, sizeof(*ti));
1313
1314         ti->tcpi_state = tp->t_state;
1315         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1316                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1317         if (tp->t_flags & TF_SACK_PERMIT)
1318                 ti->tcpi_options |= TCPI_OPT_SACK;
1319         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1320                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1321                 ti->tcpi_snd_wscale = tp->snd_scale;
1322                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1323         }
1324
1325         ti->tcpi_rto = tp->t_rxtcur * tick;
1326         ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1327         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1328         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1329
1330         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1331         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1332
1333         /*
1334          * FreeBSD-specific extension fields for tcp_info.
1335          */
1336         ti->tcpi_rcv_space = tp->rcv_wnd;
1337         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1338         ti->tcpi_snd_wnd = tp->snd_wnd;
1339         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1340         ti->tcpi_snd_nxt = tp->snd_nxt;
1341         ti->tcpi_snd_mss = tp->t_maxseg;
1342         ti->tcpi_rcv_mss = tp->t_maxseg;
1343         if (tp->t_flags & TF_TOE)
1344                 ti->tcpi_options |= TCPI_OPT_TOE;
1345         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1346         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1347         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1348 }
1349
1350 /*
1351  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1352  * socket option arguments.  When it re-acquires the lock after the copy, it
1353  * has to revalidate that the connection is still valid for the socket
1354  * option.
1355  */
1356 #define INP_WLOCK_RECHECK(inp) do {                                     \
1357         INP_WLOCK(inp);                                                 \
1358         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1359                 INP_WUNLOCK(inp);                                       \
1360                 return (ECONNRESET);                                    \
1361         }                                                               \
1362         tp = intotcpcb(inp);                                            \
1363 } while(0)
1364
1365 int
1366 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1367 {
1368         int     error, opt, optval;
1369         u_int   ui;
1370         struct  inpcb *inp;
1371         struct  tcpcb *tp;
1372         struct  tcp_info ti;
1373         char buf[TCP_CA_NAME_MAX];
1374         struct cc_algo *algo;
1375
1376         error = 0;
1377         inp = sotoinpcb(so);
1378         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1379         INP_WLOCK(inp);
1380         if (sopt->sopt_level != IPPROTO_TCP) {
1381 #ifdef INET6
1382                 if (inp->inp_vflag & INP_IPV6PROTO) {
1383                         INP_WUNLOCK(inp);
1384                         error = ip6_ctloutput(so, sopt);
1385                 }
1386 #endif /* INET6 */
1387 #if defined(INET6) && defined(INET)
1388                 else
1389 #endif
1390 #ifdef INET
1391                 {
1392                         INP_WUNLOCK(inp);
1393                         error = ip_ctloutput(so, sopt);
1394                 }
1395 #endif
1396                 return (error);
1397         }
1398         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1399                 INP_WUNLOCK(inp);
1400                 return (ECONNRESET);
1401         }
1402
1403         switch (sopt->sopt_dir) {
1404         case SOPT_SET:
1405                 switch (sopt->sopt_name) {
1406 #ifdef TCP_SIGNATURE
1407                 case TCP_MD5SIG:
1408                         INP_WUNLOCK(inp);
1409                         error = sooptcopyin(sopt, &optval, sizeof optval,
1410                             sizeof optval);
1411                         if (error)
1412                                 return (error);
1413
1414                         INP_WLOCK_RECHECK(inp);
1415                         if (optval > 0)
1416                                 tp->t_flags |= TF_SIGNATURE;
1417                         else
1418                                 tp->t_flags &= ~TF_SIGNATURE;
1419                         goto unlock_and_done;
1420 #endif /* TCP_SIGNATURE */
1421
1422                 case TCP_NODELAY:
1423                 case TCP_NOOPT:
1424                         INP_WUNLOCK(inp);
1425                         error = sooptcopyin(sopt, &optval, sizeof optval,
1426                             sizeof optval);
1427                         if (error)
1428                                 return (error);
1429
1430                         INP_WLOCK_RECHECK(inp);
1431                         switch (sopt->sopt_name) {
1432                         case TCP_NODELAY:
1433                                 opt = TF_NODELAY;
1434                                 break;
1435                         case TCP_NOOPT:
1436                                 opt = TF_NOOPT;
1437                                 break;
1438                         default:
1439                                 opt = 0; /* dead code to fool gcc */
1440                                 break;
1441                         }
1442
1443                         if (optval)
1444                                 tp->t_flags |= opt;
1445                         else
1446                                 tp->t_flags &= ~opt;
1447 unlock_and_done:
1448 #ifdef TCP_OFFLOAD
1449                         if (tp->t_flags & TF_TOE) {
1450                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1451                                     sopt->sopt_name);
1452                         }
1453 #endif
1454                         INP_WUNLOCK(inp);
1455                         break;
1456
1457                 case TCP_NOPUSH:
1458                         INP_WUNLOCK(inp);
1459                         error = sooptcopyin(sopt, &optval, sizeof optval,
1460                             sizeof optval);
1461                         if (error)
1462                                 return (error);
1463
1464                         INP_WLOCK_RECHECK(inp);
1465                         if (optval)
1466                                 tp->t_flags |= TF_NOPUSH;
1467                         else if (tp->t_flags & TF_NOPUSH) {
1468                                 tp->t_flags &= ~TF_NOPUSH;
1469                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1470                                         error = tcp_output(tp);
1471                         }
1472                         goto unlock_and_done;
1473
1474                 case TCP_MAXSEG:
1475                         INP_WUNLOCK(inp);
1476                         error = sooptcopyin(sopt, &optval, sizeof optval,
1477                             sizeof optval);
1478                         if (error)
1479                                 return (error);
1480
1481                         INP_WLOCK_RECHECK(inp);
1482                         if (optval > 0 && optval <= tp->t_maxseg &&
1483                             optval + 40 >= V_tcp_minmss)
1484                                 tp->t_maxseg = optval;
1485                         else
1486                                 error = EINVAL;
1487                         goto unlock_and_done;
1488
1489                 case TCP_INFO:
1490                         INP_WUNLOCK(inp);
1491                         error = EINVAL;
1492                         break;
1493
1494                 case TCP_CONGESTION:
1495                         INP_WUNLOCK(inp);
1496                         bzero(buf, sizeof(buf));
1497                         error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1498                         if (error)
1499                                 break;
1500                         INP_WLOCK_RECHECK(inp);
1501                         /*
1502                          * Return EINVAL if we can't find the requested cc algo.
1503                          */
1504                         error = EINVAL;
1505                         CC_LIST_RLOCK();
1506                         STAILQ_FOREACH(algo, &cc_list, entries) {
1507                                 if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1508                                     == 0) {
1509                                         /* We've found the requested algo. */
1510                                         error = 0;
1511                                         /*
1512                                          * We hold a write lock over the tcb
1513                                          * so it's safe to do these things
1514                                          * without ordering concerns.
1515                                          */
1516                                         if (CC_ALGO(tp)->cb_destroy != NULL)
1517                                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1518                                         CC_ALGO(tp) = algo;
1519                                         /*
1520                                          * If something goes pear shaped
1521                                          * initialising the new algo,
1522                                          * fall back to newreno (which
1523                                          * does not require initialisation).
1524                                          */
1525                                         if (algo->cb_init != NULL)
1526                                                 if (algo->cb_init(tp->ccv) > 0) {
1527                                                         CC_ALGO(tp) = &newreno_cc_algo;
1528                                                         /*
1529                                                          * The only reason init
1530                                                          * should fail is
1531                                                          * because of malloc.
1532                                                          */
1533                                                         error = ENOMEM;
1534                                                 }
1535                                         break; /* Break the STAILQ_FOREACH. */
1536                                 }
1537                         }
1538                         CC_LIST_RUNLOCK();
1539                         goto unlock_and_done;
1540
1541                 case TCP_KEEPIDLE:
1542                 case TCP_KEEPINTVL:
1543                 case TCP_KEEPINIT:
1544                         INP_WUNLOCK(inp);
1545                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1546                         if (error)
1547                                 return (error);
1548
1549                         if (ui > (UINT_MAX / hz)) {
1550                                 error = EINVAL;
1551                                 break;
1552                         }
1553                         ui *= hz;
1554
1555                         INP_WLOCK_RECHECK(inp);
1556                         switch (sopt->sopt_name) {
1557                         case TCP_KEEPIDLE:
1558                                 tp->t_keepidle = ui;
1559                                 /*
1560                                  * XXX: better check current remaining
1561                                  * timeout and "merge" it with new value.
1562                                  */
1563                                 if ((tp->t_state > TCPS_LISTEN) &&
1564                                     (tp->t_state <= TCPS_CLOSING))
1565                                         tcp_timer_activate(tp, TT_KEEP,
1566                                             TP_KEEPIDLE(tp));
1567                                 break;
1568                         case TCP_KEEPINTVL:
1569                                 tp->t_keepintvl = ui;
1570                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1571                                     (TP_MAXIDLE(tp) > 0))
1572                                         tcp_timer_activate(tp, TT_2MSL,
1573                                             TP_MAXIDLE(tp));
1574                                 break;
1575                         case TCP_KEEPINIT:
1576                                 tp->t_keepinit = ui;
1577                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1578                                     tp->t_state == TCPS_SYN_SENT)
1579                                         tcp_timer_activate(tp, TT_KEEP,
1580                                             TP_KEEPINIT(tp));
1581                                 break;
1582                         }
1583                         goto unlock_and_done;
1584
1585                 case TCP_KEEPCNT:
1586                         INP_WUNLOCK(inp);
1587                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1588                         if (error)
1589                                 return (error);
1590
1591                         INP_WLOCK_RECHECK(inp);
1592                         tp->t_keepcnt = ui;
1593                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1594                             (TP_MAXIDLE(tp) > 0))
1595                                 tcp_timer_activate(tp, TT_2MSL,
1596                                     TP_MAXIDLE(tp));
1597                         goto unlock_and_done;
1598
1599 #ifdef TCP_RFC7413
1600                 case TCP_FASTOPEN:
1601                         INP_WUNLOCK(inp);
1602                         if (!V_tcp_fastopen_enabled)
1603                                 return (EPERM);
1604
1605                         error = sooptcopyin(sopt, &optval, sizeof optval,
1606                             sizeof optval);
1607                         if (error)
1608                                 return (error);
1609
1610                         INP_WLOCK_RECHECK(inp);
1611                         if (optval) {
1612                                 tp->t_flags |= TF_FASTOPEN;
1613                                 if ((tp->t_state == TCPS_LISTEN) &&
1614                                     (tp->t_tfo_pending == NULL))
1615                                         tp->t_tfo_pending =
1616                                             tcp_fastopen_alloc_counter();
1617                         } else
1618                                 tp->t_flags &= ~TF_FASTOPEN;
1619                         goto unlock_and_done;
1620 #endif
1621
1622                 default:
1623                         INP_WUNLOCK(inp);
1624                         error = ENOPROTOOPT;
1625                         break;
1626                 }
1627                 break;
1628
1629         case SOPT_GET:
1630                 tp = intotcpcb(inp);
1631                 switch (sopt->sopt_name) {
1632 #ifdef TCP_SIGNATURE
1633                 case TCP_MD5SIG:
1634                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1635                         INP_WUNLOCK(inp);
1636                         error = sooptcopyout(sopt, &optval, sizeof optval);
1637                         break;
1638 #endif
1639
1640                 case TCP_NODELAY:
1641                         optval = tp->t_flags & TF_NODELAY;
1642                         INP_WUNLOCK(inp);
1643                         error = sooptcopyout(sopt, &optval, sizeof optval);
1644                         break;
1645                 case TCP_MAXSEG:
1646                         optval = tp->t_maxseg;
1647                         INP_WUNLOCK(inp);
1648                         error = sooptcopyout(sopt, &optval, sizeof optval);
1649                         break;
1650                 case TCP_NOOPT:
1651                         optval = tp->t_flags & TF_NOOPT;
1652                         INP_WUNLOCK(inp);
1653                         error = sooptcopyout(sopt, &optval, sizeof optval);
1654                         break;
1655                 case TCP_NOPUSH:
1656                         optval = tp->t_flags & TF_NOPUSH;
1657                         INP_WUNLOCK(inp);
1658                         error = sooptcopyout(sopt, &optval, sizeof optval);
1659                         break;
1660                 case TCP_INFO:
1661                         tcp_fill_info(tp, &ti);
1662                         INP_WUNLOCK(inp);
1663                         error = sooptcopyout(sopt, &ti, sizeof ti);
1664                         break;
1665                 case TCP_CONGESTION:
1666                         bzero(buf, sizeof(buf));
1667                         strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1668                         INP_WUNLOCK(inp);
1669                         error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1670                         break;
1671                 case TCP_KEEPIDLE:
1672                 case TCP_KEEPINTVL:
1673                 case TCP_KEEPINIT:
1674                 case TCP_KEEPCNT:
1675                         switch (sopt->sopt_name) {
1676                         case TCP_KEEPIDLE:
1677                                 ui = tp->t_keepidle / hz;
1678                                 break;
1679                         case TCP_KEEPINTVL:
1680                                 ui = tp->t_keepintvl / hz;
1681                                 break;
1682                         case TCP_KEEPINIT:
1683                                 ui = tp->t_keepinit / hz;
1684                                 break;
1685                         case TCP_KEEPCNT:
1686                                 ui = tp->t_keepcnt;
1687                                 break;
1688                         }
1689                         INP_WUNLOCK(inp);
1690                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1691                         break;
1692 #ifdef TCP_RFC7413
1693                 case TCP_FASTOPEN:
1694                         optval = tp->t_flags & TF_FASTOPEN;
1695                         INP_WUNLOCK(inp);
1696                         error = sooptcopyout(sopt, &optval, sizeof optval);
1697                         break;
1698 #endif
1699                 default:
1700                         INP_WUNLOCK(inp);
1701                         error = ENOPROTOOPT;
1702                         break;
1703                 }
1704                 break;
1705         }
1706         return (error);
1707 }
1708 #undef INP_WLOCK_RECHECK
1709
1710 /*
1711  * Attach TCP protocol to socket, allocating
1712  * internet protocol control block, tcp control block,
1713  * bufer space, and entering LISTEN state if to accept connections.
1714  */
1715 static int
1716 tcp_attach(struct socket *so)
1717 {
1718         struct tcpcb *tp;
1719         struct inpcb *inp;
1720         int error;
1721
1722         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1723                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1724                 if (error)
1725                         return (error);
1726         }
1727         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1728         so->so_snd.sb_flags |= SB_AUTOSIZE;
1729         INP_INFO_RLOCK(&V_tcbinfo);
1730         error = in_pcballoc(so, &V_tcbinfo);
1731         if (error) {
1732                 INP_INFO_RUNLOCK(&V_tcbinfo);
1733                 return (error);
1734         }
1735         inp = sotoinpcb(so);
1736 #ifdef INET6
1737         if (inp->inp_vflag & INP_IPV6PROTO) {
1738                 inp->inp_vflag |= INP_IPV6;
1739                 inp->in6p_hops = -1;    /* use kernel default */
1740         }
1741         else
1742 #endif
1743         inp->inp_vflag |= INP_IPV4;
1744         tp = tcp_newtcpcb(inp);
1745         if (tp == NULL) {
1746                 in_pcbdetach(inp);
1747                 in_pcbfree(inp);
1748                 INP_INFO_RUNLOCK(&V_tcbinfo);
1749                 return (ENOBUFS);
1750         }
1751         tp->t_state = TCPS_CLOSED;
1752         INP_WUNLOCK(inp);
1753         INP_INFO_RUNLOCK(&V_tcbinfo);
1754         return (0);
1755 }
1756
1757 /*
1758  * Initiate (or continue) disconnect.
1759  * If embryonic state, just send reset (once).
1760  * If in ``let data drain'' option and linger null, just drop.
1761  * Otherwise (hard), mark socket disconnecting and drop
1762  * current input data; switch states based on user close, and
1763  * send segment to peer (with FIN).
1764  */
1765 static void
1766 tcp_disconnect(struct tcpcb *tp)
1767 {
1768         struct inpcb *inp = tp->t_inpcb;
1769         struct socket *so = inp->inp_socket;
1770
1771         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1772         INP_WLOCK_ASSERT(inp);
1773
1774         /*
1775          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1776          * socket is still open.
1777          */
1778         if (tp->t_state < TCPS_ESTABLISHED) {
1779                 tp = tcp_close(tp);
1780                 KASSERT(tp != NULL,
1781                     ("tcp_disconnect: tcp_close() returned NULL"));
1782         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1783                 tp = tcp_drop(tp, 0);
1784                 KASSERT(tp != NULL,
1785                     ("tcp_disconnect: tcp_drop() returned NULL"));
1786         } else {
1787                 soisdisconnecting(so);
1788                 sbflush(&so->so_rcv);
1789                 tcp_usrclosed(tp);
1790                 if (!(inp->inp_flags & INP_DROPPED))
1791                         tcp_output(tp);
1792         }
1793 }
1794
1795 /*
1796  * User issued close, and wish to trail through shutdown states:
1797  * if never received SYN, just forget it.  If got a SYN from peer,
1798  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1799  * If already got a FIN from peer, then almost done; go to LAST_ACK
1800  * state.  In all other cases, have already sent FIN to peer (e.g.
1801  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1802  * for peer to send FIN or not respond to keep-alives, etc.
1803  * We can let the user exit from the close as soon as the FIN is acked.
1804  */
1805 static void
1806 tcp_usrclosed(struct tcpcb *tp)
1807 {
1808
1809         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1810         INP_WLOCK_ASSERT(tp->t_inpcb);
1811
1812         switch (tp->t_state) {
1813         case TCPS_LISTEN:
1814 #ifdef TCP_OFFLOAD
1815                 tcp_offload_listen_stop(tp);
1816 #endif
1817                 tcp_state_change(tp, TCPS_CLOSED);
1818                 /* FALLTHROUGH */
1819         case TCPS_CLOSED:
1820                 tp = tcp_close(tp);
1821                 /*
1822                  * tcp_close() should never return NULL here as the socket is
1823                  * still open.
1824                  */
1825                 KASSERT(tp != NULL,
1826                     ("tcp_usrclosed: tcp_close() returned NULL"));
1827                 break;
1828
1829         case TCPS_SYN_SENT:
1830         case TCPS_SYN_RECEIVED:
1831                 tp->t_flags |= TF_NEEDFIN;
1832                 break;
1833
1834         case TCPS_ESTABLISHED:
1835                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1836                 break;
1837
1838         case TCPS_CLOSE_WAIT:
1839                 tcp_state_change(tp, TCPS_LAST_ACK);
1840                 break;
1841         }
1842         if (tp->t_state >= TCPS_FIN_WAIT_2) {
1843                 soisdisconnected(tp->t_inpcb->inp_socket);
1844                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1845                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1846                         int timeout;
1847
1848                         timeout = (tcp_fast_finwait2_recycle) ?
1849                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
1850                         tcp_timer_activate(tp, TT_2MSL, timeout);
1851                 }
1852         }
1853 }
1854
1855 #ifdef DDB
1856 static void
1857 db_print_indent(int indent)
1858 {
1859         int i;
1860
1861         for (i = 0; i < indent; i++)
1862                 db_printf(" ");
1863 }
1864
1865 static void
1866 db_print_tstate(int t_state)
1867 {
1868
1869         switch (t_state) {
1870         case TCPS_CLOSED:
1871                 db_printf("TCPS_CLOSED");
1872                 return;
1873
1874         case TCPS_LISTEN:
1875                 db_printf("TCPS_LISTEN");
1876                 return;
1877
1878         case TCPS_SYN_SENT:
1879                 db_printf("TCPS_SYN_SENT");
1880                 return;
1881
1882         case TCPS_SYN_RECEIVED:
1883                 db_printf("TCPS_SYN_RECEIVED");
1884                 return;
1885
1886         case TCPS_ESTABLISHED:
1887                 db_printf("TCPS_ESTABLISHED");
1888                 return;
1889
1890         case TCPS_CLOSE_WAIT:
1891                 db_printf("TCPS_CLOSE_WAIT");
1892                 return;
1893
1894         case TCPS_FIN_WAIT_1:
1895                 db_printf("TCPS_FIN_WAIT_1");
1896                 return;
1897
1898         case TCPS_CLOSING:
1899                 db_printf("TCPS_CLOSING");
1900                 return;
1901
1902         case TCPS_LAST_ACK:
1903                 db_printf("TCPS_LAST_ACK");
1904                 return;
1905
1906         case TCPS_FIN_WAIT_2:
1907                 db_printf("TCPS_FIN_WAIT_2");
1908                 return;
1909
1910         case TCPS_TIME_WAIT:
1911                 db_printf("TCPS_TIME_WAIT");
1912                 return;
1913
1914         default:
1915                 db_printf("unknown");
1916                 return;
1917         }
1918 }
1919
1920 static void
1921 db_print_tflags(u_int t_flags)
1922 {
1923         int comma;
1924
1925         comma = 0;
1926         if (t_flags & TF_ACKNOW) {
1927                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1928                 comma = 1;
1929         }
1930         if (t_flags & TF_DELACK) {
1931                 db_printf("%sTF_DELACK", comma ? ", " : "");
1932                 comma = 1;
1933         }
1934         if (t_flags & TF_NODELAY) {
1935                 db_printf("%sTF_NODELAY", comma ? ", " : "");
1936                 comma = 1;
1937         }
1938         if (t_flags & TF_NOOPT) {
1939                 db_printf("%sTF_NOOPT", comma ? ", " : "");
1940                 comma = 1;
1941         }
1942         if (t_flags & TF_SENTFIN) {
1943                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1944                 comma = 1;
1945         }
1946         if (t_flags & TF_REQ_SCALE) {
1947                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1948                 comma = 1;
1949         }
1950         if (t_flags & TF_RCVD_SCALE) {
1951                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1952                 comma = 1;
1953         }
1954         if (t_flags & TF_REQ_TSTMP) {
1955                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1956                 comma = 1;
1957         }
1958         if (t_flags & TF_RCVD_TSTMP) {
1959                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1960                 comma = 1;
1961         }
1962         if (t_flags & TF_SACK_PERMIT) {
1963                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1964                 comma = 1;
1965         }
1966         if (t_flags & TF_NEEDSYN) {
1967                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1968                 comma = 1;
1969         }
1970         if (t_flags & TF_NEEDFIN) {
1971                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1972                 comma = 1;
1973         }
1974         if (t_flags & TF_NOPUSH) {
1975                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1976                 comma = 1;
1977         }
1978         if (t_flags & TF_MORETOCOME) {
1979                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1980                 comma = 1;
1981         }
1982         if (t_flags & TF_LQ_OVERFLOW) {
1983                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1984                 comma = 1;
1985         }
1986         if (t_flags & TF_LASTIDLE) {
1987                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1988                 comma = 1;
1989         }
1990         if (t_flags & TF_RXWIN0SENT) {
1991                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1992                 comma = 1;
1993         }
1994         if (t_flags & TF_FASTRECOVERY) {
1995                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1996                 comma = 1;
1997         }
1998         if (t_flags & TF_CONGRECOVERY) {
1999                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2000                 comma = 1;
2001         }
2002         if (t_flags & TF_WASFRECOVERY) {
2003                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2004                 comma = 1;
2005         }
2006         if (t_flags & TF_SIGNATURE) {
2007                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2008                 comma = 1;
2009         }
2010         if (t_flags & TF_FORCEDATA) {
2011                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2012                 comma = 1;
2013         }
2014         if (t_flags & TF_TSO) {
2015                 db_printf("%sTF_TSO", comma ? ", " : "");
2016                 comma = 1;
2017         }
2018         if (t_flags & TF_ECN_PERMIT) {
2019                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2020                 comma = 1;
2021         }
2022         if (t_flags & TF_FASTOPEN) {
2023                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2024                 comma = 1;
2025         }
2026 }
2027
2028 static void
2029 db_print_toobflags(char t_oobflags)
2030 {
2031         int comma;
2032
2033         comma = 0;
2034         if (t_oobflags & TCPOOB_HAVEDATA) {
2035                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2036                 comma = 1;
2037         }
2038         if (t_oobflags & TCPOOB_HADDATA) {
2039                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2040                 comma = 1;
2041         }
2042 }
2043
2044 static void
2045 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2046 {
2047
2048         db_print_indent(indent);
2049         db_printf("%s at %p\n", name, tp);
2050
2051         indent += 2;
2052
2053         db_print_indent(indent);
2054         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2055            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2056
2057         db_print_indent(indent);
2058         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2059             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2060
2061         db_print_indent(indent);
2062         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2063             &tp->t_timers->tt_delack, tp->t_inpcb);
2064
2065         db_print_indent(indent);
2066         db_printf("t_state: %d (", tp->t_state);
2067         db_print_tstate(tp->t_state);
2068         db_printf(")\n");
2069
2070         db_print_indent(indent);
2071         db_printf("t_flags: 0x%x (", tp->t_flags);
2072         db_print_tflags(tp->t_flags);
2073         db_printf(")\n");
2074
2075         db_print_indent(indent);
2076         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2077             tp->snd_una, tp->snd_max, tp->snd_nxt);
2078
2079         db_print_indent(indent);
2080         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2081            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2082
2083         db_print_indent(indent);
2084         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2085             tp->iss, tp->irs, tp->rcv_nxt);
2086
2087         db_print_indent(indent);
2088         db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2089             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2090
2091         db_print_indent(indent);
2092         db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2093            tp->snd_wnd, tp->snd_cwnd);
2094
2095         db_print_indent(indent);
2096         db_printf("snd_ssthresh: %lu   snd_recover: "
2097             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2098
2099         db_print_indent(indent);
2100         db_printf("t_maxopd: %u   t_rcvtime: %u   t_startime: %u\n",
2101             tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
2102
2103         db_print_indent(indent);
2104         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2105             tp->t_rtttime, tp->t_rtseq);
2106
2107         db_print_indent(indent);
2108         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2109             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2110
2111         db_print_indent(indent);
2112         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2113             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2114             tp->t_rttbest);
2115
2116         db_print_indent(indent);
2117         db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2118             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2119
2120         db_print_indent(indent);
2121         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2122         db_print_toobflags(tp->t_oobflags);
2123         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2124
2125         db_print_indent(indent);
2126         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2127             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2128
2129         db_print_indent(indent);
2130         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2131             tp->ts_recent, tp->ts_recent_age);
2132
2133         db_print_indent(indent);
2134         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2135             "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2136
2137         db_print_indent(indent);
2138         db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2139             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2140             tp->snd_recover_prev, tp->t_badrxtwin);
2141
2142         db_print_indent(indent);
2143         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2144             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2145
2146         db_print_indent(indent);
2147         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2148             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2149
2150         /* Skip sackblks, sackhint. */
2151
2152         db_print_indent(indent);
2153         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2154             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2155 }
2156
2157 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2158 {
2159         struct tcpcb *tp;
2160
2161         if (!have_addr) {
2162                 db_printf("usage: show tcpcb <addr>\n");
2163                 return;
2164         }
2165         tp = (struct tcpcb *)addr;
2166
2167         db_print_tcpcb(tp, "tcpcb", 0);
2168 }
2169 #endif