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