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
  62 #ifdef DDB
  63 #include <ddb/ddb.h>
  64 #endif
  65
  66 #include <net/if.h>
  67 #include <net/route.h>
  68 #include <net/vnet.h>
  69
  70 #include <netinet/cc.h>
  71 #include <netinet/in.h>
  72 #include <netinet/in_pcb.h>
  73 #include <netinet/in_systm.h>
  74 #include <netinet/in_var.h>
  75 #include <netinet/ip_var.h>
  76 #ifdef INET6
  77 #include <netinet/ip6.h>
  78 #include <netinet6/in6_pcb.h>
  79 #include <netinet6/ip6_var.h>
  80 #include <netinet6/scope6_var.h>
  81 #endif
  82 #ifdef TCP_RFC7413
  83 #include <netinet/tcp_fastopen.h>
  84 #endif
  85 #include <netinet/tcp_fsm.h>
  86 #include <netinet/tcp_seq.h>
  87 #include <netinet/tcp_timer.h>
  88 #include <netinet/tcp_var.h>
  89 #include <netinet/tcpip.h>
  90 #ifdef TCPDEBUG
  91 #include <netinet/tcp_debug.h>
  92 #endif
  93 #ifdef TCP_OFFLOAD
  94 #include <netinet/tcp_offload.h>
  95 #endif
  96
  97 /*
  98  * TCP protocol interface to socket abstraction.
  99  */
 100 static int      tcp_attach(struct socket *);
 101 #ifdef INET
 102 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 103                     struct thread *td);
 104 #endif /* INET */
 105 #ifdef INET6
 106 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 107                     struct thread *td);
 108 #endif /* INET6 */
 109 static void     tcp_disconnect(struct tcpcb *);
 110 static void     tcp_usrclosed(struct tcpcb *);
 111 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 112
 113 #ifdef TCPDEBUG
 114 #define TCPDEBUG0       int ostate = 0
 115 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 116 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 117                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 118 #else
 119 #define TCPDEBUG0
 120 #define TCPDEBUG1()
 121 #define TCPDEBUG2(req)
 122 #endif
 123
 124 /*
 125  * TCP attaches to socket via pru_attach(), reserving space,
 126  * and an internet control block.
 127  */
 128 static int
 129 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 130 {
 131         struct inpcb *inp;
 132         struct tcpcb *tp = NULL;
 133         int error;
 134         TCPDEBUG0;
 135
 136         inp = sotoinpcb(so);
 137         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 138         TCPDEBUG1();
 139
 140         error = tcp_attach(so);
 141         if (error)
 142                 goto out;
 143
 144         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 145                 so->so_linger = TCP_LINGERTIME;
 146
 147         inp = sotoinpcb(so);
 148         tp = intotcpcb(inp);
 149 out:
 150         TCPDEBUG2(PRU_ATTACH);
 151         return error;
 152 }
 153
 154 /*
 155  * tcp_detach is called when the socket layer loses its final reference
 156  * to the socket, be it a file descriptor reference, a reference from TCP,
 157  * etc.  At this point, there is only one case in which we will keep around
 158  * inpcb state: time wait.
 159  *
 160  * This function can probably be re-absorbed back into tcp_usr_detach() now
 161  * that there is a single detach path.
 162  */
 163 static void
 164 tcp_detach(struct socket *so, struct inpcb *inp)
 165 {
 166         struct tcpcb *tp;
 167
 168         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 169         INP_WLOCK_ASSERT(inp);
 170
 171         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 172         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 173
 174         tp = intotcpcb(inp);
 175
 176         if (inp->inp_flags & INP_TIMEWAIT) {
 177                 /*
 178                  * There are two cases to handle: one in which the time wait
 179                  * state is being discarded (INP_DROPPED), and one in which
 180                  * this connection will remain in timewait.  In the former,
 181                  * it is time to discard all state (except tcptw, which has
 182                  * already been discarded by the timewait close code, which
 183                  * should be further up the call stack somewhere).  In the
 184                  * latter case, we detach from the socket, but leave the pcb
 185                  * present until timewait ends.
 186                  *
 187                  * XXXRW: Would it be cleaner to free the tcptw here?
 188                  *
 189                  * Astute question indeed, from twtcp perspective there are
 190                  * three cases to consider:
 191                  *
 192                  * #1 tcp_detach is called at tcptw creation time by
 193                  *  tcp_twstart, then do not discard the newly created tcptw
 194                  *  and leave inpcb present until timewait ends
 195                  * #2 tcp_detach is called at timewait end (or reuse) by
 196                  *  tcp_twclose, then the tcptw has already been discarded
 197                  *  and inpcb is freed here
 198                  * #3 tcp_detach is called() after timewait ends (or reuse)
 199                  *  (e.g. by soclose), then tcptw has already been discarded
 200                  *  and inpcb is freed here
 201                  *
 202                  *  In all three cases the tcptw should not be freed here.
 203                  */
 204                 if (inp->inp_flags & INP_DROPPED) {
 205                         KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
 206                             "INP_DROPPED && tp != NULL"));
 207                         in_pcbdetach(inp);
 208                         in_pcbfree(inp);
 209                 } else {
 210                         in_pcbdetach(inp);
 211                         INP_WUNLOCK(inp);
 212                 }
 213         } else {
 214                 /*
 215                  * If the connection is not in timewait, we consider two
 216                  * two conditions: one in which no further processing is
 217                  * necessary (dropped || embryonic), and one in which TCP is
 218                  * not yet done, but no longer requires the socket, so the
 219                  * pcb will persist for the time being.
 220                  *
 221                  * XXXRW: Does the second case still occur?
 222                  */
 223                 if (inp->inp_flags & INP_DROPPED ||
 224                     tp->t_state < TCPS_SYN_SENT) {
 225                         tcp_discardcb(tp);
 226                         in_pcbdetach(inp);
 227                         in_pcbfree(inp);
 228                 } else {
 229                         in_pcbdetach(inp);
 230                         INP_WUNLOCK(inp);
 231                 }
 232         }
 233 }
 234
 235 /*
 236  * pru_detach() detaches the TCP protocol from the socket.
 237  * If the protocol state is non-embryonic, then can't
 238  * do this directly: have to initiate a pru_disconnect(),
 239  * which may finish later; embryonic TCB's can just
 240  * be discarded here.
 241  */
 242 static void
 243 tcp_usr_detach(struct socket *so)
 244 {
 245         struct inpcb *inp;
 246
 247         inp = sotoinpcb(so);
 248         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 249         INP_INFO_WLOCK(&V_tcbinfo);
 250         INP_WLOCK(inp);
 251         KASSERT(inp->inp_socket != NULL,
 252             ("tcp_usr_detach: inp_socket == NULL"));
 253         tcp_detach(so, inp);
 254         INP_INFO_WUNLOCK(&V_tcbinfo);
 255 }
 256
 257 #ifdef INET
 258 /*
 259  * Give the socket an address.
 260  */
 261 static int
 262 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 263 {
 264         int error = 0;
 265         struct inpcb *inp;
 266         struct tcpcb *tp = NULL;
 267         struct sockaddr_in *sinp;
 268
 269         sinp = (struct sockaddr_in *)nam;
 270         if (nam->sa_len != sizeof (*sinp))
 271                 return (EINVAL);
 272         /*
 273          * Must check for multicast addresses and disallow binding
 274          * to them.
 275          */
 276         if (sinp->sin_family == AF_INET &&
 277             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 278                 return (EAFNOSUPPORT);
 279
 280         TCPDEBUG0;
 281         inp = sotoinpcb(so);
 282         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 283         INP_WLOCK(inp);
 284         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 285                 error = EINVAL;
 286                 goto out;
 287         }
 288         tp = intotcpcb(inp);
 289         TCPDEBUG1();
 290         INP_HASH_WLOCK(&V_tcbinfo);
 291         error = in_pcbbind(inp, nam, td->td_ucred);
 292         INP_HASH_WUNLOCK(&V_tcbinfo);
 293 out:
 294         TCPDEBUG2(PRU_BIND);
 295         INP_WUNLOCK(inp);
 296
 297         return (error);
 298 }
 299 #endif /* INET */
 300
 301 #ifdef INET6
 302 static int
 303 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 304 {
 305         int error = 0;
 306         struct inpcb *inp;
 307         struct tcpcb *tp = NULL;
 308         struct sockaddr_in6 *sin6p;
 309
 310         sin6p = (struct sockaddr_in6 *)nam;
 311         if (nam->sa_len != sizeof (*sin6p))
 312                 return (EINVAL);
 313         /*
 314          * Must check for multicast addresses and disallow binding
 315          * to them.
 316          */
 317         if (sin6p->sin6_family == AF_INET6 &&
 318             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 319                 return (EAFNOSUPPORT);
 320
 321         TCPDEBUG0;
 322         inp = sotoinpcb(so);
 323         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 324         INP_WLOCK(inp);
 325         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 326                 error = EINVAL;
 327                 goto out;
 328         }
 329         tp = intotcpcb(inp);
 330         TCPDEBUG1();
 331         INP_HASH_WLOCK(&V_tcbinfo);
 332         inp->inp_vflag &= ~INP_IPV4;
 333         inp->inp_vflag |= INP_IPV6;
 334 #ifdef INET
 335         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 336                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 337                         inp->inp_vflag |= INP_IPV4;
 338                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 339                         struct sockaddr_in sin;
 340
 341                         in6_sin6_2_sin(&sin, sin6p);
 342                         inp->inp_vflag |= INP_IPV4;
 343                         inp->inp_vflag &= ~INP_IPV6;
 344                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 345                             td->td_ucred);
 346                         INP_HASH_WUNLOCK(&V_tcbinfo);
 347                         goto out;
 348                 }
 349         }
 350 #endif
 351         error = in6_pcbbind(inp, nam, td->td_ucred);
 352         INP_HASH_WUNLOCK(&V_tcbinfo);
 353 out:
 354         TCPDEBUG2(PRU_BIND);
 355         INP_WUNLOCK(inp);
 356         return (error);
 357 }
 358 #endif /* INET6 */
 359
 360 #ifdef INET
 361 /*
 362  * Prepare to accept connections.
 363  */
 364 static int
 365 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 366 {
 367         int error = 0;
 368         struct inpcb *inp;
 369         struct tcpcb *tp = NULL;
 370
 371         TCPDEBUG0;
 372         inp = sotoinpcb(so);
 373         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 374         INP_WLOCK(inp);
 375         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 376                 error = EINVAL;
 377                 goto out;
 378         }
 379         tp = intotcpcb(inp);
 380         TCPDEBUG1();
 381         SOCK_LOCK(so);
 382         error = solisten_proto_check(so);
 383         INP_HASH_WLOCK(&V_tcbinfo);
 384         if (error == 0 && inp->inp_lport == 0)
 385                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 386         INP_HASH_WUNLOCK(&V_tcbinfo);
 387         if (error == 0) {
 388                 tcp_state_change(tp, TCPS_LISTEN);
 389                 solisten_proto(so, backlog);
 390 #ifdef TCP_OFFLOAD
 391                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 392                         tcp_offload_listen_start(tp);
 393 #endif
 394         }
 395         SOCK_UNLOCK(so);
 396
 397 #ifdef TCP_RFC7413
 398         if (tp->t_flags & TF_FASTOPEN)
 399                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 400 #endif
 401 out:
 402         TCPDEBUG2(PRU_LISTEN);
 403         INP_WUNLOCK(inp);
 404         return (error);
 405 }
 406 #endif /* INET */
 407
 408 #ifdef INET6
 409 static int
 410 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 411 {
 412         int error = 0;
 413         struct inpcb *inp;
 414         struct tcpcb *tp = NULL;
 415
 416         TCPDEBUG0;
 417         inp = sotoinpcb(so);
 418         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 419         INP_WLOCK(inp);
 420         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 421                 error = EINVAL;
 422                 goto out;
 423         }
 424         tp = intotcpcb(inp);
 425         TCPDEBUG1();
 426         SOCK_LOCK(so);
 427         error = solisten_proto_check(so);
 428         INP_HASH_WLOCK(&V_tcbinfo);
 429         if (error == 0 && inp->inp_lport == 0) {
 430                 inp->inp_vflag &= ~INP_IPV4;
 431                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 432                         inp->inp_vflag |= INP_IPV4;
 433                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 434         }
 435         INP_HASH_WUNLOCK(&V_tcbinfo);
 436         if (error == 0) {
 437                 tcp_state_change(tp, TCPS_LISTEN);
 438                 solisten_proto(so, backlog);
 439 #ifdef TCP_OFFLOAD
 440                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 441                         tcp_offload_listen_start(tp);
 442 #endif
 443         }
 444         SOCK_UNLOCK(so);
 445
 446 #ifdef TCP_RFC7413
 447         if (tp->t_flags & TF_FASTOPEN)
 448                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 449 #endif
 450 out:
 451         TCPDEBUG2(PRU_LISTEN);
 452         INP_WUNLOCK(inp);
 453         return (error);
 454 }
 455 #endif /* INET6 */
 456
 457 #ifdef INET
 458 /*
 459  * Initiate connection to peer.
 460  * Create a template for use in transmissions on this connection.
 461  * Enter SYN_SENT state, and mark socket as connecting.
 462  * Start keep-alive timer, and seed output sequence space.
 463  * Send initial segment on connection.
 464  */
 465 static int
 466 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 467 {
 468         int error = 0;
 469         struct inpcb *inp;
 470         struct tcpcb *tp = NULL;
 471         struct sockaddr_in *sinp;
 472
 473         sinp = (struct sockaddr_in *)nam;
 474         if (nam->sa_len != sizeof (*sinp))
 475                 return (EINVAL);
 476         /*
 477          * Must disallow TCP ``connections'' to multicast addresses.
 478          */
 479         if (sinp->sin_family == AF_INET
 480             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 481                 return (EAFNOSUPPORT);
 482         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 483                 return (error);
 484
 485         TCPDEBUG0;
 486         inp = sotoinpcb(so);
 487         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 488         INP_WLOCK(inp);
 489         if (inp->inp_flags & INP_TIMEWAIT) {
 490                 error = EADDRINUSE;
 491                 goto out;
 492         }
 493         if (inp->inp_flags & INP_DROPPED) {
 494                 error = ECONNREFUSED;
 495                 goto out;
 496         }
 497         tp = intotcpcb(inp);
 498         TCPDEBUG1();
 499         if ((error = tcp_connect(tp, nam, td)) != 0)
 500                 goto out;
 501 #ifdef TCP_OFFLOAD
 502         if (registered_toedevs > 0 &&
 503             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 504             (error = tcp_offload_connect(so, nam)) == 0)
 505                 goto out;
 506 #endif
 507         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 508         error = tcp_output(tp);
 509 out:
 510         TCPDEBUG2(PRU_CONNECT);
 511         INP_WUNLOCK(inp);
 512         return (error);
 513 }
 514 #endif /* INET */
 515
 516 #ifdef INET6
 517 static int
 518 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 519 {
 520         int error = 0;
 521         struct inpcb *inp;
 522         struct tcpcb *tp = NULL;
 523         struct sockaddr_in6 *sin6p;
 524
 525         TCPDEBUG0;
 526
 527         sin6p = (struct sockaddr_in6 *)nam;
 528         if (nam->sa_len != sizeof (*sin6p))
 529                 return (EINVAL);
 530         /*
 531          * Must disallow TCP ``connections'' to multicast addresses.
 532          */
 533         if (sin6p->sin6_family == AF_INET6
 534             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 535                 return (EAFNOSUPPORT);
 536
 537         inp = sotoinpcb(so);
 538         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 539         INP_WLOCK(inp);
 540         if (inp->inp_flags & INP_TIMEWAIT) {
 541                 error = EADDRINUSE;
 542                 goto out;
 543         }
 544         if (inp->inp_flags & INP_DROPPED) {
 545                 error = ECONNREFUSED;
 546                 goto out;
 547         }
 548         tp = intotcpcb(inp);
 549         TCPDEBUG1();
 550 #ifdef INET
 551         /*
 552          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 553          * therefore probably require the hash lock, which isn't held here.
 554          * Is this a significant problem?
 555          */
 556         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 557                 struct sockaddr_in sin;
 558
 559                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 560                         error = EINVAL;
 561                         goto out;
 562                 }
 563
 564                 in6_sin6_2_sin(&sin, sin6p);
 565                 inp->inp_vflag |= INP_IPV4;
 566                 inp->inp_vflag &= ~INP_IPV6;
 567                 if ((error = prison_remote_ip4(td->td_ucred,
 568                     &sin.sin_addr)) != 0)
 569                         goto out;
 570                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 571                         goto out;
 572 #ifdef TCP_OFFLOAD
 573                 if (registered_toedevs > 0 &&
 574                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 575                     (error = tcp_offload_connect(so, nam)) == 0)
 576                         goto out;
 577 #endif
 578                 error = tcp_output(tp);
 579                 goto out;
 580         }
 581 #endif
 582         inp->inp_vflag &= ~INP_IPV4;
 583         inp->inp_vflag |= INP_IPV6;
 584         inp->inp_inc.inc_flags |= INC_ISIPV6;
 585         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 586                 goto out;
 587         if ((error = tcp6_connect(tp, nam, 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         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 596         error = tcp_output(tp);
 597
 598 out:
 599         TCPDEBUG2(PRU_CONNECT);
 600         INP_WUNLOCK(inp);
 601         return (error);
 602 }
 603 #endif /* INET6 */
 604
 605 /*
 606  * Initiate disconnect from peer.
 607  * If connection never passed embryonic stage, just drop;
 608  * else if don't need to let data drain, then can just drop anyways,
 609  * else have to begin TCP shutdown process: mark socket disconnecting,
 610  * drain unread data, state switch to reflect user close, and
 611  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 612  * when peer sends FIN and acks ours.
 613  *
 614  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 615  */
 616 static int
 617 tcp_usr_disconnect(struct socket *so)
 618 {
 619         struct inpcb *inp;
 620         struct tcpcb *tp = NULL;
 621         int error = 0;
 622
 623         TCPDEBUG0;
 624         INP_INFO_WLOCK(&V_tcbinfo);
 625         inp = sotoinpcb(so);
 626         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 627         INP_WLOCK(inp);
 628         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 629                 error = ECONNRESET;
 630                 goto out;
 631         }
 632         tp = intotcpcb(inp);
 633         TCPDEBUG1();
 634         tcp_disconnect(tp);
 635 out:
 636         TCPDEBUG2(PRU_DISCONNECT);
 637         INP_WUNLOCK(inp);
 638         INP_INFO_WUNLOCK(&V_tcbinfo);
 639         return (error);
 640 }
 641
 642 #ifdef INET
 643 /*
 644  * Accept a connection.  Essentially all the work is done at higher levels;
 645  * just return the address of the peer, storing through addr.
 646  *
 647  * The rationale for acquiring the tcbinfo lock here is somewhat complicated,
 648  * and is described in detail in the commit log entry for r175612.  Acquiring
 649  * it delays an accept(2) racing with sonewconn(), which inserts the socket
 650  * before the inpcb address/port fields are initialized.  A better fix would
 651  * prevent the socket from being placed in the listen queue until all fields
 652  * are fully initialized.
 653  */
 654 static int
 655 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 656 {
 657         int error = 0;
 658         struct inpcb *inp = NULL;
 659         struct tcpcb *tp = NULL;
 660         struct in_addr addr;
 661         in_port_t port = 0;
 662         TCPDEBUG0;
 663
 664         if (so->so_state & SS_ISDISCONNECTED)
 665                 return (ECONNABORTED);
 666
 667         inp = sotoinpcb(so);
 668         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 669         INP_INFO_RLOCK(&V_tcbinfo);
 670         INP_WLOCK(inp);
 671         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 672                 error = ECONNABORTED;
 673                 goto out;
 674         }
 675         tp = intotcpcb(inp);
 676         TCPDEBUG1();
 677
 678         /*
 679          * We inline in_getpeeraddr and COMMON_END here, so that we can
 680          * copy the data of interest and defer the malloc until after we
 681          * release the lock.
 682          */
 683         port = inp->inp_fport;
 684         addr = inp->inp_faddr;
 685
 686 out:
 687         TCPDEBUG2(PRU_ACCEPT);
 688         INP_WUNLOCK(inp);
 689         INP_INFO_RUNLOCK(&V_tcbinfo);
 690         if (error == 0)
 691                 *nam = in_sockaddr(port, &addr);
 692         return error;
 693 }
 694 #endif /* INET */
 695
 696 #ifdef INET6
 697 static int
 698 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 699 {
 700         struct inpcb *inp = NULL;
 701         int error = 0;
 702         struct tcpcb *tp = NULL;
 703         struct in_addr addr;
 704         struct in6_addr addr6;
 705         in_port_t port = 0;
 706         int v4 = 0;
 707         TCPDEBUG0;
 708
 709         if (so->so_state & SS_ISDISCONNECTED)
 710                 return (ECONNABORTED);
 711
 712         inp = sotoinpcb(so);
 713         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 714         INP_INFO_RLOCK(&V_tcbinfo);
 715         INP_WLOCK(inp);
 716         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 717                 error = ECONNABORTED;
 718                 goto out;
 719         }
 720         tp = intotcpcb(inp);
 721         TCPDEBUG1();
 722
 723         /*
 724          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 725          * copy the data of interest and defer the malloc until after we
 726          * release the lock.
 727          */
 728         if (inp->inp_vflag & INP_IPV4) {
 729                 v4 = 1;
 730                 port = inp->inp_fport;
 731                 addr = inp->inp_faddr;
 732         } else {
 733                 port = inp->inp_fport;
 734                 addr6 = inp->in6p_faddr;
 735         }
 736
 737 out:
 738         TCPDEBUG2(PRU_ACCEPT);
 739         INP_WUNLOCK(inp);
 740         INP_INFO_RUNLOCK(&V_tcbinfo);
 741         if (error == 0) {
 742                 if (v4)
 743                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 744                 else
 745                         *nam = in6_sockaddr(port, &addr6);
 746         }
 747         return error;
 748 }
 749 #endif /* INET6 */
 750
 751 /*
 752  * Mark the connection as being incapable of further output.
 753  */
 754 static int
 755 tcp_usr_shutdown(struct socket *so)
 756 {
 757         int error = 0;
 758         struct inpcb *inp;
 759         struct tcpcb *tp = NULL;
 760
 761         TCPDEBUG0;
 762         INP_INFO_WLOCK(&V_tcbinfo);
 763         inp = sotoinpcb(so);
 764         KASSERT(inp != NULL, ("inp == NULL"));
 765         INP_WLOCK(inp);
 766         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 767                 error = ECONNRESET;
 768                 goto out;
 769         }
 770         tp = intotcpcb(inp);
 771         TCPDEBUG1();
 772         socantsendmore(so);
 773         tcp_usrclosed(tp);
 774         if (!(inp->inp_flags & INP_DROPPED))
 775                 error = tcp_output(tp);
 776
 777 out:
 778         TCPDEBUG2(PRU_SHUTDOWN);
 779         INP_WUNLOCK(inp);
 780         INP_INFO_WUNLOCK(&V_tcbinfo);
 781
 782         return (error);
 783 }
 784
 785 /*
 786  * After a receive, possibly send window update to peer.
 787  */
 788 static int
 789 tcp_usr_rcvd(struct socket *so, int flags)
 790 {
 791         struct inpcb *inp;
 792         struct tcpcb *tp = NULL;
 793         int error = 0;
 794
 795         TCPDEBUG0;
 796         inp = sotoinpcb(so);
 797         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 798         INP_WLOCK(inp);
 799         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 800                 error = ECONNRESET;
 801                 goto out;
 802         }
 803         tp = intotcpcb(inp);
 804         TCPDEBUG1();
 805 #ifdef TCP_RFC7413
 806         /*
 807          * For passively-created TFO connections, don't attempt a window
 808          * update while still in SYN_RECEIVED as this may trigger an early
 809          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 810          * application response data, or failing that, when the DELACK timer
 811          * expires.
 812          */
 813         if ((tp->t_flags & TF_FASTOPEN) &&
 814             (tp->t_state == TCPS_SYN_RECEIVED))
 815                 goto out;
 816 #endif
 817 #ifdef TCP_OFFLOAD
 818         if (tp->t_flags & TF_TOE)
 819                 tcp_offload_rcvd(tp);
 820         else
 821 #endif
 822         tcp_output(tp);
 823
 824 out:
 825         TCPDEBUG2(PRU_RCVD);
 826         INP_WUNLOCK(inp);
 827         return (error);
 828 }
 829
 830 /*
 831  * Do a send by putting data in output queue and updating urgent
 832  * marker if URG set.  Possibly send more data.  Unlike the other
 833  * pru_*() routines, the mbuf chains are our responsibility.  We
 834  * must either enqueue them or free them.  The other pru_* routines
 835  * generally are caller-frees.
 836  */
 837 static int
 838 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 839     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 840 {
 841         int error = 0;
 842         struct inpcb *inp;
 843         struct tcpcb *tp = NULL;
 844 #ifdef INET6
 845         int isipv6;
 846 #endif
 847         TCPDEBUG0;
 848
 849         /*
 850          * We require the pcbinfo lock if we will close the socket as part of
 851          * this call.
 852          */
 853         if (flags & PRUS_EOF)
 854                 INP_INFO_WLOCK(&V_tcbinfo);
 855         inp = sotoinpcb(so);
 856         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 857         INP_WLOCK(inp);
 858         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 859                 if (control)
 860                         m_freem(control);
 861                 if (m)
 862                         m_freem(m);
 863                 error = ECONNRESET;
 864                 goto out;
 865         }
 866 #ifdef INET6
 867         isipv6 = nam && nam->sa_family == AF_INET6;
 868 #endif /* INET6 */
 869         tp = intotcpcb(inp);
 870         TCPDEBUG1();
 871         if (control) {
 872                 /* TCP doesn't do control messages (rights, creds, etc) */
 873                 if (control->m_len) {
 874                         m_freem(control);
 875                         if (m)
 876                                 m_freem(m);
 877                         error = EINVAL;
 878                         goto out;
 879                 }
 880                 m_freem(control);       /* empty control, just free it */
 881         }
 882         if (!(flags & PRUS_OOB)) {
 883                 sbappendstream(&so->so_snd, m);
 884                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 885                         /*
 886                          * Do implied connect if not yet connected,
 887                          * initialize window to default value, and
 888                          * initialize maxseg/maxopd using peer's cached
 889                          * MSS.
 890                          */
 891 #ifdef INET6
 892                         if (isipv6)
 893                                 error = tcp6_connect(tp, nam, td);
 894 #endif /* INET6 */
 895 #if defined(INET6) && defined(INET)
 896                         else
 897 #endif
 898 #ifdef INET
 899                                 error = tcp_connect(tp, nam, td);
 900 #endif
 901                         if (error)
 902                                 goto out;
 903                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 904                         tcp_mss(tp, -1);
 905                 }
 906                 if (flags & PRUS_EOF) {
 907                         /*
 908                          * Close the send side of the connection after
 909                          * the data is sent.
 910                          */
 911                         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 912                         socantsendmore(so);
 913                         tcp_usrclosed(tp);
 914                 }
 915                 if (!(inp->inp_flags & INP_DROPPED)) {
 916                         if (flags & PRUS_MORETOCOME)
 917                                 tp->t_flags |= TF_MORETOCOME;
 918                         error = tcp_output(tp);
 919                         if (flags & PRUS_MORETOCOME)
 920                                 tp->t_flags &= ~TF_MORETOCOME;
 921                 }
 922         } else {
 923                 /*
 924                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 925                  */
 926                 SOCKBUF_LOCK(&so->so_snd);
 927                 if (sbspace(&so->so_snd) < -512) {
 928                         SOCKBUF_UNLOCK(&so->so_snd);
 929                         m_freem(m);
 930                         error = ENOBUFS;
 931                         goto out;
 932                 }
 933                 /*
 934                  * According to RFC961 (Assigned Protocols),
 935                  * the urgent pointer points to the last octet
 936                  * of urgent data.  We continue, however,
 937                  * to consider it to indicate the first octet
 938                  * of data past the urgent section.
 939                  * Otherwise, snd_up should be one lower.
 940                  */
 941                 sbappendstream_locked(&so->so_snd, m);
 942                 SOCKBUF_UNLOCK(&so->so_snd);
 943                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 944                         /*
 945                          * Do implied connect if not yet connected,
 946                          * initialize window to default value, and
 947                          * initialize maxseg/maxopd using peer's cached
 948                          * MSS.
 949                          */
 950 #ifdef INET6
 951                         if (isipv6)
 952                                 error = tcp6_connect(tp, nam, td);
 953 #endif /* INET6 */
 954 #if defined(INET6) && defined(INET)
 955                         else
 956 #endif
 957 #ifdef INET
 958                                 error = tcp_connect(tp, nam, td);
 959 #endif
 960                         if (error)
 961                                 goto out;
 962                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 963                         tcp_mss(tp, -1);
 964                 }
 965                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 966                 tp->t_flags |= TF_FORCEDATA;
 967                 error = tcp_output(tp);
 968                 tp->t_flags &= ~TF_FORCEDATA;
 969         }
 970 out:
 971         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 972                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 973         INP_WUNLOCK(inp);
 974         if (flags & PRUS_EOF)
 975                 INP_INFO_WUNLOCK(&V_tcbinfo);
 976         return (error);
 977 }
 978
 979 /*
 980  * Abort the TCP.  Drop the connection abruptly.
 981  */
 982 static void
 983 tcp_usr_abort(struct socket *so)
 984 {
 985         struct inpcb *inp;
 986         struct tcpcb *tp = NULL;
 987         TCPDEBUG0;
 988
 989         inp = sotoinpcb(so);
 990         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
 991
 992         INP_INFO_WLOCK(&V_tcbinfo);
 993         INP_WLOCK(inp);
 994         KASSERT(inp->inp_socket != NULL,
 995             ("tcp_usr_abort: inp_socket == NULL"));
 996
 997         /*
 998          * If we still have full TCP state, and we're not dropped, drop.
 999          */
1000         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1001             !(inp->inp_flags & INP_DROPPED)) {
1002                 tp = intotcpcb(inp);
1003                 TCPDEBUG1();
1004                 tcp_drop(tp, ECONNABORTED);
1005                 TCPDEBUG2(PRU_ABORT);
1006         }
1007         if (!(inp->inp_flags & INP_DROPPED)) {
1008                 SOCK_LOCK(so);
1009                 so->so_state |= SS_PROTOREF;
1010                 SOCK_UNLOCK(so);
1011                 inp->inp_flags |= INP_SOCKREF;
1012         }
1013         INP_WUNLOCK(inp);
1014         INP_INFO_WUNLOCK(&V_tcbinfo);
1015 }
1016
1017 /*
1018  * TCP socket is closed.  Start friendly disconnect.
1019  */
1020 static void
1021 tcp_usr_close(struct socket *so)
1022 {
1023         struct inpcb *inp;
1024         struct tcpcb *tp = NULL;
1025         TCPDEBUG0;
1026
1027         inp = sotoinpcb(so);
1028         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1029
1030         INP_INFO_WLOCK(&V_tcbinfo);
1031         INP_WLOCK(inp);
1032         KASSERT(inp->inp_socket != NULL,
1033             ("tcp_usr_close: inp_socket == NULL"));
1034
1035         /*
1036          * If we still have full TCP state, and we're not dropped, initiate
1037          * a disconnect.
1038          */
1039         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1040             !(inp->inp_flags & INP_DROPPED)) {
1041                 tp = intotcpcb(inp);
1042                 TCPDEBUG1();
1043                 tcp_disconnect(tp);
1044                 TCPDEBUG2(PRU_CLOSE);
1045         }
1046         if (!(inp->inp_flags & INP_DROPPED)) {
1047                 SOCK_LOCK(so);
1048                 so->so_state |= SS_PROTOREF;
1049                 SOCK_UNLOCK(so);
1050                 inp->inp_flags |= INP_SOCKREF;
1051         }
1052         INP_WUNLOCK(inp);
1053         INP_INFO_WUNLOCK(&V_tcbinfo);
1054 }
1055
1056 /*
1057  * Receive out-of-band data.
1058  */
1059 static int
1060 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1061 {
1062         int error = 0;
1063         struct inpcb *inp;
1064         struct tcpcb *tp = NULL;
1065
1066         TCPDEBUG0;
1067         inp = sotoinpcb(so);
1068         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1069         INP_WLOCK(inp);
1070         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1071                 error = ECONNRESET;
1072                 goto out;
1073         }
1074         tp = intotcpcb(inp);
1075         TCPDEBUG1();
1076         if ((so->so_oobmark == 0 &&
1077              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1078             so->so_options & SO_OOBINLINE ||
1079             tp->t_oobflags & TCPOOB_HADDATA) {
1080                 error = EINVAL;
1081                 goto out;
1082         }
1083         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1084                 error = EWOULDBLOCK;
1085                 goto out;
1086         }
1087         m->m_len = 1;
1088         *mtod(m, caddr_t) = tp->t_iobc;
1089         if ((flags & MSG_PEEK) == 0)
1090                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1091
1092 out:
1093         TCPDEBUG2(PRU_RCVOOB);
1094         INP_WUNLOCK(inp);
1095         return (error);
1096 }
1097
1098 #ifdef INET
1099 struct pr_usrreqs tcp_usrreqs = {
1100         .pru_abort =            tcp_usr_abort,
1101         .pru_accept =           tcp_usr_accept,
1102         .pru_attach =           tcp_usr_attach,
1103         .pru_bind =             tcp_usr_bind,
1104         .pru_connect =          tcp_usr_connect,
1105         .pru_control =          in_control,
1106         .pru_detach =           tcp_usr_detach,
1107         .pru_disconnect =       tcp_usr_disconnect,
1108         .pru_listen =           tcp_usr_listen,
1109         .pru_peeraddr =         in_getpeeraddr,
1110         .pru_rcvd =             tcp_usr_rcvd,
1111         .pru_rcvoob =           tcp_usr_rcvoob,
1112         .pru_send =             tcp_usr_send,
1113         .pru_shutdown =         tcp_usr_shutdown,
1114         .pru_sockaddr =         in_getsockaddr,
1115         .pru_sosetlabel =       in_pcbsosetlabel,
1116         .pru_close =            tcp_usr_close,
1117 };
1118 #endif /* INET */
1119
1120 #ifdef INET6
1121 struct pr_usrreqs tcp6_usrreqs = {
1122         .pru_abort =            tcp_usr_abort,
1123         .pru_accept =           tcp6_usr_accept,
1124         .pru_attach =           tcp_usr_attach,
1125         .pru_bind =             tcp6_usr_bind,
1126         .pru_connect =          tcp6_usr_connect,
1127         .pru_control =          in6_control,
1128         .pru_detach =           tcp_usr_detach,
1129         .pru_disconnect =       tcp_usr_disconnect,
1130         .pru_listen =           tcp6_usr_listen,
1131         .pru_peeraddr =         in6_mapped_peeraddr,
1132         .pru_rcvd =             tcp_usr_rcvd,
1133         .pru_rcvoob =           tcp_usr_rcvoob,
1134         .pru_send =             tcp_usr_send,
1135         .pru_shutdown =         tcp_usr_shutdown,
1136         .pru_sockaddr =         in6_mapped_sockaddr,
1137         .pru_sosetlabel =       in_pcbsosetlabel,
1138         .pru_close =            tcp_usr_close,
1139 };
1140 #endif /* INET6 */
1141
1142 #ifdef INET
1143 /*
1144  * Common subroutine to open a TCP connection to remote host specified
1145  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1146  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1147  * to choose a local host address (interface).  If there is an existing
1148  * incarnation of the same connection in TIME-WAIT state and if the remote
1149  * host was sending CC options and if the connection duration was < MSL, then
1150  * truncate the previous TIME-WAIT state and proceed.
1151  * Initialize connection parameters and enter SYN-SENT state.
1152  */
1153 static int
1154 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1155 {
1156         struct inpcb *inp = tp->t_inpcb, *oinp;
1157         struct socket *so = inp->inp_socket;
1158         struct in_addr laddr;
1159         u_short lport;
1160         int error;
1161
1162         INP_WLOCK_ASSERT(inp);
1163         INP_HASH_WLOCK(&V_tcbinfo);
1164
1165         if (inp->inp_lport == 0) {
1166                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1167                 if (error)
1168                         goto out;
1169         }
1170
1171         /*
1172          * Cannot simply call in_pcbconnect, because there might be an
1173          * earlier incarnation of this same connection still in
1174          * TIME_WAIT state, creating an ADDRINUSE error.
1175          */
1176         laddr = inp->inp_laddr;
1177         lport = inp->inp_lport;
1178         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1179             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1180         if (error && oinp == NULL)
1181                 goto out;
1182         if (oinp) {
1183                 error = EADDRINUSE;
1184                 goto out;
1185         }
1186         inp->inp_laddr = laddr;
1187         in_pcbrehash(inp);
1188         INP_HASH_WUNLOCK(&V_tcbinfo);
1189
1190         /*
1191          * Compute window scaling to request:
1192          * Scale to fit into sweet spot.  See tcp_syncache.c.
1193          * XXX: This should move to tcp_output().
1194          */
1195         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1196             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1197                 tp->request_r_scale++;
1198
1199         soisconnecting(so);
1200         TCPSTAT_INC(tcps_connattempt);
1201         tcp_state_change(tp, TCPS_SYN_SENT);
1202         tp->iss = tcp_new_isn(tp);
1203         tcp_sendseqinit(tp);
1204
1205         return 0;
1206
1207 out:
1208         INP_HASH_WUNLOCK(&V_tcbinfo);
1209         return (error);
1210 }
1211 #endif /* INET */
1212
1213 #ifdef INET6
1214 static int
1215 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1216 {
1217         struct inpcb *inp = tp->t_inpcb, *oinp;
1218         struct socket *so = inp->inp_socket;
1219         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1220         struct in6_addr addr6;
1221         int error;
1222
1223         INP_WLOCK_ASSERT(inp);
1224         INP_HASH_WLOCK(&V_tcbinfo);
1225
1226         if (inp->inp_lport == 0) {
1227                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1228                 if (error)
1229                         goto out;
1230         }
1231
1232         /*
1233          * Cannot simply call in_pcbconnect, because there might be an
1234          * earlier incarnation of this same connection still in
1235          * TIME_WAIT state, creating an ADDRINUSE error.
1236          * in6_pcbladdr() also handles scope zone IDs.
1237          *
1238          * XXXRW: We wouldn't need to expose in6_pcblookup_hash_locked()
1239          * outside of in6_pcb.c if there were an in6_pcbconnect_setup().
1240          */
1241         error = in6_pcbladdr(inp, nam, &addr6);
1242         if (error)
1243                 goto out;
1244         oinp = in6_pcblookup_hash_locked(inp->inp_pcbinfo,
1245                                   &sin6->sin6_addr, sin6->sin6_port,
1246                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1247                                   ? &addr6
1248                                   : &inp->in6p_laddr,
1249                                   inp->inp_lport,  0, NULL);
1250         if (oinp) {
1251                 error = EADDRINUSE;
1252                 goto out;
1253         }
1254         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1255                 inp->in6p_laddr = addr6;
1256         inp->in6p_faddr = sin6->sin6_addr;
1257         inp->inp_fport = sin6->sin6_port;
1258         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1259         inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
1260         if (inp->inp_flags & IN6P_AUTOFLOWLABEL)
1261                 inp->inp_flow |=
1262                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1263         in_pcbrehash(inp);
1264         INP_HASH_WUNLOCK(&V_tcbinfo);
1265
1266         /* Compute window scaling to request.  */
1267         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1268             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1269                 tp->request_r_scale++;
1270
1271         soisconnecting(so);
1272         TCPSTAT_INC(tcps_connattempt);
1273         tcp_state_change(tp, TCPS_SYN_SENT);
1274         tp->iss = tcp_new_isn(tp);
1275         tcp_sendseqinit(tp);
1276
1277         return 0;
1278
1279 out:
1280         INP_HASH_WUNLOCK(&V_tcbinfo);
1281         return error;
1282 }
1283 #endif /* INET6 */
1284
1285 /*
1286  * Export TCP internal state information via a struct tcp_info, based on the
1287  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1288  * (TCP state machine, etc).  We export all information using FreeBSD-native
1289  * constants -- for example, the numeric values for tcpi_state will differ
1290  * from Linux.
1291  */
1292 static void
1293 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1294 {
1295
1296         INP_WLOCK_ASSERT(tp->t_inpcb);
1297         bzero(ti, sizeof(*ti));
1298
1299         ti->tcpi_state = tp->t_state;
1300         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1301                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1302         if (tp->t_flags & TF_SACK_PERMIT)
1303                 ti->tcpi_options |= TCPI_OPT_SACK;
1304         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1305                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1306                 ti->tcpi_snd_wscale = tp->snd_scale;
1307                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1308         }
1309
1310         ti->tcpi_rto = tp->t_rxtcur * tick;
1311         ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1312         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1313         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1314
1315         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1316         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1317
1318         /*
1319          * FreeBSD-specific extension fields for tcp_info.
1320          */
1321         ti->tcpi_rcv_space = tp->rcv_wnd;
1322         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1323         ti->tcpi_snd_wnd = tp->snd_wnd;
1324         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1325         ti->tcpi_snd_nxt = tp->snd_nxt;
1326         ti->tcpi_snd_mss = tp->t_maxseg;
1327         ti->tcpi_rcv_mss = tp->t_maxseg;
1328         if (tp->t_flags & TF_TOE)
1329                 ti->tcpi_options |= TCPI_OPT_TOE;
1330         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1331         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1332         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1333 }
1334
1335 /*
1336  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1337  * socket option arguments.  When it re-acquires the lock after the copy, it
1338  * has to revalidate that the connection is still valid for the socket
1339  * option.
1340  */
1341 #define INP_WLOCK_RECHECK(inp) do {                                     \
1342         INP_WLOCK(inp);                                                 \
1343         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1344                 INP_WUNLOCK(inp);                                       \
1345                 return (ECONNRESET);                                    \
1346         }                                                               \
1347         tp = intotcpcb(inp);                                            \
1348 } while(0)
1349
1350 int
1351 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1352 {
1353         int     error, opt, optval;
1354         u_int   ui;
1355         struct  inpcb *inp;
1356         struct  tcpcb *tp;
1357         struct  tcp_info ti;
1358         char buf[TCP_CA_NAME_MAX];
1359         struct cc_algo *algo;
1360
1361         error = 0;
1362         inp = sotoinpcb(so);
1363         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1364         INP_WLOCK(inp);
1365         if (sopt->sopt_level != IPPROTO_TCP) {
1366 #ifdef INET6
1367                 if (inp->inp_vflag & INP_IPV6PROTO) {
1368                         INP_WUNLOCK(inp);
1369                         error = ip6_ctloutput(so, sopt);
1370                 }
1371 #endif /* INET6 */
1372 #if defined(INET6) && defined(INET)
1373                 else
1374 #endif
1375 #ifdef INET
1376                 {
1377                         INP_WUNLOCK(inp);
1378                         error = ip_ctloutput(so, sopt);
1379                 }
1380 #endif
1381                 return (error);
1382         }
1383         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1384                 INP_WUNLOCK(inp);
1385                 return (ECONNRESET);
1386         }
1387
1388         switch (sopt->sopt_dir) {
1389         case SOPT_SET:
1390                 switch (sopt->sopt_name) {
1391 #ifdef TCP_SIGNATURE
1392                 case TCP_MD5SIG:
1393                         INP_WUNLOCK(inp);
1394                         error = sooptcopyin(sopt, &optval, sizeof optval,
1395                             sizeof optval);
1396                         if (error)
1397                                 return (error);
1398
1399                         INP_WLOCK_RECHECK(inp);
1400                         if (optval > 0)
1401                                 tp->t_flags |= TF_SIGNATURE;
1402                         else
1403                                 tp->t_flags &= ~TF_SIGNATURE;
1404                         goto unlock_and_done;
1405 #endif /* TCP_SIGNATURE */
1406
1407                 case TCP_NODELAY:
1408                 case TCP_NOOPT:
1409                         INP_WUNLOCK(inp);
1410                         error = sooptcopyin(sopt, &optval, sizeof optval,
1411                             sizeof optval);
1412                         if (error)
1413                                 return (error);
1414
1415                         INP_WLOCK_RECHECK(inp);
1416                         switch (sopt->sopt_name) {
1417                         case TCP_NODELAY:
1418                                 opt = TF_NODELAY;
1419                                 break;
1420                         case TCP_NOOPT:
1421                                 opt = TF_NOOPT;
1422                                 break;
1423                         default:
1424                                 opt = 0; /* dead code to fool gcc */
1425                                 break;
1426                         }
1427
1428                         if (optval)
1429                                 tp->t_flags |= opt;
1430                         else
1431                                 tp->t_flags &= ~opt;
1432 unlock_and_done:
1433 #ifdef TCP_OFFLOAD
1434                         if (tp->t_flags & TF_TOE) {
1435                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1436                                     sopt->sopt_name);
1437                         }
1438 #endif
1439                         INP_WUNLOCK(inp);
1440                         break;
1441
1442                 case TCP_NOPUSH:
1443                         INP_WUNLOCK(inp);
1444                         error = sooptcopyin(sopt, &optval, sizeof optval,
1445                             sizeof optval);
1446                         if (error)
1447                                 return (error);
1448
1449                         INP_WLOCK_RECHECK(inp);
1450                         if (optval)
1451                                 tp->t_flags |= TF_NOPUSH;
1452                         else if (tp->t_flags & TF_NOPUSH) {
1453                                 tp->t_flags &= ~TF_NOPUSH;
1454                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1455                                         error = tcp_output(tp);
1456                         }
1457                         goto unlock_and_done;
1458
1459                 case TCP_MAXSEG:
1460                         INP_WUNLOCK(inp);
1461                         error = sooptcopyin(sopt, &optval, sizeof optval,
1462                             sizeof optval);
1463                         if (error)
1464                                 return (error);
1465
1466                         INP_WLOCK_RECHECK(inp);
1467                         if (optval > 0 && optval <= tp->t_maxseg &&
1468                             optval + 40 >= V_tcp_minmss)
1469                                 tp->t_maxseg = optval;
1470                         else
1471                                 error = EINVAL;
1472                         goto unlock_and_done;
1473
1474                 case TCP_INFO:
1475                         INP_WUNLOCK(inp);
1476                         error = EINVAL;
1477                         break;
1478
1479                 case TCP_CONGESTION:
1480                         INP_WUNLOCK(inp);
1481                         bzero(buf, sizeof(buf));
1482                         error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1483                         if (error)
1484                                 break;
1485                         INP_WLOCK_RECHECK(inp);
1486                         /*
1487                          * Return EINVAL if we can't find the requested cc algo.
1488                          */
1489                         error = EINVAL;
1490                         CC_LIST_RLOCK();
1491                         STAILQ_FOREACH(algo, &cc_list, entries) {
1492                                 if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1493                                     == 0) {
1494                                         /* We've found the requested algo. */
1495                                         error = 0;
1496                                         /*
1497                                          * We hold a write lock over the tcb
1498                                          * so it's safe to do these things
1499                                          * without ordering concerns.
1500                                          */
1501                                         if (CC_ALGO(tp)->cb_destroy != NULL)
1502                                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1503                                         CC_ALGO(tp) = algo;
1504                                         /*
1505                                          * If something goes pear shaped
1506                                          * initialising the new algo,
1507                                          * fall back to newreno (which
1508                                          * does not require initialisation).
1509                                          */
1510                                         if (algo->cb_init != NULL)
1511                                                 if (algo->cb_init(tp->ccv) > 0) {
1512                                                         CC_ALGO(tp) = &newreno_cc_algo;
1513                                                         /*
1514                                                          * The only reason init
1515                                                          * should fail is
1516                                                          * because of malloc.
1517                                                          */
1518                                                         error = ENOMEM;
1519                                                 }
1520                                         break; /* Break the STAILQ_FOREACH. */
1521                                 }
1522                         }
1523                         CC_LIST_RUNLOCK();
1524                         goto unlock_and_done;
1525
1526                 case TCP_KEEPIDLE:
1527                 case TCP_KEEPINTVL:
1528                 case TCP_KEEPINIT:
1529                         INP_WUNLOCK(inp);
1530                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1531                         if (error)
1532                                 return (error);
1533
1534                         if (ui > (UINT_MAX / hz)) {
1535                                 error = EINVAL;
1536                                 break;
1537                         }
1538                         ui *= hz;
1539
1540                         INP_WLOCK_RECHECK(inp);
1541                         switch (sopt->sopt_name) {
1542                         case TCP_KEEPIDLE:
1543                                 tp->t_keepidle = ui;
1544                                 /*
1545                                  * XXX: better check current remaining
1546                                  * timeout and "merge" it with new value.
1547                                  */
1548                                 if ((tp->t_state > TCPS_LISTEN) &&
1549                                     (tp->t_state <= TCPS_CLOSING))
1550                                         tcp_timer_activate(tp, TT_KEEP,
1551                                             TP_KEEPIDLE(tp));
1552                                 break;
1553                         case TCP_KEEPINTVL:
1554                                 tp->t_keepintvl = ui;
1555                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1556                                     (TP_MAXIDLE(tp) > 0))
1557                                         tcp_timer_activate(tp, TT_2MSL,
1558                                             TP_MAXIDLE(tp));
1559                                 break;
1560                         case TCP_KEEPINIT:
1561                                 tp->t_keepinit = ui;
1562                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1563                                     tp->t_state == TCPS_SYN_SENT)
1564                                         tcp_timer_activate(tp, TT_KEEP,
1565                                             TP_KEEPINIT(tp));
1566                                 break;
1567                         }
1568                         goto unlock_and_done;
1569
1570                 case TCP_KEEPCNT:
1571                         INP_WUNLOCK(inp);
1572                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1573                         if (error)
1574                                 return (error);
1575
1576                         INP_WLOCK_RECHECK(inp);
1577                         tp->t_keepcnt = ui;
1578                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1579                             (TP_MAXIDLE(tp) > 0))
1580                                 tcp_timer_activate(tp, TT_2MSL,
1581                                     TP_MAXIDLE(tp));
1582                         goto unlock_and_done;
1583
1584 #ifdef TCP_RFC7413
1585                 case TCP_FASTOPEN:
1586                         INP_WUNLOCK(inp);
1587                         if (!V_tcp_fastopen_enabled)
1588                                 return (EPERM);
1589
1590                         error = sooptcopyin(sopt, &optval, sizeof optval,
1591                             sizeof optval);
1592                         if (error)
1593                                 return (error);
1594
1595                         INP_WLOCK_RECHECK(inp);
1596                         if (optval) {
1597                                 tp->t_flags |= TF_FASTOPEN;
1598                                 if ((tp->t_state == TCPS_LISTEN) &&
1599                                     (tp->t_tfo_pending == NULL))
1600                                         tp->t_tfo_pending =
1601                                             tcp_fastopen_alloc_counter();
1602                         } else
1603                                 tp->t_flags &= ~TF_FASTOPEN;
1604                         goto unlock_and_done;
1605 #endif
1606
1607                 default:
1608                         INP_WUNLOCK(inp);
1609                         error = ENOPROTOOPT;
1610                         break;
1611                 }
1612                 break;
1613
1614         case SOPT_GET:
1615                 tp = intotcpcb(inp);
1616                 switch (sopt->sopt_name) {
1617 #ifdef TCP_SIGNATURE
1618                 case TCP_MD5SIG:
1619                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1620                         INP_WUNLOCK(inp);
1621                         error = sooptcopyout(sopt, &optval, sizeof optval);
1622                         break;
1623 #endif
1624
1625                 case TCP_NODELAY:
1626                         optval = tp->t_flags & TF_NODELAY;
1627                         INP_WUNLOCK(inp);
1628                         error = sooptcopyout(sopt, &optval, sizeof optval);
1629                         break;
1630                 case TCP_MAXSEG:
1631                         optval = tp->t_maxseg;
1632                         INP_WUNLOCK(inp);
1633                         error = sooptcopyout(sopt, &optval, sizeof optval);
1634                         break;
1635                 case TCP_NOOPT:
1636                         optval = tp->t_flags & TF_NOOPT;
1637                         INP_WUNLOCK(inp);
1638                         error = sooptcopyout(sopt, &optval, sizeof optval);
1639                         break;
1640                 case TCP_NOPUSH:
1641                         optval = tp->t_flags & TF_NOPUSH;
1642                         INP_WUNLOCK(inp);
1643                         error = sooptcopyout(sopt, &optval, sizeof optval);
1644                         break;
1645                 case TCP_INFO:
1646                         tcp_fill_info(tp, &ti);
1647                         INP_WUNLOCK(inp);
1648                         error = sooptcopyout(sopt, &ti, sizeof ti);
1649                         break;
1650                 case TCP_CONGESTION:
1651                         bzero(buf, sizeof(buf));
1652                         strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1653                         INP_WUNLOCK(inp);
1654                         error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1655                         break;
1656                 case TCP_KEEPIDLE:
1657                 case TCP_KEEPINTVL:
1658                 case TCP_KEEPINIT:
1659                 case TCP_KEEPCNT:
1660                         switch (sopt->sopt_name) {
1661                         case TCP_KEEPIDLE:
1662                                 ui = tp->t_keepidle / hz;
1663                                 break;
1664                         case TCP_KEEPINTVL:
1665                                 ui = tp->t_keepintvl / hz;
1666                                 break;
1667                         case TCP_KEEPINIT:
1668                                 ui = tp->t_keepinit / hz;
1669                                 break;
1670                         case TCP_KEEPCNT:
1671                                 ui = tp->t_keepcnt;
1672                                 break;
1673                         }
1674                         INP_WUNLOCK(inp);
1675                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1676                         break;
1677 #ifdef TCP_RFC7413
1678                 case TCP_FASTOPEN:
1679                         optval = tp->t_flags & TF_FASTOPEN;
1680                         INP_WUNLOCK(inp);
1681                         error = sooptcopyout(sopt, &optval, sizeof optval);
1682                         break;
1683 #endif
1684                 default:
1685                         INP_WUNLOCK(inp);
1686                         error = ENOPROTOOPT;
1687                         break;
1688                 }
1689                 break;
1690         }
1691         return (error);
1692 }
1693 #undef INP_WLOCK_RECHECK
1694
1695 /*
1696  * Attach TCP protocol to socket, allocating
1697  * internet protocol control block, tcp control block,
1698  * bufer space, and entering LISTEN state if to accept connections.
1699  */
1700 static int
1701 tcp_attach(struct socket *so)
1702 {
1703         struct tcpcb *tp;
1704         struct inpcb *inp;
1705         int error;
1706
1707         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1708                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1709                 if (error)
1710                         return (error);
1711         }
1712         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1713         so->so_snd.sb_flags |= SB_AUTOSIZE;
1714         INP_INFO_WLOCK(&V_tcbinfo);
1715         error = in_pcballoc(so, &V_tcbinfo);
1716         if (error) {
1717                 INP_INFO_WUNLOCK(&V_tcbinfo);
1718                 return (error);
1719         }
1720         inp = sotoinpcb(so);
1721 #ifdef INET6
1722         if (inp->inp_vflag & INP_IPV6PROTO) {
1723                 inp->inp_vflag |= INP_IPV6;
1724                 inp->in6p_hops = -1;    /* use kernel default */
1725         }
1726         else
1727 #endif
1728         inp->inp_vflag |= INP_IPV4;
1729         tp = tcp_newtcpcb(inp);
1730         if (tp == NULL) {
1731                 in_pcbdetach(inp);
1732                 in_pcbfree(inp);
1733                 INP_INFO_WUNLOCK(&V_tcbinfo);
1734                 return (ENOBUFS);
1735         }
1736         tp->t_state = TCPS_CLOSED;
1737         INP_WUNLOCK(inp);
1738         INP_INFO_WUNLOCK(&V_tcbinfo);
1739         return (0);
1740 }
1741
1742 /*
1743  * Initiate (or continue) disconnect.
1744  * If embryonic state, just send reset (once).
1745  * If in ``let data drain'' option and linger null, just drop.
1746  * Otherwise (hard), mark socket disconnecting and drop
1747  * current input data; switch states based on user close, and
1748  * send segment to peer (with FIN).
1749  */
1750 static void
1751 tcp_disconnect(struct tcpcb *tp)
1752 {
1753         struct inpcb *inp = tp->t_inpcb;
1754         struct socket *so = inp->inp_socket;
1755
1756         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1757         INP_WLOCK_ASSERT(inp);
1758
1759         /*
1760          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1761          * socket is still open.
1762          */
1763         if (tp->t_state < TCPS_ESTABLISHED) {
1764                 tp = tcp_close(tp);
1765                 KASSERT(tp != NULL,
1766                     ("tcp_disconnect: tcp_close() returned NULL"));
1767         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1768                 tp = tcp_drop(tp, 0);
1769                 KASSERT(tp != NULL,
1770                     ("tcp_disconnect: tcp_drop() returned NULL"));
1771         } else {
1772                 soisdisconnecting(so);
1773                 sbflush(&so->so_rcv);
1774                 tcp_usrclosed(tp);
1775                 if (!(inp->inp_flags & INP_DROPPED))
1776                         tcp_output(tp);
1777         }
1778 }
1779
1780 /*
1781  * User issued close, and wish to trail through shutdown states:
1782  * if never received SYN, just forget it.  If got a SYN from peer,
1783  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1784  * If already got a FIN from peer, then almost done; go to LAST_ACK
1785  * state.  In all other cases, have already sent FIN to peer (e.g.
1786  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1787  * for peer to send FIN or not respond to keep-alives, etc.
1788  * We can let the user exit from the close as soon as the FIN is acked.
1789  */
1790 static void
1791 tcp_usrclosed(struct tcpcb *tp)
1792 {
1793
1794         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1795         INP_WLOCK_ASSERT(tp->t_inpcb);
1796
1797         switch (tp->t_state) {
1798         case TCPS_LISTEN:
1799 #ifdef TCP_OFFLOAD
1800                 tcp_offload_listen_stop(tp);
1801 #endif
1802                 tcp_state_change(tp, TCPS_CLOSED);
1803                 /* FALLTHROUGH */
1804         case TCPS_CLOSED:
1805                 tp = tcp_close(tp);
1806                 /*
1807                  * tcp_close() should never return NULL here as the socket is
1808                  * still open.
1809                  */
1810                 KASSERT(tp != NULL,
1811                     ("tcp_usrclosed: tcp_close() returned NULL"));
1812                 break;
1813
1814         case TCPS_SYN_SENT:
1815         case TCPS_SYN_RECEIVED:
1816                 tp->t_flags |= TF_NEEDFIN;
1817                 break;
1818
1819         case TCPS_ESTABLISHED:
1820                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1821                 break;
1822
1823         case TCPS_CLOSE_WAIT:
1824                 tcp_state_change(tp, TCPS_LAST_ACK);
1825                 break;
1826         }
1827         if (tp->t_state >= TCPS_FIN_WAIT_2) {
1828                 soisdisconnected(tp->t_inpcb->inp_socket);
1829                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1830                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1831                         int timeout;
1832
1833                         timeout = (tcp_fast_finwait2_recycle) ?
1834                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
1835                         tcp_timer_activate(tp, TT_2MSL, timeout);
1836                 }
1837         }
1838 }
1839
1840 #ifdef DDB
1841 static void
1842 db_print_indent(int indent)
1843 {
1844         int i;
1845
1846         for (i = 0; i < indent; i++)
1847                 db_printf(" ");
1848 }
1849
1850 static void
1851 db_print_tstate(int t_state)
1852 {
1853
1854         switch (t_state) {
1855         case TCPS_CLOSED:
1856                 db_printf("TCPS_CLOSED");
1857                 return;
1858
1859         case TCPS_LISTEN:
1860                 db_printf("TCPS_LISTEN");
1861                 return;
1862
1863         case TCPS_SYN_SENT:
1864                 db_printf("TCPS_SYN_SENT");
1865                 return;
1866
1867         case TCPS_SYN_RECEIVED:
1868                 db_printf("TCPS_SYN_RECEIVED");
1869                 return;
1870
1871         case TCPS_ESTABLISHED:
1872                 db_printf("TCPS_ESTABLISHED");
1873                 return;
1874
1875         case TCPS_CLOSE_WAIT:
1876                 db_printf("TCPS_CLOSE_WAIT");
1877                 return;
1878
1879         case TCPS_FIN_WAIT_1:
1880                 db_printf("TCPS_FIN_WAIT_1");
1881                 return;
1882
1883         case TCPS_CLOSING:
1884                 db_printf("TCPS_CLOSING");
1885                 return;
1886
1887         case TCPS_LAST_ACK:
1888                 db_printf("TCPS_LAST_ACK");
1889                 return;
1890
1891         case TCPS_FIN_WAIT_2:
1892                 db_printf("TCPS_FIN_WAIT_2");
1893                 return;
1894
1895         case TCPS_TIME_WAIT:
1896                 db_printf("TCPS_TIME_WAIT");
1897                 return;
1898
1899         default:
1900                 db_printf("unknown");
1901                 return;
1902         }
1903 }
1904
1905 static void
1906 db_print_tflags(u_int t_flags)
1907 {
1908         int comma;
1909
1910         comma = 0;
1911         if (t_flags & TF_ACKNOW) {
1912                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1913                 comma = 1;
1914         }
1915         if (t_flags & TF_DELACK) {
1916                 db_printf("%sTF_DELACK", comma ? ", " : "");
1917                 comma = 1;
1918         }
1919         if (t_flags & TF_NODELAY) {
1920                 db_printf("%sTF_NODELAY", comma ? ", " : "");
1921                 comma = 1;
1922         }
1923         if (t_flags & TF_NOOPT) {
1924                 db_printf("%sTF_NOOPT", comma ? ", " : "");
1925                 comma = 1;
1926         }
1927         if (t_flags & TF_SENTFIN) {
1928                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1929                 comma = 1;
1930         }
1931         if (t_flags & TF_REQ_SCALE) {
1932                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1933                 comma = 1;
1934         }
1935         if (t_flags & TF_RCVD_SCALE) {
1936                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1937                 comma = 1;
1938         }
1939         if (t_flags & TF_REQ_TSTMP) {
1940                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1941                 comma = 1;
1942         }
1943         if (t_flags & TF_RCVD_TSTMP) {
1944                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1945                 comma = 1;
1946         }
1947         if (t_flags & TF_SACK_PERMIT) {
1948                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1949                 comma = 1;
1950         }
1951         if (t_flags & TF_NEEDSYN) {
1952                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1953                 comma = 1;
1954         }
1955         if (t_flags & TF_NEEDFIN) {
1956                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1957                 comma = 1;
1958         }
1959         if (t_flags & TF_NOPUSH) {
1960                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1961                 comma = 1;
1962         }
1963         if (t_flags & TF_MORETOCOME) {
1964                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1965                 comma = 1;
1966         }
1967         if (t_flags & TF_LQ_OVERFLOW) {
1968                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1969                 comma = 1;
1970         }
1971         if (t_flags & TF_LASTIDLE) {
1972                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1973                 comma = 1;
1974         }
1975         if (t_flags & TF_RXWIN0SENT) {
1976                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1977                 comma = 1;
1978         }
1979         if (t_flags & TF_FASTRECOVERY) {
1980                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1981                 comma = 1;
1982         }
1983         if (t_flags & TF_CONGRECOVERY) {
1984                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
1985                 comma = 1;
1986         }
1987         if (t_flags & TF_WASFRECOVERY) {
1988                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
1989                 comma = 1;
1990         }
1991         if (t_flags & TF_SIGNATURE) {
1992                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
1993                 comma = 1;
1994         }
1995         if (t_flags & TF_FORCEDATA) {
1996                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
1997                 comma = 1;
1998         }
1999         if (t_flags & TF_TSO) {
2000                 db_printf("%sTF_TSO", comma ? ", " : "");
2001                 comma = 1;
2002         }
2003         if (t_flags & TF_ECN_PERMIT) {
2004                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2005                 comma = 1;
2006         }
2007         if (t_flags & TF_FASTOPEN) {
2008                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2009                 comma = 1;
2010         }
2011 }
2012
2013 static void
2014 db_print_toobflags(char t_oobflags)
2015 {
2016         int comma;
2017
2018         comma = 0;
2019         if (t_oobflags & TCPOOB_HAVEDATA) {
2020                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2021                 comma = 1;
2022         }
2023         if (t_oobflags & TCPOOB_HADDATA) {
2024                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2025                 comma = 1;
2026         }
2027 }
2028
2029 static void
2030 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2031 {
2032
2033         db_print_indent(indent);
2034         db_printf("%s at %p\n", name, tp);
2035
2036         indent += 2;
2037
2038         db_print_indent(indent);
2039         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2040            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2041
2042         db_print_indent(indent);
2043         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2044             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2045
2046         db_print_indent(indent);
2047         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2048             &tp->t_timers->tt_delack, tp->t_inpcb);
2049
2050         db_print_indent(indent);
2051         db_printf("t_state: %d (", tp->t_state);
2052         db_print_tstate(tp->t_state);
2053         db_printf(")\n");
2054
2055         db_print_indent(indent);
2056         db_printf("t_flags: 0x%x (", tp->t_flags);
2057         db_print_tflags(tp->t_flags);
2058         db_printf(")\n");
2059
2060         db_print_indent(indent);
2061         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2062             tp->snd_una, tp->snd_max, tp->snd_nxt);
2063
2064         db_print_indent(indent);
2065         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2066            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2067
2068         db_print_indent(indent);
2069         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2070             tp->iss, tp->irs, tp->rcv_nxt);
2071
2072         db_print_indent(indent);
2073         db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2074             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2075
2076         db_print_indent(indent);
2077         db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2078            tp->snd_wnd, tp->snd_cwnd);
2079
2080         db_print_indent(indent);
2081         db_printf("snd_ssthresh: %lu   snd_recover: "
2082             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2083
2084         db_print_indent(indent);
2085         db_printf("t_maxopd: %u   t_rcvtime: %u   t_startime: %u\n",
2086             tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
2087
2088         db_print_indent(indent);
2089         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2090             tp->t_rtttime, tp->t_rtseq);
2091
2092         db_print_indent(indent);
2093         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2094             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2095
2096         db_print_indent(indent);
2097         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2098             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2099             tp->t_rttbest);
2100
2101         db_print_indent(indent);
2102         db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2103             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2104
2105         db_print_indent(indent);
2106         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2107         db_print_toobflags(tp->t_oobflags);
2108         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2109
2110         db_print_indent(indent);
2111         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2112             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2113
2114         db_print_indent(indent);
2115         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2116             tp->ts_recent, tp->ts_recent_age);
2117
2118         db_print_indent(indent);
2119         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2120             "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2121
2122         db_print_indent(indent);
2123         db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2124             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2125             tp->snd_recover_prev, tp->t_badrxtwin);
2126
2127         db_print_indent(indent);
2128         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2129             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2130
2131         db_print_indent(indent);
2132         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2133             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2134
2135         /* Skip sackblks, sackhint. */
2136
2137         db_print_indent(indent);
2138         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2139             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2140 }
2141
2142 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2143 {
2144         struct tcpcb *tp;
2145
2146         if (!have_addr) {
2147                 db_printf("usage: show tcpcb <addr>\n");
2148                 return;
2149         }
2150         tp = (struct tcpcb *)addr;
2151
2152         db_print_tcpcb(tp, "tcpcb", 0);
2153 }
2154 #endif