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  * 3. 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_ipsec.h"
  45 #include "opt_tcpdebug.h"
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/limits.h>
  50 #include <sys/malloc.h>
  51 #include <sys/refcount.h>
  52 #include <sys/kernel.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/mbuf.h>
  55 #ifdef INET6
  56 #include <sys/domain.h>
  57 #endif /* INET6 */
  58 #include <sys/socket.h>
  59 #include <sys/socketvar.h>
  60 #include <sys/protosw.h>
  61 #include <sys/proc.h>
  62 #include <sys/jail.h>
  63 #include <sys/syslog.h>
  64
  65 #ifdef DDB
  66 #include <ddb/ddb.h>
  67 #endif
  68
  69 #include <net/if.h>
  70 #include <net/if_var.h>
  71 #include <net/route.h>
  72 #include <net/vnet.h>
  73
  74 #include <netinet/in.h>
  75 #include <netinet/in_kdtrace.h>
  76 #include <netinet/in_pcb.h>
  77 #include <netinet/in_systm.h>
  78 #include <netinet/in_var.h>
  79 #include <netinet/ip_var.h>
  80 #ifdef INET6
  81 #include <netinet/ip6.h>
  82 #include <netinet6/in6_pcb.h>
  83 #include <netinet6/ip6_var.h>
  84 #include <netinet6/scope6_var.h>
  85 #endif
  86 #ifdef TCP_RFC7413
  87 #include <netinet/tcp_fastopen.h>
  88 #endif
  89 #include <netinet/tcp.h>
  90 #include <netinet/tcp_fsm.h>
  91 #include <netinet/tcp_seq.h>
  92 #include <netinet/tcp_timer.h>
  93 #include <netinet/tcp_var.h>
  94 #include <netinet/tcpip.h>
  95 #include <netinet/cc/cc.h>
  96 #ifdef TCPPCAP
  97 #include <netinet/tcp_pcap.h>
  98 #endif
  99 #ifdef TCPDEBUG
 100 #include <netinet/tcp_debug.h>
 101 #endif
 102 #ifdef TCP_OFFLOAD
 103 #include <netinet/tcp_offload.h>
 104 #endif
 105 #include <netipsec/ipsec_support.h>
 106
 107 /*
 108  * TCP protocol interface to socket abstraction.
 109  */
 110 static int      tcp_attach(struct socket *);
 111 #ifdef INET
 112 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 113                     struct thread *td);
 114 #endif /* INET */
 115 #ifdef INET6
 116 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 117                     struct thread *td);
 118 #endif /* INET6 */
 119 static void     tcp_disconnect(struct tcpcb *);
 120 static void     tcp_usrclosed(struct tcpcb *);
 121 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 122
 123 #ifdef TCPDEBUG
 124 #define TCPDEBUG0       int ostate = 0
 125 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 126 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 127                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 128 #else
 129 #define TCPDEBUG0
 130 #define TCPDEBUG1()
 131 #define TCPDEBUG2(req)
 132 #endif
 133
 134 /*
 135  * TCP attaches to socket via pru_attach(), reserving space,
 136  * and an internet control block.
 137  */
 138 static int
 139 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 140 {
 141         struct inpcb *inp;
 142         struct tcpcb *tp = NULL;
 143         int error;
 144         TCPDEBUG0;
 145
 146         inp = sotoinpcb(so);
 147         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 148         TCPDEBUG1();
 149
 150         error = tcp_attach(so);
 151         if (error)
 152                 goto out;
 153
 154         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 155                 so->so_linger = TCP_LINGERTIME;
 156
 157         inp = sotoinpcb(so);
 158         tp = intotcpcb(inp);
 159 out:
 160         TCPDEBUG2(PRU_ATTACH);
 161         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 162         return error;
 163 }
 164
 165 /*
 166  * tcp_detach is called when the socket layer loses its final reference
 167  * to the socket, be it a file descriptor reference, a reference from TCP,
 168  * etc.  At this point, there is only one case in which we will keep around
 169  * inpcb state: time wait.
 170  *
 171  * This function can probably be re-absorbed back into tcp_usr_detach() now
 172  * that there is a single detach path.
 173  */
 174 static void
 175 tcp_detach(struct socket *so, struct inpcb *inp)
 176 {
 177         struct tcpcb *tp;
 178
 179         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
 180         INP_WLOCK_ASSERT(inp);
 181
 182         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 183         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 184
 185         tp = intotcpcb(inp);
 186
 187         if (inp->inp_flags & INP_TIMEWAIT) {
 188                 /*
 189                  * There are two cases to handle: one in which the time wait
 190                  * state is being discarded (INP_DROPPED), and one in which
 191                  * this connection will remain in timewait.  In the former,
 192                  * it is time to discard all state (except tcptw, which has
 193                  * already been discarded by the timewait close code, which
 194                  * should be further up the call stack somewhere).  In the
 195                  * latter case, we detach from the socket, but leave the pcb
 196                  * present until timewait ends.
 197                  *
 198                  * XXXRW: Would it be cleaner to free the tcptw here?
 199                  *
 200                  * Astute question indeed, from twtcp perspective there are
 201                  * three cases to consider:
 202                  *
 203                  * #1 tcp_detach is called at tcptw creation time by
 204                  *  tcp_twstart, then do not discard the newly created tcptw
 205                  *  and leave inpcb present until timewait ends
 206                  * #2 tcp_detach is called at timewait end (or reuse) by
 207                  *  tcp_twclose, then the tcptw has already been discarded
 208                  *  (or reused) and inpcb is freed here
 209                  * #3 tcp_detach is called() after timewait ends (or reuse)
 210                  *  (e.g. by soclose), then tcptw has already been discarded
 211                  *  (or reused) and inpcb is freed here
 212                  *
 213                  *  In all three cases the tcptw should not be freed here.
 214                  */
 215                 if (inp->inp_flags & INP_DROPPED) {
 216                         in_pcbdetach(inp);
 217                         if (__predict_true(tp == NULL)) {
 218                                 in_pcbfree(inp);
 219                         } else {
 220                                 /*
 221                                  * This case should not happen as in TIMEWAIT
 222                                  * state the inp should not be destroyed before
 223                                  * its tcptw.  If INVARIANTS is defined, panic.
 224                                  */
 225 #ifdef INVARIANTS
 226                                 panic("%s: Panic before an inp double-free: "
 227                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 228                                     , __func__);
 229 #else
 230                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 231                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 232                                     , __func__);
 233 #endif
 234                                 INP_WUNLOCK(inp);
 235                         }
 236                 } else {
 237                         in_pcbdetach(inp);
 238                         INP_WUNLOCK(inp);
 239                 }
 240         } else {
 241                 /*
 242                  * If the connection is not in timewait, we consider two
 243                  * two conditions: one in which no further processing is
 244                  * necessary (dropped || embryonic), and one in which TCP is
 245                  * not yet done, but no longer requires the socket, so the
 246                  * pcb will persist for the time being.
 247                  *
 248                  * XXXRW: Does the second case still occur?
 249                  */
 250                 if (inp->inp_flags & INP_DROPPED ||
 251                     tp->t_state < TCPS_SYN_SENT) {
 252                         tcp_discardcb(tp);
 253                         in_pcbdetach(inp);
 254                         in_pcbfree(inp);
 255                 } else {
 256                         in_pcbdetach(inp);
 257                         INP_WUNLOCK(inp);
 258                 }
 259         }
 260 }
 261
 262 /*
 263  * pru_detach() detaches the TCP protocol from the socket.
 264  * If the protocol state is non-embryonic, then can't
 265  * do this directly: have to initiate a pru_disconnect(),
 266  * which may finish later; embryonic TCB's can just
 267  * be discarded here.
 268  */
 269 static void
 270 tcp_usr_detach(struct socket *so)
 271 {
 272         struct inpcb *inp;
 273         int rlock = 0;
 274
 275         inp = sotoinpcb(so);
 276         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 277         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 278                 INP_INFO_RLOCK(&V_tcbinfo);
 279                 rlock = 1;
 280         }
 281         INP_WLOCK(inp);
 282         KASSERT(inp->inp_socket != NULL,
 283             ("tcp_usr_detach: inp_socket == NULL"));
 284         tcp_detach(so, inp);
 285         if (rlock)
 286                 INP_INFO_RUNLOCK(&V_tcbinfo);
 287 }
 288
 289 #ifdef INET
 290 /*
 291  * Give the socket an address.
 292  */
 293 static int
 294 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 295 {
 296         int error = 0;
 297         struct inpcb *inp;
 298         struct tcpcb *tp = NULL;
 299         struct sockaddr_in *sinp;
 300
 301         sinp = (struct sockaddr_in *)nam;
 302         if (nam->sa_len != sizeof (*sinp))
 303                 return (EINVAL);
 304         /*
 305          * Must check for multicast addresses and disallow binding
 306          * to them.
 307          */
 308         if (sinp->sin_family == AF_INET &&
 309             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 310                 return (EAFNOSUPPORT);
 311
 312         TCPDEBUG0;
 313         inp = sotoinpcb(so);
 314         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 315         INP_WLOCK(inp);
 316         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 317                 error = EINVAL;
 318                 goto out;
 319         }
 320         tp = intotcpcb(inp);
 321         TCPDEBUG1();
 322         INP_HASH_WLOCK(&V_tcbinfo);
 323         error = in_pcbbind(inp, nam, td->td_ucred);
 324         INP_HASH_WUNLOCK(&V_tcbinfo);
 325 out:
 326         TCPDEBUG2(PRU_BIND);
 327         TCP_PROBE2(debug__user, tp, PRU_BIND);
 328         INP_WUNLOCK(inp);
 329
 330         return (error);
 331 }
 332 #endif /* INET */
 333
 334 #ifdef INET6
 335 static int
 336 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 337 {
 338         int error = 0;
 339         struct inpcb *inp;
 340         struct tcpcb *tp = NULL;
 341         struct sockaddr_in6 *sin6p;
 342
 343         sin6p = (struct sockaddr_in6 *)nam;
 344         if (nam->sa_len != sizeof (*sin6p))
 345                 return (EINVAL);
 346         /*
 347          * Must check for multicast addresses and disallow binding
 348          * to them.
 349          */
 350         if (sin6p->sin6_family == AF_INET6 &&
 351             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 352                 return (EAFNOSUPPORT);
 353
 354         TCPDEBUG0;
 355         inp = sotoinpcb(so);
 356         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 357         INP_WLOCK(inp);
 358         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 359                 error = EINVAL;
 360                 goto out;
 361         }
 362         tp = intotcpcb(inp);
 363         TCPDEBUG1();
 364         INP_HASH_WLOCK(&V_tcbinfo);
 365         inp->inp_vflag &= ~INP_IPV4;
 366         inp->inp_vflag |= INP_IPV6;
 367 #ifdef INET
 368         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 369                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 370                         inp->inp_vflag |= INP_IPV4;
 371                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 372                         struct sockaddr_in sin;
 373
 374                         in6_sin6_2_sin(&sin, sin6p);
 375                         inp->inp_vflag |= INP_IPV4;
 376                         inp->inp_vflag &= ~INP_IPV6;
 377                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 378                             td->td_ucred);
 379                         INP_HASH_WUNLOCK(&V_tcbinfo);
 380                         goto out;
 381                 }
 382         }
 383 #endif
 384         error = in6_pcbbind(inp, nam, td->td_ucred);
 385         INP_HASH_WUNLOCK(&V_tcbinfo);
 386 out:
 387         TCPDEBUG2(PRU_BIND);
 388         TCP_PROBE2(debug__user, tp, PRU_BIND);
 389         INP_WUNLOCK(inp);
 390         return (error);
 391 }
 392 #endif /* INET6 */
 393
 394 #ifdef INET
 395 /*
 396  * Prepare to accept connections.
 397  */
 398 static int
 399 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 400 {
 401         int error = 0;
 402         struct inpcb *inp;
 403         struct tcpcb *tp = NULL;
 404
 405         TCPDEBUG0;
 406         inp = sotoinpcb(so);
 407         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 408         INP_WLOCK(inp);
 409         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 410                 error = EINVAL;
 411                 goto out;
 412         }
 413         tp = intotcpcb(inp);
 414         TCPDEBUG1();
 415         SOCK_LOCK(so);
 416         error = solisten_proto_check(so);
 417         INP_HASH_WLOCK(&V_tcbinfo);
 418         if (error == 0 && inp->inp_lport == 0)
 419                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 420         INP_HASH_WUNLOCK(&V_tcbinfo);
 421         if (error == 0) {
 422                 tcp_state_change(tp, TCPS_LISTEN);
 423                 solisten_proto(so, backlog);
 424 #ifdef TCP_OFFLOAD
 425                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 426                         tcp_offload_listen_start(tp);
 427 #endif
 428         }
 429         SOCK_UNLOCK(so);
 430
 431 #ifdef TCP_RFC7413
 432         if (IS_FASTOPEN(tp->t_flags))
 433                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 434 #endif
 435 out:
 436         TCPDEBUG2(PRU_LISTEN);
 437         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 438         INP_WUNLOCK(inp);
 439         return (error);
 440 }
 441 #endif /* INET */
 442
 443 #ifdef INET6
 444 static int
 445 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 446 {
 447         int error = 0;
 448         struct inpcb *inp;
 449         struct tcpcb *tp = NULL;
 450
 451         TCPDEBUG0;
 452         inp = sotoinpcb(so);
 453         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 454         INP_WLOCK(inp);
 455         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 456                 error = EINVAL;
 457                 goto out;
 458         }
 459         tp = intotcpcb(inp);
 460         TCPDEBUG1();
 461         SOCK_LOCK(so);
 462         error = solisten_proto_check(so);
 463         INP_HASH_WLOCK(&V_tcbinfo);
 464         if (error == 0 && inp->inp_lport == 0) {
 465                 inp->inp_vflag &= ~INP_IPV4;
 466                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 467                         inp->inp_vflag |= INP_IPV4;
 468                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 469         }
 470         INP_HASH_WUNLOCK(&V_tcbinfo);
 471         if (error == 0) {
 472                 tcp_state_change(tp, TCPS_LISTEN);
 473                 solisten_proto(so, backlog);
 474 #ifdef TCP_OFFLOAD
 475                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 476                         tcp_offload_listen_start(tp);
 477 #endif
 478         }
 479         SOCK_UNLOCK(so);
 480
 481 #ifdef TCP_RFC7413
 482         if (IS_FASTOPEN(tp->t_flags))
 483                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 484 #endif
 485 out:
 486         TCPDEBUG2(PRU_LISTEN);
 487         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 488         INP_WUNLOCK(inp);
 489         return (error);
 490 }
 491 #endif /* INET6 */
 492
 493 #ifdef INET
 494 /*
 495  * Initiate connection to peer.
 496  * Create a template for use in transmissions on this connection.
 497  * Enter SYN_SENT state, and mark socket as connecting.
 498  * Start keep-alive timer, and seed output sequence space.
 499  * Send initial segment on connection.
 500  */
 501 static int
 502 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 503 {
 504         int error = 0;
 505         struct inpcb *inp;
 506         struct tcpcb *tp = NULL;
 507         struct sockaddr_in *sinp;
 508
 509         sinp = (struct sockaddr_in *)nam;
 510         if (nam->sa_len != sizeof (*sinp))
 511                 return (EINVAL);
 512         /*
 513          * Must disallow TCP ``connections'' to multicast addresses.
 514          */
 515         if (sinp->sin_family == AF_INET
 516             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 517                 return (EAFNOSUPPORT);
 518         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 519                 return (error);
 520
 521         TCPDEBUG0;
 522         inp = sotoinpcb(so);
 523         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 524         INP_WLOCK(inp);
 525         if (inp->inp_flags & INP_TIMEWAIT) {
 526                 error = EADDRINUSE;
 527                 goto out;
 528         }
 529         if (inp->inp_flags & INP_DROPPED) {
 530                 error = ECONNREFUSED;
 531                 goto out;
 532         }
 533         tp = intotcpcb(inp);
 534         TCPDEBUG1();
 535         if ((error = tcp_connect(tp, nam, td)) != 0)
 536                 goto out;
 537 #ifdef TCP_OFFLOAD
 538         if (registered_toedevs > 0 &&
 539             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 540             (error = tcp_offload_connect(so, nam)) == 0)
 541                 goto out;
 542 #endif
 543         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 544         error = tp->t_fb->tfb_tcp_output(tp);
 545 out:
 546         TCPDEBUG2(PRU_CONNECT);
 547         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 548         INP_WUNLOCK(inp);
 549         return (error);
 550 }
 551 #endif /* INET */
 552
 553 #ifdef INET6
 554 static int
 555 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 556 {
 557         int error = 0;
 558         struct inpcb *inp;
 559         struct tcpcb *tp = NULL;
 560         struct sockaddr_in6 *sin6p;
 561
 562         TCPDEBUG0;
 563
 564         sin6p = (struct sockaddr_in6 *)nam;
 565         if (nam->sa_len != sizeof (*sin6p))
 566                 return (EINVAL);
 567         /*
 568          * Must disallow TCP ``connections'' to multicast addresses.
 569          */
 570         if (sin6p->sin6_family == AF_INET6
 571             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 572                 return (EAFNOSUPPORT);
 573
 574         inp = sotoinpcb(so);
 575         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 576         INP_WLOCK(inp);
 577         if (inp->inp_flags & INP_TIMEWAIT) {
 578                 error = EADDRINUSE;
 579                 goto out;
 580         }
 581         if (inp->inp_flags & INP_DROPPED) {
 582                 error = ECONNREFUSED;
 583                 goto out;
 584         }
 585         tp = intotcpcb(inp);
 586         TCPDEBUG1();
 587 #ifdef INET
 588         /*
 589          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 590          * therefore probably require the hash lock, which isn't held here.
 591          * Is this a significant problem?
 592          */
 593         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 594                 struct sockaddr_in sin;
 595
 596                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 597                         error = EINVAL;
 598                         goto out;
 599                 }
 600                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 601                         error = EAFNOSUPPORT;
 602                         goto out;
 603                 }
 604
 605                 in6_sin6_2_sin(&sin, sin6p);
 606                 inp->inp_vflag |= INP_IPV4;
 607                 inp->inp_vflag &= ~INP_IPV6;
 608                 if ((error = prison_remote_ip4(td->td_ucred,
 609                     &sin.sin_addr)) != 0)
 610                         goto out;
 611                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 612                         goto out;
 613 #ifdef TCP_OFFLOAD
 614                 if (registered_toedevs > 0 &&
 615                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 616                     (error = tcp_offload_connect(so, nam)) == 0)
 617                         goto out;
 618 #endif
 619                 error = tp->t_fb->tfb_tcp_output(tp);
 620                 goto out;
 621         } else {
 622                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 623                         error = EAFNOSUPPORT;
 624                         goto out;
 625                 }
 626         }
 627 #endif
 628         inp->inp_vflag &= ~INP_IPV4;
 629         inp->inp_vflag |= INP_IPV6;
 630         inp->inp_inc.inc_flags |= INC_ISIPV6;
 631         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 632                 goto out;
 633         if ((error = tcp6_connect(tp, nam, td)) != 0)
 634                 goto out;
 635 #ifdef TCP_OFFLOAD
 636         if (registered_toedevs > 0 &&
 637             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 638             (error = tcp_offload_connect(so, nam)) == 0)
 639                 goto out;
 640 #endif
 641         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 642         error = tp->t_fb->tfb_tcp_output(tp);
 643
 644 out:
 645         TCPDEBUG2(PRU_CONNECT);
 646         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 647         INP_WUNLOCK(inp);
 648         return (error);
 649 }
 650 #endif /* INET6 */
 651
 652 /*
 653  * Initiate disconnect from peer.
 654  * If connection never passed embryonic stage, just drop;
 655  * else if don't need to let data drain, then can just drop anyways,
 656  * else have to begin TCP shutdown process: mark socket disconnecting,
 657  * drain unread data, state switch to reflect user close, and
 658  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 659  * when peer sends FIN and acks ours.
 660  *
 661  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 662  */
 663 static int
 664 tcp_usr_disconnect(struct socket *so)
 665 {
 666         struct inpcb *inp;
 667         struct tcpcb *tp = NULL;
 668         int error = 0;
 669
 670         TCPDEBUG0;
 671         INP_INFO_RLOCK(&V_tcbinfo);
 672         inp = sotoinpcb(so);
 673         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 674         INP_WLOCK(inp);
 675         if (inp->inp_flags & INP_TIMEWAIT)
 676                 goto out;
 677         if (inp->inp_flags & INP_DROPPED) {
 678                 error = ECONNRESET;
 679                 goto out;
 680         }
 681         tp = intotcpcb(inp);
 682         TCPDEBUG1();
 683         tcp_disconnect(tp);
 684 out:
 685         TCPDEBUG2(PRU_DISCONNECT);
 686         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 687         INP_WUNLOCK(inp);
 688         INP_INFO_RUNLOCK(&V_tcbinfo);
 689         return (error);
 690 }
 691
 692 #ifdef INET
 693 /*
 694  * Accept a connection.  Essentially all the work is done at higher levels;
 695  * just return the address of the peer, storing through addr.
 696  */
 697 static int
 698 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 699 {
 700         int error = 0;
 701         struct inpcb *inp = NULL;
 702         struct tcpcb *tp = NULL;
 703         struct in_addr addr;
 704         in_port_t port = 0;
 705         TCPDEBUG0;
 706
 707         if (so->so_state & SS_ISDISCONNECTED)
 708                 return (ECONNABORTED);
 709
 710         inp = sotoinpcb(so);
 711         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 712         INP_WLOCK(inp);
 713         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 714                 error = ECONNABORTED;
 715                 goto out;
 716         }
 717         tp = intotcpcb(inp);
 718         TCPDEBUG1();
 719
 720         /*
 721          * We inline in_getpeeraddr and COMMON_END here, so that we can
 722          * copy the data of interest and defer the malloc until after we
 723          * release the lock.
 724          */
 725         port = inp->inp_fport;
 726         addr = inp->inp_faddr;
 727
 728 out:
 729         TCPDEBUG2(PRU_ACCEPT);
 730         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 731         INP_WUNLOCK(inp);
 732         if (error == 0)
 733                 *nam = in_sockaddr(port, &addr);
 734         return error;
 735 }
 736 #endif /* INET */
 737
 738 #ifdef INET6
 739 static int
 740 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 741 {
 742         struct inpcb *inp = NULL;
 743         int error = 0;
 744         struct tcpcb *tp = NULL;
 745         struct in_addr addr;
 746         struct in6_addr addr6;
 747         in_port_t port = 0;
 748         int v4 = 0;
 749         TCPDEBUG0;
 750
 751         if (so->so_state & SS_ISDISCONNECTED)
 752                 return (ECONNABORTED);
 753
 754         inp = sotoinpcb(so);
 755         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 756         INP_INFO_RLOCK(&V_tcbinfo);
 757         INP_WLOCK(inp);
 758         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 759                 error = ECONNABORTED;
 760                 goto out;
 761         }
 762         tp = intotcpcb(inp);
 763         TCPDEBUG1();
 764
 765         /*
 766          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 767          * copy the data of interest and defer the malloc until after we
 768          * release the lock.
 769          */
 770         if (inp->inp_vflag & INP_IPV4) {
 771                 v4 = 1;
 772                 port = inp->inp_fport;
 773                 addr = inp->inp_faddr;
 774         } else {
 775                 port = inp->inp_fport;
 776                 addr6 = inp->in6p_faddr;
 777         }
 778
 779 out:
 780         TCPDEBUG2(PRU_ACCEPT);
 781         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 782         INP_WUNLOCK(inp);
 783         INP_INFO_RUNLOCK(&V_tcbinfo);
 784         if (error == 0) {
 785                 if (v4)
 786                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 787                 else
 788                         *nam = in6_sockaddr(port, &addr6);
 789         }
 790         return error;
 791 }
 792 #endif /* INET6 */
 793
 794 /*
 795  * Mark the connection as being incapable of further output.
 796  */
 797 static int
 798 tcp_usr_shutdown(struct socket *so)
 799 {
 800         int error = 0;
 801         struct inpcb *inp;
 802         struct tcpcb *tp = NULL;
 803
 804         TCPDEBUG0;
 805         INP_INFO_RLOCK(&V_tcbinfo);
 806         inp = sotoinpcb(so);
 807         KASSERT(inp != NULL, ("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         socantsendmore(so);
 816         tcp_usrclosed(tp);
 817         if (!(inp->inp_flags & INP_DROPPED))
 818                 error = tp->t_fb->tfb_tcp_output(tp);
 819
 820 out:
 821         TCPDEBUG2(PRU_SHUTDOWN);
 822         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 823         INP_WUNLOCK(inp);
 824         INP_INFO_RUNLOCK(&V_tcbinfo);
 825
 826         return (error);
 827 }
 828
 829 /*
 830  * After a receive, possibly send window update to peer.
 831  */
 832 static int
 833 tcp_usr_rcvd(struct socket *so, int flags)
 834 {
 835         struct inpcb *inp;
 836         struct tcpcb *tp = NULL;
 837         int error = 0;
 838
 839         TCPDEBUG0;
 840         inp = sotoinpcb(so);
 841         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 842         INP_WLOCK(inp);
 843         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 844                 error = ECONNRESET;
 845                 goto out;
 846         }
 847         tp = intotcpcb(inp);
 848         TCPDEBUG1();
 849 #ifdef TCP_RFC7413
 850         /*
 851          * For passively-created TFO connections, don't attempt a window
 852          * update while still in SYN_RECEIVED as this may trigger an early
 853          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 854          * application response data, or failing that, when the DELACK timer
 855          * expires.
 856          */
 857         if (IS_FASTOPEN(tp->t_flags) &&
 858             (tp->t_state == TCPS_SYN_RECEIVED))
 859                 goto out;
 860 #endif
 861 #ifdef TCP_OFFLOAD
 862         if (tp->t_flags & TF_TOE)
 863                 tcp_offload_rcvd(tp);
 864         else
 865 #endif
 866         tp->t_fb->tfb_tcp_output(tp);
 867
 868 out:
 869         TCPDEBUG2(PRU_RCVD);
 870         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 871         INP_WUNLOCK(inp);
 872         return (error);
 873 }
 874
 875 /*
 876  * Do a send by putting data in output queue and updating urgent
 877  * marker if URG set.  Possibly send more data.  Unlike the other
 878  * pru_*() routines, the mbuf chains are our responsibility.  We
 879  * must either enqueue them or free them.  The other pru_* routines
 880  * generally are caller-frees.
 881  */
 882 static int
 883 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 884     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 885 {
 886         int error = 0;
 887         struct inpcb *inp;
 888         struct tcpcb *tp = NULL;
 889 #ifdef INET6
 890         int isipv6;
 891 #endif
 892         TCPDEBUG0;
 893
 894         /*
 895          * We require the pcbinfo lock if we will close the socket as part of
 896          * this call.
 897          */
 898         if (flags & PRUS_EOF)
 899                 INP_INFO_RLOCK(&V_tcbinfo);
 900         inp = sotoinpcb(so);
 901         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 902         INP_WLOCK(inp);
 903         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 904                 if (control)
 905                         m_freem(control);
 906                 /*
 907                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 908                  * for freeing memory.
 909                  */
 910                 if (m && (flags & PRUS_NOTREADY) == 0)
 911                         m_freem(m);
 912                 error = ECONNRESET;
 913                 goto out;
 914         }
 915 #ifdef INET6
 916         isipv6 = nam && nam->sa_family == AF_INET6;
 917 #endif /* INET6 */
 918         tp = intotcpcb(inp);
 919         TCPDEBUG1();
 920         if (control) {
 921                 /* TCP doesn't do control messages (rights, creds, etc) */
 922                 if (control->m_len) {
 923                         m_freem(control);
 924                         if (m)
 925                                 m_freem(m);
 926                         error = EINVAL;
 927                         goto out;
 928                 }
 929                 m_freem(control);       /* empty control, just free it */
 930         }
 931         if (!(flags & PRUS_OOB)) {
 932                 sbappendstream(&so->so_snd, m, flags);
 933                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 934                         /*
 935                          * Do implied connect if not yet connected,
 936                          * initialize window to default value, and
 937                          * initialize maxseg using peer's cached MSS.
 938                          */
 939 #ifdef INET6
 940                         if (isipv6)
 941                                 error = tcp6_connect(tp, nam, td);
 942 #endif /* INET6 */
 943 #if defined(INET6) && defined(INET)
 944                         else
 945 #endif
 946 #ifdef INET
 947                                 error = tcp_connect(tp, nam, td);
 948 #endif
 949                         if (error)
 950                                 goto out;
 951                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 952                         tcp_mss(tp, -1);
 953                 }
 954                 if (flags & PRUS_EOF) {
 955                         /*
 956                          * Close the send side of the connection after
 957                          * the data is sent.
 958                          */
 959                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 960                         socantsendmore(so);
 961                         tcp_usrclosed(tp);
 962                 }
 963                 if (!(inp->inp_flags & INP_DROPPED) &&
 964                     !(flags & PRUS_NOTREADY)) {
 965                         if (flags & PRUS_MORETOCOME)
 966                                 tp->t_flags |= TF_MORETOCOME;
 967                         error = tp->t_fb->tfb_tcp_output(tp);
 968                         if (flags & PRUS_MORETOCOME)
 969                                 tp->t_flags &= ~TF_MORETOCOME;
 970                 }
 971         } else {
 972                 /*
 973                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 974                  */
 975                 SOCKBUF_LOCK(&so->so_snd);
 976                 if (sbspace(&so->so_snd) < -512) {
 977                         SOCKBUF_UNLOCK(&so->so_snd);
 978                         m_freem(m);
 979                         error = ENOBUFS;
 980                         goto out;
 981                 }
 982                 /*
 983                  * According to RFC961 (Assigned Protocols),
 984                  * the urgent pointer points to the last octet
 985                  * of urgent data.  We continue, however,
 986                  * to consider it to indicate the first octet
 987                  * of data past the urgent section.
 988                  * Otherwise, snd_up should be one lower.
 989                  */
 990                 sbappendstream_locked(&so->so_snd, m, flags);
 991                 SOCKBUF_UNLOCK(&so->so_snd);
 992                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 993                         /*
 994                          * Do implied connect if not yet connected,
 995                          * initialize window to default value, and
 996                          * initialize maxseg using peer's cached MSS.
 997                          */
 998 #ifdef INET6
 999                         if (isipv6)
1000                                 error = tcp6_connect(tp, nam, td);
1001 #endif /* INET6 */
1002 #if defined(INET6) && defined(INET)
1003                         else
1004 #endif
1005 #ifdef INET
1006                                 error = tcp_connect(tp, nam, td);
1007 #endif
1008                         if (error)
1009                                 goto out;
1010                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1011                         tcp_mss(tp, -1);
1012                 }
1013                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1014                 if (!(flags & PRUS_NOTREADY)) {
1015                         tp->t_flags |= TF_FORCEDATA;
1016                         error = tp->t_fb->tfb_tcp_output(tp);
1017                         tp->t_flags &= ~TF_FORCEDATA;
1018                 }
1019         }
1020 out:
1021         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1022                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1023         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1024                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1025         INP_WUNLOCK(inp);
1026         if (flags & PRUS_EOF)
1027                 INP_INFO_RUNLOCK(&V_tcbinfo);
1028         return (error);
1029 }
1030
1031 static int
1032 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1033 {
1034         struct inpcb *inp;
1035         struct tcpcb *tp;
1036         int error;
1037
1038         inp = sotoinpcb(so);
1039         INP_WLOCK(inp);
1040         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1041                 INP_WUNLOCK(inp);
1042                 for (int i = 0; i < count; i++)
1043                         m = m_free(m);
1044                 return (ECONNRESET);
1045         }
1046         tp = intotcpcb(inp);
1047
1048         SOCKBUF_LOCK(&so->so_snd);
1049         error = sbready(&so->so_snd, m, count);
1050         SOCKBUF_UNLOCK(&so->so_snd);
1051         if (error == 0)
1052                 error = tp->t_fb->tfb_tcp_output(tp);
1053         INP_WUNLOCK(inp);
1054
1055         return (error);
1056 }
1057
1058 /*
1059  * Abort the TCP.  Drop the connection abruptly.
1060  */
1061 static void
1062 tcp_usr_abort(struct socket *so)
1063 {
1064         struct inpcb *inp;
1065         struct tcpcb *tp = NULL;
1066         TCPDEBUG0;
1067
1068         inp = sotoinpcb(so);
1069         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1070
1071         INP_INFO_RLOCK(&V_tcbinfo);
1072         INP_WLOCK(inp);
1073         KASSERT(inp->inp_socket != NULL,
1074             ("tcp_usr_abort: inp_socket == NULL"));
1075
1076         /*
1077          * If we still have full TCP state, and we're not dropped, drop.
1078          */
1079         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1080             !(inp->inp_flags & INP_DROPPED)) {
1081                 tp = intotcpcb(inp);
1082                 TCPDEBUG1();
1083                 tcp_drop(tp, ECONNABORTED);
1084                 TCPDEBUG2(PRU_ABORT);
1085                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1086         }
1087         if (!(inp->inp_flags & INP_DROPPED)) {
1088                 SOCK_LOCK(so);
1089                 so->so_state |= SS_PROTOREF;
1090                 SOCK_UNLOCK(so);
1091                 inp->inp_flags |= INP_SOCKREF;
1092         }
1093         INP_WUNLOCK(inp);
1094         INP_INFO_RUNLOCK(&V_tcbinfo);
1095 }
1096
1097 /*
1098  * TCP socket is closed.  Start friendly disconnect.
1099  */
1100 static void
1101 tcp_usr_close(struct socket *so)
1102 {
1103         struct inpcb *inp;
1104         struct tcpcb *tp = NULL;
1105         TCPDEBUG0;
1106
1107         inp = sotoinpcb(so);
1108         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1109
1110         INP_INFO_RLOCK(&V_tcbinfo);
1111         INP_WLOCK(inp);
1112         KASSERT(inp->inp_socket != NULL,
1113             ("tcp_usr_close: inp_socket == NULL"));
1114
1115         /*
1116          * If we still have full TCP state, and we're not dropped, initiate
1117          * a disconnect.
1118          */
1119         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1120             !(inp->inp_flags & INP_DROPPED)) {
1121                 tp = intotcpcb(inp);
1122                 TCPDEBUG1();
1123                 tcp_disconnect(tp);
1124                 TCPDEBUG2(PRU_CLOSE);
1125                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1126         }
1127         if (!(inp->inp_flags & INP_DROPPED)) {
1128                 SOCK_LOCK(so);
1129                 so->so_state |= SS_PROTOREF;
1130                 SOCK_UNLOCK(so);
1131                 inp->inp_flags |= INP_SOCKREF;
1132         }
1133         INP_WUNLOCK(inp);
1134         INP_INFO_RUNLOCK(&V_tcbinfo);
1135 }
1136
1137 /*
1138  * Receive out-of-band data.
1139  */
1140 static int
1141 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1142 {
1143         int error = 0;
1144         struct inpcb *inp;
1145         struct tcpcb *tp = NULL;
1146
1147         TCPDEBUG0;
1148         inp = sotoinpcb(so);
1149         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1150         INP_WLOCK(inp);
1151         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1152                 error = ECONNRESET;
1153                 goto out;
1154         }
1155         tp = intotcpcb(inp);
1156         TCPDEBUG1();
1157         if ((so->so_oobmark == 0 &&
1158              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1159             so->so_options & SO_OOBINLINE ||
1160             tp->t_oobflags & TCPOOB_HADDATA) {
1161                 error = EINVAL;
1162                 goto out;
1163         }
1164         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1165                 error = EWOULDBLOCK;
1166                 goto out;
1167         }
1168         m->m_len = 1;
1169         *mtod(m, caddr_t) = tp->t_iobc;
1170         if ((flags & MSG_PEEK) == 0)
1171                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1172
1173 out:
1174         TCPDEBUG2(PRU_RCVOOB);
1175         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1176         INP_WUNLOCK(inp);
1177         return (error);
1178 }
1179
1180 #ifdef INET
1181 struct pr_usrreqs tcp_usrreqs = {
1182         .pru_abort =            tcp_usr_abort,
1183         .pru_accept =           tcp_usr_accept,
1184         .pru_attach =           tcp_usr_attach,
1185         .pru_bind =             tcp_usr_bind,
1186         .pru_connect =          tcp_usr_connect,
1187         .pru_control =          in_control,
1188         .pru_detach =           tcp_usr_detach,
1189         .pru_disconnect =       tcp_usr_disconnect,
1190         .pru_listen =           tcp_usr_listen,
1191         .pru_peeraddr =         in_getpeeraddr,
1192         .pru_rcvd =             tcp_usr_rcvd,
1193         .pru_rcvoob =           tcp_usr_rcvoob,
1194         .pru_send =             tcp_usr_send,
1195         .pru_ready =            tcp_usr_ready,
1196         .pru_shutdown =         tcp_usr_shutdown,
1197         .pru_sockaddr =         in_getsockaddr,
1198         .pru_sosetlabel =       in_pcbsosetlabel,
1199         .pru_close =            tcp_usr_close,
1200 };
1201 #endif /* INET */
1202
1203 #ifdef INET6
1204 struct pr_usrreqs tcp6_usrreqs = {
1205         .pru_abort =            tcp_usr_abort,
1206         .pru_accept =           tcp6_usr_accept,
1207         .pru_attach =           tcp_usr_attach,
1208         .pru_bind =             tcp6_usr_bind,
1209         .pru_connect =          tcp6_usr_connect,
1210         .pru_control =          in6_control,
1211         .pru_detach =           tcp_usr_detach,
1212         .pru_disconnect =       tcp_usr_disconnect,
1213         .pru_listen =           tcp6_usr_listen,
1214         .pru_peeraddr =         in6_mapped_peeraddr,
1215         .pru_rcvd =             tcp_usr_rcvd,
1216         .pru_rcvoob =           tcp_usr_rcvoob,
1217         .pru_send =             tcp_usr_send,
1218         .pru_ready =            tcp_usr_ready,
1219         .pru_shutdown =         tcp_usr_shutdown,
1220         .pru_sockaddr =         in6_mapped_sockaddr,
1221         .pru_sosetlabel =       in_pcbsosetlabel,
1222         .pru_close =            tcp_usr_close,
1223 };
1224 #endif /* INET6 */
1225
1226 #ifdef INET
1227 /*
1228  * Common subroutine to open a TCP connection to remote host specified
1229  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1230  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1231  * to choose a local host address (interface).  If there is an existing
1232  * incarnation of the same connection in TIME-WAIT state and if the remote
1233  * host was sending CC options and if the connection duration was < MSL, then
1234  * truncate the previous TIME-WAIT state and proceed.
1235  * Initialize connection parameters and enter SYN-SENT state.
1236  */
1237 static int
1238 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1239 {
1240         struct inpcb *inp = tp->t_inpcb, *oinp;
1241         struct socket *so = inp->inp_socket;
1242         struct in_addr laddr;
1243         u_short lport;
1244         int error;
1245
1246         INP_WLOCK_ASSERT(inp);
1247         INP_HASH_WLOCK(&V_tcbinfo);
1248
1249         if (inp->inp_lport == 0) {
1250                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1251                 if (error)
1252                         goto out;
1253         }
1254
1255         /*
1256          * Cannot simply call in_pcbconnect, because there might be an
1257          * earlier incarnation of this same connection still in
1258          * TIME_WAIT state, creating an ADDRINUSE error.
1259          */
1260         laddr = inp->inp_laddr;
1261         lport = inp->inp_lport;
1262         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1263             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1264         if (error && oinp == NULL)
1265                 goto out;
1266         if (oinp) {
1267                 error = EADDRINUSE;
1268                 goto out;
1269         }
1270         inp->inp_laddr = laddr;
1271         in_pcbrehash(inp);
1272         INP_HASH_WUNLOCK(&V_tcbinfo);
1273
1274         /*
1275          * Compute window scaling to request:
1276          * Scale to fit into sweet spot.  See tcp_syncache.c.
1277          * XXX: This should move to tcp_output().
1278          */
1279         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1280             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1281                 tp->request_r_scale++;
1282
1283         soisconnecting(so);
1284         TCPSTAT_INC(tcps_connattempt);
1285         tcp_state_change(tp, TCPS_SYN_SENT);
1286         tp->iss = tcp_new_isn(tp);
1287         tcp_sendseqinit(tp);
1288
1289         return 0;
1290
1291 out:
1292         INP_HASH_WUNLOCK(&V_tcbinfo);
1293         return (error);
1294 }
1295 #endif /* INET */
1296
1297 #ifdef INET6
1298 static int
1299 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1300 {
1301         struct inpcb *inp = tp->t_inpcb;
1302         int error;
1303
1304         INP_WLOCK_ASSERT(inp);
1305         INP_HASH_WLOCK(&V_tcbinfo);
1306
1307         if (inp->inp_lport == 0) {
1308                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1309                 if (error)
1310                         goto out;
1311         }
1312         error = in6_pcbconnect(inp, nam, td->td_ucred);
1313         if (error != 0)
1314                 goto out;
1315         INP_HASH_WUNLOCK(&V_tcbinfo);
1316
1317         /* Compute window scaling to request.  */
1318         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1319             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1320                 tp->request_r_scale++;
1321
1322         soisconnecting(inp->inp_socket);
1323         TCPSTAT_INC(tcps_connattempt);
1324         tcp_state_change(tp, TCPS_SYN_SENT);
1325         tp->iss = tcp_new_isn(tp);
1326         tcp_sendseqinit(tp);
1327
1328         return 0;
1329
1330 out:
1331         INP_HASH_WUNLOCK(&V_tcbinfo);
1332         return error;
1333 }
1334 #endif /* INET6 */
1335
1336 /*
1337  * Export TCP internal state information via a struct tcp_info, based on the
1338  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1339  * (TCP state machine, etc).  We export all information using FreeBSD-native
1340  * constants -- for example, the numeric values for tcpi_state will differ
1341  * from Linux.
1342  */
1343 static void
1344 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1345 {
1346
1347         INP_WLOCK_ASSERT(tp->t_inpcb);
1348         bzero(ti, sizeof(*ti));
1349
1350         ti->tcpi_state = tp->t_state;
1351         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1352                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1353         if (tp->t_flags & TF_SACK_PERMIT)
1354                 ti->tcpi_options |= TCPI_OPT_SACK;
1355         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1356                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1357                 ti->tcpi_snd_wscale = tp->snd_scale;
1358                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1359         }
1360         if (tp->t_flags & TF_ECN_PERMIT)
1361                 ti->tcpi_options |= TCPI_OPT_ECN;
1362
1363         ti->tcpi_rto = tp->t_rxtcur * tick;
1364         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1365         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1366         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1367
1368         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1369         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1370
1371         /*
1372          * FreeBSD-specific extension fields for tcp_info.
1373          */
1374         ti->tcpi_rcv_space = tp->rcv_wnd;
1375         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1376         ti->tcpi_snd_wnd = tp->snd_wnd;
1377         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1378         ti->tcpi_snd_nxt = tp->snd_nxt;
1379         ti->tcpi_snd_mss = tp->t_maxseg;
1380         ti->tcpi_rcv_mss = tp->t_maxseg;
1381         if (tp->t_flags & TF_TOE)
1382                 ti->tcpi_options |= TCPI_OPT_TOE;
1383         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1384         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1385         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1386 }
1387
1388 /*
1389  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1390  * socket option arguments.  When it re-acquires the lock after the copy, it
1391  * has to revalidate that the connection is still valid for the socket
1392  * option.
1393  */
1394 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1395         INP_WLOCK(inp);                                                 \
1396         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1397                 INP_WUNLOCK(inp);                                       \
1398                 cleanup;                                                \
1399                 return (ECONNRESET);                                    \
1400         }                                                               \
1401         tp = intotcpcb(inp);                                            \
1402 } while(0)
1403 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1404
1405 int
1406 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1407 {
1408         int     error;
1409         struct  inpcb *inp;
1410         struct  tcpcb *tp;
1411         struct tcp_function_block *blk;
1412         struct tcp_function_set fsn;
1413
1414         error = 0;
1415         inp = sotoinpcb(so);
1416         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1417         INP_WLOCK(inp);
1418         if (sopt->sopt_level != IPPROTO_TCP) {
1419 #ifdef INET6
1420                 if (inp->inp_vflag & INP_IPV6PROTO) {
1421                         INP_WUNLOCK(inp);
1422                         error = ip6_ctloutput(so, sopt);
1423                 }
1424 #endif /* INET6 */
1425 #if defined(INET6) && defined(INET)
1426                 else
1427 #endif
1428 #ifdef INET
1429                 {
1430                         INP_WUNLOCK(inp);
1431                         error = ip_ctloutput(so, sopt);
1432                 }
1433 #endif
1434                 return (error);
1435         }
1436         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1437                 INP_WUNLOCK(inp);
1438                 return (ECONNRESET);
1439         }
1440         tp = intotcpcb(inp);
1441         /*
1442          * Protect the TCP option TCP_FUNCTION_BLK so
1443          * that a sub-function can *never* overwrite this.
1444          */
1445         if ((sopt->sopt_dir == SOPT_SET) &&
1446             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1447                 INP_WUNLOCK(inp);
1448                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1449                     sizeof fsn);
1450                 if (error)
1451                         return (error);
1452                 INP_WLOCK_RECHECK(inp);
1453                 blk = find_and_ref_tcp_functions(&fsn);
1454                 if (blk == NULL) {
1455                         INP_WUNLOCK(inp);
1456                         return (ENOENT);
1457                 }
1458                 if (tp->t_fb == blk) {
1459                         /* You already have this */
1460                         refcount_release(&blk->tfb_refcnt);
1461                         INP_WUNLOCK(inp);
1462                         return (0);
1463                 }
1464                 if (tp->t_state != TCPS_CLOSED) {
1465                         int error=EINVAL;
1466                         /*
1467                          * The user has advanced the state
1468                          * past the initial point, we may not
1469                          * be able to switch.
1470                          */
1471                         if (blk->tfb_tcp_handoff_ok != NULL) {
1472                                 /*
1473                                  * Does the stack provide a
1474                                  * query mechanism, if so it may
1475                                  * still be possible?
1476                                  */
1477                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1478                         }
1479                         if (error) {
1480                                 refcount_release(&blk->tfb_refcnt);
1481                                 INP_WUNLOCK(inp);
1482                                 return(error);
1483                         }
1484                 }
1485                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1486                         refcount_release(&blk->tfb_refcnt);
1487                         INP_WUNLOCK(inp);
1488                         return (ENOENT);
1489                 }
1490                 /*
1491                  * Release the old refcnt, the
1492                  * lookup acquired a ref on the
1493                  * new one already.
1494                  */
1495                 if (tp->t_fb->tfb_tcp_fb_fini) {
1496                         /*
1497                          * Tell the stack to cleanup with 0 i.e.
1498                          * the tcb is not going away.
1499                          */
1500                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1501                 }
1502                 refcount_release(&tp->t_fb->tfb_refcnt);
1503                 tp->t_fb = blk;
1504                 if (tp->t_fb->tfb_tcp_fb_init) {
1505                         (*tp->t_fb->tfb_tcp_fb_init)(tp);
1506                 }
1507 #ifdef TCP_OFFLOAD
1508                 if (tp->t_flags & TF_TOE) {
1509                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1510                              sopt->sopt_name);
1511                 }
1512 #endif
1513                 INP_WUNLOCK(inp);
1514                 return (error);
1515         } else if ((sopt->sopt_dir == SOPT_GET) &&
1516             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1517                 strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1518                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1519                 INP_WUNLOCK(inp);
1520                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1521                 return (error);
1522         }
1523         /* Pass in the INP locked, called must unlock it */
1524         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1525 }
1526
1527 int
1528 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1529 {
1530         int     error, opt, optval;
1531         u_int   ui;
1532         struct  tcp_info ti;
1533         struct cc_algo *algo;
1534         char    *pbuf, buf[TCP_CA_NAME_MAX];
1535         size_t  len;
1536
1537         /*
1538          * For TCP_CCALGOOPT forward the control to CC module, for both
1539          * SOPT_SET and SOPT_GET.
1540          */
1541         switch (sopt->sopt_name) {
1542         case TCP_CCALGOOPT:
1543                 INP_WUNLOCK(inp);
1544                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1545                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1546                     sopt->sopt_valsize);
1547                 if (error) {
1548                         free(pbuf, M_TEMP);
1549                         return (error);
1550                 }
1551                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1552                 if (CC_ALGO(tp)->ctl_output != NULL)
1553                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1554                 else
1555                         error = ENOENT;
1556                 INP_WUNLOCK(inp);
1557                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1558                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1559                 free(pbuf, M_TEMP);
1560                 return (error);
1561         }
1562
1563         switch (sopt->sopt_dir) {
1564         case SOPT_SET:
1565                 switch (sopt->sopt_name) {
1566 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1567                 case TCP_MD5SIG:
1568                         if (!TCPMD5_ENABLED()) {
1569                                 INP_WUNLOCK(inp);
1570                                 return (ENOPROTOOPT);
1571                         }
1572                         error = TCPMD5_PCBCTL(inp, sopt);
1573                         if (error)
1574                                 return (error);
1575                         goto unlock_and_done;
1576 #endif /* IPSEC */
1577
1578                 case TCP_NODELAY:
1579                 case TCP_NOOPT:
1580                         INP_WUNLOCK(inp);
1581                         error = sooptcopyin(sopt, &optval, sizeof optval,
1582                             sizeof optval);
1583                         if (error)
1584                                 return (error);
1585
1586                         INP_WLOCK_RECHECK(inp);
1587                         switch (sopt->sopt_name) {
1588                         case TCP_NODELAY:
1589                                 opt = TF_NODELAY;
1590                                 break;
1591                         case TCP_NOOPT:
1592                                 opt = TF_NOOPT;
1593                                 break;
1594                         default:
1595                                 opt = 0; /* dead code to fool gcc */
1596                                 break;
1597                         }
1598
1599                         if (optval)
1600                                 tp->t_flags |= opt;
1601                         else
1602                                 tp->t_flags &= ~opt;
1603 unlock_and_done:
1604 #ifdef TCP_OFFLOAD
1605                         if (tp->t_flags & TF_TOE) {
1606                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1607                                     sopt->sopt_name);
1608                         }
1609 #endif
1610                         INP_WUNLOCK(inp);
1611                         break;
1612
1613                 case TCP_NOPUSH:
1614                         INP_WUNLOCK(inp);
1615                         error = sooptcopyin(sopt, &optval, sizeof optval,
1616                             sizeof optval);
1617                         if (error)
1618                                 return (error);
1619
1620                         INP_WLOCK_RECHECK(inp);
1621                         if (optval)
1622                                 tp->t_flags |= TF_NOPUSH;
1623                         else if (tp->t_flags & TF_NOPUSH) {
1624                                 tp->t_flags &= ~TF_NOPUSH;
1625                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1626                                         error = tp->t_fb->tfb_tcp_output(tp);
1627                         }
1628                         goto unlock_and_done;
1629
1630                 case TCP_MAXSEG:
1631                         INP_WUNLOCK(inp);
1632                         error = sooptcopyin(sopt, &optval, sizeof optval,
1633                             sizeof optval);
1634                         if (error)
1635                                 return (error);
1636
1637                         INP_WLOCK_RECHECK(inp);
1638                         if (optval > 0 && optval <= tp->t_maxseg &&
1639                             optval + 40 >= V_tcp_minmss)
1640                                 tp->t_maxseg = optval;
1641                         else
1642                                 error = EINVAL;
1643                         goto unlock_and_done;
1644
1645                 case TCP_INFO:
1646                         INP_WUNLOCK(inp);
1647                         error = EINVAL;
1648                         break;
1649
1650                 case TCP_CONGESTION:
1651                         INP_WUNLOCK(inp);
1652                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1653                         if (error)
1654                                 break;
1655                         buf[sopt->sopt_valsize] = '\0';
1656                         INP_WLOCK_RECHECK(inp);
1657                         CC_LIST_RLOCK();
1658                         STAILQ_FOREACH(algo, &cc_list, entries)
1659                                 if (strncmp(buf, algo->name,
1660                                     TCP_CA_NAME_MAX) == 0)
1661                                         break;
1662                         CC_LIST_RUNLOCK();
1663                         if (algo == NULL) {
1664                                 INP_WUNLOCK(inp);
1665                                 error = EINVAL;
1666                                 break;
1667                         }
1668                         /*
1669                          * We hold a write lock over the tcb so it's safe to
1670                          * do these things without ordering concerns.
1671                          */
1672                         if (CC_ALGO(tp)->cb_destroy != NULL)
1673                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1674                         CC_ALGO(tp) = algo;
1675                         /*
1676                          * If something goes pear shaped initialising the new
1677                          * algo, fall back to newreno (which does not
1678                          * require initialisation).
1679                          */
1680                         if (algo->cb_init != NULL &&
1681                             algo->cb_init(tp->ccv) != 0) {
1682                                 CC_ALGO(tp) = &newreno_cc_algo;
1683                                 /*
1684                                  * The only reason init should fail is
1685                                  * because of malloc.
1686                                  */
1687                                 error = ENOMEM;
1688                         }
1689                         INP_WUNLOCK(inp);
1690                         break;
1691
1692                 case TCP_KEEPIDLE:
1693                 case TCP_KEEPINTVL:
1694                 case TCP_KEEPINIT:
1695                         INP_WUNLOCK(inp);
1696                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1697                         if (error)
1698                                 return (error);
1699
1700                         if (ui > (UINT_MAX / hz)) {
1701                                 error = EINVAL;
1702                                 break;
1703                         }
1704                         ui *= hz;
1705
1706                         INP_WLOCK_RECHECK(inp);
1707                         switch (sopt->sopt_name) {
1708                         case TCP_KEEPIDLE:
1709                                 tp->t_keepidle = ui;
1710                                 /*
1711                                  * XXX: better check current remaining
1712                                  * timeout and "merge" it with new value.
1713                                  */
1714                                 if ((tp->t_state > TCPS_LISTEN) &&
1715                                     (tp->t_state <= TCPS_CLOSING))
1716                                         tcp_timer_activate(tp, TT_KEEP,
1717                                             TP_KEEPIDLE(tp));
1718                                 break;
1719                         case TCP_KEEPINTVL:
1720                                 tp->t_keepintvl = ui;
1721                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1722                                     (TP_MAXIDLE(tp) > 0))
1723                                         tcp_timer_activate(tp, TT_2MSL,
1724                                             TP_MAXIDLE(tp));
1725                                 break;
1726                         case TCP_KEEPINIT:
1727                                 tp->t_keepinit = ui;
1728                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1729                                     tp->t_state == TCPS_SYN_SENT)
1730                                         tcp_timer_activate(tp, TT_KEEP,
1731                                             TP_KEEPINIT(tp));
1732                                 break;
1733                         }
1734                         goto unlock_and_done;
1735
1736                 case TCP_KEEPCNT:
1737                         INP_WUNLOCK(inp);
1738                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1739                         if (error)
1740                                 return (error);
1741
1742                         INP_WLOCK_RECHECK(inp);
1743                         tp->t_keepcnt = ui;
1744                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1745                             (TP_MAXIDLE(tp) > 0))
1746                                 tcp_timer_activate(tp, TT_2MSL,
1747                                     TP_MAXIDLE(tp));
1748                         goto unlock_and_done;
1749
1750 #ifdef TCPPCAP
1751                 case TCP_PCAP_OUT:
1752                 case TCP_PCAP_IN:
1753                         INP_WUNLOCK(inp);
1754                         error = sooptcopyin(sopt, &optval, sizeof optval,
1755                             sizeof optval);
1756                         if (error)
1757                                 return (error);
1758
1759                         INP_WLOCK_RECHECK(inp);
1760                         if (optval >= 0)
1761                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1762                                         &(tp->t_outpkts) : &(tp->t_inpkts),
1763                                         optval);
1764                         else
1765                                 error = EINVAL;
1766                         goto unlock_and_done;
1767 #endif
1768
1769 #ifdef TCP_RFC7413
1770                 case TCP_FASTOPEN:
1771                         INP_WUNLOCK(inp);
1772                         if (!V_tcp_fastopen_enabled)
1773                                 return (EPERM);
1774
1775                         error = sooptcopyin(sopt, &optval, sizeof optval,
1776                             sizeof optval);
1777                         if (error)
1778                                 return (error);
1779
1780                         INP_WLOCK_RECHECK(inp);
1781                         if (optval) {
1782                                 tp->t_flags |= TF_FASTOPEN;
1783                                 if ((tp->t_state == TCPS_LISTEN) &&
1784                                     (tp->t_tfo_pending == NULL))
1785                                         tp->t_tfo_pending =
1786                                             tcp_fastopen_alloc_counter();
1787                         } else
1788                                 tp->t_flags &= ~TF_FASTOPEN;
1789                         goto unlock_and_done;
1790 #endif
1791
1792                 default:
1793                         INP_WUNLOCK(inp);
1794                         error = ENOPROTOOPT;
1795                         break;
1796                 }
1797                 break;
1798
1799         case SOPT_GET:
1800                 tp = intotcpcb(inp);
1801                 switch (sopt->sopt_name) {
1802 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1803                 case TCP_MD5SIG:
1804                         if (!TCPMD5_ENABLED()) {
1805                                 INP_WUNLOCK(inp);
1806                                 return (ENOPROTOOPT);
1807                         }
1808                         error = TCPMD5_PCBCTL(inp, sopt);
1809                         break;
1810 #endif
1811
1812                 case TCP_NODELAY:
1813                         optval = tp->t_flags & TF_NODELAY;
1814                         INP_WUNLOCK(inp);
1815                         error = sooptcopyout(sopt, &optval, sizeof optval);
1816                         break;
1817                 case TCP_MAXSEG:
1818                         optval = tp->t_maxseg;
1819                         INP_WUNLOCK(inp);
1820                         error = sooptcopyout(sopt, &optval, sizeof optval);
1821                         break;
1822                 case TCP_NOOPT:
1823                         optval = tp->t_flags & TF_NOOPT;
1824                         INP_WUNLOCK(inp);
1825                         error = sooptcopyout(sopt, &optval, sizeof optval);
1826                         break;
1827                 case TCP_NOPUSH:
1828                         optval = tp->t_flags & TF_NOPUSH;
1829                         INP_WUNLOCK(inp);
1830                         error = sooptcopyout(sopt, &optval, sizeof optval);
1831                         break;
1832                 case TCP_INFO:
1833                         tcp_fill_info(tp, &ti);
1834                         INP_WUNLOCK(inp);
1835                         error = sooptcopyout(sopt, &ti, sizeof ti);
1836                         break;
1837                 case TCP_CONGESTION:
1838                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1839                         INP_WUNLOCK(inp);
1840                         error = sooptcopyout(sopt, buf, len + 1);
1841                         break;
1842                 case TCP_KEEPIDLE:
1843                 case TCP_KEEPINTVL:
1844                 case TCP_KEEPINIT:
1845                 case TCP_KEEPCNT:
1846                         switch (sopt->sopt_name) {
1847                         case TCP_KEEPIDLE:
1848                                 ui = TP_KEEPIDLE(tp) / hz;
1849                                 break;
1850                         case TCP_KEEPINTVL:
1851                                 ui = TP_KEEPINTVL(tp) / hz;
1852                                 break;
1853                         case TCP_KEEPINIT:
1854                                 ui = TP_KEEPINIT(tp) / hz;
1855                                 break;
1856                         case TCP_KEEPCNT:
1857                                 ui = TP_KEEPCNT(tp);
1858                                 break;
1859                         }
1860                         INP_WUNLOCK(inp);
1861                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1862                         break;
1863 #ifdef TCPPCAP
1864                 case TCP_PCAP_OUT:
1865                 case TCP_PCAP_IN:
1866                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1867                                         &(tp->t_outpkts) : &(tp->t_inpkts));
1868                         INP_WUNLOCK(inp);
1869                         error = sooptcopyout(sopt, &optval, sizeof optval);
1870                         break;
1871 #endif
1872
1873 #ifdef TCP_RFC7413
1874                 case TCP_FASTOPEN:
1875                         optval = tp->t_flags & TF_FASTOPEN;
1876                         INP_WUNLOCK(inp);
1877                         error = sooptcopyout(sopt, &optval, sizeof optval);
1878                         break;
1879 #endif
1880                 default:
1881                         INP_WUNLOCK(inp);
1882                         error = ENOPROTOOPT;
1883                         break;
1884                 }
1885                 break;
1886         }
1887         return (error);
1888 }
1889 #undef INP_WLOCK_RECHECK
1890 #undef INP_WLOCK_RECHECK_CLEANUP
1891
1892 /*
1893  * Attach TCP protocol to socket, allocating
1894  * internet protocol control block, tcp control block,
1895  * bufer space, and entering LISTEN state if to accept connections.
1896  */
1897 static int
1898 tcp_attach(struct socket *so)
1899 {
1900         struct tcpcb *tp;
1901         struct inpcb *inp;
1902         int error;
1903
1904         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1905                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1906                 if (error)
1907                         return (error);
1908         }
1909         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1910         so->so_snd.sb_flags |= SB_AUTOSIZE;
1911         INP_INFO_RLOCK(&V_tcbinfo);
1912         error = in_pcballoc(so, &V_tcbinfo);
1913         if (error) {
1914                 INP_INFO_RUNLOCK(&V_tcbinfo);
1915                 return (error);
1916         }
1917         inp = sotoinpcb(so);
1918 #ifdef INET6
1919         if (inp->inp_vflag & INP_IPV6PROTO) {
1920                 inp->inp_vflag |= INP_IPV6;
1921                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1922                         inp->inp_vflag |= INP_IPV4;
1923                 inp->in6p_hops = -1;    /* use kernel default */
1924         }
1925         else
1926 #endif
1927         inp->inp_vflag |= INP_IPV4;
1928         tp = tcp_newtcpcb(inp);
1929         if (tp == NULL) {
1930                 in_pcbdetach(inp);
1931                 in_pcbfree(inp);
1932                 INP_INFO_RUNLOCK(&V_tcbinfo);
1933                 return (ENOBUFS);
1934         }
1935         tp->t_state = TCPS_CLOSED;
1936         INP_WUNLOCK(inp);
1937         INP_INFO_RUNLOCK(&V_tcbinfo);
1938         TCPSTATES_INC(TCPS_CLOSED);
1939         return (0);
1940 }
1941
1942 /*
1943  * Initiate (or continue) disconnect.
1944  * If embryonic state, just send reset (once).
1945  * If in ``let data drain'' option and linger null, just drop.
1946  * Otherwise (hard), mark socket disconnecting and drop
1947  * current input data; switch states based on user close, and
1948  * send segment to peer (with FIN).
1949  */
1950 static void
1951 tcp_disconnect(struct tcpcb *tp)
1952 {
1953         struct inpcb *inp = tp->t_inpcb;
1954         struct socket *so = inp->inp_socket;
1955
1956         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1957         INP_WLOCK_ASSERT(inp);
1958
1959         /*
1960          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1961          * socket is still open.
1962          */
1963         if (tp->t_state < TCPS_ESTABLISHED) {
1964                 tp = tcp_close(tp);
1965                 KASSERT(tp != NULL,
1966                     ("tcp_disconnect: tcp_close() returned NULL"));
1967         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1968                 tp = tcp_drop(tp, 0);
1969                 KASSERT(tp != NULL,
1970                     ("tcp_disconnect: tcp_drop() returned NULL"));
1971         } else {
1972                 soisdisconnecting(so);
1973                 sbflush(&so->so_rcv);
1974                 tcp_usrclosed(tp);
1975                 if (!(inp->inp_flags & INP_DROPPED))
1976                         tp->t_fb->tfb_tcp_output(tp);
1977         }
1978 }
1979
1980 /*
1981  * User issued close, and wish to trail through shutdown states:
1982  * if never received SYN, just forget it.  If got a SYN from peer,
1983  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1984  * If already got a FIN from peer, then almost done; go to LAST_ACK
1985  * state.  In all other cases, have already sent FIN to peer (e.g.
1986  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1987  * for peer to send FIN or not respond to keep-alives, etc.
1988  * We can let the user exit from the close as soon as the FIN is acked.
1989  */
1990 static void
1991 tcp_usrclosed(struct tcpcb *tp)
1992 {
1993
1994         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1995         INP_WLOCK_ASSERT(tp->t_inpcb);
1996
1997         switch (tp->t_state) {
1998         case TCPS_LISTEN:
1999 #ifdef TCP_OFFLOAD
2000                 tcp_offload_listen_stop(tp);
2001 #endif
2002                 tcp_state_change(tp, TCPS_CLOSED);
2003                 /* FALLTHROUGH */
2004         case TCPS_CLOSED:
2005                 tp = tcp_close(tp);
2006                 /*
2007                  * tcp_close() should never return NULL here as the socket is
2008                  * still open.
2009                  */
2010                 KASSERT(tp != NULL,
2011                     ("tcp_usrclosed: tcp_close() returned NULL"));
2012                 break;
2013
2014         case TCPS_SYN_SENT:
2015         case TCPS_SYN_RECEIVED:
2016                 tp->t_flags |= TF_NEEDFIN;
2017                 break;
2018
2019         case TCPS_ESTABLISHED:
2020                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2021                 break;
2022
2023         case TCPS_CLOSE_WAIT:
2024                 tcp_state_change(tp, TCPS_LAST_ACK);
2025                 break;
2026         }
2027         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2028                 soisdisconnected(tp->t_inpcb->inp_socket);
2029                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2030                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2031                         int timeout;
2032
2033                         timeout = (tcp_fast_finwait2_recycle) ?
2034                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2035                         tcp_timer_activate(tp, TT_2MSL, timeout);
2036                 }
2037         }
2038 }
2039
2040 #ifdef DDB
2041 static void
2042 db_print_indent(int indent)
2043 {
2044         int i;
2045
2046         for (i = 0; i < indent; i++)
2047                 db_printf(" ");
2048 }
2049
2050 static void
2051 db_print_tstate(int t_state)
2052 {
2053
2054         switch (t_state) {
2055         case TCPS_CLOSED:
2056                 db_printf("TCPS_CLOSED");
2057                 return;
2058
2059         case TCPS_LISTEN:
2060                 db_printf("TCPS_LISTEN");
2061                 return;
2062
2063         case TCPS_SYN_SENT:
2064                 db_printf("TCPS_SYN_SENT");
2065                 return;
2066
2067         case TCPS_SYN_RECEIVED:
2068                 db_printf("TCPS_SYN_RECEIVED");
2069                 return;
2070
2071         case TCPS_ESTABLISHED:
2072                 db_printf("TCPS_ESTABLISHED");
2073                 return;
2074
2075         case TCPS_CLOSE_WAIT:
2076                 db_printf("TCPS_CLOSE_WAIT");
2077                 return;
2078
2079         case TCPS_FIN_WAIT_1:
2080                 db_printf("TCPS_FIN_WAIT_1");
2081                 return;
2082
2083         case TCPS_CLOSING:
2084                 db_printf("TCPS_CLOSING");
2085                 return;
2086
2087         case TCPS_LAST_ACK:
2088                 db_printf("TCPS_LAST_ACK");
2089                 return;
2090
2091         case TCPS_FIN_WAIT_2:
2092                 db_printf("TCPS_FIN_WAIT_2");
2093                 return;
2094
2095         case TCPS_TIME_WAIT:
2096                 db_printf("TCPS_TIME_WAIT");
2097                 return;
2098
2099         default:
2100                 db_printf("unknown");
2101                 return;
2102         }
2103 }
2104
2105 static void
2106 db_print_tflags(u_int t_flags)
2107 {
2108         int comma;
2109
2110         comma = 0;
2111         if (t_flags & TF_ACKNOW) {
2112                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2113                 comma = 1;
2114         }
2115         if (t_flags & TF_DELACK) {
2116                 db_printf("%sTF_DELACK", comma ? ", " : "");
2117                 comma = 1;
2118         }
2119         if (t_flags & TF_NODELAY) {
2120                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2121                 comma = 1;
2122         }
2123         if (t_flags & TF_NOOPT) {
2124                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2125                 comma = 1;
2126         }
2127         if (t_flags & TF_SENTFIN) {
2128                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2129                 comma = 1;
2130         }
2131         if (t_flags & TF_REQ_SCALE) {
2132                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2133                 comma = 1;
2134         }
2135         if (t_flags & TF_RCVD_SCALE) {
2136                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2137                 comma = 1;
2138         }
2139         if (t_flags & TF_REQ_TSTMP) {
2140                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2141                 comma = 1;
2142         }
2143         if (t_flags & TF_RCVD_TSTMP) {
2144                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2145                 comma = 1;
2146         }
2147         if (t_flags & TF_SACK_PERMIT) {
2148                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2149                 comma = 1;
2150         }
2151         if (t_flags & TF_NEEDSYN) {
2152                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2153                 comma = 1;
2154         }
2155         if (t_flags & TF_NEEDFIN) {
2156                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2157                 comma = 1;
2158         }
2159         if (t_flags & TF_NOPUSH) {
2160                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2161                 comma = 1;
2162         }
2163         if (t_flags & TF_MORETOCOME) {
2164                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2165                 comma = 1;
2166         }
2167         if (t_flags & TF_LQ_OVERFLOW) {
2168                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2169                 comma = 1;
2170         }
2171         if (t_flags & TF_LASTIDLE) {
2172                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2173                 comma = 1;
2174         }
2175         if (t_flags & TF_RXWIN0SENT) {
2176                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2177                 comma = 1;
2178         }
2179         if (t_flags & TF_FASTRECOVERY) {
2180                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2181                 comma = 1;
2182         }
2183         if (t_flags & TF_CONGRECOVERY) {
2184                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2185                 comma = 1;
2186         }
2187         if (t_flags & TF_WASFRECOVERY) {
2188                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2189                 comma = 1;
2190         }
2191         if (t_flags & TF_SIGNATURE) {
2192                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2193                 comma = 1;
2194         }
2195         if (t_flags & TF_FORCEDATA) {
2196                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2197                 comma = 1;
2198         }
2199         if (t_flags & TF_TSO) {
2200                 db_printf("%sTF_TSO", comma ? ", " : "");
2201                 comma = 1;
2202         }
2203         if (t_flags & TF_ECN_PERMIT) {
2204                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2205                 comma = 1;
2206         }
2207         if (t_flags & TF_FASTOPEN) {
2208                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2209                 comma = 1;
2210         }
2211 }
2212
2213 static void
2214 db_print_toobflags(char t_oobflags)
2215 {
2216         int comma;
2217
2218         comma = 0;
2219         if (t_oobflags & TCPOOB_HAVEDATA) {
2220                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2221                 comma = 1;
2222         }
2223         if (t_oobflags & TCPOOB_HADDATA) {
2224                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2225                 comma = 1;
2226         }
2227 }
2228
2229 static void
2230 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2231 {
2232
2233         db_print_indent(indent);
2234         db_printf("%s at %p\n", name, tp);
2235
2236         indent += 2;
2237
2238         db_print_indent(indent);
2239         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2240            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2241
2242         db_print_indent(indent);
2243         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2244             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2245
2246         db_print_indent(indent);
2247         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2248             &tp->t_timers->tt_delack, tp->t_inpcb);
2249
2250         db_print_indent(indent);
2251         db_printf("t_state: %d (", tp->t_state);
2252         db_print_tstate(tp->t_state);
2253         db_printf(")\n");
2254
2255         db_print_indent(indent);
2256         db_printf("t_flags: 0x%x (", tp->t_flags);
2257         db_print_tflags(tp->t_flags);
2258         db_printf(")\n");
2259
2260         db_print_indent(indent);
2261         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2262             tp->snd_una, tp->snd_max, tp->snd_nxt);
2263
2264         db_print_indent(indent);
2265         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2266            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2267
2268         db_print_indent(indent);
2269         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2270             tp->iss, tp->irs, tp->rcv_nxt);
2271
2272         db_print_indent(indent);
2273         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2274             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2275
2276         db_print_indent(indent);
2277         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2278            tp->snd_wnd, tp->snd_cwnd);
2279
2280         db_print_indent(indent);
2281         db_printf("snd_ssthresh: %u   snd_recover: "
2282             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2283
2284         db_print_indent(indent);
2285         db_printf("t_rcvtime: %u   t_startime: %u\n",
2286             tp->t_rcvtime, tp->t_starttime);
2287
2288         db_print_indent(indent);
2289         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2290             tp->t_rtttime, tp->t_rtseq);
2291
2292         db_print_indent(indent);
2293         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2294             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2295
2296         db_print_indent(indent);
2297         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2298             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2299             tp->t_rttbest);
2300
2301         db_print_indent(indent);
2302         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2303             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2304
2305         db_print_indent(indent);
2306         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2307         db_print_toobflags(tp->t_oobflags);
2308         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2309
2310         db_print_indent(indent);
2311         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2312             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2313
2314         db_print_indent(indent);
2315         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2316             tp->ts_recent, tp->ts_recent_age);
2317
2318         db_print_indent(indent);
2319         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2320             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2321
2322         db_print_indent(indent);
2323         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2324             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2325             tp->snd_recover_prev, tp->t_badrxtwin);
2326
2327         db_print_indent(indent);
2328         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2329             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2330
2331         db_print_indent(indent);
2332         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2333             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2334
2335         /* Skip sackblks, sackhint. */
2336
2337         db_print_indent(indent);
2338         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2339             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2340 }
2341
2342 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2343 {
2344         struct tcpcb *tp;
2345
2346         if (!have_addr) {
2347                 db_printf("usage: show tcpcb <addr>\n");
2348                 return;
2349         }
2350         tp = (struct tcpcb *)addr;
2351
2352         db_print_tcpcb(tp, "tcpcb", 0);
2353 }
2354 #endif