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_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 (tp->t_flags & TF_FASTOPEN)
 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 (tp->t_flags & TF_FASTOPEN)
 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
 601                 in6_sin6_2_sin(&sin, sin6p);
 602                 inp->inp_vflag |= INP_IPV4;
 603                 inp->inp_vflag &= ~INP_IPV6;
 604                 if ((error = prison_remote_ip4(td->td_ucred,
 605                     &sin.sin_addr)) != 0)
 606                         goto out;
 607                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 608                         goto out;
 609 #ifdef TCP_OFFLOAD
 610                 if (registered_toedevs > 0 &&
 611                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 612                     (error = tcp_offload_connect(so, nam)) == 0)
 613                         goto out;
 614 #endif
 615                 error = tp->t_fb->tfb_tcp_output(tp);
 616                 goto out;
 617         }
 618 #endif
 619         inp->inp_vflag &= ~INP_IPV4;
 620         inp->inp_vflag |= INP_IPV6;
 621         inp->inp_inc.inc_flags |= INC_ISIPV6;
 622         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 623                 goto out;
 624         if ((error = tcp6_connect(tp, nam, td)) != 0)
 625                 goto out;
 626 #ifdef TCP_OFFLOAD
 627         if (registered_toedevs > 0 &&
 628             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 629             (error = tcp_offload_connect(so, nam)) == 0)
 630                 goto out;
 631 #endif
 632         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 633         error = tp->t_fb->tfb_tcp_output(tp);
 634
 635 out:
 636         TCPDEBUG2(PRU_CONNECT);
 637         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 638         INP_WUNLOCK(inp);
 639         return (error);
 640 }
 641 #endif /* INET6 */
 642
 643 /*
 644  * Initiate disconnect from peer.
 645  * If connection never passed embryonic stage, just drop;
 646  * else if don't need to let data drain, then can just drop anyways,
 647  * else have to begin TCP shutdown process: mark socket disconnecting,
 648  * drain unread data, state switch to reflect user close, and
 649  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 650  * when peer sends FIN and acks ours.
 651  *
 652  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 653  */
 654 static int
 655 tcp_usr_disconnect(struct socket *so)
 656 {
 657         struct inpcb *inp;
 658         struct tcpcb *tp = NULL;
 659         int error = 0;
 660
 661         TCPDEBUG0;
 662         INP_INFO_RLOCK(&V_tcbinfo);
 663         inp = sotoinpcb(so);
 664         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 665         INP_WLOCK(inp);
 666         if (inp->inp_flags & INP_TIMEWAIT)
 667                 goto out;
 668         if (inp->inp_flags & INP_DROPPED) {
 669                 error = ECONNRESET;
 670                 goto out;
 671         }
 672         tp = intotcpcb(inp);
 673         TCPDEBUG1();
 674         tcp_disconnect(tp);
 675 out:
 676         TCPDEBUG2(PRU_DISCONNECT);
 677         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 678         INP_WUNLOCK(inp);
 679         INP_INFO_RUNLOCK(&V_tcbinfo);
 680         return (error);
 681 }
 682
 683 #ifdef INET
 684 /*
 685  * Accept a connection.  Essentially all the work is done at higher levels;
 686  * just return the address of the peer, storing through addr.
 687  */
 688 static int
 689 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 690 {
 691         int error = 0;
 692         struct inpcb *inp = NULL;
 693         struct tcpcb *tp = NULL;
 694         struct in_addr addr;
 695         in_port_t port = 0;
 696         TCPDEBUG0;
 697
 698         if (so->so_state & SS_ISDISCONNECTED)
 699                 return (ECONNABORTED);
 700
 701         inp = sotoinpcb(so);
 702         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 703         INP_WLOCK(inp);
 704         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 705                 error = ECONNABORTED;
 706                 goto out;
 707         }
 708         tp = intotcpcb(inp);
 709         TCPDEBUG1();
 710
 711         /*
 712          * We inline in_getpeeraddr and COMMON_END here, so that we can
 713          * copy the data of interest and defer the malloc until after we
 714          * release the lock.
 715          */
 716         port = inp->inp_fport;
 717         addr = inp->inp_faddr;
 718
 719 out:
 720         TCPDEBUG2(PRU_ACCEPT);
 721         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 722         INP_WUNLOCK(inp);
 723         if (error == 0)
 724                 *nam = in_sockaddr(port, &addr);
 725         return error;
 726 }
 727 #endif /* INET */
 728
 729 #ifdef INET6
 730 static int
 731 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 732 {
 733         struct inpcb *inp = NULL;
 734         int error = 0;
 735         struct tcpcb *tp = NULL;
 736         struct in_addr addr;
 737         struct in6_addr addr6;
 738         in_port_t port = 0;
 739         int v4 = 0;
 740         TCPDEBUG0;
 741
 742         if (so->so_state & SS_ISDISCONNECTED)
 743                 return (ECONNABORTED);
 744
 745         inp = sotoinpcb(so);
 746         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 747         INP_INFO_RLOCK(&V_tcbinfo);
 748         INP_WLOCK(inp);
 749         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 750                 error = ECONNABORTED;
 751                 goto out;
 752         }
 753         tp = intotcpcb(inp);
 754         TCPDEBUG1();
 755
 756         /*
 757          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 758          * copy the data of interest and defer the malloc until after we
 759          * release the lock.
 760          */
 761         if (inp->inp_vflag & INP_IPV4) {
 762                 v4 = 1;
 763                 port = inp->inp_fport;
 764                 addr = inp->inp_faddr;
 765         } else {
 766                 port = inp->inp_fport;
 767                 addr6 = inp->in6p_faddr;
 768         }
 769
 770 out:
 771         TCPDEBUG2(PRU_ACCEPT);
 772         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 773         INP_WUNLOCK(inp);
 774         INP_INFO_RUNLOCK(&V_tcbinfo);
 775         if (error == 0) {
 776                 if (v4)
 777                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 778                 else
 779                         *nam = in6_sockaddr(port, &addr6);
 780         }
 781         return error;
 782 }
 783 #endif /* INET6 */
 784
 785 /*
 786  * Mark the connection as being incapable of further output.
 787  */
 788 static int
 789 tcp_usr_shutdown(struct socket *so)
 790 {
 791         int error = 0;
 792         struct inpcb *inp;
 793         struct tcpcb *tp = NULL;
 794
 795         TCPDEBUG0;
 796         INP_INFO_RLOCK(&V_tcbinfo);
 797         inp = sotoinpcb(so);
 798         KASSERT(inp != NULL, ("inp == NULL"));
 799         INP_WLOCK(inp);
 800         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 801                 error = ECONNRESET;
 802                 goto out;
 803         }
 804         tp = intotcpcb(inp);
 805         TCPDEBUG1();
 806         socantsendmore(so);
 807         tcp_usrclosed(tp);
 808         if (!(inp->inp_flags & INP_DROPPED))
 809                 error = tp->t_fb->tfb_tcp_output(tp);
 810
 811 out:
 812         TCPDEBUG2(PRU_SHUTDOWN);
 813         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 814         INP_WUNLOCK(inp);
 815         INP_INFO_RUNLOCK(&V_tcbinfo);
 816
 817         return (error);
 818 }
 819
 820 /*
 821  * After a receive, possibly send window update to peer.
 822  */
 823 static int
 824 tcp_usr_rcvd(struct socket *so, int flags)
 825 {
 826         struct inpcb *inp;
 827         struct tcpcb *tp = NULL;
 828         int error = 0;
 829
 830         TCPDEBUG0;
 831         inp = sotoinpcb(so);
 832         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 833         INP_WLOCK(inp);
 834         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 835                 error = ECONNRESET;
 836                 goto out;
 837         }
 838         tp = intotcpcb(inp);
 839         TCPDEBUG1();
 840 #ifdef TCP_RFC7413
 841         /*
 842          * For passively-created TFO connections, don't attempt a window
 843          * update while still in SYN_RECEIVED as this may trigger an early
 844          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 845          * application response data, or failing that, when the DELACK timer
 846          * expires.
 847          */
 848         if ((tp->t_flags & TF_FASTOPEN) &&
 849             (tp->t_state == TCPS_SYN_RECEIVED))
 850                 goto out;
 851 #endif
 852 #ifdef TCP_OFFLOAD
 853         if (tp->t_flags & TF_TOE)
 854                 tcp_offload_rcvd(tp);
 855         else
 856 #endif
 857         tp->t_fb->tfb_tcp_output(tp);
 858
 859 out:
 860         TCPDEBUG2(PRU_RCVD);
 861         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 862         INP_WUNLOCK(inp);
 863         return (error);
 864 }
 865
 866 /*
 867  * Do a send by putting data in output queue and updating urgent
 868  * marker if URG set.  Possibly send more data.  Unlike the other
 869  * pru_*() routines, the mbuf chains are our responsibility.  We
 870  * must either enqueue them or free them.  The other pru_* routines
 871  * generally are caller-frees.
 872  */
 873 static int
 874 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 875     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 876 {
 877         int error = 0;
 878         struct inpcb *inp;
 879         struct tcpcb *tp = NULL;
 880 #ifdef INET6
 881         int isipv6;
 882 #endif
 883         TCPDEBUG0;
 884
 885         /*
 886          * We require the pcbinfo lock if we will close the socket as part of
 887          * this call.
 888          */
 889         if (flags & PRUS_EOF)
 890                 INP_INFO_RLOCK(&V_tcbinfo);
 891         inp = sotoinpcb(so);
 892         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 893         INP_WLOCK(inp);
 894         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 895                 if (control)
 896                         m_freem(control);
 897                 /*
 898                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 899                  * for freeing memory.
 900                  */
 901                 if (m && (flags & PRUS_NOTREADY) == 0)
 902                         m_freem(m);
 903                 error = ECONNRESET;
 904                 goto out;
 905         }
 906 #ifdef INET6
 907         isipv6 = nam && nam->sa_family == AF_INET6;
 908 #endif /* INET6 */
 909         tp = intotcpcb(inp);
 910         TCPDEBUG1();
 911         if (control) {
 912                 /* TCP doesn't do control messages (rights, creds, etc) */
 913                 if (control->m_len) {
 914                         m_freem(control);
 915                         if (m)
 916                                 m_freem(m);
 917                         error = EINVAL;
 918                         goto out;
 919                 }
 920                 m_freem(control);       /* empty control, just free it */
 921         }
 922         if (!(flags & PRUS_OOB)) {
 923                 sbappendstream(&so->so_snd, m, flags);
 924                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 925                         /*
 926                          * Do implied connect if not yet connected,
 927                          * initialize window to default value, and
 928                          * initialize maxseg using peer's cached MSS.
 929                          */
 930 #ifdef INET6
 931                         if (isipv6)
 932                                 error = tcp6_connect(tp, nam, td);
 933 #endif /* INET6 */
 934 #if defined(INET6) && defined(INET)
 935                         else
 936 #endif
 937 #ifdef INET
 938                                 error = tcp_connect(tp, nam, td);
 939 #endif
 940                         if (error)
 941                                 goto out;
 942                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 943                         tcp_mss(tp, -1);
 944                 }
 945                 if (flags & PRUS_EOF) {
 946                         /*
 947                          * Close the send side of the connection after
 948                          * the data is sent.
 949                          */
 950                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 951                         socantsendmore(so);
 952                         tcp_usrclosed(tp);
 953                 }
 954                 if (!(inp->inp_flags & INP_DROPPED) &&
 955                     !(flags & PRUS_NOTREADY)) {
 956                         if (flags & PRUS_MORETOCOME)
 957                                 tp->t_flags |= TF_MORETOCOME;
 958                         error = tp->t_fb->tfb_tcp_output(tp);
 959                         if (flags & PRUS_MORETOCOME)
 960                                 tp->t_flags &= ~TF_MORETOCOME;
 961                 }
 962         } else {
 963                 /*
 964                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 965                  */
 966                 SOCKBUF_LOCK(&so->so_snd);
 967                 if (sbspace(&so->so_snd) < -512) {
 968                         SOCKBUF_UNLOCK(&so->so_snd);
 969                         m_freem(m);
 970                         error = ENOBUFS;
 971                         goto out;
 972                 }
 973                 /*
 974                  * According to RFC961 (Assigned Protocols),
 975                  * the urgent pointer points to the last octet
 976                  * of urgent data.  We continue, however,
 977                  * to consider it to indicate the first octet
 978                  * of data past the urgent section.
 979                  * Otherwise, snd_up should be one lower.
 980                  */
 981                 sbappendstream_locked(&so->so_snd, m, flags);
 982                 SOCKBUF_UNLOCK(&so->so_snd);
 983                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 984                         /*
 985                          * Do implied connect if not yet connected,
 986                          * initialize window to default value, and
 987                          * initialize maxseg using peer's cached MSS.
 988                          */
 989 #ifdef INET6
 990                         if (isipv6)
 991                                 error = tcp6_connect(tp, nam, td);
 992 #endif /* INET6 */
 993 #if defined(INET6) && defined(INET)
 994                         else
 995 #endif
 996 #ifdef INET
 997                                 error = tcp_connect(tp, nam, td);
 998 #endif
 999                         if (error)
1000                                 goto out;
1001                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1002                         tcp_mss(tp, -1);
1003                 }
1004                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1005                 if (!(flags & PRUS_NOTREADY)) {
1006                         tp->t_flags |= TF_FORCEDATA;
1007                         error = tp->t_fb->tfb_tcp_output(tp);
1008                         tp->t_flags &= ~TF_FORCEDATA;
1009                 }
1010         }
1011 out:
1012         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1013                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1014         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1015                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1016         INP_WUNLOCK(inp);
1017         if (flags & PRUS_EOF)
1018                 INP_INFO_RUNLOCK(&V_tcbinfo);
1019         return (error);
1020 }
1021
1022 static int
1023 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1024 {
1025         struct inpcb *inp;
1026         struct tcpcb *tp;
1027         int error;
1028
1029         inp = sotoinpcb(so);
1030         INP_WLOCK(inp);
1031         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1032                 INP_WUNLOCK(inp);
1033                 for (int i = 0; i < count; i++)
1034                         m = m_free(m);
1035                 return (ECONNRESET);
1036         }
1037         tp = intotcpcb(inp);
1038
1039         SOCKBUF_LOCK(&so->so_snd);
1040         error = sbready(&so->so_snd, m, count);
1041         SOCKBUF_UNLOCK(&so->so_snd);
1042         if (error == 0)
1043                 error = tp->t_fb->tfb_tcp_output(tp);
1044         INP_WUNLOCK(inp);
1045
1046         return (error);
1047 }
1048
1049 /*
1050  * Abort the TCP.  Drop the connection abruptly.
1051  */
1052 static void
1053 tcp_usr_abort(struct socket *so)
1054 {
1055         struct inpcb *inp;
1056         struct tcpcb *tp = NULL;
1057         TCPDEBUG0;
1058
1059         inp = sotoinpcb(so);
1060         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1061
1062         INP_INFO_RLOCK(&V_tcbinfo);
1063         INP_WLOCK(inp);
1064         KASSERT(inp->inp_socket != NULL,
1065             ("tcp_usr_abort: inp_socket == NULL"));
1066
1067         /*
1068          * If we still have full TCP state, and we're not dropped, drop.
1069          */
1070         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1071             !(inp->inp_flags & INP_DROPPED)) {
1072                 tp = intotcpcb(inp);
1073                 TCPDEBUG1();
1074                 tcp_drop(tp, ECONNABORTED);
1075                 TCPDEBUG2(PRU_ABORT);
1076                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1077         }
1078         if (!(inp->inp_flags & INP_DROPPED)) {
1079                 SOCK_LOCK(so);
1080                 so->so_state |= SS_PROTOREF;
1081                 SOCK_UNLOCK(so);
1082                 inp->inp_flags |= INP_SOCKREF;
1083         }
1084         INP_WUNLOCK(inp);
1085         INP_INFO_RUNLOCK(&V_tcbinfo);
1086 }
1087
1088 /*
1089  * TCP socket is closed.  Start friendly disconnect.
1090  */
1091 static void
1092 tcp_usr_close(struct socket *so)
1093 {
1094         struct inpcb *inp;
1095         struct tcpcb *tp = NULL;
1096         TCPDEBUG0;
1097
1098         inp = sotoinpcb(so);
1099         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1100
1101         INP_INFO_RLOCK(&V_tcbinfo);
1102         INP_WLOCK(inp);
1103         KASSERT(inp->inp_socket != NULL,
1104             ("tcp_usr_close: inp_socket == NULL"));
1105
1106         /*
1107          * If we still have full TCP state, and we're not dropped, initiate
1108          * a disconnect.
1109          */
1110         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1111             !(inp->inp_flags & INP_DROPPED)) {
1112                 tp = intotcpcb(inp);
1113                 TCPDEBUG1();
1114                 tcp_disconnect(tp);
1115                 TCPDEBUG2(PRU_CLOSE);
1116                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1117         }
1118         if (!(inp->inp_flags & INP_DROPPED)) {
1119                 SOCK_LOCK(so);
1120                 so->so_state |= SS_PROTOREF;
1121                 SOCK_UNLOCK(so);
1122                 inp->inp_flags |= INP_SOCKREF;
1123         }
1124         INP_WUNLOCK(inp);
1125         INP_INFO_RUNLOCK(&V_tcbinfo);
1126 }
1127
1128 /*
1129  * Receive out-of-band data.
1130  */
1131 static int
1132 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1133 {
1134         int error = 0;
1135         struct inpcb *inp;
1136         struct tcpcb *tp = NULL;
1137
1138         TCPDEBUG0;
1139         inp = sotoinpcb(so);
1140         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1141         INP_WLOCK(inp);
1142         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1143                 error = ECONNRESET;
1144                 goto out;
1145         }
1146         tp = intotcpcb(inp);
1147         TCPDEBUG1();
1148         if ((so->so_oobmark == 0 &&
1149              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1150             so->so_options & SO_OOBINLINE ||
1151             tp->t_oobflags & TCPOOB_HADDATA) {
1152                 error = EINVAL;
1153                 goto out;
1154         }
1155         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1156                 error = EWOULDBLOCK;
1157                 goto out;
1158         }
1159         m->m_len = 1;
1160         *mtod(m, caddr_t) = tp->t_iobc;
1161         if ((flags & MSG_PEEK) == 0)
1162                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1163
1164 out:
1165         TCPDEBUG2(PRU_RCVOOB);
1166         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1167         INP_WUNLOCK(inp);
1168         return (error);
1169 }
1170
1171 #ifdef INET
1172 struct pr_usrreqs tcp_usrreqs = {
1173         .pru_abort =            tcp_usr_abort,
1174         .pru_accept =           tcp_usr_accept,
1175         .pru_attach =           tcp_usr_attach,
1176         .pru_bind =             tcp_usr_bind,
1177         .pru_connect =          tcp_usr_connect,
1178         .pru_control =          in_control,
1179         .pru_detach =           tcp_usr_detach,
1180         .pru_disconnect =       tcp_usr_disconnect,
1181         .pru_listen =           tcp_usr_listen,
1182         .pru_peeraddr =         in_getpeeraddr,
1183         .pru_rcvd =             tcp_usr_rcvd,
1184         .pru_rcvoob =           tcp_usr_rcvoob,
1185         .pru_send =             tcp_usr_send,
1186         .pru_ready =            tcp_usr_ready,
1187         .pru_shutdown =         tcp_usr_shutdown,
1188         .pru_sockaddr =         in_getsockaddr,
1189         .pru_sosetlabel =       in_pcbsosetlabel,
1190         .pru_close =            tcp_usr_close,
1191 };
1192 #endif /* INET */
1193
1194 #ifdef INET6
1195 struct pr_usrreqs tcp6_usrreqs = {
1196         .pru_abort =            tcp_usr_abort,
1197         .pru_accept =           tcp6_usr_accept,
1198         .pru_attach =           tcp_usr_attach,
1199         .pru_bind =             tcp6_usr_bind,
1200         .pru_connect =          tcp6_usr_connect,
1201         .pru_control =          in6_control,
1202         .pru_detach =           tcp_usr_detach,
1203         .pru_disconnect =       tcp_usr_disconnect,
1204         .pru_listen =           tcp6_usr_listen,
1205         .pru_peeraddr =         in6_mapped_peeraddr,
1206         .pru_rcvd =             tcp_usr_rcvd,
1207         .pru_rcvoob =           tcp_usr_rcvoob,
1208         .pru_send =             tcp_usr_send,
1209         .pru_ready =            tcp_usr_ready,
1210         .pru_shutdown =         tcp_usr_shutdown,
1211         .pru_sockaddr =         in6_mapped_sockaddr,
1212         .pru_sosetlabel =       in_pcbsosetlabel,
1213         .pru_close =            tcp_usr_close,
1214 };
1215 #endif /* INET6 */
1216
1217 #ifdef INET
1218 /*
1219  * Common subroutine to open a TCP connection to remote host specified
1220  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1221  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1222  * to choose a local host address (interface).  If there is an existing
1223  * incarnation of the same connection in TIME-WAIT state and if the remote
1224  * host was sending CC options and if the connection duration was < MSL, then
1225  * truncate the previous TIME-WAIT state and proceed.
1226  * Initialize connection parameters and enter SYN-SENT state.
1227  */
1228 static int
1229 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1230 {
1231         struct inpcb *inp = tp->t_inpcb, *oinp;
1232         struct socket *so = inp->inp_socket;
1233         struct in_addr laddr;
1234         u_short lport;
1235         int error;
1236
1237         INP_WLOCK_ASSERT(inp);
1238         INP_HASH_WLOCK(&V_tcbinfo);
1239
1240         if (inp->inp_lport == 0) {
1241                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1242                 if (error)
1243                         goto out;
1244         }
1245
1246         /*
1247          * Cannot simply call in_pcbconnect, because there might be an
1248          * earlier incarnation of this same connection still in
1249          * TIME_WAIT state, creating an ADDRINUSE error.
1250          */
1251         laddr = inp->inp_laddr;
1252         lport = inp->inp_lport;
1253         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1254             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1255         if (error && oinp == NULL)
1256                 goto out;
1257         if (oinp) {
1258                 error = EADDRINUSE;
1259                 goto out;
1260         }
1261         inp->inp_laddr = laddr;
1262         in_pcbrehash(inp);
1263         INP_HASH_WUNLOCK(&V_tcbinfo);
1264
1265         /*
1266          * Compute window scaling to request:
1267          * Scale to fit into sweet spot.  See tcp_syncache.c.
1268          * XXX: This should move to tcp_output().
1269          */
1270         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1271             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1272                 tp->request_r_scale++;
1273
1274         soisconnecting(so);
1275         TCPSTAT_INC(tcps_connattempt);
1276         tcp_state_change(tp, TCPS_SYN_SENT);
1277         tp->iss = tcp_new_isn(tp);
1278         tcp_sendseqinit(tp);
1279
1280         return 0;
1281
1282 out:
1283         INP_HASH_WUNLOCK(&V_tcbinfo);
1284         return (error);
1285 }
1286 #endif /* INET */
1287
1288 #ifdef INET6
1289 static int
1290 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1291 {
1292         struct inpcb *inp = tp->t_inpcb;
1293         int error;
1294
1295         INP_WLOCK_ASSERT(inp);
1296         INP_HASH_WLOCK(&V_tcbinfo);
1297
1298         if (inp->inp_lport == 0) {
1299                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1300                 if (error)
1301                         goto out;
1302         }
1303         error = in6_pcbconnect(inp, nam, td->td_ucred);
1304         if (error != 0)
1305                 goto out;
1306         INP_HASH_WUNLOCK(&V_tcbinfo);
1307
1308         /* Compute window scaling to request.  */
1309         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1310             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1311                 tp->request_r_scale++;
1312
1313         soisconnecting(inp->inp_socket);
1314         TCPSTAT_INC(tcps_connattempt);
1315         tcp_state_change(tp, TCPS_SYN_SENT);
1316         tp->iss = tcp_new_isn(tp);
1317         tcp_sendseqinit(tp);
1318
1319         return 0;
1320
1321 out:
1322         INP_HASH_WUNLOCK(&V_tcbinfo);
1323         return error;
1324 }
1325 #endif /* INET6 */
1326
1327 /*
1328  * Export TCP internal state information via a struct tcp_info, based on the
1329  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1330  * (TCP state machine, etc).  We export all information using FreeBSD-native
1331  * constants -- for example, the numeric values for tcpi_state will differ
1332  * from Linux.
1333  */
1334 static void
1335 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1336 {
1337
1338         INP_WLOCK_ASSERT(tp->t_inpcb);
1339         bzero(ti, sizeof(*ti));
1340
1341         ti->tcpi_state = tp->t_state;
1342         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1343                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1344         if (tp->t_flags & TF_SACK_PERMIT)
1345                 ti->tcpi_options |= TCPI_OPT_SACK;
1346         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1347                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1348                 ti->tcpi_snd_wscale = tp->snd_scale;
1349                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1350         }
1351         if (tp->t_flags & TF_ECN_PERMIT)
1352                 ti->tcpi_options |= TCPI_OPT_ECN;
1353
1354         ti->tcpi_rto = tp->t_rxtcur * tick;
1355         ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1356         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1357         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1358
1359         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1360         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1361
1362         /*
1363          * FreeBSD-specific extension fields for tcp_info.
1364          */
1365         ti->tcpi_rcv_space = tp->rcv_wnd;
1366         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1367         ti->tcpi_snd_wnd = tp->snd_wnd;
1368         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1369         ti->tcpi_snd_nxt = tp->snd_nxt;
1370         ti->tcpi_snd_mss = tp->t_maxseg;
1371         ti->tcpi_rcv_mss = tp->t_maxseg;
1372         if (tp->t_flags & TF_TOE)
1373                 ti->tcpi_options |= TCPI_OPT_TOE;
1374         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1375         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1376         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1377 }
1378
1379 /*
1380  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1381  * socket option arguments.  When it re-acquires the lock after the copy, it
1382  * has to revalidate that the connection is still valid for the socket
1383  * option.
1384  */
1385 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1386         INP_WLOCK(inp);                                                 \
1387         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1388                 INP_WUNLOCK(inp);                                       \
1389                 cleanup;                                                \
1390                 return (ECONNRESET);                                    \
1391         }                                                               \
1392         tp = intotcpcb(inp);                                            \
1393 } while(0)
1394 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1395
1396 int
1397 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1398 {
1399         int     error;
1400         struct  inpcb *inp;
1401         struct  tcpcb *tp;
1402         struct tcp_function_block *blk;
1403         struct tcp_function_set fsn;
1404
1405         error = 0;
1406         inp = sotoinpcb(so);
1407         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1408         INP_WLOCK(inp);
1409         if (sopt->sopt_level != IPPROTO_TCP) {
1410 #ifdef INET6
1411                 if (inp->inp_vflag & INP_IPV6PROTO) {
1412                         INP_WUNLOCK(inp);
1413                         error = ip6_ctloutput(so, sopt);
1414                 }
1415 #endif /* INET6 */
1416 #if defined(INET6) && defined(INET)
1417                 else
1418 #endif
1419 #ifdef INET
1420                 {
1421                         INP_WUNLOCK(inp);
1422                         error = ip_ctloutput(so, sopt);
1423                 }
1424 #endif
1425                 return (error);
1426         }
1427         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1428                 INP_WUNLOCK(inp);
1429                 return (ECONNRESET);
1430         }
1431         tp = intotcpcb(inp);
1432         /*
1433          * Protect the TCP option TCP_FUNCTION_BLK so
1434          * that a sub-function can *never* overwrite this.
1435          */
1436         if ((sopt->sopt_dir == SOPT_SET) &&
1437             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1438                 INP_WUNLOCK(inp);
1439                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1440                     sizeof fsn);
1441                 if (error)
1442                         return (error);
1443                 INP_WLOCK_RECHECK(inp);
1444                 if (tp->t_state != TCPS_CLOSED) {
1445                         /*
1446                          * The user has advanced the state
1447                          * past the initial point, we can't
1448                          * switch since we are down the road
1449                          * and a new set of functions may
1450                          * not be compatibile.
1451                          */
1452                         INP_WUNLOCK(inp);
1453                         return(EINVAL);
1454                 }
1455                 blk = find_and_ref_tcp_functions(&fsn);
1456                 if (blk == NULL) {
1457                         INP_WUNLOCK(inp);
1458                         return (ENOENT);
1459                 }
1460                 if (tp->t_fb != blk) {
1461                         if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1462                                 refcount_release(&blk->tfb_refcnt);
1463                                 INP_WUNLOCK(inp);
1464                                 return (ENOENT);
1465                         }
1466                         /*
1467                          * Release the old refcnt, the
1468                          * lookup acquires a ref on the
1469                          * new one.
1470                          */
1471                         if (tp->t_fb->tfb_tcp_fb_fini)
1472                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1473                         refcount_release(&tp->t_fb->tfb_refcnt);
1474                         tp->t_fb = blk;
1475                         if (tp->t_fb->tfb_tcp_fb_init) {
1476                                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1477                         }
1478                 }
1479 #ifdef TCP_OFFLOAD
1480                 if (tp->t_flags & TF_TOE) {
1481                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1482                              sopt->sopt_name);
1483                 }
1484 #endif
1485                 INP_WUNLOCK(inp);
1486                 return (error);
1487         } else if ((sopt->sopt_dir == SOPT_GET) &&
1488             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1489                 strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1490                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1491                 INP_WUNLOCK(inp);
1492                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1493                 return (error);
1494         }
1495         /* Pass in the INP locked, called must unlock it */
1496         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1497 }
1498
1499 int
1500 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1501 {
1502         int     error, opt, optval;
1503         u_int   ui;
1504         struct  tcp_info ti;
1505         struct cc_algo *algo;
1506         char    *pbuf, buf[TCP_CA_NAME_MAX];
1507         size_t  len;
1508
1509         /*
1510          * For TCP_CCALGOOPT forward the control to CC module, for both
1511          * SOPT_SET and SOPT_GET.
1512          */
1513         switch (sopt->sopt_name) {
1514         case TCP_CCALGOOPT:
1515                 INP_WUNLOCK(inp);
1516                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1517                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1518                     sopt->sopt_valsize);
1519                 if (error) {
1520                         free(pbuf, M_TEMP);
1521                         return (error);
1522                 }
1523                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1524                 if (CC_ALGO(tp)->ctl_output != NULL)
1525                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1526                 else
1527                         error = ENOENT;
1528                 INP_WUNLOCK(inp);
1529                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1530                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1531                 free(pbuf, M_TEMP);
1532                 return (error);
1533         }
1534
1535         switch (sopt->sopt_dir) {
1536         case SOPT_SET:
1537                 switch (sopt->sopt_name) {
1538 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1539                 case TCP_MD5SIG:
1540                         if (!TCPMD5_ENABLED()) {
1541                                 INP_WUNLOCK(inp);
1542                                 return (ENOPROTOOPT);
1543                         }
1544                         error = TCPMD5_PCBCTL(inp, sopt);
1545                         if (error)
1546                                 return (error);
1547                         goto unlock_and_done;
1548 #endif /* IPSEC */
1549
1550                 case TCP_NODELAY:
1551                 case TCP_NOOPT:
1552                         INP_WUNLOCK(inp);
1553                         error = sooptcopyin(sopt, &optval, sizeof optval,
1554                             sizeof optval);
1555                         if (error)
1556                                 return (error);
1557
1558                         INP_WLOCK_RECHECK(inp);
1559                         switch (sopt->sopt_name) {
1560                         case TCP_NODELAY:
1561                                 opt = TF_NODELAY;
1562                                 break;
1563                         case TCP_NOOPT:
1564                                 opt = TF_NOOPT;
1565                                 break;
1566                         default:
1567                                 opt = 0; /* dead code to fool gcc */
1568                                 break;
1569                         }
1570
1571                         if (optval)
1572                                 tp->t_flags |= opt;
1573                         else
1574                                 tp->t_flags &= ~opt;
1575 unlock_and_done:
1576 #ifdef TCP_OFFLOAD
1577                         if (tp->t_flags & TF_TOE) {
1578                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1579                                     sopt->sopt_name);
1580                         }
1581 #endif
1582                         INP_WUNLOCK(inp);
1583                         break;
1584
1585                 case TCP_NOPUSH:
1586                         INP_WUNLOCK(inp);
1587                         error = sooptcopyin(sopt, &optval, sizeof optval,
1588                             sizeof optval);
1589                         if (error)
1590                                 return (error);
1591
1592                         INP_WLOCK_RECHECK(inp);
1593                         if (optval)
1594                                 tp->t_flags |= TF_NOPUSH;
1595                         else if (tp->t_flags & TF_NOPUSH) {
1596                                 tp->t_flags &= ~TF_NOPUSH;
1597                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1598                                         error = tp->t_fb->tfb_tcp_output(tp);
1599                         }
1600                         goto unlock_and_done;
1601
1602                 case TCP_MAXSEG:
1603                         INP_WUNLOCK(inp);
1604                         error = sooptcopyin(sopt, &optval, sizeof optval,
1605                             sizeof optval);
1606                         if (error)
1607                                 return (error);
1608
1609                         INP_WLOCK_RECHECK(inp);
1610                         if (optval > 0 && optval <= tp->t_maxseg &&
1611                             optval + 40 >= V_tcp_minmss)
1612                                 tp->t_maxseg = optval;
1613                         else
1614                                 error = EINVAL;
1615                         goto unlock_and_done;
1616
1617                 case TCP_INFO:
1618                         INP_WUNLOCK(inp);
1619                         error = EINVAL;
1620                         break;
1621
1622                 case TCP_CONGESTION:
1623                         INP_WUNLOCK(inp);
1624                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1625                         if (error)
1626                                 break;
1627                         buf[sopt->sopt_valsize] = '\0';
1628                         INP_WLOCK_RECHECK(inp);
1629                         CC_LIST_RLOCK();
1630                         STAILQ_FOREACH(algo, &cc_list, entries)
1631                                 if (strncmp(buf, algo->name,
1632                                     TCP_CA_NAME_MAX) == 0)
1633                                         break;
1634                         CC_LIST_RUNLOCK();
1635                         if (algo == NULL) {
1636                                 INP_WUNLOCK(inp);
1637                                 error = EINVAL;
1638                                 break;
1639                         }
1640                         /*
1641                          * We hold a write lock over the tcb so it's safe to
1642                          * do these things without ordering concerns.
1643                          */
1644                         if (CC_ALGO(tp)->cb_destroy != NULL)
1645                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1646                         CC_ALGO(tp) = algo;
1647                         /*
1648                          * If something goes pear shaped initialising the new
1649                          * algo, fall back to newreno (which does not
1650                          * require initialisation).
1651                          */
1652                         if (algo->cb_init != NULL &&
1653                             algo->cb_init(tp->ccv) != 0) {
1654                                 CC_ALGO(tp) = &newreno_cc_algo;
1655                                 /*
1656                                  * The only reason init should fail is
1657                                  * because of malloc.
1658                                  */
1659                                 error = ENOMEM;
1660                         }
1661                         INP_WUNLOCK(inp);
1662                         break;
1663
1664                 case TCP_KEEPIDLE:
1665                 case TCP_KEEPINTVL:
1666                 case TCP_KEEPINIT:
1667                         INP_WUNLOCK(inp);
1668                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1669                         if (error)
1670                                 return (error);
1671
1672                         if (ui > (UINT_MAX / hz)) {
1673                                 error = EINVAL;
1674                                 break;
1675                         }
1676                         ui *= hz;
1677
1678                         INP_WLOCK_RECHECK(inp);
1679                         switch (sopt->sopt_name) {
1680                         case TCP_KEEPIDLE:
1681                                 tp->t_keepidle = ui;
1682                                 /*
1683                                  * XXX: better check current remaining
1684                                  * timeout and "merge" it with new value.
1685                                  */
1686                                 if ((tp->t_state > TCPS_LISTEN) &&
1687                                     (tp->t_state <= TCPS_CLOSING))
1688                                         tcp_timer_activate(tp, TT_KEEP,
1689                                             TP_KEEPIDLE(tp));
1690                                 break;
1691                         case TCP_KEEPINTVL:
1692                                 tp->t_keepintvl = ui;
1693                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1694                                     (TP_MAXIDLE(tp) > 0))
1695                                         tcp_timer_activate(tp, TT_2MSL,
1696                                             TP_MAXIDLE(tp));
1697                                 break;
1698                         case TCP_KEEPINIT:
1699                                 tp->t_keepinit = ui;
1700                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1701                                     tp->t_state == TCPS_SYN_SENT)
1702                                         tcp_timer_activate(tp, TT_KEEP,
1703                                             TP_KEEPINIT(tp));
1704                                 break;
1705                         }
1706                         goto unlock_and_done;
1707
1708                 case TCP_KEEPCNT:
1709                         INP_WUNLOCK(inp);
1710                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1711                         if (error)
1712                                 return (error);
1713
1714                         INP_WLOCK_RECHECK(inp);
1715                         tp->t_keepcnt = ui;
1716                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1717                             (TP_MAXIDLE(tp) > 0))
1718                                 tcp_timer_activate(tp, TT_2MSL,
1719                                     TP_MAXIDLE(tp));
1720                         goto unlock_and_done;
1721
1722 #ifdef TCPPCAP
1723                 case TCP_PCAP_OUT:
1724                 case TCP_PCAP_IN:
1725                         INP_WUNLOCK(inp);
1726                         error = sooptcopyin(sopt, &optval, sizeof optval,
1727                             sizeof optval);
1728                         if (error)
1729                                 return (error);
1730
1731                         INP_WLOCK_RECHECK(inp);
1732                         if (optval >= 0)
1733                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1734                                         &(tp->t_outpkts) : &(tp->t_inpkts),
1735                                         optval);
1736                         else
1737                                 error = EINVAL;
1738                         goto unlock_and_done;
1739 #endif
1740
1741 #ifdef TCP_RFC7413
1742                 case TCP_FASTOPEN:
1743                         INP_WUNLOCK(inp);
1744                         if (!V_tcp_fastopen_enabled)
1745                                 return (EPERM);
1746
1747                         error = sooptcopyin(sopt, &optval, sizeof optval,
1748                             sizeof optval);
1749                         if (error)
1750                                 return (error);
1751
1752                         INP_WLOCK_RECHECK(inp);
1753                         if (optval) {
1754                                 tp->t_flags |= TF_FASTOPEN;
1755                                 if ((tp->t_state == TCPS_LISTEN) &&
1756                                     (tp->t_tfo_pending == NULL))
1757                                         tp->t_tfo_pending =
1758                                             tcp_fastopen_alloc_counter();
1759                         } else
1760                                 tp->t_flags &= ~TF_FASTOPEN;
1761                         goto unlock_and_done;
1762 #endif
1763
1764                 default:
1765                         INP_WUNLOCK(inp);
1766                         error = ENOPROTOOPT;
1767                         break;
1768                 }
1769                 break;
1770
1771         case SOPT_GET:
1772                 tp = intotcpcb(inp);
1773                 switch (sopt->sopt_name) {
1774 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1775                 case TCP_MD5SIG:
1776                         if (!TCPMD5_ENABLED()) {
1777                                 INP_WUNLOCK(inp);
1778                                 return (ENOPROTOOPT);
1779                         }
1780                         error = TCPMD5_PCBCTL(inp, sopt);
1781                         break;
1782 #endif
1783
1784                 case TCP_NODELAY:
1785                         optval = tp->t_flags & TF_NODELAY;
1786                         INP_WUNLOCK(inp);
1787                         error = sooptcopyout(sopt, &optval, sizeof optval);
1788                         break;
1789                 case TCP_MAXSEG:
1790                         optval = tp->t_maxseg;
1791                         INP_WUNLOCK(inp);
1792                         error = sooptcopyout(sopt, &optval, sizeof optval);
1793                         break;
1794                 case TCP_NOOPT:
1795                         optval = tp->t_flags & TF_NOOPT;
1796                         INP_WUNLOCK(inp);
1797                         error = sooptcopyout(sopt, &optval, sizeof optval);
1798                         break;
1799                 case TCP_NOPUSH:
1800                         optval = tp->t_flags & TF_NOPUSH;
1801                         INP_WUNLOCK(inp);
1802                         error = sooptcopyout(sopt, &optval, sizeof optval);
1803                         break;
1804                 case TCP_INFO:
1805                         tcp_fill_info(tp, &ti);
1806                         INP_WUNLOCK(inp);
1807                         error = sooptcopyout(sopt, &ti, sizeof ti);
1808                         break;
1809                 case TCP_CONGESTION:
1810                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1811                         INP_WUNLOCK(inp);
1812                         error = sooptcopyout(sopt, buf, len + 1);
1813                         break;
1814                 case TCP_KEEPIDLE:
1815                 case TCP_KEEPINTVL:
1816                 case TCP_KEEPINIT:
1817                 case TCP_KEEPCNT:
1818                         switch (sopt->sopt_name) {
1819                         case TCP_KEEPIDLE:
1820                                 ui = TP_KEEPIDLE(tp) / hz;
1821                                 break;
1822                         case TCP_KEEPINTVL:
1823                                 ui = TP_KEEPINTVL(tp) / hz;
1824                                 break;
1825                         case TCP_KEEPINIT:
1826                                 ui = TP_KEEPINIT(tp) / hz;
1827                                 break;
1828                         case TCP_KEEPCNT:
1829                                 ui = TP_KEEPCNT(tp);
1830                                 break;
1831                         }
1832                         INP_WUNLOCK(inp);
1833                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1834                         break;
1835 #ifdef TCPPCAP
1836                 case TCP_PCAP_OUT:
1837                 case TCP_PCAP_IN:
1838                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1839                                         &(tp->t_outpkts) : &(tp->t_inpkts));
1840                         INP_WUNLOCK(inp);
1841                         error = sooptcopyout(sopt, &optval, sizeof optval);
1842                         break;
1843 #endif
1844
1845 #ifdef TCP_RFC7413
1846                 case TCP_FASTOPEN:
1847                         optval = tp->t_flags & TF_FASTOPEN;
1848                         INP_WUNLOCK(inp);
1849                         error = sooptcopyout(sopt, &optval, sizeof optval);
1850                         break;
1851 #endif
1852                 default:
1853                         INP_WUNLOCK(inp);
1854                         error = ENOPROTOOPT;
1855                         break;
1856                 }
1857                 break;
1858         }
1859         return (error);
1860 }
1861 #undef INP_WLOCK_RECHECK
1862 #undef INP_WLOCK_RECHECK_CLEANUP
1863
1864 /*
1865  * Attach TCP protocol to socket, allocating
1866  * internet protocol control block, tcp control block,
1867  * bufer space, and entering LISTEN state if to accept connections.
1868  */
1869 static int
1870 tcp_attach(struct socket *so)
1871 {
1872         struct tcpcb *tp;
1873         struct inpcb *inp;
1874         int error;
1875
1876         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1877                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1878                 if (error)
1879                         return (error);
1880         }
1881         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1882         so->so_snd.sb_flags |= SB_AUTOSIZE;
1883         INP_INFO_RLOCK(&V_tcbinfo);
1884         error = in_pcballoc(so, &V_tcbinfo);
1885         if (error) {
1886                 INP_INFO_RUNLOCK(&V_tcbinfo);
1887                 return (error);
1888         }
1889         inp = sotoinpcb(so);
1890 #ifdef INET6
1891         if (inp->inp_vflag & INP_IPV6PROTO) {
1892                 inp->inp_vflag |= INP_IPV6;
1893                 inp->in6p_hops = -1;    /* use kernel default */
1894         }
1895         else
1896 #endif
1897         inp->inp_vflag |= INP_IPV4;
1898         tp = tcp_newtcpcb(inp);
1899         if (tp == NULL) {
1900                 in_pcbdetach(inp);
1901                 in_pcbfree(inp);
1902                 INP_INFO_RUNLOCK(&V_tcbinfo);
1903                 return (ENOBUFS);
1904         }
1905         tp->t_state = TCPS_CLOSED;
1906         INP_WUNLOCK(inp);
1907         INP_INFO_RUNLOCK(&V_tcbinfo);
1908         TCPSTATES_INC(TCPS_CLOSED);
1909         return (0);
1910 }
1911
1912 /*
1913  * Initiate (or continue) disconnect.
1914  * If embryonic state, just send reset (once).
1915  * If in ``let data drain'' option and linger null, just drop.
1916  * Otherwise (hard), mark socket disconnecting and drop
1917  * current input data; switch states based on user close, and
1918  * send segment to peer (with FIN).
1919  */
1920 static void
1921 tcp_disconnect(struct tcpcb *tp)
1922 {
1923         struct inpcb *inp = tp->t_inpcb;
1924         struct socket *so = inp->inp_socket;
1925
1926         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1927         INP_WLOCK_ASSERT(inp);
1928
1929         /*
1930          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1931          * socket is still open.
1932          */
1933         if (tp->t_state < TCPS_ESTABLISHED) {
1934                 tp = tcp_close(tp);
1935                 KASSERT(tp != NULL,
1936                     ("tcp_disconnect: tcp_close() returned NULL"));
1937         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1938                 tp = tcp_drop(tp, 0);
1939                 KASSERT(tp != NULL,
1940                     ("tcp_disconnect: tcp_drop() returned NULL"));
1941         } else {
1942                 soisdisconnecting(so);
1943                 sbflush(&so->so_rcv);
1944                 tcp_usrclosed(tp);
1945                 if (!(inp->inp_flags & INP_DROPPED))
1946                         tp->t_fb->tfb_tcp_output(tp);
1947         }
1948 }
1949
1950 /*
1951  * User issued close, and wish to trail through shutdown states:
1952  * if never received SYN, just forget it.  If got a SYN from peer,
1953  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1954  * If already got a FIN from peer, then almost done; go to LAST_ACK
1955  * state.  In all other cases, have already sent FIN to peer (e.g.
1956  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1957  * for peer to send FIN or not respond to keep-alives, etc.
1958  * We can let the user exit from the close as soon as the FIN is acked.
1959  */
1960 static void
1961 tcp_usrclosed(struct tcpcb *tp)
1962 {
1963
1964         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1965         INP_WLOCK_ASSERT(tp->t_inpcb);
1966
1967         switch (tp->t_state) {
1968         case TCPS_LISTEN:
1969 #ifdef TCP_OFFLOAD
1970                 tcp_offload_listen_stop(tp);
1971 #endif
1972                 tcp_state_change(tp, TCPS_CLOSED);
1973                 /* FALLTHROUGH */
1974         case TCPS_CLOSED:
1975                 tp = tcp_close(tp);
1976                 /*
1977                  * tcp_close() should never return NULL here as the socket is
1978                  * still open.
1979                  */
1980                 KASSERT(tp != NULL,
1981                     ("tcp_usrclosed: tcp_close() returned NULL"));
1982                 break;
1983
1984         case TCPS_SYN_SENT:
1985         case TCPS_SYN_RECEIVED:
1986                 tp->t_flags |= TF_NEEDFIN;
1987                 break;
1988
1989         case TCPS_ESTABLISHED:
1990                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1991                 break;
1992
1993         case TCPS_CLOSE_WAIT:
1994                 tcp_state_change(tp, TCPS_LAST_ACK);
1995                 break;
1996         }
1997         if (tp->t_state >= TCPS_FIN_WAIT_2) {
1998                 soisdisconnected(tp->t_inpcb->inp_socket);
1999                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2000                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2001                         int timeout;
2002
2003                         timeout = (tcp_fast_finwait2_recycle) ?
2004                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2005                         tcp_timer_activate(tp, TT_2MSL, timeout);
2006                 }
2007         }
2008 }
2009
2010 #ifdef DDB
2011 static void
2012 db_print_indent(int indent)
2013 {
2014         int i;
2015
2016         for (i = 0; i < indent; i++)
2017                 db_printf(" ");
2018 }
2019
2020 static void
2021 db_print_tstate(int t_state)
2022 {
2023
2024         switch (t_state) {
2025         case TCPS_CLOSED:
2026                 db_printf("TCPS_CLOSED");
2027                 return;
2028
2029         case TCPS_LISTEN:
2030                 db_printf("TCPS_LISTEN");
2031                 return;
2032
2033         case TCPS_SYN_SENT:
2034                 db_printf("TCPS_SYN_SENT");
2035                 return;
2036
2037         case TCPS_SYN_RECEIVED:
2038                 db_printf("TCPS_SYN_RECEIVED");
2039                 return;
2040
2041         case TCPS_ESTABLISHED:
2042                 db_printf("TCPS_ESTABLISHED");
2043                 return;
2044
2045         case TCPS_CLOSE_WAIT:
2046                 db_printf("TCPS_CLOSE_WAIT");
2047                 return;
2048
2049         case TCPS_FIN_WAIT_1:
2050                 db_printf("TCPS_FIN_WAIT_1");
2051                 return;
2052
2053         case TCPS_CLOSING:
2054                 db_printf("TCPS_CLOSING");
2055                 return;
2056
2057         case TCPS_LAST_ACK:
2058                 db_printf("TCPS_LAST_ACK");
2059                 return;
2060
2061         case TCPS_FIN_WAIT_2:
2062                 db_printf("TCPS_FIN_WAIT_2");
2063                 return;
2064
2065         case TCPS_TIME_WAIT:
2066                 db_printf("TCPS_TIME_WAIT");
2067                 return;
2068
2069         default:
2070                 db_printf("unknown");
2071                 return;
2072         }
2073 }
2074
2075 static void
2076 db_print_tflags(u_int t_flags)
2077 {
2078         int comma;
2079
2080         comma = 0;
2081         if (t_flags & TF_ACKNOW) {
2082                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2083                 comma = 1;
2084         }
2085         if (t_flags & TF_DELACK) {
2086                 db_printf("%sTF_DELACK", comma ? ", " : "");
2087                 comma = 1;
2088         }
2089         if (t_flags & TF_NODELAY) {
2090                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2091                 comma = 1;
2092         }
2093         if (t_flags & TF_NOOPT) {
2094                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2095                 comma = 1;
2096         }
2097         if (t_flags & TF_SENTFIN) {
2098                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2099                 comma = 1;
2100         }
2101         if (t_flags & TF_REQ_SCALE) {
2102                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2103                 comma = 1;
2104         }
2105         if (t_flags & TF_RCVD_SCALE) {
2106                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2107                 comma = 1;
2108         }
2109         if (t_flags & TF_REQ_TSTMP) {
2110                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2111                 comma = 1;
2112         }
2113         if (t_flags & TF_RCVD_TSTMP) {
2114                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2115                 comma = 1;
2116         }
2117         if (t_flags & TF_SACK_PERMIT) {
2118                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2119                 comma = 1;
2120         }
2121         if (t_flags & TF_NEEDSYN) {
2122                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2123                 comma = 1;
2124         }
2125         if (t_flags & TF_NEEDFIN) {
2126                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2127                 comma = 1;
2128         }
2129         if (t_flags & TF_NOPUSH) {
2130                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2131                 comma = 1;
2132         }
2133         if (t_flags & TF_MORETOCOME) {
2134                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2135                 comma = 1;
2136         }
2137         if (t_flags & TF_LQ_OVERFLOW) {
2138                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2139                 comma = 1;
2140         }
2141         if (t_flags & TF_LASTIDLE) {
2142                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2143                 comma = 1;
2144         }
2145         if (t_flags & TF_RXWIN0SENT) {
2146                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2147                 comma = 1;
2148         }
2149         if (t_flags & TF_FASTRECOVERY) {
2150                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2151                 comma = 1;
2152         }
2153         if (t_flags & TF_CONGRECOVERY) {
2154                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2155                 comma = 1;
2156         }
2157         if (t_flags & TF_WASFRECOVERY) {
2158                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2159                 comma = 1;
2160         }
2161         if (t_flags & TF_SIGNATURE) {
2162                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2163                 comma = 1;
2164         }
2165         if (t_flags & TF_FORCEDATA) {
2166                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2167                 comma = 1;
2168         }
2169         if (t_flags & TF_TSO) {
2170                 db_printf("%sTF_TSO", comma ? ", " : "");
2171                 comma = 1;
2172         }
2173         if (t_flags & TF_ECN_PERMIT) {
2174                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2175                 comma = 1;
2176         }
2177         if (t_flags & TF_FASTOPEN) {
2178                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2179                 comma = 1;
2180         }
2181 }
2182
2183 static void
2184 db_print_toobflags(char t_oobflags)
2185 {
2186         int comma;
2187
2188         comma = 0;
2189         if (t_oobflags & TCPOOB_HAVEDATA) {
2190                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2191                 comma = 1;
2192         }
2193         if (t_oobflags & TCPOOB_HADDATA) {
2194                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2195                 comma = 1;
2196         }
2197 }
2198
2199 static void
2200 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2201 {
2202
2203         db_print_indent(indent);
2204         db_printf("%s at %p\n", name, tp);
2205
2206         indent += 2;
2207
2208         db_print_indent(indent);
2209         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2210            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2211
2212         db_print_indent(indent);
2213         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2214             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2215
2216         db_print_indent(indent);
2217         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2218             &tp->t_timers->tt_delack, tp->t_inpcb);
2219
2220         db_print_indent(indent);
2221         db_printf("t_state: %d (", tp->t_state);
2222         db_print_tstate(tp->t_state);
2223         db_printf(")\n");
2224
2225         db_print_indent(indent);
2226         db_printf("t_flags: 0x%x (", tp->t_flags);
2227         db_print_tflags(tp->t_flags);
2228         db_printf(")\n");
2229
2230         db_print_indent(indent);
2231         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2232             tp->snd_una, tp->snd_max, tp->snd_nxt);
2233
2234         db_print_indent(indent);
2235         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2236            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2237
2238         db_print_indent(indent);
2239         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2240             tp->iss, tp->irs, tp->rcv_nxt);
2241
2242         db_print_indent(indent);
2243         db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2244             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2245
2246         db_print_indent(indent);
2247         db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2248            tp->snd_wnd, tp->snd_cwnd);
2249
2250         db_print_indent(indent);
2251         db_printf("snd_ssthresh: %lu   snd_recover: "
2252             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2253
2254         db_print_indent(indent);
2255         db_printf("t_rcvtime: %u   t_startime: %u\n",
2256             tp->t_rcvtime, tp->t_starttime);
2257
2258         db_print_indent(indent);
2259         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2260             tp->t_rtttime, tp->t_rtseq);
2261
2262         db_print_indent(indent);
2263         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2264             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2265
2266         db_print_indent(indent);
2267         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2268             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2269             tp->t_rttbest);
2270
2271         db_print_indent(indent);
2272         db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2273             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2274
2275         db_print_indent(indent);
2276         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2277         db_print_toobflags(tp->t_oobflags);
2278         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2279
2280         db_print_indent(indent);
2281         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2282             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2283
2284         db_print_indent(indent);
2285         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2286             tp->ts_recent, tp->ts_recent_age);
2287
2288         db_print_indent(indent);
2289         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2290             "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2291
2292         db_print_indent(indent);
2293         db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2294             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2295             tp->snd_recover_prev, tp->t_badrxtwin);
2296
2297         db_print_indent(indent);
2298         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2299             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2300
2301         db_print_indent(indent);
2302         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2303             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2304
2305         /* Skip sackblks, sackhint. */
2306
2307         db_print_indent(indent);
2308         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2309             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2310 }
2311
2312 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2313 {
2314         struct tcpcb *tp;
2315
2316         if (!have_addr) {
2317                 db_printf("usage: show tcpcb <addr>\n");
2318                 return;
2319         }
2320         tp = (struct tcpcb *)addr;
2321
2322         db_print_tcpcb(tp, "tcpcb", 0);
2323 }
2324 #endif