sys/netinet/tcp_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1993
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2006-2007 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Robert N. M. Watson under
  11  * contract to Juniper Networks, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_ddb.h"
  44 #include "opt_inet.h"
  45 #include "opt_inet6.h"
  46 #include "opt_ipsec.h"
  47 #include "opt_tcpdebug.h"
  48
  49 #include <sys/param.h>
  50 #include <sys/systm.h>
  51 #include <sys/limits.h>
  52 #include <sys/malloc.h>
  53 #include <sys/refcount.h>
  54 #include <sys/kernel.h>
  55 #include <sys/sysctl.h>
  56 #include <sys/mbuf.h>
  57 #ifdef INET6
  58 #include <sys/domain.h>
  59 #endif /* INET6 */
  60 #include <sys/socket.h>
  61 #include <sys/socketvar.h>
  62 #include <sys/protosw.h>
  63 #include <sys/proc.h>
  64 #include <sys/jail.h>
  65 #include <sys/syslog.h>
  66
  67 #ifdef DDB
  68 #include <ddb/ddb.h>
  69 #endif
  70
  71 #include <net/if.h>
  72 #include <net/if_var.h>
  73 #include <net/route.h>
  74 #include <net/vnet.h>
  75
  76 #include <netinet/in.h>
  77 #include <netinet/in_kdtrace.h>
  78 #include <netinet/in_pcb.h>
  79 #include <netinet/in_systm.h>
  80 #include <netinet/in_var.h>
  81 #include <netinet/ip_var.h>
  82 #ifdef INET6
  83 #include <netinet/ip6.h>
  84 #include <netinet6/in6_pcb.h>
  85 #include <netinet6/ip6_var.h>
  86 #include <netinet6/scope6_var.h>
  87 #endif
  88 #ifdef TCP_RFC7413
  89 #include <netinet/tcp_fastopen.h>
  90 #endif
  91 #include <netinet/tcp.h>
  92 #include <netinet/tcp_fsm.h>
  93 #include <netinet/tcp_seq.h>
  94 #include <netinet/tcp_timer.h>
  95 #include <netinet/tcp_var.h>
  96 #include <netinet/tcpip.h>
  97 #include <netinet/cc/cc.h>
  98 #ifdef TCPPCAP
  99 #include <netinet/tcp_pcap.h>
 100 #endif
 101 #ifdef TCPDEBUG
 102 #include <netinet/tcp_debug.h>
 103 #endif
 104 #ifdef TCP_OFFLOAD
 105 #include <netinet/tcp_offload.h>
 106 #endif
 107 #include <netipsec/ipsec_support.h>
 108
 109 /*
 110  * TCP protocol interface to socket abstraction.
 111  */
 112 static int      tcp_attach(struct socket *);
 113 #ifdef INET
 114 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 115                     struct thread *td);
 116 #endif /* INET */
 117 #ifdef INET6
 118 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 119                     struct thread *td);
 120 #endif /* INET6 */
 121 static void     tcp_disconnect(struct tcpcb *);
 122 static void     tcp_usrclosed(struct tcpcb *);
 123 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 124
 125 #ifdef TCPDEBUG
 126 #define TCPDEBUG0       int ostate = 0
 127 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 128 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 129                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 130 #else
 131 #define TCPDEBUG0
 132 #define TCPDEBUG1()
 133 #define TCPDEBUG2(req)
 134 #endif
 135
 136 /*
 137  * TCP attaches to socket via pru_attach(), reserving space,
 138  * and an internet control block.
 139  */
 140 static int
 141 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 142 {
 143         struct inpcb *inp;
 144         struct tcpcb *tp = NULL;
 145         int error;
 146         TCPDEBUG0;
 147
 148         inp = sotoinpcb(so);
 149         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 150         TCPDEBUG1();
 151
 152         error = tcp_attach(so);
 153         if (error)
 154                 goto out;
 155
 156         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 157                 so->so_linger = TCP_LINGERTIME;
 158
 159         inp = sotoinpcb(so);
 160         tp = intotcpcb(inp);
 161 out:
 162         TCPDEBUG2(PRU_ATTACH);
 163         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 164         return error;
 165 }
 166
 167 /*
 168  * tcp_detach is called when the socket layer loses its final reference
 169  * to the socket, be it a file descriptor reference, a reference from TCP,
 170  * etc.  At this point, there is only one case in which we will keep around
 171  * inpcb state: time wait.
 172  *
 173  * This function can probably be re-absorbed back into tcp_usr_detach() now
 174  * that there is a single detach path.
 175  */
 176 static void
 177 tcp_detach(struct socket *so, struct inpcb *inp)
 178 {
 179         struct tcpcb *tp;
 180
 181         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
 182         INP_WLOCK_ASSERT(inp);
 183
 184         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 185         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 186
 187         tp = intotcpcb(inp);
 188
 189         if (inp->inp_flags & INP_TIMEWAIT) {
 190                 /*
 191                  * There are two cases to handle: one in which the time wait
 192                  * state is being discarded (INP_DROPPED), and one in which
 193                  * this connection will remain in timewait.  In the former,
 194                  * it is time to discard all state (except tcptw, which has
 195                  * already been discarded by the timewait close code, which
 196                  * should be further up the call stack somewhere).  In the
 197                  * latter case, we detach from the socket, but leave the pcb
 198                  * present until timewait ends.
 199                  *
 200                  * XXXRW: Would it be cleaner to free the tcptw here?
 201                  *
 202                  * Astute question indeed, from twtcp perspective there are
 203                  * three cases to consider:
 204                  *
 205                  * #1 tcp_detach is called at tcptw creation time by
 206                  *  tcp_twstart, then do not discard the newly created tcptw
 207                  *  and leave inpcb present until timewait ends
 208                  * #2 tcp_detach is called at timewait end (or reuse) by
 209                  *  tcp_twclose, then the tcptw has already been discarded
 210                  *  (or reused) and inpcb is freed here
 211                  * #3 tcp_detach is called() after timewait ends (or reuse)
 212                  *  (e.g. by soclose), then tcptw has already been discarded
 213                  *  (or reused) and inpcb is freed here
 214                  *
 215                  *  In all three cases the tcptw should not be freed here.
 216                  */
 217                 if (inp->inp_flags & INP_DROPPED) {
 218                         in_pcbdetach(inp);
 219                         if (__predict_true(tp == NULL)) {
 220                                 in_pcbfree(inp);
 221                         } else {
 222                                 /*
 223                                  * This case should not happen as in TIMEWAIT
 224                                  * state the inp should not be destroyed before
 225                                  * its tcptw.  If INVARIANTS is defined, panic.
 226                                  */
 227 #ifdef INVARIANTS
 228                                 panic("%s: Panic before an inp double-free: "
 229                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 230                                     , __func__);
 231 #else
 232                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 233                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 234                                     , __func__);
 235 #endif
 236                                 INP_WUNLOCK(inp);
 237                         }
 238                 } else {
 239                         in_pcbdetach(inp);
 240                         INP_WUNLOCK(inp);
 241                 }
 242         } else {
 243                 /*
 244                  * If the connection is not in timewait, we consider two
 245                  * two conditions: one in which no further processing is
 246                  * necessary (dropped || embryonic), and one in which TCP is
 247                  * not yet done, but no longer requires the socket, so the
 248                  * pcb will persist for the time being.
 249                  *
 250                  * XXXRW: Does the second case still occur?
 251                  */
 252                 if (inp->inp_flags & INP_DROPPED ||
 253                     tp->t_state < TCPS_SYN_SENT) {
 254                         tcp_discardcb(tp);
 255                         in_pcbdetach(inp);
 256                         in_pcbfree(inp);
 257                 } else {
 258                         in_pcbdetach(inp);
 259                         INP_WUNLOCK(inp);
 260                 }
 261         }
 262 }
 263
 264 /*
 265  * pru_detach() detaches the TCP protocol from the socket.
 266  * If the protocol state is non-embryonic, then can't
 267  * do this directly: have to initiate a pru_disconnect(),
 268  * which may finish later; embryonic TCB's can just
 269  * be discarded here.
 270  */
 271 static void
 272 tcp_usr_detach(struct socket *so)
 273 {
 274         struct inpcb *inp;
 275         int rlock = 0;
 276
 277         inp = sotoinpcb(so);
 278         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 279         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 280                 INP_INFO_RLOCK(&V_tcbinfo);
 281                 rlock = 1;
 282         }
 283         INP_WLOCK(inp);
 284         KASSERT(inp->inp_socket != NULL,
 285             ("tcp_usr_detach: inp_socket == NULL"));
 286         tcp_detach(so, inp);
 287         if (rlock)
 288                 INP_INFO_RUNLOCK(&V_tcbinfo);
 289 }
 290
 291 #ifdef INET
 292 /*
 293  * Give the socket an address.
 294  */
 295 static int
 296 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 297 {
 298         int error = 0;
 299         struct inpcb *inp;
 300         struct tcpcb *tp = NULL;
 301         struct sockaddr_in *sinp;
 302
 303         sinp = (struct sockaddr_in *)nam;
 304         if (nam->sa_len != sizeof (*sinp))
 305                 return (EINVAL);
 306         /*
 307          * Must check for multicast addresses and disallow binding
 308          * to them.
 309          */
 310         if (sinp->sin_family == AF_INET &&
 311             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 312                 return (EAFNOSUPPORT);
 313
 314         TCPDEBUG0;
 315         inp = sotoinpcb(so);
 316         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 317         INP_WLOCK(inp);
 318         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 319                 error = EINVAL;
 320                 goto out;
 321         }
 322         tp = intotcpcb(inp);
 323         TCPDEBUG1();
 324         INP_HASH_WLOCK(&V_tcbinfo);
 325         error = in_pcbbind(inp, nam, td->td_ucred);
 326         INP_HASH_WUNLOCK(&V_tcbinfo);
 327 out:
 328         TCPDEBUG2(PRU_BIND);
 329         TCP_PROBE2(debug__user, tp, PRU_BIND);
 330         INP_WUNLOCK(inp);
 331
 332         return (error);
 333 }
 334 #endif /* INET */
 335
 336 #ifdef INET6
 337 static int
 338 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 339 {
 340         int error = 0;
 341         struct inpcb *inp;
 342         struct tcpcb *tp = NULL;
 343         struct sockaddr_in6 *sin6p;
 344
 345         sin6p = (struct sockaddr_in6 *)nam;
 346         if (nam->sa_len != sizeof (*sin6p))
 347                 return (EINVAL);
 348         /*
 349          * Must check for multicast addresses and disallow binding
 350          * to them.
 351          */
 352         if (sin6p->sin6_family == AF_INET6 &&
 353             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 354                 return (EAFNOSUPPORT);
 355
 356         TCPDEBUG0;
 357         inp = sotoinpcb(so);
 358         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 359         INP_WLOCK(inp);
 360         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 361                 error = EINVAL;
 362                 goto out;
 363         }
 364         tp = intotcpcb(inp);
 365         TCPDEBUG1();
 366         INP_HASH_WLOCK(&V_tcbinfo);
 367         inp->inp_vflag &= ~INP_IPV4;
 368         inp->inp_vflag |= INP_IPV6;
 369 #ifdef INET
 370         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 371                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 372                         inp->inp_vflag |= INP_IPV4;
 373                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 374                         struct sockaddr_in sin;
 375
 376                         in6_sin6_2_sin(&sin, sin6p);
 377                         inp->inp_vflag |= INP_IPV4;
 378                         inp->inp_vflag &= ~INP_IPV6;
 379                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 380                             td->td_ucred);
 381                         INP_HASH_WUNLOCK(&V_tcbinfo);
 382                         goto out;
 383                 }
 384         }
 385 #endif
 386         error = in6_pcbbind(inp, nam, td->td_ucred);
 387         INP_HASH_WUNLOCK(&V_tcbinfo);
 388 out:
 389         TCPDEBUG2(PRU_BIND);
 390         TCP_PROBE2(debug__user, tp, PRU_BIND);
 391         INP_WUNLOCK(inp);
 392         return (error);
 393 }
 394 #endif /* INET6 */
 395
 396 #ifdef INET
 397 /*
 398  * Prepare to accept connections.
 399  */
 400 static int
 401 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 402 {
 403         int error = 0;
 404         struct inpcb *inp;
 405         struct tcpcb *tp = NULL;
 406
 407         TCPDEBUG0;
 408         inp = sotoinpcb(so);
 409         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 410         INP_WLOCK(inp);
 411         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 412                 error = EINVAL;
 413                 goto out;
 414         }
 415         tp = intotcpcb(inp);
 416         TCPDEBUG1();
 417         SOCK_LOCK(so);
 418         error = solisten_proto_check(so);
 419         INP_HASH_WLOCK(&V_tcbinfo);
 420         if (error == 0 && inp->inp_lport == 0)
 421                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 422         INP_HASH_WUNLOCK(&V_tcbinfo);
 423         if (error == 0) {
 424                 tcp_state_change(tp, TCPS_LISTEN);
 425                 solisten_proto(so, backlog);
 426 #ifdef TCP_OFFLOAD
 427                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 428                         tcp_offload_listen_start(tp);
 429 #endif
 430         }
 431         SOCK_UNLOCK(so);
 432
 433 #ifdef TCP_RFC7413
 434         if (IS_FASTOPEN(tp->t_flags))
 435                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 436 #endif
 437 out:
 438         TCPDEBUG2(PRU_LISTEN);
 439         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 440         INP_WUNLOCK(inp);
 441         return (error);
 442 }
 443 #endif /* INET */
 444
 445 #ifdef INET6
 446 static int
 447 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 448 {
 449         int error = 0;
 450         struct inpcb *inp;
 451         struct tcpcb *tp = NULL;
 452
 453         TCPDEBUG0;
 454         inp = sotoinpcb(so);
 455         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 456         INP_WLOCK(inp);
 457         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 458                 error = EINVAL;
 459                 goto out;
 460         }
 461         tp = intotcpcb(inp);
 462         TCPDEBUG1();
 463         SOCK_LOCK(so);
 464         error = solisten_proto_check(so);
 465         INP_HASH_WLOCK(&V_tcbinfo);
 466         if (error == 0 && inp->inp_lport == 0) {
 467                 inp->inp_vflag &= ~INP_IPV4;
 468                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 469                         inp->inp_vflag |= INP_IPV4;
 470                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 471         }
 472         INP_HASH_WUNLOCK(&V_tcbinfo);
 473         if (error == 0) {
 474                 tcp_state_change(tp, TCPS_LISTEN);
 475                 solisten_proto(so, backlog);
 476 #ifdef TCP_OFFLOAD
 477                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 478                         tcp_offload_listen_start(tp);
 479 #endif
 480         }
 481         SOCK_UNLOCK(so);
 482
 483 #ifdef TCP_RFC7413
 484         if (IS_FASTOPEN(tp->t_flags))
 485                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 486 #endif
 487 out:
 488         TCPDEBUG2(PRU_LISTEN);
 489         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 490         INP_WUNLOCK(inp);
 491         return (error);
 492 }
 493 #endif /* INET6 */
 494
 495 #ifdef INET
 496 /*
 497  * Initiate connection to peer.
 498  * Create a template for use in transmissions on this connection.
 499  * Enter SYN_SENT state, and mark socket as connecting.
 500  * Start keep-alive timer, and seed output sequence space.
 501  * Send initial segment on connection.
 502  */
 503 static int
 504 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 505 {
 506         int error = 0;
 507         struct inpcb *inp;
 508         struct tcpcb *tp = NULL;
 509         struct sockaddr_in *sinp;
 510
 511         sinp = (struct sockaddr_in *)nam;
 512         if (nam->sa_len != sizeof (*sinp))
 513                 return (EINVAL);
 514         /*
 515          * Must disallow TCP ``connections'' to multicast addresses.
 516          */
 517         if (sinp->sin_family == AF_INET
 518             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 519                 return (EAFNOSUPPORT);
 520         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 521                 return (error);
 522
 523         TCPDEBUG0;
 524         inp = sotoinpcb(so);
 525         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 526         INP_WLOCK(inp);
 527         if (inp->inp_flags & INP_TIMEWAIT) {
 528                 error = EADDRINUSE;
 529                 goto out;
 530         }
 531         if (inp->inp_flags & INP_DROPPED) {
 532                 error = ECONNREFUSED;
 533                 goto out;
 534         }
 535         tp = intotcpcb(inp);
 536         TCPDEBUG1();
 537         if ((error = tcp_connect(tp, nam, td)) != 0)
 538                 goto out;
 539 #ifdef TCP_OFFLOAD
 540         if (registered_toedevs > 0 &&
 541             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 542             (error = tcp_offload_connect(so, nam)) == 0)
 543                 goto out;
 544 #endif
 545         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 546         error = tp->t_fb->tfb_tcp_output(tp);
 547 out:
 548         TCPDEBUG2(PRU_CONNECT);
 549         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 550         INP_WUNLOCK(inp);
 551         return (error);
 552 }
 553 #endif /* INET */
 554
 555 #ifdef INET6
 556 static int
 557 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 558 {
 559         int error = 0;
 560         struct inpcb *inp;
 561         struct tcpcb *tp = NULL;
 562         struct sockaddr_in6 *sin6p;
 563
 564         TCPDEBUG0;
 565
 566         sin6p = (struct sockaddr_in6 *)nam;
 567         if (nam->sa_len != sizeof (*sin6p))
 568                 return (EINVAL);
 569         /*
 570          * Must disallow TCP ``connections'' to multicast addresses.
 571          */
 572         if (sin6p->sin6_family == AF_INET6
 573             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 574                 return (EAFNOSUPPORT);
 575
 576         inp = sotoinpcb(so);
 577         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 578         INP_WLOCK(inp);
 579         if (inp->inp_flags & INP_TIMEWAIT) {
 580                 error = EADDRINUSE;
 581                 goto out;
 582         }
 583         if (inp->inp_flags & INP_DROPPED) {
 584                 error = ECONNREFUSED;
 585                 goto out;
 586         }
 587         tp = intotcpcb(inp);
 588         TCPDEBUG1();
 589 #ifdef INET
 590         /*
 591          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 592          * therefore probably require the hash lock, which isn't held here.
 593          * Is this a significant problem?
 594          */
 595         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 596                 struct sockaddr_in sin;
 597
 598                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 599                         error = EINVAL;
 600                         goto out;
 601                 }
 602                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 603                         error = EAFNOSUPPORT;
 604                         goto out;
 605                 }
 606
 607                 in6_sin6_2_sin(&sin, sin6p);
 608                 inp->inp_vflag |= INP_IPV4;
 609                 inp->inp_vflag &= ~INP_IPV6;
 610                 if ((error = prison_remote_ip4(td->td_ucred,
 611                     &sin.sin_addr)) != 0)
 612                         goto out;
 613                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 614                         goto out;
 615 #ifdef TCP_OFFLOAD
 616                 if (registered_toedevs > 0 &&
 617                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 618                     (error = tcp_offload_connect(so, nam)) == 0)
 619                         goto out;
 620 #endif
 621                 error = tp->t_fb->tfb_tcp_output(tp);
 622                 goto out;
 623         } else {
 624                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 625                         error = EAFNOSUPPORT;
 626                         goto out;
 627                 }
 628         }
 629 #endif
 630         inp->inp_vflag &= ~INP_IPV4;
 631         inp->inp_vflag |= INP_IPV6;
 632         inp->inp_inc.inc_flags |= INC_ISIPV6;
 633         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 634                 goto out;
 635         if ((error = tcp6_connect(tp, nam, td)) != 0)
 636                 goto out;
 637 #ifdef TCP_OFFLOAD
 638         if (registered_toedevs > 0 &&
 639             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 640             (error = tcp_offload_connect(so, nam)) == 0)
 641                 goto out;
 642 #endif
 643         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 644         error = tp->t_fb->tfb_tcp_output(tp);
 645
 646 out:
 647         TCPDEBUG2(PRU_CONNECT);
 648         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 649         INP_WUNLOCK(inp);
 650         return (error);
 651 }
 652 #endif /* INET6 */
 653
 654 /*
 655  * Initiate disconnect from peer.
 656  * If connection never passed embryonic stage, just drop;
 657  * else if don't need to let data drain, then can just drop anyways,
 658  * else have to begin TCP shutdown process: mark socket disconnecting,
 659  * drain unread data, state switch to reflect user close, and
 660  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 661  * when peer sends FIN and acks ours.
 662  *
 663  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 664  */
 665 static int
 666 tcp_usr_disconnect(struct socket *so)
 667 {
 668         struct inpcb *inp;
 669         struct tcpcb *tp = NULL;
 670         int error = 0;
 671
 672         TCPDEBUG0;
 673         INP_INFO_RLOCK(&V_tcbinfo);
 674         inp = sotoinpcb(so);
 675         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 676         INP_WLOCK(inp);
 677         if (inp->inp_flags & INP_TIMEWAIT)
 678                 goto out;
 679         if (inp->inp_flags & INP_DROPPED) {
 680                 error = ECONNRESET;
 681                 goto out;
 682         }
 683         tp = intotcpcb(inp);
 684         TCPDEBUG1();
 685         tcp_disconnect(tp);
 686 out:
 687         TCPDEBUG2(PRU_DISCONNECT);
 688         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 689         INP_WUNLOCK(inp);
 690         INP_INFO_RUNLOCK(&V_tcbinfo);
 691         return (error);
 692 }
 693
 694 #ifdef INET
 695 /*
 696  * Accept a connection.  Essentially all the work is done at higher levels;
 697  * just return the address of the peer, storing through addr.
 698  */
 699 static int
 700 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 701 {
 702         int error = 0;
 703         struct inpcb *inp = NULL;
 704         struct tcpcb *tp = NULL;
 705         struct in_addr addr;
 706         in_port_t port = 0;
 707         TCPDEBUG0;
 708
 709         if (so->so_state & SS_ISDISCONNECTED)
 710                 return (ECONNABORTED);
 711
 712         inp = sotoinpcb(so);
 713         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 714         INP_WLOCK(inp);
 715         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 716                 error = ECONNABORTED;
 717                 goto out;
 718         }
 719         tp = intotcpcb(inp);
 720         TCPDEBUG1();
 721
 722         /*
 723          * We inline in_getpeeraddr and COMMON_END here, so that we can
 724          * copy the data of interest and defer the malloc until after we
 725          * release the lock.
 726          */
 727         port = inp->inp_fport;
 728         addr = inp->inp_faddr;
 729
 730 out:
 731         TCPDEBUG2(PRU_ACCEPT);
 732         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 733         INP_WUNLOCK(inp);
 734         if (error == 0)
 735                 *nam = in_sockaddr(port, &addr);
 736         return error;
 737 }
 738 #endif /* INET */
 739
 740 #ifdef INET6
 741 static int
 742 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 743 {
 744         struct inpcb *inp = NULL;
 745         int error = 0;
 746         struct tcpcb *tp = NULL;
 747         struct in_addr addr;
 748         struct in6_addr addr6;
 749         in_port_t port = 0;
 750         int v4 = 0;
 751         TCPDEBUG0;
 752
 753         if (so->so_state & SS_ISDISCONNECTED)
 754                 return (ECONNABORTED);
 755
 756         inp = sotoinpcb(so);
 757         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 758         INP_INFO_RLOCK(&V_tcbinfo);
 759         INP_WLOCK(inp);
 760         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 761                 error = ECONNABORTED;
 762                 goto out;
 763         }
 764         tp = intotcpcb(inp);
 765         TCPDEBUG1();
 766
 767         /*
 768          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 769          * copy the data of interest and defer the malloc until after we
 770          * release the lock.
 771          */
 772         if (inp->inp_vflag & INP_IPV4) {
 773                 v4 = 1;
 774                 port = inp->inp_fport;
 775                 addr = inp->inp_faddr;
 776         } else {
 777                 port = inp->inp_fport;
 778                 addr6 = inp->in6p_faddr;
 779         }
 780
 781 out:
 782         TCPDEBUG2(PRU_ACCEPT);
 783         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 784         INP_WUNLOCK(inp);
 785         INP_INFO_RUNLOCK(&V_tcbinfo);
 786         if (error == 0) {
 787                 if (v4)
 788                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 789                 else
 790                         *nam = in6_sockaddr(port, &addr6);
 791         }
 792         return error;
 793 }
 794 #endif /* INET6 */
 795
 796 /*
 797  * Mark the connection as being incapable of further output.
 798  */
 799 static int
 800 tcp_usr_shutdown(struct socket *so)
 801 {
 802         int error = 0;
 803         struct inpcb *inp;
 804         struct tcpcb *tp = NULL;
 805
 806         TCPDEBUG0;
 807         INP_INFO_RLOCK(&V_tcbinfo);
 808         inp = sotoinpcb(so);
 809         KASSERT(inp != NULL, ("inp == NULL"));
 810         INP_WLOCK(inp);
 811         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 812                 error = ECONNRESET;
 813                 goto out;
 814         }
 815         tp = intotcpcb(inp);
 816         TCPDEBUG1();
 817         socantsendmore(so);
 818         tcp_usrclosed(tp);
 819         if (!(inp->inp_flags & INP_DROPPED))
 820                 error = tp->t_fb->tfb_tcp_output(tp);
 821
 822 out:
 823         TCPDEBUG2(PRU_SHUTDOWN);
 824         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 825         INP_WUNLOCK(inp);
 826         INP_INFO_RUNLOCK(&V_tcbinfo);
 827
 828         return (error);
 829 }
 830
 831 /*
 832  * After a receive, possibly send window update to peer.
 833  */
 834 static int
 835 tcp_usr_rcvd(struct socket *so, int flags)
 836 {
 837         struct inpcb *inp;
 838         struct tcpcb *tp = NULL;
 839         int error = 0;
 840
 841         TCPDEBUG0;
 842         inp = sotoinpcb(so);
 843         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 844         INP_WLOCK(inp);
 845         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 846                 error = ECONNRESET;
 847                 goto out;
 848         }
 849         tp = intotcpcb(inp);
 850         TCPDEBUG1();
 851 #ifdef TCP_RFC7413
 852         /*
 853          * For passively-created TFO connections, don't attempt a window
 854          * update while still in SYN_RECEIVED as this may trigger an early
 855          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 856          * application response data, or failing that, when the DELACK timer
 857          * expires.
 858          */
 859         if (IS_FASTOPEN(tp->t_flags) &&
 860             (tp->t_state == TCPS_SYN_RECEIVED))
 861                 goto out;
 862 #endif
 863 #ifdef TCP_OFFLOAD
 864         if (tp->t_flags & TF_TOE)
 865                 tcp_offload_rcvd(tp);
 866         else
 867 #endif
 868         tp->t_fb->tfb_tcp_output(tp);
 869
 870 out:
 871         TCPDEBUG2(PRU_RCVD);
 872         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 873         INP_WUNLOCK(inp);
 874         return (error);
 875 }
 876
 877 /*
 878  * Do a send by putting data in output queue and updating urgent
 879  * marker if URG set.  Possibly send more data.  Unlike the other
 880  * pru_*() routines, the mbuf chains are our responsibility.  We
 881  * must either enqueue them or free them.  The other pru_* routines
 882  * generally are caller-frees.
 883  */
 884 static int
 885 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 886     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 887 {
 888         int error = 0;
 889         struct inpcb *inp;
 890         struct tcpcb *tp = NULL;
 891 #ifdef INET6
 892         int isipv6;
 893 #endif
 894         TCPDEBUG0;
 895
 896         /*
 897          * We require the pcbinfo lock if we will close the socket as part of
 898          * this call.
 899          */
 900         if (flags & PRUS_EOF)
 901                 INP_INFO_RLOCK(&V_tcbinfo);
 902         inp = sotoinpcb(so);
 903         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 904         INP_WLOCK(inp);
 905         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 906                 if (control)
 907                         m_freem(control);
 908                 /*
 909                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 910                  * for freeing memory.
 911                  */
 912                 if (m && (flags & PRUS_NOTREADY) == 0)
 913                         m_freem(m);
 914                 error = ECONNRESET;
 915                 goto out;
 916         }
 917 #ifdef INET6
 918         isipv6 = nam && nam->sa_family == AF_INET6;
 919 #endif /* INET6 */
 920         tp = intotcpcb(inp);
 921         TCPDEBUG1();
 922         if (control) {
 923                 /* TCP doesn't do control messages (rights, creds, etc) */
 924                 if (control->m_len) {
 925                         m_freem(control);
 926                         if (m)
 927                                 m_freem(m);
 928                         error = EINVAL;
 929                         goto out;
 930                 }
 931                 m_freem(control);       /* empty control, just free it */
 932         }
 933         if (!(flags & PRUS_OOB)) {
 934                 sbappendstream(&so->so_snd, m, flags);
 935                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 936                         /*
 937                          * Do implied connect if not yet connected,
 938                          * initialize window to default value, and
 939                          * initialize maxseg using peer's cached MSS.
 940                          */
 941 #ifdef INET6
 942                         if (isipv6)
 943                                 error = tcp6_connect(tp, nam, td);
 944 #endif /* INET6 */
 945 #if defined(INET6) && defined(INET)
 946                         else
 947 #endif
 948 #ifdef INET
 949                                 error = tcp_connect(tp, nam, td);
 950 #endif
 951                         if (error)
 952                                 goto out;
 953                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 954                         tcp_mss(tp, -1);
 955                 }
 956                 if (flags & PRUS_EOF) {
 957                         /*
 958                          * Close the send side of the connection after
 959                          * the data is sent.
 960                          */
 961                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 962                         socantsendmore(so);
 963                         tcp_usrclosed(tp);
 964                 }
 965                 if (!(inp->inp_flags & INP_DROPPED) &&
 966                     !(flags & PRUS_NOTREADY)) {
 967                         if (flags & PRUS_MORETOCOME)
 968                                 tp->t_flags |= TF_MORETOCOME;
 969                         error = tp->t_fb->tfb_tcp_output(tp);
 970                         if (flags & PRUS_MORETOCOME)
 971                                 tp->t_flags &= ~TF_MORETOCOME;
 972                 }
 973         } else {
 974                 /*
 975                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 976                  */
 977                 SOCKBUF_LOCK(&so->so_snd);
 978                 if (sbspace(&so->so_snd) < -512) {
 979                         SOCKBUF_UNLOCK(&so->so_snd);
 980                         m_freem(m);
 981                         error = ENOBUFS;
 982                         goto out;
 983                 }
 984                 /*
 985                  * According to RFC961 (Assigned Protocols),
 986                  * the urgent pointer points to the last octet
 987                  * of urgent data.  We continue, however,
 988                  * to consider it to indicate the first octet
 989                  * of data past the urgent section.
 990                  * Otherwise, snd_up should be one lower.
 991                  */
 992                 sbappendstream_locked(&so->so_snd, m, flags);
 993                 SOCKBUF_UNLOCK(&so->so_snd);
 994                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 995                         /*
 996                          * Do implied connect if not yet connected,
 997                          * initialize window to default value, and
 998                          * initialize maxseg using peer's cached MSS.
 999                          */
1000 #ifdef INET6
1001                         if (isipv6)
1002                                 error = tcp6_connect(tp, nam, td);
1003 #endif /* INET6 */
1004 #if defined(INET6) && defined(INET)
1005                         else
1006 #endif
1007 #ifdef INET
1008                                 error = tcp_connect(tp, nam, td);
1009 #endif
1010                         if (error)
1011                                 goto out;
1012                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1013                         tcp_mss(tp, -1);
1014                 }
1015                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1016                 if (!(flags & PRUS_NOTREADY)) {
1017                         tp->t_flags |= TF_FORCEDATA;
1018                         error = tp->t_fb->tfb_tcp_output(tp);
1019                         tp->t_flags &= ~TF_FORCEDATA;
1020                 }
1021         }
1022 out:
1023         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1024                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1025         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1026                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1027         INP_WUNLOCK(inp);
1028         if (flags & PRUS_EOF)
1029                 INP_INFO_RUNLOCK(&V_tcbinfo);
1030         return (error);
1031 }
1032
1033 static int
1034 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1035 {
1036         struct inpcb *inp;
1037         struct tcpcb *tp;
1038         int error;
1039
1040         inp = sotoinpcb(so);
1041         INP_WLOCK(inp);
1042         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1043                 INP_WUNLOCK(inp);
1044                 for (int i = 0; i < count; i++)
1045                         m = m_free(m);
1046                 return (ECONNRESET);
1047         }
1048         tp = intotcpcb(inp);
1049
1050         SOCKBUF_LOCK(&so->so_snd);
1051         error = sbready(&so->so_snd, m, count);
1052         SOCKBUF_UNLOCK(&so->so_snd);
1053         if (error == 0)
1054                 error = tp->t_fb->tfb_tcp_output(tp);
1055         INP_WUNLOCK(inp);
1056
1057         return (error);
1058 }
1059
1060 /*
1061  * Abort the TCP.  Drop the connection abruptly.
1062  */
1063 static void
1064 tcp_usr_abort(struct socket *so)
1065 {
1066         struct inpcb *inp;
1067         struct tcpcb *tp = NULL;
1068         TCPDEBUG0;
1069
1070         inp = sotoinpcb(so);
1071         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1072
1073         INP_INFO_RLOCK(&V_tcbinfo);
1074         INP_WLOCK(inp);
1075         KASSERT(inp->inp_socket != NULL,
1076             ("tcp_usr_abort: inp_socket == NULL"));
1077
1078         /*
1079          * If we still have full TCP state, and we're not dropped, drop.
1080          */
1081         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1082             !(inp->inp_flags & INP_DROPPED)) {
1083                 tp = intotcpcb(inp);
1084                 TCPDEBUG1();
1085                 tcp_drop(tp, ECONNABORTED);
1086                 TCPDEBUG2(PRU_ABORT);
1087                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1088         }
1089         if (!(inp->inp_flags & INP_DROPPED)) {
1090                 SOCK_LOCK(so);
1091                 so->so_state |= SS_PROTOREF;
1092                 SOCK_UNLOCK(so);
1093                 inp->inp_flags |= INP_SOCKREF;
1094         }
1095         INP_WUNLOCK(inp);
1096         INP_INFO_RUNLOCK(&V_tcbinfo);
1097 }
1098
1099 /*
1100  * TCP socket is closed.  Start friendly disconnect.
1101  */
1102 static void
1103 tcp_usr_close(struct socket *so)
1104 {
1105         struct inpcb *inp;
1106         struct tcpcb *tp = NULL;
1107         TCPDEBUG0;
1108
1109         inp = sotoinpcb(so);
1110         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1111
1112         INP_INFO_RLOCK(&V_tcbinfo);
1113         INP_WLOCK(inp);
1114         KASSERT(inp->inp_socket != NULL,
1115             ("tcp_usr_close: inp_socket == NULL"));
1116
1117         /*
1118          * If we still have full TCP state, and we're not dropped, initiate
1119          * a disconnect.
1120          */
1121         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1122             !(inp->inp_flags & INP_DROPPED)) {
1123                 tp = intotcpcb(inp);
1124                 TCPDEBUG1();
1125                 tcp_disconnect(tp);
1126                 TCPDEBUG2(PRU_CLOSE);
1127                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1128         }
1129         if (!(inp->inp_flags & INP_DROPPED)) {
1130                 SOCK_LOCK(so);
1131                 so->so_state |= SS_PROTOREF;
1132                 SOCK_UNLOCK(so);
1133                 inp->inp_flags |= INP_SOCKREF;
1134         }
1135         INP_WUNLOCK(inp);
1136         INP_INFO_RUNLOCK(&V_tcbinfo);
1137 }
1138
1139 /*
1140  * Receive out-of-band data.
1141  */
1142 static int
1143 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1144 {
1145         int error = 0;
1146         struct inpcb *inp;
1147         struct tcpcb *tp = NULL;
1148
1149         TCPDEBUG0;
1150         inp = sotoinpcb(so);
1151         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1152         INP_WLOCK(inp);
1153         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1154                 error = ECONNRESET;
1155                 goto out;
1156         }
1157         tp = intotcpcb(inp);
1158         TCPDEBUG1();
1159         if ((so->so_oobmark == 0 &&
1160              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1161             so->so_options & SO_OOBINLINE ||
1162             tp->t_oobflags & TCPOOB_HADDATA) {
1163                 error = EINVAL;
1164                 goto out;
1165         }
1166         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1167                 error = EWOULDBLOCK;
1168                 goto out;
1169         }
1170         m->m_len = 1;
1171         *mtod(m, caddr_t) = tp->t_iobc;
1172         if ((flags & MSG_PEEK) == 0)
1173                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1174
1175 out:
1176         TCPDEBUG2(PRU_RCVOOB);
1177         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1178         INP_WUNLOCK(inp);
1179         return (error);
1180 }
1181
1182 #ifdef INET
1183 struct pr_usrreqs tcp_usrreqs = {
1184         .pru_abort =            tcp_usr_abort,
1185         .pru_accept =           tcp_usr_accept,
1186         .pru_attach =           tcp_usr_attach,
1187         .pru_bind =             tcp_usr_bind,
1188         .pru_connect =          tcp_usr_connect,
1189         .pru_control =          in_control,
1190         .pru_detach =           tcp_usr_detach,
1191         .pru_disconnect =       tcp_usr_disconnect,
1192         .pru_listen =           tcp_usr_listen,
1193         .pru_peeraddr =         in_getpeeraddr,
1194         .pru_rcvd =             tcp_usr_rcvd,
1195         .pru_rcvoob =           tcp_usr_rcvoob,
1196         .pru_send =             tcp_usr_send,
1197         .pru_ready =            tcp_usr_ready,
1198         .pru_shutdown =         tcp_usr_shutdown,
1199         .pru_sockaddr =         in_getsockaddr,
1200         .pru_sosetlabel =       in_pcbsosetlabel,
1201         .pru_close =            tcp_usr_close,
1202 };
1203 #endif /* INET */
1204
1205 #ifdef INET6
1206 struct pr_usrreqs tcp6_usrreqs = {
1207         .pru_abort =            tcp_usr_abort,
1208         .pru_accept =           tcp6_usr_accept,
1209         .pru_attach =           tcp_usr_attach,
1210         .pru_bind =             tcp6_usr_bind,
1211         .pru_connect =          tcp6_usr_connect,
1212         .pru_control =          in6_control,
1213         .pru_detach =           tcp_usr_detach,
1214         .pru_disconnect =       tcp_usr_disconnect,
1215         .pru_listen =           tcp6_usr_listen,
1216         .pru_peeraddr =         in6_mapped_peeraddr,
1217         .pru_rcvd =             tcp_usr_rcvd,
1218         .pru_rcvoob =           tcp_usr_rcvoob,
1219         .pru_send =             tcp_usr_send,
1220         .pru_ready =            tcp_usr_ready,
1221         .pru_shutdown =         tcp_usr_shutdown,
1222         .pru_sockaddr =         in6_mapped_sockaddr,
1223         .pru_sosetlabel =       in_pcbsosetlabel,
1224         .pru_close =            tcp_usr_close,
1225 };
1226 #endif /* INET6 */
1227
1228 #ifdef INET
1229 /*
1230  * Common subroutine to open a TCP connection to remote host specified
1231  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1232  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1233  * to choose a local host address (interface).  If there is an existing
1234  * incarnation of the same connection in TIME-WAIT state and if the remote
1235  * host was sending CC options and if the connection duration was < MSL, then
1236  * truncate the previous TIME-WAIT state and proceed.
1237  * Initialize connection parameters and enter SYN-SENT state.
1238  */
1239 static int
1240 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1241 {
1242         struct inpcb *inp = tp->t_inpcb, *oinp;
1243         struct socket *so = inp->inp_socket;
1244         struct in_addr laddr;
1245         u_short lport;
1246         int error;
1247
1248         INP_WLOCK_ASSERT(inp);
1249         INP_HASH_WLOCK(&V_tcbinfo);
1250
1251         if (inp->inp_lport == 0) {
1252                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1253                 if (error)
1254                         goto out;
1255         }
1256
1257         /*
1258          * Cannot simply call in_pcbconnect, because there might be an
1259          * earlier incarnation of this same connection still in
1260          * TIME_WAIT state, creating an ADDRINUSE error.
1261          */
1262         laddr = inp->inp_laddr;
1263         lport = inp->inp_lport;
1264         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1265             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1266         if (error && oinp == NULL)
1267                 goto out;
1268         if (oinp) {
1269                 error = EADDRINUSE;
1270                 goto out;
1271         }
1272         inp->inp_laddr = laddr;
1273         in_pcbrehash(inp);
1274         INP_HASH_WUNLOCK(&V_tcbinfo);
1275
1276         /*
1277          * Compute window scaling to request:
1278          * Scale to fit into sweet spot.  See tcp_syncache.c.
1279          * XXX: This should move to tcp_output().
1280          */
1281         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1282             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1283                 tp->request_r_scale++;
1284
1285         soisconnecting(so);
1286         TCPSTAT_INC(tcps_connattempt);
1287         tcp_state_change(tp, TCPS_SYN_SENT);
1288         tp->iss = tcp_new_isn(tp);
1289         tcp_sendseqinit(tp);
1290
1291         return 0;
1292
1293 out:
1294         INP_HASH_WUNLOCK(&V_tcbinfo);
1295         return (error);
1296 }
1297 #endif /* INET */
1298
1299 #ifdef INET6
1300 static int
1301 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1302 {
1303         struct inpcb *inp = tp->t_inpcb;
1304         int error;
1305
1306         INP_WLOCK_ASSERT(inp);
1307         INP_HASH_WLOCK(&V_tcbinfo);
1308
1309         if (inp->inp_lport == 0) {
1310                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1311                 if (error)
1312                         goto out;
1313         }
1314         error = in6_pcbconnect(inp, nam, td->td_ucred);
1315         if (error != 0)
1316                 goto out;
1317         INP_HASH_WUNLOCK(&V_tcbinfo);
1318
1319         /* Compute window scaling to request.  */
1320         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1321             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1322                 tp->request_r_scale++;
1323
1324         soisconnecting(inp->inp_socket);
1325         TCPSTAT_INC(tcps_connattempt);
1326         tcp_state_change(tp, TCPS_SYN_SENT);
1327         tp->iss = tcp_new_isn(tp);
1328         tcp_sendseqinit(tp);
1329
1330         return 0;
1331
1332 out:
1333         INP_HASH_WUNLOCK(&V_tcbinfo);
1334         return error;
1335 }
1336 #endif /* INET6 */
1337
1338 /*
1339  * Export TCP internal state information via a struct tcp_info, based on the
1340  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1341  * (TCP state machine, etc).  We export all information using FreeBSD-native
1342  * constants -- for example, the numeric values for tcpi_state will differ
1343  * from Linux.
1344  */
1345 static void
1346 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1347 {
1348
1349         INP_WLOCK_ASSERT(tp->t_inpcb);
1350         bzero(ti, sizeof(*ti));
1351
1352         ti->tcpi_state = tp->t_state;
1353         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1354                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1355         if (tp->t_flags & TF_SACK_PERMIT)
1356                 ti->tcpi_options |= TCPI_OPT_SACK;
1357         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1358                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1359                 ti->tcpi_snd_wscale = tp->snd_scale;
1360                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1361         }
1362         if (tp->t_flags & TF_ECN_PERMIT)
1363                 ti->tcpi_options |= TCPI_OPT_ECN;
1364
1365         ti->tcpi_rto = tp->t_rxtcur * tick;
1366         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1367         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1368         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1369
1370         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1371         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1372
1373         /*
1374          * FreeBSD-specific extension fields for tcp_info.
1375          */
1376         ti->tcpi_rcv_space = tp->rcv_wnd;
1377         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1378         ti->tcpi_snd_wnd = tp->snd_wnd;
1379         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1380         ti->tcpi_snd_nxt = tp->snd_nxt;
1381         ti->tcpi_snd_mss = tp->t_maxseg;
1382         ti->tcpi_rcv_mss = tp->t_maxseg;
1383         if (tp->t_flags & TF_TOE)
1384                 ti->tcpi_options |= TCPI_OPT_TOE;
1385         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1386         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1387         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1388 }
1389
1390 /*
1391  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1392  * socket option arguments.  When it re-acquires the lock after the copy, it
1393  * has to revalidate that the connection is still valid for the socket
1394  * option.
1395  */
1396 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1397         INP_WLOCK(inp);                                                 \
1398         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1399                 INP_WUNLOCK(inp);                                       \
1400                 cleanup;                                                \
1401                 return (ECONNRESET);                                    \
1402         }                                                               \
1403         tp = intotcpcb(inp);                                            \
1404 } while(0)
1405 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1406
1407 int
1408 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1409 {
1410         int     error;
1411         struct  inpcb *inp;
1412         struct  tcpcb *tp;
1413         struct tcp_function_block *blk;
1414         struct tcp_function_set fsn;
1415
1416         error = 0;
1417         inp = sotoinpcb(so);
1418         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1419         INP_WLOCK(inp);
1420         if (sopt->sopt_level != IPPROTO_TCP) {
1421 #ifdef INET6
1422                 if (inp->inp_vflag & INP_IPV6PROTO) {
1423                         INP_WUNLOCK(inp);
1424                         error = ip6_ctloutput(so, sopt);
1425                 }
1426 #endif /* INET6 */
1427 #if defined(INET6) && defined(INET)
1428                 else
1429 #endif
1430 #ifdef INET
1431                 {
1432                         INP_WUNLOCK(inp);
1433                         error = ip_ctloutput(so, sopt);
1434                 }
1435 #endif
1436                 return (error);
1437         }
1438         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1439                 INP_WUNLOCK(inp);
1440                 return (ECONNRESET);
1441         }
1442         tp = intotcpcb(inp);
1443         /*
1444          * Protect the TCP option TCP_FUNCTION_BLK so
1445          * that a sub-function can *never* overwrite this.
1446          */
1447         if ((sopt->sopt_dir == SOPT_SET) &&
1448             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1449                 INP_WUNLOCK(inp);
1450                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1451                     sizeof fsn);
1452                 if (error)
1453                         return (error);
1454                 INP_WLOCK_RECHECK(inp);
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                         /* You already have this */
1462                         refcount_release(&blk->tfb_refcnt);
1463                         INP_WUNLOCK(inp);
1464                         return (0);
1465                 }
1466                 if (tp->t_state != TCPS_CLOSED) {
1467                         int error=EINVAL;
1468                         /*
1469                          * The user has advanced the state
1470                          * past the initial point, we may not
1471                          * be able to switch.
1472                          */
1473                         if (blk->tfb_tcp_handoff_ok != NULL) {
1474                                 /*
1475                                  * Does the stack provide a
1476                                  * query mechanism, if so it may
1477                                  * still be possible?
1478                                  */
1479                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1480                         }
1481                         if (error) {
1482                                 refcount_release(&blk->tfb_refcnt);
1483                                 INP_WUNLOCK(inp);
1484                                 return(error);
1485                         }
1486                 }
1487                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1488                         refcount_release(&blk->tfb_refcnt);
1489                         INP_WUNLOCK(inp);
1490                         return (ENOENT);
1491                 }
1492                 /*
1493                  * Release the old refcnt, the
1494                  * lookup acquired a ref on the
1495                  * new one already.
1496                  */
1497                 if (tp->t_fb->tfb_tcp_fb_fini) {
1498                         /*
1499                          * Tell the stack to cleanup with 0 i.e.
1500                          * the tcb is not going away.
1501                          */
1502                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1503                 }
1504                 refcount_release(&tp->t_fb->tfb_refcnt);
1505                 tp->t_fb = blk;
1506                 if (tp->t_fb->tfb_tcp_fb_init) {
1507                         (*tp->t_fb->tfb_tcp_fb_init)(tp);
1508                 }
1509 #ifdef TCP_OFFLOAD
1510                 if (tp->t_flags & TF_TOE) {
1511                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1512                              sopt->sopt_name);
1513                 }
1514 #endif
1515                 INP_WUNLOCK(inp);
1516                 return (error);
1517         } else if ((sopt->sopt_dir == SOPT_GET) &&
1518             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1519                 strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1520                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1521                 INP_WUNLOCK(inp);
1522                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1523                 return (error);
1524         }
1525         /* Pass in the INP locked, called must unlock it */
1526         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1527 }
1528
1529 int
1530 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1531 {
1532         int     error, opt, optval;
1533         u_int   ui;
1534         struct  tcp_info ti;
1535         struct cc_algo *algo;
1536         char    *pbuf, buf[TCP_CA_NAME_MAX];
1537         size_t  len;
1538
1539         /*
1540          * For TCP_CCALGOOPT forward the control to CC module, for both
1541          * SOPT_SET and SOPT_GET.
1542          */
1543         switch (sopt->sopt_name) {
1544         case TCP_CCALGOOPT:
1545                 INP_WUNLOCK(inp);
1546                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1547                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1548                     sopt->sopt_valsize);
1549                 if (error) {
1550                         free(pbuf, M_TEMP);
1551                         return (error);
1552                 }
1553                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1554                 if (CC_ALGO(tp)->ctl_output != NULL)
1555                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1556                 else
1557                         error = ENOENT;
1558                 INP_WUNLOCK(inp);
1559                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1560                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1561                 free(pbuf, M_TEMP);
1562                 return (error);
1563         }
1564
1565         switch (sopt->sopt_dir) {
1566         case SOPT_SET:
1567                 switch (sopt->sopt_name) {
1568 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1569                 case TCP_MD5SIG:
1570                         if (!TCPMD5_ENABLED()) {
1571                                 INP_WUNLOCK(inp);
1572                                 return (ENOPROTOOPT);
1573                         }
1574                         error = TCPMD5_PCBCTL(inp, sopt);
1575                         if (error)
1576                                 return (error);
1577                         goto unlock_and_done;
1578 #endif /* IPSEC */
1579
1580                 case TCP_NODELAY:
1581                 case TCP_NOOPT:
1582                         INP_WUNLOCK(inp);
1583                         error = sooptcopyin(sopt, &optval, sizeof optval,
1584                             sizeof optval);
1585                         if (error)
1586                                 return (error);
1587
1588                         INP_WLOCK_RECHECK(inp);
1589                         switch (sopt->sopt_name) {
1590                         case TCP_NODELAY:
1591                                 opt = TF_NODELAY;
1592                                 break;
1593                         case TCP_NOOPT:
1594                                 opt = TF_NOOPT;
1595                                 break;
1596                         default:
1597                                 opt = 0; /* dead code to fool gcc */
1598                                 break;
1599                         }
1600
1601                         if (optval)
1602                                 tp->t_flags |= opt;
1603                         else
1604                                 tp->t_flags &= ~opt;
1605 unlock_and_done:
1606 #ifdef TCP_OFFLOAD
1607                         if (tp->t_flags & TF_TOE) {
1608                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1609                                     sopt->sopt_name);
1610                         }
1611 #endif
1612                         INP_WUNLOCK(inp);
1613                         break;
1614
1615                 case TCP_NOPUSH:
1616                         INP_WUNLOCK(inp);
1617                         error = sooptcopyin(sopt, &optval, sizeof optval,
1618                             sizeof optval);
1619                         if (error)
1620                                 return (error);
1621
1622                         INP_WLOCK_RECHECK(inp);
1623                         if (optval)
1624                                 tp->t_flags |= TF_NOPUSH;
1625                         else if (tp->t_flags & TF_NOPUSH) {
1626                                 tp->t_flags &= ~TF_NOPUSH;
1627                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1628                                         error = tp->t_fb->tfb_tcp_output(tp);
1629                         }
1630                         goto unlock_and_done;
1631
1632                 case TCP_MAXSEG:
1633                         INP_WUNLOCK(inp);
1634                         error = sooptcopyin(sopt, &optval, sizeof optval,
1635                             sizeof optval);
1636                         if (error)
1637                                 return (error);
1638
1639                         INP_WLOCK_RECHECK(inp);
1640                         if (optval > 0 && optval <= tp->t_maxseg &&
1641                             optval + 40 >= V_tcp_minmss)
1642                                 tp->t_maxseg = optval;
1643                         else
1644                                 error = EINVAL;
1645                         goto unlock_and_done;
1646
1647                 case TCP_INFO:
1648                         INP_WUNLOCK(inp);
1649                         error = EINVAL;
1650                         break;
1651
1652                 case TCP_CONGESTION:
1653                         INP_WUNLOCK(inp);
1654                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1655                         if (error)
1656                                 break;
1657                         buf[sopt->sopt_valsize] = '\0';
1658                         INP_WLOCK_RECHECK(inp);
1659                         CC_LIST_RLOCK();
1660                         STAILQ_FOREACH(algo, &cc_list, entries)
1661                                 if (strncmp(buf, algo->name,
1662                                     TCP_CA_NAME_MAX) == 0)
1663                                         break;
1664                         CC_LIST_RUNLOCK();
1665                         if (algo == NULL) {
1666                                 INP_WUNLOCK(inp);
1667                                 error = EINVAL;
1668                                 break;
1669                         }
1670                         /*
1671                          * We hold a write lock over the tcb so it's safe to
1672                          * do these things without ordering concerns.
1673                          */
1674                         if (CC_ALGO(tp)->cb_destroy != NULL)
1675                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1676                         CC_ALGO(tp) = algo;
1677                         /*
1678                          * If something goes pear shaped initialising the new
1679                          * algo, fall back to newreno (which does not
1680                          * require initialisation).
1681                          */
1682                         if (algo->cb_init != NULL &&
1683                             algo->cb_init(tp->ccv) != 0) {
1684                                 CC_ALGO(tp) = &newreno_cc_algo;
1685                                 /*
1686                                  * The only reason init should fail is
1687                                  * because of malloc.
1688                                  */
1689                                 error = ENOMEM;
1690                         }
1691                         INP_WUNLOCK(inp);
1692                         break;
1693
1694                 case TCP_KEEPIDLE:
1695                 case TCP_KEEPINTVL:
1696                 case TCP_KEEPINIT:
1697                         INP_WUNLOCK(inp);
1698                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1699                         if (error)
1700                                 return (error);
1701
1702                         if (ui > (UINT_MAX / hz)) {
1703                                 error = EINVAL;
1704                                 break;
1705                         }
1706                         ui *= hz;
1707
1708                         INP_WLOCK_RECHECK(inp);
1709                         switch (sopt->sopt_name) {
1710                         case TCP_KEEPIDLE:
1711                                 tp->t_keepidle = ui;
1712                                 /*
1713                                  * XXX: better check current remaining
1714                                  * timeout and "merge" it with new value.
1715                                  */
1716                                 if ((tp->t_state > TCPS_LISTEN) &&
1717                                     (tp->t_state <= TCPS_CLOSING))
1718                                         tcp_timer_activate(tp, TT_KEEP,
1719                                             TP_KEEPIDLE(tp));
1720                                 break;
1721                         case TCP_KEEPINTVL:
1722                                 tp->t_keepintvl = ui;
1723                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1724                                     (TP_MAXIDLE(tp) > 0))
1725                                         tcp_timer_activate(tp, TT_2MSL,
1726                                             TP_MAXIDLE(tp));
1727                                 break;
1728                         case TCP_KEEPINIT:
1729                                 tp->t_keepinit = ui;
1730                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1731                                     tp->t_state == TCPS_SYN_SENT)
1732                                         tcp_timer_activate(tp, TT_KEEP,
1733                                             TP_KEEPINIT(tp));
1734                                 break;
1735                         }
1736                         goto unlock_and_done;
1737
1738                 case TCP_KEEPCNT:
1739                         INP_WUNLOCK(inp);
1740                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1741                         if (error)
1742                                 return (error);
1743
1744                         INP_WLOCK_RECHECK(inp);
1745                         tp->t_keepcnt = ui;
1746                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1747                             (TP_MAXIDLE(tp) > 0))
1748                                 tcp_timer_activate(tp, TT_2MSL,
1749                                     TP_MAXIDLE(tp));
1750                         goto unlock_and_done;
1751
1752 #ifdef TCPPCAP
1753                 case TCP_PCAP_OUT:
1754                 case TCP_PCAP_IN:
1755                         INP_WUNLOCK(inp);
1756                         error = sooptcopyin(sopt, &optval, sizeof optval,
1757                             sizeof optval);
1758                         if (error)
1759                                 return (error);
1760
1761                         INP_WLOCK_RECHECK(inp);
1762                         if (optval >= 0)
1763                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1764                                         &(tp->t_outpkts) : &(tp->t_inpkts),
1765                                         optval);
1766                         else
1767                                 error = EINVAL;
1768                         goto unlock_and_done;
1769 #endif
1770
1771 #ifdef TCP_RFC7413
1772                 case TCP_FASTOPEN:
1773                         INP_WUNLOCK(inp);
1774                         if (!V_tcp_fastopen_enabled)
1775                                 return (EPERM);
1776
1777                         error = sooptcopyin(sopt, &optval, sizeof optval,
1778                             sizeof optval);
1779                         if (error)
1780                                 return (error);
1781
1782                         INP_WLOCK_RECHECK(inp);
1783                         if (optval) {
1784                                 tp->t_flags |= TF_FASTOPEN;
1785                                 if ((tp->t_state == TCPS_LISTEN) &&
1786                                     (tp->t_tfo_pending == NULL))
1787                                         tp->t_tfo_pending =
1788                                             tcp_fastopen_alloc_counter();
1789                         } else
1790                                 tp->t_flags &= ~TF_FASTOPEN;
1791                         goto unlock_and_done;
1792 #endif
1793
1794                 default:
1795                         INP_WUNLOCK(inp);
1796                         error = ENOPROTOOPT;
1797                         break;
1798                 }
1799                 break;
1800
1801         case SOPT_GET:
1802                 tp = intotcpcb(inp);
1803                 switch (sopt->sopt_name) {
1804 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1805                 case TCP_MD5SIG:
1806                         if (!TCPMD5_ENABLED()) {
1807                                 INP_WUNLOCK(inp);
1808                                 return (ENOPROTOOPT);
1809                         }
1810                         error = TCPMD5_PCBCTL(inp, sopt);
1811                         break;
1812 #endif
1813
1814                 case TCP_NODELAY:
1815                         optval = tp->t_flags & TF_NODELAY;
1816                         INP_WUNLOCK(inp);
1817                         error = sooptcopyout(sopt, &optval, sizeof optval);
1818                         break;
1819                 case TCP_MAXSEG:
1820                         optval = tp->t_maxseg;
1821                         INP_WUNLOCK(inp);
1822                         error = sooptcopyout(sopt, &optval, sizeof optval);
1823                         break;
1824                 case TCP_NOOPT:
1825                         optval = tp->t_flags & TF_NOOPT;
1826                         INP_WUNLOCK(inp);
1827                         error = sooptcopyout(sopt, &optval, sizeof optval);
1828                         break;
1829                 case TCP_NOPUSH:
1830                         optval = tp->t_flags & TF_NOPUSH;
1831                         INP_WUNLOCK(inp);
1832                         error = sooptcopyout(sopt, &optval, sizeof optval);
1833                         break;
1834                 case TCP_INFO:
1835                         tcp_fill_info(tp, &ti);
1836                         INP_WUNLOCK(inp);
1837                         error = sooptcopyout(sopt, &ti, sizeof ti);
1838                         break;
1839                 case TCP_CONGESTION:
1840                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1841                         INP_WUNLOCK(inp);
1842                         error = sooptcopyout(sopt, buf, len + 1);
1843                         break;
1844                 case TCP_KEEPIDLE:
1845                 case TCP_KEEPINTVL:
1846                 case TCP_KEEPINIT:
1847                 case TCP_KEEPCNT:
1848                         switch (sopt->sopt_name) {
1849                         case TCP_KEEPIDLE:
1850                                 ui = TP_KEEPIDLE(tp) / hz;
1851                                 break;
1852                         case TCP_KEEPINTVL:
1853                                 ui = TP_KEEPINTVL(tp) / hz;
1854                                 break;
1855                         case TCP_KEEPINIT:
1856                                 ui = TP_KEEPINIT(tp) / hz;
1857                                 break;
1858                         case TCP_KEEPCNT:
1859                                 ui = TP_KEEPCNT(tp);
1860                                 break;
1861                         }
1862                         INP_WUNLOCK(inp);
1863                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1864                         break;
1865 #ifdef TCPPCAP
1866                 case TCP_PCAP_OUT:
1867                 case TCP_PCAP_IN:
1868                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1869                                         &(tp->t_outpkts) : &(tp->t_inpkts));
1870                         INP_WUNLOCK(inp);
1871                         error = sooptcopyout(sopt, &optval, sizeof optval);
1872                         break;
1873 #endif
1874
1875 #ifdef TCP_RFC7413
1876                 case TCP_FASTOPEN:
1877                         optval = tp->t_flags & TF_FASTOPEN;
1878                         INP_WUNLOCK(inp);
1879                         error = sooptcopyout(sopt, &optval, sizeof optval);
1880                         break;
1881 #endif
1882                 default:
1883                         INP_WUNLOCK(inp);
1884                         error = ENOPROTOOPT;
1885                         break;
1886                 }
1887                 break;
1888         }
1889         return (error);
1890 }
1891 #undef INP_WLOCK_RECHECK
1892 #undef INP_WLOCK_RECHECK_CLEANUP
1893
1894 /*
1895  * Attach TCP protocol to socket, allocating
1896  * internet protocol control block, tcp control block,
1897  * bufer space, and entering LISTEN state if to accept connections.
1898  */
1899 static int
1900 tcp_attach(struct socket *so)
1901 {
1902         struct tcpcb *tp;
1903         struct inpcb *inp;
1904         int error;
1905
1906         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1907                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1908                 if (error)
1909                         return (error);
1910         }
1911         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1912         so->so_snd.sb_flags |= SB_AUTOSIZE;
1913         INP_INFO_RLOCK(&V_tcbinfo);
1914         error = in_pcballoc(so, &V_tcbinfo);
1915         if (error) {
1916                 INP_INFO_RUNLOCK(&V_tcbinfo);
1917                 return (error);
1918         }
1919         inp = sotoinpcb(so);
1920 #ifdef INET6
1921         if (inp->inp_vflag & INP_IPV6PROTO) {
1922                 inp->inp_vflag |= INP_IPV6;
1923                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1924                         inp->inp_vflag |= INP_IPV4;
1925                 inp->in6p_hops = -1;    /* use kernel default */
1926         }
1927         else
1928 #endif
1929         inp->inp_vflag |= INP_IPV4;
1930         tp = tcp_newtcpcb(inp);
1931         if (tp == NULL) {
1932                 in_pcbdetach(inp);
1933                 in_pcbfree(inp);
1934                 INP_INFO_RUNLOCK(&V_tcbinfo);
1935                 return (ENOBUFS);
1936         }
1937         tp->t_state = TCPS_CLOSED;
1938         INP_WUNLOCK(inp);
1939         INP_INFO_RUNLOCK(&V_tcbinfo);
1940         TCPSTATES_INC(TCPS_CLOSED);
1941         return (0);
1942 }
1943
1944 /*
1945  * Initiate (or continue) disconnect.
1946  * If embryonic state, just send reset (once).
1947  * If in ``let data drain'' option and linger null, just drop.
1948  * Otherwise (hard), mark socket disconnecting and drop
1949  * current input data; switch states based on user close, and
1950  * send segment to peer (with FIN).
1951  */
1952 static void
1953 tcp_disconnect(struct tcpcb *tp)
1954 {
1955         struct inpcb *inp = tp->t_inpcb;
1956         struct socket *so = inp->inp_socket;
1957
1958         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1959         INP_WLOCK_ASSERT(inp);
1960
1961         /*
1962          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1963          * socket is still open.
1964          */
1965         if (tp->t_state < TCPS_ESTABLISHED) {
1966                 tp = tcp_close(tp);
1967                 KASSERT(tp != NULL,
1968                     ("tcp_disconnect: tcp_close() returned NULL"));
1969         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1970                 tp = tcp_drop(tp, 0);
1971                 KASSERT(tp != NULL,
1972                     ("tcp_disconnect: tcp_drop() returned NULL"));
1973         } else {
1974                 soisdisconnecting(so);
1975                 sbflush(&so->so_rcv);
1976                 tcp_usrclosed(tp);
1977                 if (!(inp->inp_flags & INP_DROPPED))
1978                         tp->t_fb->tfb_tcp_output(tp);
1979         }
1980 }
1981
1982 /*
1983  * User issued close, and wish to trail through shutdown states:
1984  * if never received SYN, just forget it.  If got a SYN from peer,
1985  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1986  * If already got a FIN from peer, then almost done; go to LAST_ACK
1987  * state.  In all other cases, have already sent FIN to peer (e.g.
1988  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1989  * for peer to send FIN or not respond to keep-alives, etc.
1990  * We can let the user exit from the close as soon as the FIN is acked.
1991  */
1992 static void
1993 tcp_usrclosed(struct tcpcb *tp)
1994 {
1995
1996         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1997         INP_WLOCK_ASSERT(tp->t_inpcb);
1998
1999         switch (tp->t_state) {
2000         case TCPS_LISTEN:
2001 #ifdef TCP_OFFLOAD
2002                 tcp_offload_listen_stop(tp);
2003 #endif
2004                 tcp_state_change(tp, TCPS_CLOSED);
2005                 /* FALLTHROUGH */
2006         case TCPS_CLOSED:
2007                 tp = tcp_close(tp);
2008                 /*
2009                  * tcp_close() should never return NULL here as the socket is
2010                  * still open.
2011                  */
2012                 KASSERT(tp != NULL,
2013                     ("tcp_usrclosed: tcp_close() returned NULL"));
2014                 break;
2015
2016         case TCPS_SYN_SENT:
2017         case TCPS_SYN_RECEIVED:
2018                 tp->t_flags |= TF_NEEDFIN;
2019                 break;
2020
2021         case TCPS_ESTABLISHED:
2022                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2023                 break;
2024
2025         case TCPS_CLOSE_WAIT:
2026                 tcp_state_change(tp, TCPS_LAST_ACK);
2027                 break;
2028         }
2029         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2030                 soisdisconnected(tp->t_inpcb->inp_socket);
2031                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2032                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2033                         int timeout;
2034
2035                         timeout = (tcp_fast_finwait2_recycle) ?
2036                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2037                         tcp_timer_activate(tp, TT_2MSL, timeout);
2038                 }
2039         }
2040 }
2041
2042 #ifdef DDB
2043 static void
2044 db_print_indent(int indent)
2045 {
2046         int i;
2047
2048         for (i = 0; i < indent; i++)
2049                 db_printf(" ");
2050 }
2051
2052 static void
2053 db_print_tstate(int t_state)
2054 {
2055
2056         switch (t_state) {
2057         case TCPS_CLOSED:
2058                 db_printf("TCPS_CLOSED");
2059                 return;
2060
2061         case TCPS_LISTEN:
2062                 db_printf("TCPS_LISTEN");
2063                 return;
2064
2065         case TCPS_SYN_SENT:
2066                 db_printf("TCPS_SYN_SENT");
2067                 return;
2068
2069         case TCPS_SYN_RECEIVED:
2070                 db_printf("TCPS_SYN_RECEIVED");
2071                 return;
2072
2073         case TCPS_ESTABLISHED:
2074                 db_printf("TCPS_ESTABLISHED");
2075                 return;
2076
2077         case TCPS_CLOSE_WAIT:
2078                 db_printf("TCPS_CLOSE_WAIT");
2079                 return;
2080
2081         case TCPS_FIN_WAIT_1:
2082                 db_printf("TCPS_FIN_WAIT_1");
2083                 return;
2084
2085         case TCPS_CLOSING:
2086                 db_printf("TCPS_CLOSING");
2087                 return;
2088
2089         case TCPS_LAST_ACK:
2090                 db_printf("TCPS_LAST_ACK");
2091                 return;
2092
2093         case TCPS_FIN_WAIT_2:
2094                 db_printf("TCPS_FIN_WAIT_2");
2095                 return;
2096
2097         case TCPS_TIME_WAIT:
2098                 db_printf("TCPS_TIME_WAIT");
2099                 return;
2100
2101         default:
2102                 db_printf("unknown");
2103                 return;
2104         }
2105 }
2106
2107 static void
2108 db_print_tflags(u_int t_flags)
2109 {
2110         int comma;
2111
2112         comma = 0;
2113         if (t_flags & TF_ACKNOW) {
2114                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2115                 comma = 1;
2116         }
2117         if (t_flags & TF_DELACK) {
2118                 db_printf("%sTF_DELACK", comma ? ", " : "");
2119                 comma = 1;
2120         }
2121         if (t_flags & TF_NODELAY) {
2122                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2123                 comma = 1;
2124         }
2125         if (t_flags & TF_NOOPT) {
2126                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2127                 comma = 1;
2128         }
2129         if (t_flags & TF_SENTFIN) {
2130                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2131                 comma = 1;
2132         }
2133         if (t_flags & TF_REQ_SCALE) {
2134                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2135                 comma = 1;
2136         }
2137         if (t_flags & TF_RCVD_SCALE) {
2138                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2139                 comma = 1;
2140         }
2141         if (t_flags & TF_REQ_TSTMP) {
2142                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2143                 comma = 1;
2144         }
2145         if (t_flags & TF_RCVD_TSTMP) {
2146                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2147                 comma = 1;
2148         }
2149         if (t_flags & TF_SACK_PERMIT) {
2150                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2151                 comma = 1;
2152         }
2153         if (t_flags & TF_NEEDSYN) {
2154                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2155                 comma = 1;
2156         }
2157         if (t_flags & TF_NEEDFIN) {
2158                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2159                 comma = 1;
2160         }
2161         if (t_flags & TF_NOPUSH) {
2162                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2163                 comma = 1;
2164         }
2165         if (t_flags & TF_MORETOCOME) {
2166                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2167                 comma = 1;
2168         }
2169         if (t_flags & TF_LQ_OVERFLOW) {
2170                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2171                 comma = 1;
2172         }
2173         if (t_flags & TF_LASTIDLE) {
2174                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2175                 comma = 1;
2176         }
2177         if (t_flags & TF_RXWIN0SENT) {
2178                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2179                 comma = 1;
2180         }
2181         if (t_flags & TF_FASTRECOVERY) {
2182                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2183                 comma = 1;
2184         }
2185         if (t_flags & TF_CONGRECOVERY) {
2186                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2187                 comma = 1;
2188         }
2189         if (t_flags & TF_WASFRECOVERY) {
2190                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2191                 comma = 1;
2192         }
2193         if (t_flags & TF_SIGNATURE) {
2194                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2195                 comma = 1;
2196         }
2197         if (t_flags & TF_FORCEDATA) {
2198                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2199                 comma = 1;
2200         }
2201         if (t_flags & TF_TSO) {
2202                 db_printf("%sTF_TSO", comma ? ", " : "");
2203                 comma = 1;
2204         }
2205         if (t_flags & TF_ECN_PERMIT) {
2206                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2207                 comma = 1;
2208         }
2209         if (t_flags & TF_FASTOPEN) {
2210                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2211                 comma = 1;
2212         }
2213 }
2214
2215 static void
2216 db_print_toobflags(char t_oobflags)
2217 {
2218         int comma;
2219
2220         comma = 0;
2221         if (t_oobflags & TCPOOB_HAVEDATA) {
2222                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2223                 comma = 1;
2224         }
2225         if (t_oobflags & TCPOOB_HADDATA) {
2226                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2227                 comma = 1;
2228         }
2229 }
2230
2231 static void
2232 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2233 {
2234
2235         db_print_indent(indent);
2236         db_printf("%s at %p\n", name, tp);
2237
2238         indent += 2;
2239
2240         db_print_indent(indent);
2241         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2242            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2243
2244         db_print_indent(indent);
2245         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2246             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2247
2248         db_print_indent(indent);
2249         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2250             &tp->t_timers->tt_delack, tp->t_inpcb);
2251
2252         db_print_indent(indent);
2253         db_printf("t_state: %d (", tp->t_state);
2254         db_print_tstate(tp->t_state);
2255         db_printf(")\n");
2256
2257         db_print_indent(indent);
2258         db_printf("t_flags: 0x%x (", tp->t_flags);
2259         db_print_tflags(tp->t_flags);
2260         db_printf(")\n");
2261
2262         db_print_indent(indent);
2263         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2264             tp->snd_una, tp->snd_max, tp->snd_nxt);
2265
2266         db_print_indent(indent);
2267         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2268            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2269
2270         db_print_indent(indent);
2271         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2272             tp->iss, tp->irs, tp->rcv_nxt);
2273
2274         db_print_indent(indent);
2275         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2276             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2277
2278         db_print_indent(indent);
2279         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2280            tp->snd_wnd, tp->snd_cwnd);
2281
2282         db_print_indent(indent);
2283         db_printf("snd_ssthresh: %u   snd_recover: "
2284             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2285
2286         db_print_indent(indent);
2287         db_printf("t_rcvtime: %u   t_startime: %u\n",
2288             tp->t_rcvtime, tp->t_starttime);
2289
2290         db_print_indent(indent);
2291         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2292             tp->t_rtttime, tp->t_rtseq);
2293
2294         db_print_indent(indent);
2295         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2296             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2297
2298         db_print_indent(indent);
2299         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2300             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2301             tp->t_rttbest);
2302
2303         db_print_indent(indent);
2304         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2305             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2306
2307         db_print_indent(indent);
2308         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2309         db_print_toobflags(tp->t_oobflags);
2310         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2311
2312         db_print_indent(indent);
2313         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2314             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2315
2316         db_print_indent(indent);
2317         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2318             tp->ts_recent, tp->ts_recent_age);
2319
2320         db_print_indent(indent);
2321         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2322             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2323
2324         db_print_indent(indent);
2325         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2326             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2327             tp->snd_recover_prev, tp->t_badrxtwin);
2328
2329         db_print_indent(indent);
2330         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2331             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2332
2333         db_print_indent(indent);
2334         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2335             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2336
2337         /* Skip sackblks, sackhint. */
2338
2339         db_print_indent(indent);
2340         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2341             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2342 }
2343
2344 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2345 {
2346         struct tcpcb *tp;
2347
2348         if (!have_addr) {
2349                 db_printf("usage: show tcpcb <addr>\n");
2350                 return;
2351         }
2352         tp = (struct tcpcb *)addr;
2353
2354         db_print_tcpcb(tp, "tcpcb", 0);
2355 }
2356 #endif