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 #include <netinet/tcp.h>
  89 #include <netinet/tcp_fsm.h>
  90 #include <netinet/tcp_seq.h>
  91 #include <netinet/tcp_timer.h>
  92 #include <netinet/tcp_var.h>
  93 #include <netinet/tcp_log_buf.h>
  94 #include <netinet/tcpip.h>
  95 #include <netinet/cc/cc.h>
  96 #include <netinet/tcp_fastopen.h>
  97 #include <netinet/tcp_hpts.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                  * four 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 tcptw creation time by
 209                  *  tcp_twstart, but connection is local and tw will be
 210                  *  discarded immediately
 211                  * #3 tcp_detach is called at timewait end (or reuse) by
 212                  *  tcp_twclose, then the tcptw has already been discarded
 213                  *  (or reused) and inpcb is freed here
 214                  * #4 tcp_detach is called() after timewait ends (or reuse)
 215                  *  (e.g. by soclose), then tcptw has already been discarded
 216                  *  (or reused) and inpcb is freed here
 217                  *
 218                  *  In all three cases the tcptw should not be freed here.
 219                  */
 220                 if (inp->inp_flags & INP_DROPPED) {
 221                         in_pcbdetach(inp);
 222                         if (__predict_true(tp == NULL)) {
 223                                 in_pcbfree(inp);
 224                         } else {
 225                                 /*
 226                                  * This case should not happen as in TIMEWAIT
 227                                  * state the inp should not be destroyed before
 228                                  * its tcptw.  If INVARIANTS is defined, panic.
 229                                  */
 230 #ifdef INVARIANTS
 231                                 panic("%s: Panic before an inp double-free: "
 232                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 233                                     , __func__);
 234 #else
 235                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 236                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 237                                     , __func__);
 238 #endif
 239                                 INP_WUNLOCK(inp);
 240                         }
 241                 } else {
 242                         in_pcbdetach(inp);
 243                         INP_WUNLOCK(inp);
 244                 }
 245         } else {
 246                 /*
 247                  * If the connection is not in timewait, we consider two
 248                  * two conditions: one in which no further processing is
 249                  * necessary (dropped || embryonic), and one in which TCP is
 250                  * not yet done, but no longer requires the socket, so the
 251                  * pcb will persist for the time being.
 252                  *
 253                  * XXXRW: Does the second case still occur?
 254                  */
 255                 if (inp->inp_flags & INP_DROPPED ||
 256                     tp->t_state < TCPS_SYN_SENT) {
 257                         tcp_discardcb(tp);
 258                         in_pcbdetach(inp);
 259                         in_pcbfree(inp);
 260                 } else {
 261                         in_pcbdetach(inp);
 262                         INP_WUNLOCK(inp);
 263                 }
 264         }
 265 }
 266
 267 /*
 268  * pru_detach() detaches the TCP protocol from the socket.
 269  * If the protocol state is non-embryonic, then can't
 270  * do this directly: have to initiate a pru_disconnect(),
 271  * which may finish later; embryonic TCB's can just
 272  * be discarded here.
 273  */
 274 static void
 275 tcp_usr_detach(struct socket *so)
 276 {
 277         struct inpcb *inp;
 278         int rlock = 0;
 279         struct epoch_tracker et;
 280
 281         inp = sotoinpcb(so);
 282         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 283         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 284                 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 285                 rlock = 1;
 286         }
 287         INP_WLOCK(inp);
 288         KASSERT(inp->inp_socket != NULL,
 289             ("tcp_usr_detach: inp_socket == NULL"));
 290         tcp_detach(so, inp);
 291         if (rlock)
 292                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 293 }
 294
 295 #ifdef INET
 296 /*
 297  * Give the socket an address.
 298  */
 299 static int
 300 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 301 {
 302         int error = 0;
 303         struct inpcb *inp;
 304         struct tcpcb *tp = NULL;
 305         struct sockaddr_in *sinp;
 306
 307         sinp = (struct sockaddr_in *)nam;
 308         if (nam->sa_len != sizeof (*sinp))
 309                 return (EINVAL);
 310         /*
 311          * Must check for multicast addresses and disallow binding
 312          * to them.
 313          */
 314         if (sinp->sin_family == AF_INET &&
 315             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 316                 return (EAFNOSUPPORT);
 317
 318         TCPDEBUG0;
 319         inp = sotoinpcb(so);
 320         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 321         INP_WLOCK(inp);
 322         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 323                 error = EINVAL;
 324                 goto out;
 325         }
 326         tp = intotcpcb(inp);
 327         TCPDEBUG1();
 328         INP_HASH_WLOCK(&V_tcbinfo);
 329         error = in_pcbbind(inp, nam, td->td_ucred);
 330         INP_HASH_WUNLOCK(&V_tcbinfo);
 331 out:
 332         TCPDEBUG2(PRU_BIND);
 333         TCP_PROBE2(debug__user, tp, PRU_BIND);
 334         INP_WUNLOCK(inp);
 335
 336         return (error);
 337 }
 338 #endif /* INET */
 339
 340 #ifdef INET6
 341 static int
 342 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 343 {
 344         int error = 0;
 345         struct inpcb *inp;
 346         struct tcpcb *tp = NULL;
 347         struct sockaddr_in6 *sin6p;
 348         u_char vflagsav;
 349
 350         sin6p = (struct sockaddr_in6 *)nam;
 351         if (nam->sa_len != sizeof (*sin6p))
 352                 return (EINVAL);
 353         /*
 354          * Must check for multicast addresses and disallow binding
 355          * to them.
 356          */
 357         if (sin6p->sin6_family == AF_INET6 &&
 358             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 359                 return (EAFNOSUPPORT);
 360
 361         TCPDEBUG0;
 362         inp = sotoinpcb(so);
 363         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 364         INP_WLOCK(inp);
 365         vflagsav = inp->inp_vflag;
 366         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 367                 error = EINVAL;
 368                 goto out;
 369         }
 370         tp = intotcpcb(inp);
 371         TCPDEBUG1();
 372         INP_HASH_WLOCK(&V_tcbinfo);
 373         inp->inp_vflag &= ~INP_IPV4;
 374         inp->inp_vflag |= INP_IPV6;
 375 #ifdef INET
 376         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 377                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 378                         inp->inp_vflag |= INP_IPV4;
 379                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 380                         struct sockaddr_in sin;
 381
 382                         in6_sin6_2_sin(&sin, sin6p);
 383                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 384                                 error = EAFNOSUPPORT;
 385                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 386                                 goto out;
 387                         }
 388                         inp->inp_vflag |= INP_IPV4;
 389                         inp->inp_vflag &= ~INP_IPV6;
 390                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 391                             td->td_ucred);
 392                         INP_HASH_WUNLOCK(&V_tcbinfo);
 393                         goto out;
 394                 }
 395         }
 396 #endif
 397         error = in6_pcbbind(inp, nam, td->td_ucred);
 398         INP_HASH_WUNLOCK(&V_tcbinfo);
 399 out:
 400         if (error != 0)
 401                 inp->inp_vflag = vflagsav;
 402         TCPDEBUG2(PRU_BIND);
 403         TCP_PROBE2(debug__user, tp, PRU_BIND);
 404         INP_WUNLOCK(inp);
 405         return (error);
 406 }
 407 #endif /* INET6 */
 408
 409 #ifdef INET
 410 /*
 411  * Prepare to accept connections.
 412  */
 413 static int
 414 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 415 {
 416         int error = 0;
 417         struct inpcb *inp;
 418         struct tcpcb *tp = NULL;
 419
 420         TCPDEBUG0;
 421         inp = sotoinpcb(so);
 422         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 423         INP_WLOCK(inp);
 424         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 425                 error = EINVAL;
 426                 goto out;
 427         }
 428         tp = intotcpcb(inp);
 429         TCPDEBUG1();
 430         SOCK_LOCK(so);
 431         error = solisten_proto_check(so);
 432         INP_HASH_WLOCK(&V_tcbinfo);
 433         if (error == 0 && inp->inp_lport == 0)
 434                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 435         INP_HASH_WUNLOCK(&V_tcbinfo);
 436         if (error == 0) {
 437                 tcp_state_change(tp, TCPS_LISTEN);
 438                 solisten_proto(so, backlog);
 439 #ifdef TCP_OFFLOAD
 440                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 441                         tcp_offload_listen_start(tp);
 442 #endif
 443         }
 444         SOCK_UNLOCK(so);
 445
 446         if (IS_FASTOPEN(tp->t_flags))
 447                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 448
 449 out:
 450         TCPDEBUG2(PRU_LISTEN);
 451         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 452         INP_WUNLOCK(inp);
 453         return (error);
 454 }
 455 #endif /* INET */
 456
 457 #ifdef INET6
 458 static int
 459 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 460 {
 461         int error = 0;
 462         struct inpcb *inp;
 463         struct tcpcb *tp = NULL;
 464         u_char vflagsav;
 465
 466         TCPDEBUG0;
 467         inp = sotoinpcb(so);
 468         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 469         INP_WLOCK(inp);
 470         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 471                 error = EINVAL;
 472                 goto out;
 473         }
 474         vflagsav = inp->inp_vflag;
 475         tp = intotcpcb(inp);
 476         TCPDEBUG1();
 477         SOCK_LOCK(so);
 478         error = solisten_proto_check(so);
 479         INP_HASH_WLOCK(&V_tcbinfo);
 480         if (error == 0 && inp->inp_lport == 0) {
 481                 inp->inp_vflag &= ~INP_IPV4;
 482                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 483                         inp->inp_vflag |= INP_IPV4;
 484                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 485         }
 486         INP_HASH_WUNLOCK(&V_tcbinfo);
 487         if (error == 0) {
 488                 tcp_state_change(tp, TCPS_LISTEN);
 489                 solisten_proto(so, backlog);
 490 #ifdef TCP_OFFLOAD
 491                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 492                         tcp_offload_listen_start(tp);
 493 #endif
 494         }
 495         SOCK_UNLOCK(so);
 496
 497         if (IS_FASTOPEN(tp->t_flags))
 498                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 499
 500         if (error != 0)
 501                 inp->inp_vflag = vflagsav;
 502
 503 out:
 504         TCPDEBUG2(PRU_LISTEN);
 505         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 506         INP_WUNLOCK(inp);
 507         return (error);
 508 }
 509 #endif /* INET6 */
 510
 511 #ifdef INET
 512 /*
 513  * Initiate connection to peer.
 514  * Create a template for use in transmissions on this connection.
 515  * Enter SYN_SENT state, and mark socket as connecting.
 516  * Start keep-alive timer, and seed output sequence space.
 517  * Send initial segment on connection.
 518  */
 519 static int
 520 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 521 {
 522         int error = 0;
 523         struct inpcb *inp;
 524         struct tcpcb *tp = NULL;
 525         struct sockaddr_in *sinp;
 526
 527         sinp = (struct sockaddr_in *)nam;
 528         if (nam->sa_len != sizeof (*sinp))
 529                 return (EINVAL);
 530         /*
 531          * Must disallow TCP ``connections'' to multicast addresses.
 532          */
 533         if (sinp->sin_family == AF_INET
 534             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 535                 return (EAFNOSUPPORT);
 536         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 537                 return (error);
 538
 539         TCPDEBUG0;
 540         inp = sotoinpcb(so);
 541         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 542         INP_WLOCK(inp);
 543         if (inp->inp_flags & INP_TIMEWAIT) {
 544                 error = EADDRINUSE;
 545                 goto out;
 546         }
 547         if (inp->inp_flags & INP_DROPPED) {
 548                 error = ECONNREFUSED;
 549                 goto out;
 550         }
 551         tp = intotcpcb(inp);
 552         TCPDEBUG1();
 553         if ((error = tcp_connect(tp, nam, td)) != 0)
 554                 goto out;
 555 #ifdef TCP_OFFLOAD
 556         if (registered_toedevs > 0 &&
 557             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 558             (error = tcp_offload_connect(so, nam)) == 0)
 559                 goto out;
 560 #endif
 561         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 562         error = tp->t_fb->tfb_tcp_output(tp);
 563 out:
 564         TCPDEBUG2(PRU_CONNECT);
 565         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 566         INP_WUNLOCK(inp);
 567         return (error);
 568 }
 569 #endif /* INET */
 570
 571 #ifdef INET6
 572 static int
 573 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 574 {
 575         int error = 0;
 576         struct inpcb *inp;
 577         struct tcpcb *tp = NULL;
 578         struct sockaddr_in6 *sin6p;
 579         u_int8_t incflagsav;
 580         u_char vflagsav;
 581
 582         TCPDEBUG0;
 583
 584         sin6p = (struct sockaddr_in6 *)nam;
 585         if (nam->sa_len != sizeof (*sin6p))
 586                 return (EINVAL);
 587         /*
 588          * Must disallow TCP ``connections'' to multicast addresses.
 589          */
 590         if (sin6p->sin6_family == AF_INET6
 591             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 592                 return (EAFNOSUPPORT);
 593
 594         inp = sotoinpcb(so);
 595         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 596         INP_WLOCK(inp);
 597         vflagsav = inp->inp_vflag;
 598         incflagsav = inp->inp_inc.inc_flags;
 599         if (inp->inp_flags & INP_TIMEWAIT) {
 600                 error = EADDRINUSE;
 601                 goto out;
 602         }
 603         if (inp->inp_flags & INP_DROPPED) {
 604                 error = ECONNREFUSED;
 605                 goto out;
 606         }
 607         tp = intotcpcb(inp);
 608         TCPDEBUG1();
 609 #ifdef INET
 610         /*
 611          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 612          * therefore probably require the hash lock, which isn't held here.
 613          * Is this a significant problem?
 614          */
 615         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 616                 struct sockaddr_in sin;
 617
 618                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 619                         error = EINVAL;
 620                         goto out;
 621                 }
 622                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 623                         error = EAFNOSUPPORT;
 624                         goto out;
 625                 }
 626
 627                 in6_sin6_2_sin(&sin, sin6p);
 628                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 629                         error = EAFNOSUPPORT;
 630                         goto out;
 631                 }
 632                 if ((error = prison_remote_ip4(td->td_ucred,
 633                     &sin.sin_addr)) != 0)
 634                         goto out;
 635                 inp->inp_vflag |= INP_IPV4;
 636                 inp->inp_vflag &= ~INP_IPV6;
 637                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 638                         goto out;
 639 #ifdef TCP_OFFLOAD
 640                 if (registered_toedevs > 0 &&
 641                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 642                     (error = tcp_offload_connect(so, nam)) == 0)
 643                         goto out;
 644 #endif
 645                 error = tp->t_fb->tfb_tcp_output(tp);
 646                 goto out;
 647         } else {
 648                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 649                         error = EAFNOSUPPORT;
 650                         goto out;
 651                 }
 652         }
 653 #endif
 654         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 655                 goto out;
 656         inp->inp_vflag &= ~INP_IPV4;
 657         inp->inp_vflag |= INP_IPV6;
 658         inp->inp_inc.inc_flags |= INC_ISIPV6;
 659         if ((error = tcp6_connect(tp, nam, td)) != 0)
 660                 goto out;
 661 #ifdef TCP_OFFLOAD
 662         if (registered_toedevs > 0 &&
 663             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 664             (error = tcp_offload_connect(so, nam)) == 0)
 665                 goto out;
 666 #endif
 667         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 668         error = tp->t_fb->tfb_tcp_output(tp);
 669
 670 out:
 671         /*
 672          * If the implicit bind in the connect call fails, restore
 673          * the flags we modified.
 674          */
 675         if (error != 0 && inp->inp_lport == 0) {
 676                 inp->inp_vflag = vflagsav;
 677                 inp->inp_inc.inc_flags = incflagsav;
 678         }
 679
 680         TCPDEBUG2(PRU_CONNECT);
 681         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 682         INP_WUNLOCK(inp);
 683         return (error);
 684 }
 685 #endif /* INET6 */
 686
 687 /*
 688  * Initiate disconnect from peer.
 689  * If connection never passed embryonic stage, just drop;
 690  * else if don't need to let data drain, then can just drop anyways,
 691  * else have to begin TCP shutdown process: mark socket disconnecting,
 692  * drain unread data, state switch to reflect user close, and
 693  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 694  * when peer sends FIN and acks ours.
 695  *
 696  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 697  */
 698 static int
 699 tcp_usr_disconnect(struct socket *so)
 700 {
 701         struct inpcb *inp;
 702         struct tcpcb *tp = NULL;
 703         struct epoch_tracker et;
 704         int error = 0;
 705
 706         TCPDEBUG0;
 707         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 708         inp = sotoinpcb(so);
 709         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 710         INP_WLOCK(inp);
 711         if (inp->inp_flags & INP_TIMEWAIT)
 712                 goto out;
 713         if (inp->inp_flags & INP_DROPPED) {
 714                 error = ECONNRESET;
 715                 goto out;
 716         }
 717         tp = intotcpcb(inp);
 718         TCPDEBUG1();
 719         tcp_disconnect(tp);
 720 out:
 721         TCPDEBUG2(PRU_DISCONNECT);
 722         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 723         INP_WUNLOCK(inp);
 724         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 725         return (error);
 726 }
 727
 728 #ifdef INET
 729 /*
 730  * Accept a connection.  Essentially all the work is done at higher levels;
 731  * just return the address of the peer, storing through addr.
 732  */
 733 static int
 734 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 735 {
 736         int error = 0;
 737         struct inpcb *inp = NULL;
 738         struct tcpcb *tp = NULL;
 739         struct in_addr addr;
 740         in_port_t port = 0;
 741         TCPDEBUG0;
 742
 743         if (so->so_state & SS_ISDISCONNECTED)
 744                 return (ECONNABORTED);
 745
 746         inp = sotoinpcb(so);
 747         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 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 in_getpeeraddr 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         port = inp->inp_fport;
 762         addr = inp->inp_faddr;
 763
 764 out:
 765         TCPDEBUG2(PRU_ACCEPT);
 766         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 767         INP_WUNLOCK(inp);
 768         if (error == 0)
 769                 *nam = in_sockaddr(port, &addr);
 770         return error;
 771 }
 772 #endif /* INET */
 773
 774 #ifdef INET6
 775 static int
 776 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 777 {
 778         struct inpcb *inp = NULL;
 779         int error = 0;
 780         struct tcpcb *tp = NULL;
 781         struct in_addr addr;
 782         struct in6_addr addr6;
 783         struct epoch_tracker et;
 784         in_port_t port = 0;
 785         int v4 = 0;
 786         TCPDEBUG0;
 787
 788         if (so->so_state & SS_ISDISCONNECTED)
 789                 return (ECONNABORTED);
 790
 791         inp = sotoinpcb(so);
 792         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 793         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 794         INP_WLOCK(inp);
 795         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 796                 error = ECONNABORTED;
 797                 goto out;
 798         }
 799         tp = intotcpcb(inp);
 800         TCPDEBUG1();
 801
 802         /*
 803          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 804          * copy the data of interest and defer the malloc until after we
 805          * release the lock.
 806          */
 807         if (inp->inp_vflag & INP_IPV4) {
 808                 v4 = 1;
 809                 port = inp->inp_fport;
 810                 addr = inp->inp_faddr;
 811         } else {
 812                 port = inp->inp_fport;
 813                 addr6 = inp->in6p_faddr;
 814         }
 815
 816 out:
 817         TCPDEBUG2(PRU_ACCEPT);
 818         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 819         INP_WUNLOCK(inp);
 820         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 821         if (error == 0) {
 822                 if (v4)
 823                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 824                 else
 825                         *nam = in6_sockaddr(port, &addr6);
 826         }
 827         return error;
 828 }
 829 #endif /* INET6 */
 830
 831 /*
 832  * Mark the connection as being incapable of further output.
 833  */
 834 static int
 835 tcp_usr_shutdown(struct socket *so)
 836 {
 837         int error = 0;
 838         struct inpcb *inp;
 839         struct tcpcb *tp = NULL;
 840         struct epoch_tracker et;
 841
 842         TCPDEBUG0;
 843         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 844         inp = sotoinpcb(so);
 845         KASSERT(inp != NULL, ("inp == NULL"));
 846         INP_WLOCK(inp);
 847         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 848                 error = ECONNRESET;
 849                 goto out;
 850         }
 851         tp = intotcpcb(inp);
 852         TCPDEBUG1();
 853         socantsendmore(so);
 854         tcp_usrclosed(tp);
 855         if (!(inp->inp_flags & INP_DROPPED))
 856                 error = tp->t_fb->tfb_tcp_output(tp);
 857
 858 out:
 859         TCPDEBUG2(PRU_SHUTDOWN);
 860         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 861         INP_WUNLOCK(inp);
 862         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 863
 864         return (error);
 865 }
 866
 867 /*
 868  * After a receive, possibly send window update to peer.
 869  */
 870 static int
 871 tcp_usr_rcvd(struct socket *so, int flags)
 872 {
 873         struct inpcb *inp;
 874         struct tcpcb *tp = NULL;
 875         int error = 0;
 876
 877         TCPDEBUG0;
 878         inp = sotoinpcb(so);
 879         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 880         INP_WLOCK(inp);
 881         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 882                 error = ECONNRESET;
 883                 goto out;
 884         }
 885         tp = intotcpcb(inp);
 886         TCPDEBUG1();
 887         /*
 888          * For passively-created TFO connections, don't attempt a window
 889          * update while still in SYN_RECEIVED as this may trigger an early
 890          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 891          * application response data, or failing that, when the DELACK timer
 892          * expires.
 893          */
 894         if (IS_FASTOPEN(tp->t_flags) &&
 895             (tp->t_state == TCPS_SYN_RECEIVED))
 896                 goto out;
 897 #ifdef TCP_OFFLOAD
 898         if (tp->t_flags & TF_TOE)
 899                 tcp_offload_rcvd(tp);
 900         else
 901 #endif
 902         tp->t_fb->tfb_tcp_output(tp);
 903
 904 out:
 905         TCPDEBUG2(PRU_RCVD);
 906         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 907         INP_WUNLOCK(inp);
 908         return (error);
 909 }
 910
 911 /*
 912  * Do a send by putting data in output queue and updating urgent
 913  * marker if URG set.  Possibly send more data.  Unlike the other
 914  * pru_*() routines, the mbuf chains are our responsibility.  We
 915  * must either enqueue them or free them.  The other pru_* routines
 916  * generally are caller-frees.
 917  */
 918 static int
 919 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 920     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 921 {
 922         int error = 0;
 923         struct inpcb *inp;
 924         struct tcpcb *tp = NULL;
 925         struct epoch_tracker net_et;
 926 #ifdef INET
 927 #ifdef INET6
 928         struct sockaddr_in sin;
 929 #endif
 930         struct sockaddr_in *sinp;
 931 #endif
 932 #ifdef INET6
 933         int isipv6;
 934 #endif
 935         u_int8_t incflagsav;
 936         u_char vflagsav;
 937         bool restoreflags;
 938         TCPDEBUG0;
 939
 940         /*
 941          * We require the pcbinfo lock if we will close the socket as part of
 942          * this call.
 943          */
 944         if (flags & PRUS_EOF)
 945                 INP_INFO_RLOCK_ET(&V_tcbinfo, net_et);
 946         inp = sotoinpcb(so);
 947         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 948         INP_WLOCK(inp);
 949         vflagsav = inp->inp_vflag;
 950         incflagsav = inp->inp_inc.inc_flags;
 951         restoreflags = false;
 952         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 953                 if (control)
 954                         m_freem(control);
 955                 /*
 956                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 957                  * for freeing memory.
 958                  */
 959                 if (m && (flags & PRUS_NOTREADY) == 0)
 960                         m_freem(m);
 961                 error = ECONNRESET;
 962                 goto out;
 963         }
 964         tp = intotcpcb(inp);
 965         TCPDEBUG1();
 966         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
 967                 switch (nam->sa_family) {
 968 #ifdef INET
 969                 case AF_INET:
 970                         sinp = (struct sockaddr_in *)nam;
 971                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
 972                                 if (m)
 973                                         m_freem(m);
 974                                 error = EINVAL;
 975                                 goto out;
 976                         }
 977                         if ((inp->inp_vflag & INP_IPV6) != 0) {
 978                                 if (m)
 979                                         m_freem(m);
 980                                 error = EAFNOSUPPORT;
 981                                 goto out;
 982                         }
 983                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 984                                 if (m)
 985                                         m_freem(m);
 986                                 error = EAFNOSUPPORT;
 987                                 goto out;
 988                         }
 989                         if ((error = prison_remote_ip4(td->td_ucred,
 990                             &sinp->sin_addr))) {
 991                                 if (m)
 992                                         m_freem(m);
 993                                 goto out;
 994                         }
 995 #ifdef INET6
 996                         isipv6 = 0;
 997 #endif
 998                         break;
 999 #endif /* INET */
1000 #ifdef INET6
1001                 case AF_INET6:
1002                 {
1003                         struct sockaddr_in6 *sin6p;
1004
1005                         sin6p = (struct sockaddr_in6 *)nam;
1006                         if (sin6p->sin6_len != sizeof(struct sockaddr_in6)) {
1007                                 if (m)
1008                                         m_freem(m);
1009                                 error = EINVAL;
1010                                 goto out;
1011                         }
1012                         if (IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
1013                                 if (m)
1014                                         m_freem(m);
1015                                 error = EAFNOSUPPORT;
1016                                 goto out;
1017                         }
1018                         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
1019 #ifdef INET
1020                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1021                                         error = EINVAL;
1022                                         if (m)
1023                                                 m_freem(m);
1024                                         goto out;
1025                                 }
1026                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1027                                         error = EAFNOSUPPORT;
1028                                         if (m)
1029                                                 m_freem(m);
1030                                         goto out;
1031                                 }
1032                                 restoreflags = true;
1033                                 inp->inp_vflag &= ~INP_IPV6;
1034                                 sinp = &sin;
1035                                 in6_sin6_2_sin(sinp, sin6p);
1036                                 if (IN_MULTICAST(
1037                                     ntohl(sinp->sin_addr.s_addr))) {
1038                                         error = EAFNOSUPPORT;
1039                                         if (m)
1040                                                 m_freem(m);
1041                                         goto out;
1042                                 }
1043                                 if ((error = prison_remote_ip4(td->td_ucred,
1044                                     &sinp->sin_addr))) {
1045                                         if (m)
1046                                                 m_freem(m);
1047                                         goto out;
1048                                 }
1049                                 isipv6 = 0;
1050 #else /* !INET */
1051                                 error = EAFNOSUPPORT;
1052                                 if (m)
1053                                         m_freem(m);
1054                                 goto out;
1055 #endif /* INET */
1056                         } else {
1057                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1058                                         if (m)
1059                                                 m_freem(m);
1060                                         error = EAFNOSUPPORT;
1061                                         goto out;
1062                                 }
1063                                 restoreflags = true;
1064                                 inp->inp_vflag &= ~INP_IPV4;
1065                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1066                                 if ((error = prison_remote_ip6(td->td_ucred,
1067                                     &sin6p->sin6_addr))) {
1068                                         if (m)
1069                                                 m_freem(m);
1070                                         goto out;
1071                                 }
1072                                 isipv6 = 1;
1073                         }
1074                         break;
1075                 }
1076 #endif /* INET6 */
1077                 default:
1078                         if (m)
1079                                 m_freem(m);
1080                         error = EAFNOSUPPORT;
1081                         goto out;
1082                 }
1083         }
1084         if (control) {
1085                 /* TCP doesn't do control messages (rights, creds, etc) */
1086                 if (control->m_len) {
1087                         m_freem(control);
1088                         if (m)
1089                                 m_freem(m);
1090                         error = EINVAL;
1091                         goto out;
1092                 }
1093                 m_freem(control);       /* empty control, just free it */
1094         }
1095         if (!(flags & PRUS_OOB)) {
1096                 sbappendstream(&so->so_snd, m, flags);
1097                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1098                         /*
1099                          * Do implied connect if not yet connected,
1100                          * initialize window to default value, and
1101                          * initialize maxseg using peer's cached MSS.
1102                          */
1103 #ifdef INET6
1104                         if (isipv6)
1105                                 error = tcp6_connect(tp, nam, td);
1106 #endif /* INET6 */
1107 #if defined(INET6) && defined(INET)
1108                         else
1109 #endif
1110 #ifdef INET
1111                                 error = tcp_connect(tp,
1112                                     (struct sockaddr *)sinp, td);
1113 #endif
1114                         /*
1115                          * The bind operation in tcp_connect succeeded. We
1116                          * no longer want to restore the flags if later
1117                          * operations fail.
1118                          */
1119                         if (error == 0 || inp->inp_lport != 0)
1120                                 restoreflags = false;
1121
1122                         if (error)
1123                                 goto out;
1124                         if (IS_FASTOPEN(tp->t_flags))
1125                                 tcp_fastopen_connect(tp);
1126                         else {
1127                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1128                                 tcp_mss(tp, -1);
1129                         }
1130                 }
1131                 if (flags & PRUS_EOF) {
1132                         /*
1133                          * Close the send side of the connection after
1134                          * the data is sent.
1135                          */
1136                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1137                         socantsendmore(so);
1138                         tcp_usrclosed(tp);
1139                 }
1140                 if (!(inp->inp_flags & INP_DROPPED) &&
1141                     !(flags & PRUS_NOTREADY)) {
1142                         if (flags & PRUS_MORETOCOME)
1143                                 tp->t_flags |= TF_MORETOCOME;
1144                         error = tp->t_fb->tfb_tcp_output(tp);
1145                         if (flags & PRUS_MORETOCOME)
1146                                 tp->t_flags &= ~TF_MORETOCOME;
1147                 }
1148         } else {
1149                 /*
1150                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1151                  */
1152                 SOCKBUF_LOCK(&so->so_snd);
1153                 if (sbspace(&so->so_snd) < -512) {
1154                         SOCKBUF_UNLOCK(&so->so_snd);
1155                         m_freem(m);
1156                         error = ENOBUFS;
1157                         goto out;
1158                 }
1159                 /*
1160                  * According to RFC961 (Assigned Protocols),
1161                  * the urgent pointer points to the last octet
1162                  * of urgent data.  We continue, however,
1163                  * to consider it to indicate the first octet
1164                  * of data past the urgent section.
1165                  * Otherwise, snd_up should be one lower.
1166                  */
1167                 sbappendstream_locked(&so->so_snd, m, flags);
1168                 SOCKBUF_UNLOCK(&so->so_snd);
1169                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1170                         /*
1171                          * Do implied connect if not yet connected,
1172                          * initialize window to default value, and
1173                          * initialize maxseg using peer's cached MSS.
1174                          */
1175
1176                         /*
1177                          * Not going to contemplate SYN|URG
1178                          */
1179                         if (IS_FASTOPEN(tp->t_flags))
1180                                 tp->t_flags &= ~TF_FASTOPEN;
1181 #ifdef INET6
1182                         if (isipv6)
1183                                 error = tcp6_connect(tp, nam, td);
1184 #endif /* INET6 */
1185 #if defined(INET6) && defined(INET)
1186                         else
1187 #endif
1188 #ifdef INET
1189                                 error = tcp_connect(tp,
1190                                     (struct sockaddr *)sinp, td);
1191 #endif
1192                         /*
1193                          * The bind operation in tcp_connect succeeded. We
1194                          * no longer want to restore the flags if later
1195                          * operations fail.
1196                          */
1197                         if (error == 0 || inp->inp_lport != 0)
1198                                 restoreflags = false;
1199
1200                         if (error)
1201                                 goto out;
1202                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1203                         tcp_mss(tp, -1);
1204                 }
1205                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1206                 if (!(flags & PRUS_NOTREADY)) {
1207                         tp->t_flags |= TF_FORCEDATA;
1208                         error = tp->t_fb->tfb_tcp_output(tp);
1209                         tp->t_flags &= ~TF_FORCEDATA;
1210                 }
1211         }
1212         TCP_LOG_EVENT(tp, NULL,
1213             &inp->inp_socket->so_rcv,
1214             &inp->inp_socket->so_snd,
1215             TCP_LOG_USERSEND, error,
1216             0, NULL, false);
1217 out:
1218         /*
1219          * If the request was unsuccessful and we changed flags,
1220          * restore the original flags.
1221          */
1222         if (error != 0 && restoreflags) {
1223                 inp->inp_vflag = vflagsav;
1224                 inp->inp_inc.inc_flags = incflagsav;
1225         }
1226         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1227                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1228         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1229                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1230         INP_WUNLOCK(inp);
1231         if (flags & PRUS_EOF)
1232                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, net_et);
1233         return (error);
1234 }
1235
1236 static int
1237 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1238 {
1239         struct inpcb *inp;
1240         struct tcpcb *tp;
1241         int error;
1242
1243         inp = sotoinpcb(so);
1244         INP_WLOCK(inp);
1245         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1246                 INP_WUNLOCK(inp);
1247                 for (int i = 0; i < count; i++)
1248                         m = m_free(m);
1249                 return (ECONNRESET);
1250         }
1251         tp = intotcpcb(inp);
1252
1253         SOCKBUF_LOCK(&so->so_snd);
1254         error = sbready(&so->so_snd, m, count);
1255         SOCKBUF_UNLOCK(&so->so_snd);
1256         if (error == 0)
1257                 error = tp->t_fb->tfb_tcp_output(tp);
1258         INP_WUNLOCK(inp);
1259
1260         return (error);
1261 }
1262
1263 /*
1264  * Abort the TCP.  Drop the connection abruptly.
1265  */
1266 static void
1267 tcp_usr_abort(struct socket *so)
1268 {
1269         struct inpcb *inp;
1270         struct tcpcb *tp = NULL;
1271         struct epoch_tracker et;
1272         TCPDEBUG0;
1273
1274         inp = sotoinpcb(so);
1275         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1276
1277         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1278         INP_WLOCK(inp);
1279         KASSERT(inp->inp_socket != NULL,
1280             ("tcp_usr_abort: inp_socket == NULL"));
1281
1282         /*
1283          * If we still have full TCP state, and we're not dropped, drop.
1284          */
1285         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1286             !(inp->inp_flags & INP_DROPPED)) {
1287                 tp = intotcpcb(inp);
1288                 TCPDEBUG1();
1289                 tp = tcp_drop(tp, ECONNABORTED);
1290                 if (tp == NULL)
1291                         goto dropped;
1292                 TCPDEBUG2(PRU_ABORT);
1293                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1294         }
1295         if (!(inp->inp_flags & INP_DROPPED)) {
1296                 SOCK_LOCK(so);
1297                 so->so_state |= SS_PROTOREF;
1298                 SOCK_UNLOCK(so);
1299                 inp->inp_flags |= INP_SOCKREF;
1300         }
1301         INP_WUNLOCK(inp);
1302 dropped:
1303         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1304 }
1305
1306 /*
1307  * TCP socket is closed.  Start friendly disconnect.
1308  */
1309 static void
1310 tcp_usr_close(struct socket *so)
1311 {
1312         struct inpcb *inp;
1313         struct tcpcb *tp = NULL;
1314         struct epoch_tracker et;
1315         TCPDEBUG0;
1316
1317         inp = sotoinpcb(so);
1318         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1319
1320         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1321         INP_WLOCK(inp);
1322         KASSERT(inp->inp_socket != NULL,
1323             ("tcp_usr_close: inp_socket == NULL"));
1324
1325         /*
1326          * If we still have full TCP state, and we're not dropped, initiate
1327          * a disconnect.
1328          */
1329         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1330             !(inp->inp_flags & INP_DROPPED)) {
1331                 tp = intotcpcb(inp);
1332                 TCPDEBUG1();
1333                 tcp_disconnect(tp);
1334                 TCPDEBUG2(PRU_CLOSE);
1335                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1336         }
1337         if (!(inp->inp_flags & INP_DROPPED)) {
1338                 SOCK_LOCK(so);
1339                 so->so_state |= SS_PROTOREF;
1340                 SOCK_UNLOCK(so);
1341                 inp->inp_flags |= INP_SOCKREF;
1342         }
1343         INP_WUNLOCK(inp);
1344         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1345 }
1346
1347 /*
1348  * Receive out-of-band data.
1349  */
1350 static int
1351 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1352 {
1353         int error = 0;
1354         struct inpcb *inp;
1355         struct tcpcb *tp = NULL;
1356
1357         TCPDEBUG0;
1358         inp = sotoinpcb(so);
1359         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1360         INP_WLOCK(inp);
1361         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1362                 error = ECONNRESET;
1363                 goto out;
1364         }
1365         tp = intotcpcb(inp);
1366         TCPDEBUG1();
1367         if ((so->so_oobmark == 0 &&
1368              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1369             so->so_options & SO_OOBINLINE ||
1370             tp->t_oobflags & TCPOOB_HADDATA) {
1371                 error = EINVAL;
1372                 goto out;
1373         }
1374         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1375                 error = EWOULDBLOCK;
1376                 goto out;
1377         }
1378         m->m_len = 1;
1379         *mtod(m, caddr_t) = tp->t_iobc;
1380         if ((flags & MSG_PEEK) == 0)
1381                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1382
1383 out:
1384         TCPDEBUG2(PRU_RCVOOB);
1385         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1386         INP_WUNLOCK(inp);
1387         return (error);
1388 }
1389
1390 #ifdef INET
1391 struct pr_usrreqs tcp_usrreqs = {
1392         .pru_abort =            tcp_usr_abort,
1393         .pru_accept =           tcp_usr_accept,
1394         .pru_attach =           tcp_usr_attach,
1395         .pru_bind =             tcp_usr_bind,
1396         .pru_connect =          tcp_usr_connect,
1397         .pru_control =          in_control,
1398         .pru_detach =           tcp_usr_detach,
1399         .pru_disconnect =       tcp_usr_disconnect,
1400         .pru_listen =           tcp_usr_listen,
1401         .pru_peeraddr =         in_getpeeraddr,
1402         .pru_rcvd =             tcp_usr_rcvd,
1403         .pru_rcvoob =           tcp_usr_rcvoob,
1404         .pru_send =             tcp_usr_send,
1405         .pru_ready =            tcp_usr_ready,
1406         .pru_shutdown =         tcp_usr_shutdown,
1407         .pru_sockaddr =         in_getsockaddr,
1408         .pru_sosetlabel =       in_pcbsosetlabel,
1409         .pru_close =            tcp_usr_close,
1410 };
1411 #endif /* INET */
1412
1413 #ifdef INET6
1414 struct pr_usrreqs tcp6_usrreqs = {
1415         .pru_abort =            tcp_usr_abort,
1416         .pru_accept =           tcp6_usr_accept,
1417         .pru_attach =           tcp_usr_attach,
1418         .pru_bind =             tcp6_usr_bind,
1419         .pru_connect =          tcp6_usr_connect,
1420         .pru_control =          in6_control,
1421         .pru_detach =           tcp_usr_detach,
1422         .pru_disconnect =       tcp_usr_disconnect,
1423         .pru_listen =           tcp6_usr_listen,
1424         .pru_peeraddr =         in6_mapped_peeraddr,
1425         .pru_rcvd =             tcp_usr_rcvd,
1426         .pru_rcvoob =           tcp_usr_rcvoob,
1427         .pru_send =             tcp_usr_send,
1428         .pru_ready =            tcp_usr_ready,
1429         .pru_shutdown =         tcp_usr_shutdown,
1430         .pru_sockaddr =         in6_mapped_sockaddr,
1431         .pru_sosetlabel =       in_pcbsosetlabel,
1432         .pru_close =            tcp_usr_close,
1433 };
1434 #endif /* INET6 */
1435
1436 #ifdef INET
1437 /*
1438  * Common subroutine to open a TCP connection to remote host specified
1439  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1440  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1441  * to choose a local host address (interface).  If there is an existing
1442  * incarnation of the same connection in TIME-WAIT state and if the remote
1443  * host was sending CC options and if the connection duration was < MSL, then
1444  * truncate the previous TIME-WAIT state and proceed.
1445  * Initialize connection parameters and enter SYN-SENT state.
1446  */
1447 static int
1448 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1449 {
1450         struct inpcb *inp = tp->t_inpcb, *oinp;
1451         struct socket *so = inp->inp_socket;
1452         struct in_addr laddr;
1453         u_short lport;
1454         int error;
1455
1456         INP_WLOCK_ASSERT(inp);
1457         INP_HASH_WLOCK(&V_tcbinfo);
1458
1459         if (inp->inp_lport == 0) {
1460                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1461                 if (error)
1462                         goto out;
1463         }
1464
1465         /*
1466          * Cannot simply call in_pcbconnect, because there might be an
1467          * earlier incarnation of this same connection still in
1468          * TIME_WAIT state, creating an ADDRINUSE error.
1469          */
1470         laddr = inp->inp_laddr;
1471         lport = inp->inp_lport;
1472         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1473             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1474         if (error && oinp == NULL)
1475                 goto out;
1476         if (oinp) {
1477                 error = EADDRINUSE;
1478                 goto out;
1479         }
1480         inp->inp_laddr = laddr;
1481         in_pcbrehash(inp);
1482         INP_HASH_WUNLOCK(&V_tcbinfo);
1483
1484         /*
1485          * Compute window scaling to request:
1486          * Scale to fit into sweet spot.  See tcp_syncache.c.
1487          * XXX: This should move to tcp_output().
1488          */
1489         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1490             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1491                 tp->request_r_scale++;
1492
1493         soisconnecting(so);
1494         TCPSTAT_INC(tcps_connattempt);
1495         tcp_state_change(tp, TCPS_SYN_SENT);
1496         tp->iss = tcp_new_isn(&inp->inp_inc);
1497         if (tp->t_flags & TF_REQ_TSTMP)
1498                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1499         tcp_sendseqinit(tp);
1500
1501         return 0;
1502
1503 out:
1504         INP_HASH_WUNLOCK(&V_tcbinfo);
1505         return (error);
1506 }
1507 #endif /* INET */
1508
1509 #ifdef INET6
1510 static int
1511 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1512 {
1513         struct inpcb *inp = tp->t_inpcb;
1514         int error;
1515
1516         INP_WLOCK_ASSERT(inp);
1517         INP_HASH_WLOCK(&V_tcbinfo);
1518
1519         if (inp->inp_lport == 0) {
1520                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1521                 if (error)
1522                         goto out;
1523         }
1524         error = in6_pcbconnect(inp, nam, td->td_ucred);
1525         if (error != 0)
1526                 goto out;
1527         INP_HASH_WUNLOCK(&V_tcbinfo);
1528
1529         /* Compute window scaling to request.  */
1530         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1531             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1532                 tp->request_r_scale++;
1533
1534         soisconnecting(inp->inp_socket);
1535         TCPSTAT_INC(tcps_connattempt);
1536         tcp_state_change(tp, TCPS_SYN_SENT);
1537         tp->iss = tcp_new_isn(&inp->inp_inc);
1538         if (tp->t_flags & TF_REQ_TSTMP)
1539                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1540         tcp_sendseqinit(tp);
1541
1542         return 0;
1543
1544 out:
1545         INP_HASH_WUNLOCK(&V_tcbinfo);
1546         return error;
1547 }
1548 #endif /* INET6 */
1549
1550 /*
1551  * Export TCP internal state information via a struct tcp_info, based on the
1552  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1553  * (TCP state machine, etc).  We export all information using FreeBSD-native
1554  * constants -- for example, the numeric values for tcpi_state will differ
1555  * from Linux.
1556  */
1557 static void
1558 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1559 {
1560
1561         INP_WLOCK_ASSERT(tp->t_inpcb);
1562         bzero(ti, sizeof(*ti));
1563
1564         ti->tcpi_state = tp->t_state;
1565         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1566                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1567         if (tp->t_flags & TF_SACK_PERMIT)
1568                 ti->tcpi_options |= TCPI_OPT_SACK;
1569         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1570                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1571                 ti->tcpi_snd_wscale = tp->snd_scale;
1572                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1573         }
1574         if (tp->t_flags & TF_ECN_PERMIT)
1575                 ti->tcpi_options |= TCPI_OPT_ECN;
1576
1577         ti->tcpi_rto = tp->t_rxtcur * tick;
1578         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1579         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1580         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1581
1582         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1583         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1584
1585         /*
1586          * FreeBSD-specific extension fields for tcp_info.
1587          */
1588         ti->tcpi_rcv_space = tp->rcv_wnd;
1589         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1590         ti->tcpi_snd_wnd = tp->snd_wnd;
1591         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1592         ti->tcpi_snd_nxt = tp->snd_nxt;
1593         ti->tcpi_snd_mss = tp->t_maxseg;
1594         ti->tcpi_rcv_mss = tp->t_maxseg;
1595         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1596         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1597         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1598 #ifdef TCP_OFFLOAD
1599         if (tp->t_flags & TF_TOE) {
1600                 ti->tcpi_options |= TCPI_OPT_TOE;
1601                 tcp_offload_tcp_info(tp, ti);
1602         }
1603 #endif
1604 }
1605
1606 /*
1607  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1608  * socket option arguments.  When it re-acquires the lock after the copy, it
1609  * has to revalidate that the connection is still valid for the socket
1610  * option.
1611  */
1612 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1613         INP_WLOCK(inp);                                                 \
1614         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1615                 INP_WUNLOCK(inp);                                       \
1616                 cleanup;                                                \
1617                 return (ECONNRESET);                                    \
1618         }                                                               \
1619         tp = intotcpcb(inp);                                            \
1620 } while(0)
1621 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1622
1623 int
1624 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1625 {
1626         int     error;
1627         struct  inpcb *inp;
1628         struct  tcpcb *tp;
1629         struct tcp_function_block *blk;
1630         struct tcp_function_set fsn;
1631
1632         error = 0;
1633         inp = sotoinpcb(so);
1634         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1635         if (sopt->sopt_level != IPPROTO_TCP) {
1636 #ifdef INET6
1637                 if (inp->inp_vflag & INP_IPV6PROTO) {
1638                         error = ip6_ctloutput(so, sopt);
1639                         /*
1640                          * In case of the IPV6_USE_MIN_MTU socket option,
1641                          * the INC_IPV6MINMTU flag to announce a corresponding
1642                          * MSS during the initial handshake.
1643                          * If the TCP connection is not in the front states,
1644                          * just reduce the MSS being used.
1645                          * This avoids the sending of TCP segments which will
1646                          * be fragmented at the IPv6 layer.
1647                          */
1648                         if ((error == 0) &&
1649                             (sopt->sopt_dir == SOPT_SET) &&
1650                             (sopt->sopt_level == IPPROTO_IPV6) &&
1651                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1652                                 INP_WLOCK(inp);
1653                                 if ((inp->inp_flags &
1654                                     (INP_TIMEWAIT | INP_DROPPED))) {
1655                                         INP_WUNLOCK(inp);
1656                                         return (ECONNRESET);
1657                                 }
1658                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1659                                 tp = intotcpcb(inp);
1660                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1661                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1662                                         struct ip6_pktopts *opt;
1663
1664                                         opt = inp->in6p_outputopts;
1665                                         if ((opt != NULL) &&
1666                                             (opt->ip6po_minmtu ==
1667                                             IP6PO_MINMTU_ALL)) {
1668                                                 if (tp->t_maxseg > TCP6_MSS) {
1669                                                         tp->t_maxseg = TCP6_MSS;
1670                                                 }
1671                                         }
1672                                 }
1673                                 INP_WUNLOCK(inp);
1674                         }
1675                 }
1676 #endif /* INET6 */
1677 #if defined(INET6) && defined(INET)
1678                 else
1679 #endif
1680 #ifdef INET
1681                 {
1682                         error = ip_ctloutput(so, sopt);
1683                 }
1684 #endif
1685                 return (error);
1686         }
1687         INP_WLOCK(inp);
1688         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1689                 INP_WUNLOCK(inp);
1690                 return (ECONNRESET);
1691         }
1692         tp = intotcpcb(inp);
1693         /*
1694          * Protect the TCP option TCP_FUNCTION_BLK so
1695          * that a sub-function can *never* overwrite this.
1696          */
1697         if ((sopt->sopt_dir == SOPT_SET) &&
1698             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1699                 INP_WUNLOCK(inp);
1700                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1701                     sizeof fsn);
1702                 if (error)
1703                         return (error);
1704                 INP_WLOCK_RECHECK(inp);
1705                 blk = find_and_ref_tcp_functions(&fsn);
1706                 if (blk == NULL) {
1707                         INP_WUNLOCK(inp);
1708                         return (ENOENT);
1709                 }
1710                 if (tp->t_fb == blk) {
1711                         /* You already have this */
1712                         refcount_release(&blk->tfb_refcnt);
1713                         INP_WUNLOCK(inp);
1714                         return (0);
1715                 }
1716                 if (tp->t_state != TCPS_CLOSED) {
1717                         /*
1718                          * The user has advanced the state
1719                          * past the initial point, we may not
1720                          * be able to switch.
1721                          */
1722                         if (blk->tfb_tcp_handoff_ok != NULL) {
1723                                 /*
1724                                  * Does the stack provide a
1725                                  * query mechanism, if so it may
1726                                  * still be possible?
1727                                  */
1728                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1729                         } else
1730                                 error = EINVAL;
1731                         if (error) {
1732                                 refcount_release(&blk->tfb_refcnt);
1733                                 INP_WUNLOCK(inp);
1734                                 return(error);
1735                         }
1736                 }
1737                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1738                         refcount_release(&blk->tfb_refcnt);
1739                         INP_WUNLOCK(inp);
1740                         return (ENOENT);
1741                 }
1742                 /*
1743                  * Release the old refcnt, the
1744                  * lookup acquired a ref on the
1745                  * new one already.
1746                  */
1747                 if (tp->t_fb->tfb_tcp_fb_fini) {
1748                         /*
1749                          * Tell the stack to cleanup with 0 i.e.
1750                          * the tcb is not going away.
1751                          */
1752                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1753                 }
1754 #ifdef TCPHPTS
1755                 /* Assure that we are not on any hpts */
1756                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1757 #endif
1758                 if (blk->tfb_tcp_fb_init) {
1759                         error = (*blk->tfb_tcp_fb_init)(tp);
1760                         if (error) {
1761                                 refcount_release(&blk->tfb_refcnt);
1762                                 if (tp->t_fb->tfb_tcp_fb_init) {
1763                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1764                                                 /* Fall back failed, drop the connection */
1765                                                 INP_WUNLOCK(inp);
1766                                                 soabort(so);
1767                                                 return(error);
1768                                         }
1769                                 }
1770                                 goto err_out;
1771                         }
1772                 }
1773                 refcount_release(&tp->t_fb->tfb_refcnt);
1774                 tp->t_fb = blk;
1775 #ifdef TCP_OFFLOAD
1776                 if (tp->t_flags & TF_TOE) {
1777                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1778                              sopt->sopt_name);
1779                 }
1780 #endif
1781 err_out:
1782                 INP_WUNLOCK(inp);
1783                 return (error);
1784         } else if ((sopt->sopt_dir == SOPT_GET) &&
1785             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1786                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1787                     TCP_FUNCTION_NAME_LEN_MAX);
1788                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1789                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1790                 INP_WUNLOCK(inp);
1791                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1792                 return (error);
1793         }
1794         /* Pass in the INP locked, called must unlock it */
1795         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1796 }
1797
1798 /*
1799  * If this assert becomes untrue, we need to change the size of the buf
1800  * variable in tcp_default_ctloutput().
1801  */
1802 #ifdef CTASSERT
1803 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1804 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1805 #endif
1806
1807 int
1808 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1809 {
1810         int     error, opt, optval;
1811         u_int   ui;
1812         struct  tcp_info ti;
1813         struct cc_algo *algo;
1814         char    *pbuf, buf[TCP_LOG_ID_LEN];
1815         size_t  len;
1816
1817         /*
1818          * For TCP_CCALGOOPT forward the control to CC module, for both
1819          * SOPT_SET and SOPT_GET.
1820          */
1821         switch (sopt->sopt_name) {
1822         case TCP_CCALGOOPT:
1823                 INP_WUNLOCK(inp);
1824                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1825                         return (EINVAL);
1826                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1827                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1828                     sopt->sopt_valsize);
1829                 if (error) {
1830                         free(pbuf, M_TEMP);
1831                         return (error);
1832                 }
1833                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1834                 if (CC_ALGO(tp)->ctl_output != NULL)
1835                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1836                 else
1837                         error = ENOENT;
1838                 INP_WUNLOCK(inp);
1839                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1840                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1841                 free(pbuf, M_TEMP);
1842                 return (error);
1843         }
1844
1845         switch (sopt->sopt_dir) {
1846         case SOPT_SET:
1847                 switch (sopt->sopt_name) {
1848 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1849                 case TCP_MD5SIG:
1850                         if (!TCPMD5_ENABLED()) {
1851                                 INP_WUNLOCK(inp);
1852                                 return (ENOPROTOOPT);
1853                         }
1854                         error = TCPMD5_PCBCTL(inp, sopt);
1855                         if (error)
1856                                 return (error);
1857                         goto unlock_and_done;
1858 #endif /* IPSEC */
1859
1860                 case TCP_NODELAY:
1861                 case TCP_NOOPT:
1862                         INP_WUNLOCK(inp);
1863                         error = sooptcopyin(sopt, &optval, sizeof optval,
1864                             sizeof optval);
1865                         if (error)
1866                                 return (error);
1867
1868                         INP_WLOCK_RECHECK(inp);
1869                         switch (sopt->sopt_name) {
1870                         case TCP_NODELAY:
1871                                 opt = TF_NODELAY;
1872                                 break;
1873                         case TCP_NOOPT:
1874                                 opt = TF_NOOPT;
1875                                 break;
1876                         default:
1877                                 opt = 0; /* dead code to fool gcc */
1878                                 break;
1879                         }
1880
1881                         if (optval)
1882                                 tp->t_flags |= opt;
1883                         else
1884                                 tp->t_flags &= ~opt;
1885 unlock_and_done:
1886 #ifdef TCP_OFFLOAD
1887                         if (tp->t_flags & TF_TOE) {
1888                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1889                                     sopt->sopt_name);
1890                         }
1891 #endif
1892                         INP_WUNLOCK(inp);
1893                         break;
1894
1895                 case TCP_NOPUSH:
1896                         INP_WUNLOCK(inp);
1897                         error = sooptcopyin(sopt, &optval, sizeof optval,
1898                             sizeof optval);
1899                         if (error)
1900                                 return (error);
1901
1902                         INP_WLOCK_RECHECK(inp);
1903                         if (optval)
1904                                 tp->t_flags |= TF_NOPUSH;
1905                         else if (tp->t_flags & TF_NOPUSH) {
1906                                 tp->t_flags &= ~TF_NOPUSH;
1907                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1908                                         error = tp->t_fb->tfb_tcp_output(tp);
1909                         }
1910                         goto unlock_and_done;
1911
1912                 case TCP_MAXSEG:
1913                         INP_WUNLOCK(inp);
1914                         error = sooptcopyin(sopt, &optval, sizeof optval,
1915                             sizeof optval);
1916                         if (error)
1917                                 return (error);
1918
1919                         INP_WLOCK_RECHECK(inp);
1920                         if (optval > 0 && optval <= tp->t_maxseg &&
1921                             optval + 40 >= V_tcp_minmss)
1922                                 tp->t_maxseg = optval;
1923                         else
1924                                 error = EINVAL;
1925                         goto unlock_and_done;
1926
1927                 case TCP_INFO:
1928                         INP_WUNLOCK(inp);
1929                         error = EINVAL;
1930                         break;
1931
1932                 case TCP_CONGESTION:
1933                         INP_WUNLOCK(inp);
1934                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1935                         if (error)
1936                                 break;
1937                         buf[sopt->sopt_valsize] = '\0';
1938                         INP_WLOCK_RECHECK(inp);
1939                         CC_LIST_RLOCK();
1940                         STAILQ_FOREACH(algo, &cc_list, entries)
1941                                 if (strncmp(buf, algo->name,
1942                                     TCP_CA_NAME_MAX) == 0)
1943                                         break;
1944                         CC_LIST_RUNLOCK();
1945                         if (algo == NULL) {
1946                                 INP_WUNLOCK(inp);
1947                                 error = EINVAL;
1948                                 break;
1949                         }
1950                         /*
1951                          * We hold a write lock over the tcb so it's safe to
1952                          * do these things without ordering concerns.
1953                          */
1954                         if (CC_ALGO(tp)->cb_destroy != NULL)
1955                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1956                         CC_DATA(tp) = NULL;
1957                         CC_ALGO(tp) = algo;
1958                         /*
1959                          * If something goes pear shaped initialising the new
1960                          * algo, fall back to newreno (which does not
1961                          * require initialisation).
1962                          */
1963                         if (algo->cb_init != NULL &&
1964                             algo->cb_init(tp->ccv) != 0) {
1965                                 CC_ALGO(tp) = &newreno_cc_algo;
1966                                 /*
1967                                  * The only reason init should fail is
1968                                  * because of malloc.
1969                                  */
1970                                 error = ENOMEM;
1971                         }
1972                         INP_WUNLOCK(inp);
1973                         break;
1974
1975                 case TCP_KEEPIDLE:
1976                 case TCP_KEEPINTVL:
1977                 case TCP_KEEPINIT:
1978                         INP_WUNLOCK(inp);
1979                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1980                         if (error)
1981                                 return (error);
1982
1983                         if (ui > (UINT_MAX / hz)) {
1984                                 error = EINVAL;
1985                                 break;
1986                         }
1987                         ui *= hz;
1988
1989                         INP_WLOCK_RECHECK(inp);
1990                         switch (sopt->sopt_name) {
1991                         case TCP_KEEPIDLE:
1992                                 tp->t_keepidle = ui;
1993                                 /*
1994                                  * XXX: better check current remaining
1995                                  * timeout and "merge" it with new value.
1996                                  */
1997                                 if ((tp->t_state > TCPS_LISTEN) &&
1998                                     (tp->t_state <= TCPS_CLOSING))
1999                                         tcp_timer_activate(tp, TT_KEEP,
2000                                             TP_KEEPIDLE(tp));
2001                                 break;
2002                         case TCP_KEEPINTVL:
2003                                 tp->t_keepintvl = ui;
2004                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2005                                     (TP_MAXIDLE(tp) > 0))
2006                                         tcp_timer_activate(tp, TT_2MSL,
2007                                             TP_MAXIDLE(tp));
2008                                 break;
2009                         case TCP_KEEPINIT:
2010                                 tp->t_keepinit = ui;
2011                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
2012                                     tp->t_state == TCPS_SYN_SENT)
2013                                         tcp_timer_activate(tp, TT_KEEP,
2014                                             TP_KEEPINIT(tp));
2015                                 break;
2016                         }
2017                         goto unlock_and_done;
2018
2019                 case TCP_KEEPCNT:
2020                         INP_WUNLOCK(inp);
2021                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2022                         if (error)
2023                                 return (error);
2024
2025                         INP_WLOCK_RECHECK(inp);
2026                         tp->t_keepcnt = ui;
2027                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2028                             (TP_MAXIDLE(tp) > 0))
2029                                 tcp_timer_activate(tp, TT_2MSL,
2030                                     TP_MAXIDLE(tp));
2031                         goto unlock_and_done;
2032
2033 #ifdef TCPPCAP
2034                 case TCP_PCAP_OUT:
2035                 case TCP_PCAP_IN:
2036                         INP_WUNLOCK(inp);
2037                         error = sooptcopyin(sopt, &optval, sizeof optval,
2038                             sizeof optval);
2039                         if (error)
2040                                 return (error);
2041
2042                         INP_WLOCK_RECHECK(inp);
2043                         if (optval >= 0)
2044                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2045                                         &(tp->t_outpkts) : &(tp->t_inpkts),
2046                                         optval);
2047                         else
2048                                 error = EINVAL;
2049                         goto unlock_and_done;
2050 #endif
2051
2052                 case TCP_FASTOPEN: {
2053                         struct tcp_fastopen tfo_optval;
2054
2055                         INP_WUNLOCK(inp);
2056                         if (!V_tcp_fastopen_client_enable &&
2057                             !V_tcp_fastopen_server_enable)
2058                                 return (EPERM);
2059
2060                         error = sooptcopyin(sopt, &tfo_optval,
2061                                     sizeof(tfo_optval), sizeof(int));
2062                         if (error)
2063                                 return (error);
2064
2065                         INP_WLOCK_RECHECK(inp);
2066                         if (tfo_optval.enable) {
2067                                 if (tp->t_state == TCPS_LISTEN) {
2068                                         if (!V_tcp_fastopen_server_enable) {
2069                                                 error = EPERM;
2070                                                 goto unlock_and_done;
2071                                         }
2072
2073                                         tp->t_flags |= TF_FASTOPEN;
2074                                         if (tp->t_tfo_pending == NULL)
2075                                                 tp->t_tfo_pending =
2076                                                     tcp_fastopen_alloc_counter();
2077                                 } else {
2078                                         /*
2079                                          * If a pre-shared key was provided,
2080                                          * stash it in the client cookie
2081                                          * field of the tcpcb for use during
2082                                          * connect.
2083                                          */
2084                                         if (sopt->sopt_valsize ==
2085                                             sizeof(tfo_optval)) {
2086                                                 memcpy(tp->t_tfo_cookie.client,
2087                                                        tfo_optval.psk,
2088                                                        TCP_FASTOPEN_PSK_LEN);
2089                                                 tp->t_tfo_client_cookie_len =
2090                                                     TCP_FASTOPEN_PSK_LEN;
2091                                         }
2092                                         tp->t_flags |= TF_FASTOPEN;
2093                                 }
2094                         } else
2095                                 tp->t_flags &= ~TF_FASTOPEN;
2096                         goto unlock_and_done;
2097                 }
2098
2099 #ifdef TCP_BLACKBOX
2100                 case TCP_LOG:
2101                         INP_WUNLOCK(inp);
2102                         error = sooptcopyin(sopt, &optval, sizeof optval,
2103                             sizeof optval);
2104                         if (error)
2105                                 return (error);
2106
2107                         INP_WLOCK_RECHECK(inp);
2108                         error = tcp_log_state_change(tp, optval);
2109                         goto unlock_and_done;
2110
2111                 case TCP_LOGBUF:
2112                         INP_WUNLOCK(inp);
2113                         error = EINVAL;
2114                         break;
2115
2116                 case TCP_LOGID:
2117                         INP_WUNLOCK(inp);
2118                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2119                         if (error)
2120                                 break;
2121                         buf[sopt->sopt_valsize] = '\0';
2122                         INP_WLOCK_RECHECK(inp);
2123                         error = tcp_log_set_id(tp, buf);
2124                         /* tcp_log_set_id() unlocks the INP. */
2125                         break;
2126
2127                 case TCP_LOGDUMP:
2128                 case TCP_LOGDUMPID:
2129                         INP_WUNLOCK(inp);
2130                         error =
2131                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2132                         if (error)
2133                                 break;
2134                         buf[sopt->sopt_valsize] = '\0';
2135                         INP_WLOCK_RECHECK(inp);
2136                         if (sopt->sopt_name == TCP_LOGDUMP) {
2137                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2138                                     M_WAITOK, true);
2139                                 INP_WUNLOCK(inp);
2140                         } else {
2141                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2142                                 /*
2143                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2144                                  * INP lock.
2145                                  */
2146                         }
2147                         break;
2148 #endif
2149
2150                 default:
2151                         INP_WUNLOCK(inp);
2152                         error = ENOPROTOOPT;
2153                         break;
2154                 }
2155                 break;
2156
2157         case SOPT_GET:
2158                 tp = intotcpcb(inp);
2159                 switch (sopt->sopt_name) {
2160 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2161                 case TCP_MD5SIG:
2162                         if (!TCPMD5_ENABLED()) {
2163                                 INP_WUNLOCK(inp);
2164                                 return (ENOPROTOOPT);
2165                         }
2166                         error = TCPMD5_PCBCTL(inp, sopt);
2167                         break;
2168 #endif
2169
2170                 case TCP_NODELAY:
2171                         optval = tp->t_flags & TF_NODELAY;
2172                         INP_WUNLOCK(inp);
2173                         error = sooptcopyout(sopt, &optval, sizeof optval);
2174                         break;
2175                 case TCP_MAXSEG:
2176                         optval = tp->t_maxseg;
2177                         INP_WUNLOCK(inp);
2178                         error = sooptcopyout(sopt, &optval, sizeof optval);
2179                         break;
2180                 case TCP_NOOPT:
2181                         optval = tp->t_flags & TF_NOOPT;
2182                         INP_WUNLOCK(inp);
2183                         error = sooptcopyout(sopt, &optval, sizeof optval);
2184                         break;
2185                 case TCP_NOPUSH:
2186                         optval = tp->t_flags & TF_NOPUSH;
2187                         INP_WUNLOCK(inp);
2188                         error = sooptcopyout(sopt, &optval, sizeof optval);
2189                         break;
2190                 case TCP_INFO:
2191                         tcp_fill_info(tp, &ti);
2192                         INP_WUNLOCK(inp);
2193                         error = sooptcopyout(sopt, &ti, sizeof ti);
2194                         break;
2195                 case TCP_CONGESTION:
2196                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2197                         INP_WUNLOCK(inp);
2198                         error = sooptcopyout(sopt, buf, len + 1);
2199                         break;
2200                 case TCP_KEEPIDLE:
2201                 case TCP_KEEPINTVL:
2202                 case TCP_KEEPINIT:
2203                 case TCP_KEEPCNT:
2204                         switch (sopt->sopt_name) {
2205                         case TCP_KEEPIDLE:
2206                                 ui = TP_KEEPIDLE(tp) / hz;
2207                                 break;
2208                         case TCP_KEEPINTVL:
2209                                 ui = TP_KEEPINTVL(tp) / hz;
2210                                 break;
2211                         case TCP_KEEPINIT:
2212                                 ui = TP_KEEPINIT(tp) / hz;
2213                                 break;
2214                         case TCP_KEEPCNT:
2215                                 ui = TP_KEEPCNT(tp);
2216                                 break;
2217                         }
2218                         INP_WUNLOCK(inp);
2219                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2220                         break;
2221 #ifdef TCPPCAP
2222                 case TCP_PCAP_OUT:
2223                 case TCP_PCAP_IN:
2224                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2225                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2226                         INP_WUNLOCK(inp);
2227                         error = sooptcopyout(sopt, &optval, sizeof optval);
2228                         break;
2229 #endif
2230                 case TCP_FASTOPEN:
2231                         optval = tp->t_flags & TF_FASTOPEN;
2232                         INP_WUNLOCK(inp);
2233                         error = sooptcopyout(sopt, &optval, sizeof optval);
2234                         break;
2235 #ifdef TCP_BLACKBOX
2236                 case TCP_LOG:
2237                         optval = tp->t_logstate;
2238                         INP_WUNLOCK(inp);
2239                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2240                         break;
2241                 case TCP_LOGBUF:
2242                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2243                         error = tcp_log_getlogbuf(sopt, tp);
2244                         break;
2245                 case TCP_LOGID:
2246                         len = tcp_log_get_id(tp, buf);
2247                         INP_WUNLOCK(inp);
2248                         error = sooptcopyout(sopt, buf, len + 1);
2249                         break;
2250                 case TCP_LOGDUMP:
2251                 case TCP_LOGDUMPID:
2252                         INP_WUNLOCK(inp);
2253                         error = EINVAL;
2254                         break;
2255 #endif
2256                 default:
2257                         INP_WUNLOCK(inp);
2258                         error = ENOPROTOOPT;
2259                         break;
2260                 }
2261                 break;
2262         }
2263         return (error);
2264 }
2265 #undef INP_WLOCK_RECHECK
2266 #undef INP_WLOCK_RECHECK_CLEANUP
2267
2268 /*
2269  * Attach TCP protocol to socket, allocating
2270  * internet protocol control block, tcp control block,
2271  * bufer space, and entering LISTEN state if to accept connections.
2272  */
2273 static int
2274 tcp_attach(struct socket *so)
2275 {
2276         struct tcpcb *tp;
2277         struct inpcb *inp;
2278         struct epoch_tracker et;
2279         int error;
2280
2281         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2282                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
2283                 if (error)
2284                         return (error);
2285         }
2286         so->so_rcv.sb_flags |= SB_AUTOSIZE;
2287         so->so_snd.sb_flags |= SB_AUTOSIZE;
2288         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2289         error = in_pcballoc(so, &V_tcbinfo);
2290         if (error) {
2291                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2292                 return (error);
2293         }
2294         inp = sotoinpcb(so);
2295 #ifdef INET6
2296         if (inp->inp_vflag & INP_IPV6PROTO) {
2297                 inp->inp_vflag |= INP_IPV6;
2298                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
2299                         inp->inp_vflag |= INP_IPV4;
2300                 inp->in6p_hops = -1;    /* use kernel default */
2301         }
2302         else
2303 #endif
2304         inp->inp_vflag |= INP_IPV4;
2305         tp = tcp_newtcpcb(inp);
2306         if (tp == NULL) {
2307                 in_pcbdetach(inp);
2308                 in_pcbfree(inp);
2309                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2310                 return (ENOBUFS);
2311         }
2312         tp->t_state = TCPS_CLOSED;
2313         INP_WUNLOCK(inp);
2314         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2315         TCPSTATES_INC(TCPS_CLOSED);
2316         return (0);
2317 }
2318
2319 /*
2320  * Initiate (or continue) disconnect.
2321  * If embryonic state, just send reset (once).
2322  * If in ``let data drain'' option and linger null, just drop.
2323  * Otherwise (hard), mark socket disconnecting and drop
2324  * current input data; switch states based on user close, and
2325  * send segment to peer (with FIN).
2326  */
2327 static void
2328 tcp_disconnect(struct tcpcb *tp)
2329 {
2330         struct inpcb *inp = tp->t_inpcb;
2331         struct socket *so = inp->inp_socket;
2332
2333         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2334         INP_WLOCK_ASSERT(inp);
2335
2336         /*
2337          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2338          * socket is still open.
2339          */
2340         if (tp->t_state < TCPS_ESTABLISHED &&
2341             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2342                 tp = tcp_close(tp);
2343                 KASSERT(tp != NULL,
2344                     ("tcp_disconnect: tcp_close() returned NULL"));
2345         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2346                 tp = tcp_drop(tp, 0);
2347                 KASSERT(tp != NULL,
2348                     ("tcp_disconnect: tcp_drop() returned NULL"));
2349         } else {
2350                 soisdisconnecting(so);
2351                 sbflush(&so->so_rcv);
2352                 tcp_usrclosed(tp);
2353                 if (!(inp->inp_flags & INP_DROPPED))
2354                         tp->t_fb->tfb_tcp_output(tp);
2355         }
2356 }
2357
2358 /*
2359  * User issued close, and wish to trail through shutdown states:
2360  * if never received SYN, just forget it.  If got a SYN from peer,
2361  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2362  * If already got a FIN from peer, then almost done; go to LAST_ACK
2363  * state.  In all other cases, have already sent FIN to peer (e.g.
2364  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2365  * for peer to send FIN or not respond to keep-alives, etc.
2366  * We can let the user exit from the close as soon as the FIN is acked.
2367  */
2368 static void
2369 tcp_usrclosed(struct tcpcb *tp)
2370 {
2371
2372         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2373         INP_WLOCK_ASSERT(tp->t_inpcb);
2374
2375         switch (tp->t_state) {
2376         case TCPS_LISTEN:
2377 #ifdef TCP_OFFLOAD
2378                 tcp_offload_listen_stop(tp);
2379 #endif
2380                 tcp_state_change(tp, TCPS_CLOSED);
2381                 /* FALLTHROUGH */
2382         case TCPS_CLOSED:
2383                 tp = tcp_close(tp);
2384                 /*
2385                  * tcp_close() should never return NULL here as the socket is
2386                  * still open.
2387                  */
2388                 KASSERT(tp != NULL,
2389                     ("tcp_usrclosed: tcp_close() returned NULL"));
2390                 break;
2391
2392         case TCPS_SYN_SENT:
2393         case TCPS_SYN_RECEIVED:
2394                 tp->t_flags |= TF_NEEDFIN;
2395                 break;
2396
2397         case TCPS_ESTABLISHED:
2398                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2399                 break;
2400
2401         case TCPS_CLOSE_WAIT:
2402                 tcp_state_change(tp, TCPS_LAST_ACK);
2403                 break;
2404         }
2405         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2406                 soisdisconnected(tp->t_inpcb->inp_socket);
2407                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2408                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2409                         int timeout;
2410
2411                         timeout = (tcp_fast_finwait2_recycle) ?
2412                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2413                         tcp_timer_activate(tp, TT_2MSL, timeout);
2414                 }
2415         }
2416 }
2417
2418 #ifdef DDB
2419 static void
2420 db_print_indent(int indent)
2421 {
2422         int i;
2423
2424         for (i = 0; i < indent; i++)
2425                 db_printf(" ");
2426 }
2427
2428 static void
2429 db_print_tstate(int t_state)
2430 {
2431
2432         switch (t_state) {
2433         case TCPS_CLOSED:
2434                 db_printf("TCPS_CLOSED");
2435                 return;
2436
2437         case TCPS_LISTEN:
2438                 db_printf("TCPS_LISTEN");
2439                 return;
2440
2441         case TCPS_SYN_SENT:
2442                 db_printf("TCPS_SYN_SENT");
2443                 return;
2444
2445         case TCPS_SYN_RECEIVED:
2446                 db_printf("TCPS_SYN_RECEIVED");
2447                 return;
2448
2449         case TCPS_ESTABLISHED:
2450                 db_printf("TCPS_ESTABLISHED");
2451                 return;
2452
2453         case TCPS_CLOSE_WAIT:
2454                 db_printf("TCPS_CLOSE_WAIT");
2455                 return;
2456
2457         case TCPS_FIN_WAIT_1:
2458                 db_printf("TCPS_FIN_WAIT_1");
2459                 return;
2460
2461         case TCPS_CLOSING:
2462                 db_printf("TCPS_CLOSING");
2463                 return;
2464
2465         case TCPS_LAST_ACK:
2466                 db_printf("TCPS_LAST_ACK");
2467                 return;
2468
2469         case TCPS_FIN_WAIT_2:
2470                 db_printf("TCPS_FIN_WAIT_2");
2471                 return;
2472
2473         case TCPS_TIME_WAIT:
2474                 db_printf("TCPS_TIME_WAIT");
2475                 return;
2476
2477         default:
2478                 db_printf("unknown");
2479                 return;
2480         }
2481 }
2482
2483 static void
2484 db_print_tflags(u_int t_flags)
2485 {
2486         int comma;
2487
2488         comma = 0;
2489         if (t_flags & TF_ACKNOW) {
2490                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2491                 comma = 1;
2492         }
2493         if (t_flags & TF_DELACK) {
2494                 db_printf("%sTF_DELACK", comma ? ", " : "");
2495                 comma = 1;
2496         }
2497         if (t_flags & TF_NODELAY) {
2498                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2499                 comma = 1;
2500         }
2501         if (t_flags & TF_NOOPT) {
2502                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2503                 comma = 1;
2504         }
2505         if (t_flags & TF_SENTFIN) {
2506                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2507                 comma = 1;
2508         }
2509         if (t_flags & TF_REQ_SCALE) {
2510                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2511                 comma = 1;
2512         }
2513         if (t_flags & TF_RCVD_SCALE) {
2514                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2515                 comma = 1;
2516         }
2517         if (t_flags & TF_REQ_TSTMP) {
2518                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2519                 comma = 1;
2520         }
2521         if (t_flags & TF_RCVD_TSTMP) {
2522                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2523                 comma = 1;
2524         }
2525         if (t_flags & TF_SACK_PERMIT) {
2526                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2527                 comma = 1;
2528         }
2529         if (t_flags & TF_NEEDSYN) {
2530                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2531                 comma = 1;
2532         }
2533         if (t_flags & TF_NEEDFIN) {
2534                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2535                 comma = 1;
2536         }
2537         if (t_flags & TF_NOPUSH) {
2538                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2539                 comma = 1;
2540         }
2541         if (t_flags & TF_MORETOCOME) {
2542                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2543                 comma = 1;
2544         }
2545         if (t_flags & TF_LQ_OVERFLOW) {
2546                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2547                 comma = 1;
2548         }
2549         if (t_flags & TF_LASTIDLE) {
2550                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2551                 comma = 1;
2552         }
2553         if (t_flags & TF_RXWIN0SENT) {
2554                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2555                 comma = 1;
2556         }
2557         if (t_flags & TF_FASTRECOVERY) {
2558                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2559                 comma = 1;
2560         }
2561         if (t_flags & TF_CONGRECOVERY) {
2562                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2563                 comma = 1;
2564         }
2565         if (t_flags & TF_WASFRECOVERY) {
2566                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2567                 comma = 1;
2568         }
2569         if (t_flags & TF_SIGNATURE) {
2570                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2571                 comma = 1;
2572         }
2573         if (t_flags & TF_FORCEDATA) {
2574                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2575                 comma = 1;
2576         }
2577         if (t_flags & TF_TSO) {
2578                 db_printf("%sTF_TSO", comma ? ", " : "");
2579                 comma = 1;
2580         }
2581         if (t_flags & TF_ECN_PERMIT) {
2582                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2583                 comma = 1;
2584         }
2585         if (t_flags & TF_FASTOPEN) {
2586                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2587                 comma = 1;
2588         }
2589 }
2590
2591 static void
2592 db_print_toobflags(char t_oobflags)
2593 {
2594         int comma;
2595
2596         comma = 0;
2597         if (t_oobflags & TCPOOB_HAVEDATA) {
2598                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2599                 comma = 1;
2600         }
2601         if (t_oobflags & TCPOOB_HADDATA) {
2602                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2603                 comma = 1;
2604         }
2605 }
2606
2607 static void
2608 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2609 {
2610
2611         db_print_indent(indent);
2612         db_printf("%s at %p\n", name, tp);
2613
2614         indent += 2;
2615
2616         db_print_indent(indent);
2617         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2618            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2619
2620         db_print_indent(indent);
2621         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2622             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2623
2624         db_print_indent(indent);
2625         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2626             &tp->t_timers->tt_delack, tp->t_inpcb);
2627
2628         db_print_indent(indent);
2629         db_printf("t_state: %d (", tp->t_state);
2630         db_print_tstate(tp->t_state);
2631         db_printf(")\n");
2632
2633         db_print_indent(indent);
2634         db_printf("t_flags: 0x%x (", tp->t_flags);
2635         db_print_tflags(tp->t_flags);
2636         db_printf(")\n");
2637
2638         db_print_indent(indent);
2639         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2640             tp->snd_una, tp->snd_max, tp->snd_nxt);
2641
2642         db_print_indent(indent);
2643         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2644            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2645
2646         db_print_indent(indent);
2647         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2648             tp->iss, tp->irs, tp->rcv_nxt);
2649
2650         db_print_indent(indent);
2651         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2652             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2653
2654         db_print_indent(indent);
2655         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2656            tp->snd_wnd, tp->snd_cwnd);
2657
2658         db_print_indent(indent);
2659         db_printf("snd_ssthresh: %u   snd_recover: "
2660             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2661
2662         db_print_indent(indent);
2663         db_printf("t_rcvtime: %u   t_startime: %u\n",
2664             tp->t_rcvtime, tp->t_starttime);
2665
2666         db_print_indent(indent);
2667         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2668             tp->t_rtttime, tp->t_rtseq);
2669
2670         db_print_indent(indent);
2671         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2672             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2673
2674         db_print_indent(indent);
2675         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2676             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2677             tp->t_rttbest);
2678
2679         db_print_indent(indent);
2680         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2681             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2682
2683         db_print_indent(indent);
2684         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2685         db_print_toobflags(tp->t_oobflags);
2686         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2687
2688         db_print_indent(indent);
2689         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2690             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2691
2692         db_print_indent(indent);
2693         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2694             tp->ts_recent, tp->ts_recent_age);
2695
2696         db_print_indent(indent);
2697         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2698             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2699
2700         db_print_indent(indent);
2701         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2702             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2703             tp->snd_recover_prev, tp->t_badrxtwin);
2704
2705         db_print_indent(indent);
2706         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2707             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2708
2709         db_print_indent(indent);
2710         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2711             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2712
2713         /* Skip sackblks, sackhint. */
2714
2715         db_print_indent(indent);
2716         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2717             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2718 }
2719
2720 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2721 {
2722         struct tcpcb *tp;
2723
2724         if (!have_addr) {
2725                 db_printf("usage: show tcpcb <addr>\n");
2726                 return;
2727         }
2728         tp = (struct tcpcb *)addr;
2729
2730         db_print_tcpcb(tp, "tcpcb", 0);
2731 }
2732 #endif