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 *sin6;
 348
 349         sin6 = (struct sockaddr_in6 *)nam;
 350         if (nam->sa_len != sizeof (*sin6))
 351                 return (EINVAL);
 352         /*
 353          * Must check for multicast addresses and disallow binding
 354          * to them.
 355          */
 356         if (sin6->sin6_family == AF_INET6 &&
 357             IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 358                 return (EAFNOSUPPORT);
 359
 360         TCPDEBUG0;
 361         inp = sotoinpcb(so);
 362         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 363         INP_WLOCK(inp);
 364         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 365                 error = EINVAL;
 366                 goto out;
 367         }
 368         tp = intotcpcb(inp);
 369         TCPDEBUG1();
 370         INP_HASH_WLOCK(&V_tcbinfo);
 371         inp->inp_vflag &= ~INP_IPV4;
 372         inp->inp_vflag |= INP_IPV6;
 373 #ifdef INET
 374         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 375                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
 376                         inp->inp_vflag |= INP_IPV4;
 377                 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 378                         struct sockaddr_in sin;
 379
 380                         in6_sin6_2_sin(&sin, sin6);
 381                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 382                                 error = EAFNOSUPPORT;
 383                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 384                                 goto out;
 385                         }
 386                         inp->inp_vflag |= INP_IPV4;
 387                         inp->inp_vflag &= ~INP_IPV6;
 388                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 389                             td->td_ucred);
 390                         INP_HASH_WUNLOCK(&V_tcbinfo);
 391                         goto out;
 392                 }
 393         }
 394 #endif
 395         error = in6_pcbbind(inp, nam, td->td_ucred);
 396         INP_HASH_WUNLOCK(&V_tcbinfo);
 397 out:
 398         TCPDEBUG2(PRU_BIND);
 399         TCP_PROBE2(debug__user, tp, PRU_BIND);
 400         INP_WUNLOCK(inp);
 401         return (error);
 402 }
 403 #endif /* INET6 */
 404
 405 #ifdef INET
 406 /*
 407  * Prepare to accept connections.
 408  */
 409 static int
 410 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 411 {
 412         int error = 0;
 413         struct inpcb *inp;
 414         struct tcpcb *tp = NULL;
 415
 416         TCPDEBUG0;
 417         inp = sotoinpcb(so);
 418         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 419         INP_WLOCK(inp);
 420         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 421                 error = EINVAL;
 422                 goto out;
 423         }
 424         tp = intotcpcb(inp);
 425         TCPDEBUG1();
 426         SOCK_LOCK(so);
 427         error = solisten_proto_check(so);
 428         INP_HASH_WLOCK(&V_tcbinfo);
 429         if (error == 0 && inp->inp_lport == 0)
 430                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 431         INP_HASH_WUNLOCK(&V_tcbinfo);
 432         if (error == 0) {
 433                 tcp_state_change(tp, TCPS_LISTEN);
 434                 solisten_proto(so, backlog);
 435 #ifdef TCP_OFFLOAD
 436                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 437                         tcp_offload_listen_start(tp);
 438 #endif
 439         }
 440         SOCK_UNLOCK(so);
 441
 442         if (IS_FASTOPEN(tp->t_flags))
 443                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 444
 445 out:
 446         TCPDEBUG2(PRU_LISTEN);
 447         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 448         INP_WUNLOCK(inp);
 449         return (error);
 450 }
 451 #endif /* INET */
 452
 453 #ifdef INET6
 454 static int
 455 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 456 {
 457         int error = 0;
 458         struct inpcb *inp;
 459         struct tcpcb *tp = NULL;
 460
 461         TCPDEBUG0;
 462         inp = sotoinpcb(so);
 463         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 464         INP_WLOCK(inp);
 465         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 466                 error = EINVAL;
 467                 goto out;
 468         }
 469         tp = intotcpcb(inp);
 470         TCPDEBUG1();
 471         SOCK_LOCK(so);
 472         error = solisten_proto_check(so);
 473         INP_HASH_WLOCK(&V_tcbinfo);
 474         if (error == 0 && inp->inp_lport == 0) {
 475                 inp->inp_vflag &= ~INP_IPV4;
 476                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 477                         inp->inp_vflag |= INP_IPV4;
 478                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 479         }
 480         INP_HASH_WUNLOCK(&V_tcbinfo);
 481         if (error == 0) {
 482                 tcp_state_change(tp, TCPS_LISTEN);
 483                 solisten_proto(so, backlog);
 484 #ifdef TCP_OFFLOAD
 485                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 486                         tcp_offload_listen_start(tp);
 487 #endif
 488         }
 489         SOCK_UNLOCK(so);
 490
 491         if (IS_FASTOPEN(tp->t_flags))
 492                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 493
 494 out:
 495         TCPDEBUG2(PRU_LISTEN);
 496         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 497         INP_WUNLOCK(inp);
 498         return (error);
 499 }
 500 #endif /* INET6 */
 501
 502 #ifdef INET
 503 /*
 504  * Initiate connection to peer.
 505  * Create a template for use in transmissions on this connection.
 506  * Enter SYN_SENT state, and mark socket as connecting.
 507  * Start keep-alive timer, and seed output sequence space.
 508  * Send initial segment on connection.
 509  */
 510 static int
 511 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 512 {
 513         int error = 0;
 514         struct inpcb *inp;
 515         struct tcpcb *tp = NULL;
 516         struct sockaddr_in *sinp;
 517
 518         sinp = (struct sockaddr_in *)nam;
 519         if (nam->sa_len != sizeof (*sinp))
 520                 return (EINVAL);
 521         /*
 522          * Must disallow TCP ``connections'' to multicast addresses.
 523          */
 524         if (sinp->sin_family == AF_INET
 525             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 526                 return (EAFNOSUPPORT);
 527         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 528                 return (error);
 529
 530         TCPDEBUG0;
 531         inp = sotoinpcb(so);
 532         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 533         INP_WLOCK(inp);
 534         if (inp->inp_flags & INP_TIMEWAIT) {
 535                 error = EADDRINUSE;
 536                 goto out;
 537         }
 538         if (inp->inp_flags & INP_DROPPED) {
 539                 error = ECONNREFUSED;
 540                 goto out;
 541         }
 542         tp = intotcpcb(inp);
 543         TCPDEBUG1();
 544         if ((error = tcp_connect(tp, nam, td)) != 0)
 545                 goto out;
 546 #ifdef TCP_OFFLOAD
 547         if (registered_toedevs > 0 &&
 548             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 549             (error = tcp_offload_connect(so, nam)) == 0)
 550                 goto out;
 551 #endif
 552         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 553         error = tp->t_fb->tfb_tcp_output(tp);
 554 out:
 555         TCPDEBUG2(PRU_CONNECT);
 556         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 557         INP_WUNLOCK(inp);
 558         return (error);
 559 }
 560 #endif /* INET */
 561
 562 #ifdef INET6
 563 static int
 564 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 565 {
 566         int error = 0;
 567         struct inpcb *inp;
 568         struct tcpcb *tp = NULL;
 569         struct sockaddr_in6 *sin6;
 570
 571         TCPDEBUG0;
 572
 573         sin6 = (struct sockaddr_in6 *)nam;
 574         if (nam->sa_len != sizeof (*sin6))
 575                 return (EINVAL);
 576         /*
 577          * Must disallow TCP ``connections'' to multicast addresses.
 578          */
 579         if (sin6->sin6_family == AF_INET6
 580             && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 581                 return (EAFNOSUPPORT);
 582
 583         inp = sotoinpcb(so);
 584         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 585         INP_WLOCK(inp);
 586         if (inp->inp_flags & INP_TIMEWAIT) {
 587                 error = EADDRINUSE;
 588                 goto out;
 589         }
 590         if (inp->inp_flags & INP_DROPPED) {
 591                 error = ECONNREFUSED;
 592                 goto out;
 593         }
 594         tp = intotcpcb(inp);
 595         TCPDEBUG1();
 596 #ifdef INET
 597         /*
 598          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 599          * therefore probably require the hash lock, which isn't held here.
 600          * Is this a significant problem?
 601          */
 602         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 603                 struct sockaddr_in sin;
 604
 605                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 606                         error = EINVAL;
 607                         goto out;
 608                 }
 609                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 610                         error = EAFNOSUPPORT;
 611                         goto out;
 612                 }
 613
 614                 in6_sin6_2_sin(&sin, sin6);
 615                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 616                         error = EAFNOSUPPORT;
 617                         goto out;
 618                 }
 619                 inp->inp_vflag |= INP_IPV4;
 620                 inp->inp_vflag &= ~INP_IPV6;
 621                 if ((error = prison_remote_ip4(td->td_ucred,
 622                     &sin.sin_addr)) != 0)
 623                         goto out;
 624                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 625                         goto out;
 626 #ifdef TCP_OFFLOAD
 627                 if (registered_toedevs > 0 &&
 628                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 629                     (error = tcp_offload_connect(so, nam)) == 0)
 630                         goto out;
 631 #endif
 632                 error = tp->t_fb->tfb_tcp_output(tp);
 633                 goto out;
 634         } else {
 635                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 636                         error = EAFNOSUPPORT;
 637                         goto out;
 638                 }
 639         }
 640 #endif
 641         inp->inp_vflag &= ~INP_IPV4;
 642         inp->inp_vflag |= INP_IPV6;
 643         inp->inp_inc.inc_flags |= INC_ISIPV6;
 644         if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
 645                 goto out;
 646         if ((error = tcp6_connect(tp, nam, td)) != 0)
 647                 goto out;
 648 #ifdef TCP_OFFLOAD
 649         if (registered_toedevs > 0 &&
 650             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 651             (error = tcp_offload_connect(so, nam)) == 0)
 652                 goto out;
 653 #endif
 654         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 655         error = tp->t_fb->tfb_tcp_output(tp);
 656
 657 out:
 658         TCPDEBUG2(PRU_CONNECT);
 659         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 660         INP_WUNLOCK(inp);
 661         return (error);
 662 }
 663 #endif /* INET6 */
 664
 665 /*
 666  * Initiate disconnect from peer.
 667  * If connection never passed embryonic stage, just drop;
 668  * else if don't need to let data drain, then can just drop anyways,
 669  * else have to begin TCP shutdown process: mark socket disconnecting,
 670  * drain unread data, state switch to reflect user close, and
 671  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 672  * when peer sends FIN and acks ours.
 673  *
 674  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 675  */
 676 static int
 677 tcp_usr_disconnect(struct socket *so)
 678 {
 679         struct inpcb *inp;
 680         struct tcpcb *tp = NULL;
 681         struct epoch_tracker et;
 682         int error = 0;
 683
 684         TCPDEBUG0;
 685         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 686         inp = sotoinpcb(so);
 687         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 688         INP_WLOCK(inp);
 689         if (inp->inp_flags & INP_TIMEWAIT)
 690                 goto out;
 691         if (inp->inp_flags & INP_DROPPED) {
 692                 error = ECONNRESET;
 693                 goto out;
 694         }
 695         tp = intotcpcb(inp);
 696         TCPDEBUG1();
 697         tcp_disconnect(tp);
 698 out:
 699         TCPDEBUG2(PRU_DISCONNECT);
 700         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 701         INP_WUNLOCK(inp);
 702         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 703         return (error);
 704 }
 705
 706 #ifdef INET
 707 /*
 708  * Accept a connection.  Essentially all the work is done at higher levels;
 709  * just return the address of the peer, storing through addr.
 710  */
 711 static int
 712 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 713 {
 714         int error = 0;
 715         struct inpcb *inp = NULL;
 716         struct tcpcb *tp = NULL;
 717         struct in_addr addr;
 718         in_port_t port = 0;
 719         TCPDEBUG0;
 720
 721         if (so->so_state & SS_ISDISCONNECTED)
 722                 return (ECONNABORTED);
 723
 724         inp = sotoinpcb(so);
 725         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 726         INP_WLOCK(inp);
 727         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 728                 error = ECONNABORTED;
 729                 goto out;
 730         }
 731         tp = intotcpcb(inp);
 732         TCPDEBUG1();
 733
 734         /*
 735          * We inline in_getpeeraddr and COMMON_END here, so that we can
 736          * copy the data of interest and defer the malloc until after we
 737          * release the lock.
 738          */
 739         port = inp->inp_fport;
 740         addr = inp->inp_faddr;
 741
 742 out:
 743         TCPDEBUG2(PRU_ACCEPT);
 744         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 745         INP_WUNLOCK(inp);
 746         if (error == 0)
 747                 *nam = in_sockaddr(port, &addr);
 748         return error;
 749 }
 750 #endif /* INET */
 751
 752 #ifdef INET6
 753 static int
 754 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 755 {
 756         struct inpcb *inp = NULL;
 757         int error = 0;
 758         struct tcpcb *tp = NULL;
 759         struct in_addr addr;
 760         struct in6_addr addr6;
 761         struct epoch_tracker et;
 762         in_port_t port = 0;
 763         int v4 = 0;
 764         TCPDEBUG0;
 765
 766         if (so->so_state & SS_ISDISCONNECTED)
 767                 return (ECONNABORTED);
 768
 769         inp = sotoinpcb(so);
 770         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 771         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 772         INP_WLOCK(inp);
 773         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 774                 error = ECONNABORTED;
 775                 goto out;
 776         }
 777         tp = intotcpcb(inp);
 778         TCPDEBUG1();
 779
 780         /*
 781          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 782          * copy the data of interest and defer the malloc until after we
 783          * release the lock.
 784          */
 785         if (inp->inp_vflag & INP_IPV4) {
 786                 v4 = 1;
 787                 port = inp->inp_fport;
 788                 addr = inp->inp_faddr;
 789         } else {
 790                 port = inp->inp_fport;
 791                 addr6 = inp->in6p_faddr;
 792         }
 793
 794 out:
 795         TCPDEBUG2(PRU_ACCEPT);
 796         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 797         INP_WUNLOCK(inp);
 798         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 799         if (error == 0) {
 800                 if (v4)
 801                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 802                 else
 803                         *nam = in6_sockaddr(port, &addr6);
 804         }
 805         return error;
 806 }
 807 #endif /* INET6 */
 808
 809 /*
 810  * Mark the connection as being incapable of further output.
 811  */
 812 static int
 813 tcp_usr_shutdown(struct socket *so)
 814 {
 815         int error = 0;
 816         struct inpcb *inp;
 817         struct tcpcb *tp = NULL;
 818         struct epoch_tracker et;
 819
 820         TCPDEBUG0;
 821         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 822         inp = sotoinpcb(so);
 823         KASSERT(inp != NULL, ("inp == NULL"));
 824         INP_WLOCK(inp);
 825         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 826                 error = ECONNRESET;
 827                 goto out;
 828         }
 829         tp = intotcpcb(inp);
 830         TCPDEBUG1();
 831         socantsendmore(so);
 832         tcp_usrclosed(tp);
 833         if (!(inp->inp_flags & INP_DROPPED))
 834                 error = tp->t_fb->tfb_tcp_output(tp);
 835
 836 out:
 837         TCPDEBUG2(PRU_SHUTDOWN);
 838         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 839         INP_WUNLOCK(inp);
 840         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 841
 842         return (error);
 843 }
 844
 845 /*
 846  * After a receive, possibly send window update to peer.
 847  */
 848 static int
 849 tcp_usr_rcvd(struct socket *so, int flags)
 850 {
 851         struct inpcb *inp;
 852         struct tcpcb *tp = NULL;
 853         int error = 0;
 854
 855         TCPDEBUG0;
 856         inp = sotoinpcb(so);
 857         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 858         INP_WLOCK(inp);
 859         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 860                 error = ECONNRESET;
 861                 goto out;
 862         }
 863         tp = intotcpcb(inp);
 864         TCPDEBUG1();
 865         /*
 866          * For passively-created TFO connections, don't attempt a window
 867          * update while still in SYN_RECEIVED as this may trigger an early
 868          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 869          * application response data, or failing that, when the DELACK timer
 870          * expires.
 871          */
 872         if (IS_FASTOPEN(tp->t_flags) &&
 873             (tp->t_state == TCPS_SYN_RECEIVED))
 874                 goto out;
 875 #ifdef TCP_OFFLOAD
 876         if (tp->t_flags & TF_TOE)
 877                 tcp_offload_rcvd(tp);
 878         else
 879 #endif
 880         tp->t_fb->tfb_tcp_output(tp);
 881
 882 out:
 883         TCPDEBUG2(PRU_RCVD);
 884         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 885         INP_WUNLOCK(inp);
 886         return (error);
 887 }
 888
 889 /*
 890  * Do a send by putting data in output queue and updating urgent
 891  * marker if URG set.  Possibly send more data.  Unlike the other
 892  * pru_*() routines, the mbuf chains are our responsibility.  We
 893  * must either enqueue them or free them.  The other pru_* routines
 894  * generally are caller-frees.
 895  */
 896 static int
 897 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 898     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 899 {
 900         int error = 0;
 901         struct inpcb *inp;
 902         struct tcpcb *tp = NULL;
 903         struct epoch_tracker net_et;
 904 #ifdef INET
 905 #ifdef INET6
 906         struct sockaddr_in sin;
 907 #endif
 908         struct sockaddr_in *sinp;
 909 #endif
 910 #ifdef INET6
 911         int isipv6;
 912 #endif
 913         TCPDEBUG0;
 914
 915         /*
 916          * We require the pcbinfo lock if we will close the socket as part of
 917          * this call.
 918          */
 919         if (flags & PRUS_EOF)
 920                 INP_INFO_RLOCK_ET(&V_tcbinfo, net_et);
 921         inp = sotoinpcb(so);
 922         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 923         INP_WLOCK(inp);
 924         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 925                 if (control)
 926                         m_freem(control);
 927                 /*
 928                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 929                  * for freeing memory.
 930                  */
 931                 if (m && (flags & PRUS_NOTREADY) == 0)
 932                         m_freem(m);
 933                 error = ECONNRESET;
 934                 goto out;
 935         }
 936         tp = intotcpcb(inp);
 937         TCPDEBUG1();
 938         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
 939                 switch (nam->sa_family) {
 940 #ifdef INET
 941                 case AF_INET:
 942                         sinp = (struct sockaddr_in *)nam;
 943                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
 944                                 if (m)
 945                                         m_freem(m);
 946                                 error = EINVAL;
 947                                 goto out;
 948                         }
 949                         if ((inp->inp_vflag & INP_IPV6) != 0) {
 950                                 if (m)
 951                                         m_freem(m);
 952                                 error = EAFNOSUPPORT;
 953                                 goto out;
 954                         }
 955                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 956                                 if (m)
 957                                         m_freem(m);
 958                                 error = EAFNOSUPPORT;
 959                                 goto out;
 960                         }
 961                         if ((error = prison_remote_ip4(td->td_ucred,
 962                             &sinp->sin_addr))) {
 963                                 if (m)
 964                                         m_freem(m);
 965                                 goto out;
 966                         }
 967 #ifdef INET6
 968                         isipv6 = 0;
 969 #endif
 970                         break;
 971 #endif /* INET */
 972 #ifdef INET6
 973                 case AF_INET6:
 974                 {
 975                         struct sockaddr_in6 *sin6;
 976
 977                         sin6 = (struct sockaddr_in6 *)nam;
 978                         if (sin6->sin6_len != sizeof(*sin6)) {
 979                                 if (m)
 980                                         m_freem(m);
 981                                 error = EINVAL;
 982                                 goto out;
 983                         }
 984                         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
 985                                 if (m)
 986                                         m_freem(m);
 987                                 error = EAFNOSUPPORT;
 988                                 goto out;
 989                         }
 990                         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 991 #ifdef INET
 992                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 993                                         error = EINVAL;
 994                                         if (m)
 995                                                 m_freem(m);
 996                                         goto out;
 997                                 }
 998                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 999                                         error = EAFNOSUPPORT;
1000                                         if (m)
1001                                                 m_freem(m);
1002                                         goto out;
1003                                 }
1004                                 inp->inp_vflag &= ~INP_IPV6;
1005                                 sinp = &sin;
1006                                 in6_sin6_2_sin(sinp, sin6);
1007                                 if (IN_MULTICAST(
1008                                     ntohl(sinp->sin_addr.s_addr))) {
1009                                         error = EAFNOSUPPORT;
1010                                         if (m)
1011                                                 m_freem(m);
1012                                         goto out;
1013                                 }
1014                                 if ((error = prison_remote_ip4(td->td_ucred,
1015                                     &sinp->sin_addr))) {
1016                                         if (m)
1017                                                 m_freem(m);
1018                                         goto out;
1019                                 }
1020                                 isipv6 = 0;
1021 #else /* !INET */
1022                                 error = EAFNOSUPPORT;
1023                                 if (m)
1024                                         m_freem(m);
1025                                 goto out;
1026 #endif /* INET */
1027                         } else {
1028                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1029                                         if (m)
1030                                                 m_freem(m);
1031                                         error = EAFNOSUPPORT;
1032                                         goto out;
1033                                 }
1034                                 inp->inp_vflag &= ~INP_IPV4;
1035                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1036                                 if ((error = prison_remote_ip6(td->td_ucred,
1037                                     &sin6->sin6_addr))) {
1038                                         if (m)
1039                                                 m_freem(m);
1040                                         goto out;
1041                                 }
1042                                 isipv6 = 1;
1043                         }
1044                         break;
1045                 }
1046 #endif /* INET6 */
1047                 default:
1048                         if (m)
1049                                 m_freem(m);
1050                         error = EAFNOSUPPORT;
1051                         goto out;
1052                 }
1053         }
1054         if (control) {
1055                 /* TCP doesn't do control messages (rights, creds, etc) */
1056                 if (control->m_len) {
1057                         m_freem(control);
1058                         if (m)
1059                                 m_freem(m);
1060                         error = EINVAL;
1061                         goto out;
1062                 }
1063                 m_freem(control);       /* empty control, just free it */
1064         }
1065         if (!(flags & PRUS_OOB)) {
1066                 sbappendstream(&so->so_snd, m, flags);
1067                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1068                         /*
1069                          * Do implied connect if not yet connected,
1070                          * initialize window to default value, and
1071                          * initialize maxseg using peer's cached MSS.
1072                          */
1073 #ifdef INET6
1074                         if (isipv6)
1075                                 error = tcp6_connect(tp, nam, td);
1076 #endif /* INET6 */
1077 #if defined(INET6) && defined(INET)
1078                         else
1079 #endif
1080 #ifdef INET
1081                                 error = tcp_connect(tp,
1082                                     (struct sockaddr *)sinp, td);
1083 #endif
1084                         if (error)
1085                                 goto out;
1086                         if (IS_FASTOPEN(tp->t_flags))
1087                                 tcp_fastopen_connect(tp);
1088                         else {
1089                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1090                                 tcp_mss(tp, -1);
1091                         }
1092                 }
1093                 if (flags & PRUS_EOF) {
1094                         /*
1095                          * Close the send side of the connection after
1096                          * the data is sent.
1097                          */
1098                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1099                         socantsendmore(so);
1100                         tcp_usrclosed(tp);
1101                 }
1102                 if (!(inp->inp_flags & INP_DROPPED) &&
1103                     !(flags & PRUS_NOTREADY)) {
1104                         if (flags & PRUS_MORETOCOME)
1105                                 tp->t_flags |= TF_MORETOCOME;
1106                         error = tp->t_fb->tfb_tcp_output(tp);
1107                         if (flags & PRUS_MORETOCOME)
1108                                 tp->t_flags &= ~TF_MORETOCOME;
1109                 }
1110         } else {
1111                 /*
1112                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1113                  */
1114                 SOCKBUF_LOCK(&so->so_snd);
1115                 if (sbspace(&so->so_snd) < -512) {
1116                         SOCKBUF_UNLOCK(&so->so_snd);
1117                         m_freem(m);
1118                         error = ENOBUFS;
1119                         goto out;
1120                 }
1121                 /*
1122                  * According to RFC961 (Assigned Protocols),
1123                  * the urgent pointer points to the last octet
1124                  * of urgent data.  We continue, however,
1125                  * to consider it to indicate the first octet
1126                  * of data past the urgent section.
1127                  * Otherwise, snd_up should be one lower.
1128                  */
1129                 sbappendstream_locked(&so->so_snd, m, flags);
1130                 SOCKBUF_UNLOCK(&so->so_snd);
1131                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1132                         /*
1133                          * Do implied connect if not yet connected,
1134                          * initialize window to default value, and
1135                          * initialize maxseg using peer's cached MSS.
1136                          */
1137
1138                         /*
1139                          * Not going to contemplate SYN|URG
1140                          */
1141                         if (IS_FASTOPEN(tp->t_flags))
1142                                 tp->t_flags &= ~TF_FASTOPEN;
1143 #ifdef INET6
1144                         if (isipv6)
1145                                 error = tcp6_connect(tp, nam, td);
1146 #endif /* INET6 */
1147 #if defined(INET6) && defined(INET)
1148                         else
1149 #endif
1150 #ifdef INET
1151                                 error = tcp_connect(tp,
1152                                     (struct sockaddr *)sinp, td);
1153 #endif
1154                         if (error)
1155                                 goto out;
1156                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1157                         tcp_mss(tp, -1);
1158                 }
1159                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1160                 if (!(flags & PRUS_NOTREADY)) {
1161                         tp->t_flags |= TF_FORCEDATA;
1162                         error = tp->t_fb->tfb_tcp_output(tp);
1163                         tp->t_flags &= ~TF_FORCEDATA;
1164                 }
1165         }
1166         TCP_LOG_EVENT(tp, NULL,
1167             &inp->inp_socket->so_rcv,
1168             &inp->inp_socket->so_snd,
1169             TCP_LOG_USERSEND, error,
1170             0, NULL, false);
1171 out:
1172         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1173                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1174         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1175                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1176         INP_WUNLOCK(inp);
1177         if (flags & PRUS_EOF)
1178                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, net_et);
1179         return (error);
1180 }
1181
1182 static int
1183 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1184 {
1185         struct inpcb *inp;
1186         struct tcpcb *tp;
1187         int error;
1188
1189         inp = sotoinpcb(so);
1190         INP_WLOCK(inp);
1191         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1192                 INP_WUNLOCK(inp);
1193                 mb_free_notready(m, count);
1194                 return (ECONNRESET);
1195         }
1196         tp = intotcpcb(inp);
1197
1198         SOCKBUF_LOCK(&so->so_snd);
1199         error = sbready(&so->so_snd, m, count);
1200         SOCKBUF_UNLOCK(&so->so_snd);
1201         if (error == 0)
1202                 error = tp->t_fb->tfb_tcp_output(tp);
1203         INP_WUNLOCK(inp);
1204
1205         return (error);
1206 }
1207
1208 /*
1209  * Abort the TCP.  Drop the connection abruptly.
1210  */
1211 static void
1212 tcp_usr_abort(struct socket *so)
1213 {
1214         struct inpcb *inp;
1215         struct tcpcb *tp = NULL;
1216         struct epoch_tracker et;
1217         TCPDEBUG0;
1218
1219         inp = sotoinpcb(so);
1220         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1221
1222         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1223         INP_WLOCK(inp);
1224         KASSERT(inp->inp_socket != NULL,
1225             ("tcp_usr_abort: inp_socket == NULL"));
1226
1227         /*
1228          * If we still have full TCP state, and we're not dropped, drop.
1229          */
1230         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1231             !(inp->inp_flags & INP_DROPPED)) {
1232                 tp = intotcpcb(inp);
1233                 TCPDEBUG1();
1234                 tp = tcp_drop(tp, ECONNABORTED);
1235                 if (tp == NULL)
1236                         goto dropped;
1237                 TCPDEBUG2(PRU_ABORT);
1238                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1239         }
1240         if (!(inp->inp_flags & INP_DROPPED)) {
1241                 SOCK_LOCK(so);
1242                 so->so_state |= SS_PROTOREF;
1243                 SOCK_UNLOCK(so);
1244                 inp->inp_flags |= INP_SOCKREF;
1245         }
1246         INP_WUNLOCK(inp);
1247 dropped:
1248         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1249 }
1250
1251 /*
1252  * TCP socket is closed.  Start friendly disconnect.
1253  */
1254 static void
1255 tcp_usr_close(struct socket *so)
1256 {
1257         struct inpcb *inp;
1258         struct tcpcb *tp = NULL;
1259         struct epoch_tracker et;
1260         TCPDEBUG0;
1261
1262         inp = sotoinpcb(so);
1263         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1264
1265         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1266         INP_WLOCK(inp);
1267         KASSERT(inp->inp_socket != NULL,
1268             ("tcp_usr_close: inp_socket == NULL"));
1269
1270         /*
1271          * If we still have full TCP state, and we're not dropped, initiate
1272          * a disconnect.
1273          */
1274         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1275             !(inp->inp_flags & INP_DROPPED)) {
1276                 tp = intotcpcb(inp);
1277                 TCPDEBUG1();
1278                 tcp_disconnect(tp);
1279                 TCPDEBUG2(PRU_CLOSE);
1280                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1281         }
1282         if (!(inp->inp_flags & INP_DROPPED)) {
1283                 SOCK_LOCK(so);
1284                 so->so_state |= SS_PROTOREF;
1285                 SOCK_UNLOCK(so);
1286                 inp->inp_flags |= INP_SOCKREF;
1287         }
1288         INP_WUNLOCK(inp);
1289         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1290 }
1291
1292 /*
1293  * Receive out-of-band data.
1294  */
1295 static int
1296 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1297 {
1298         int error = 0;
1299         struct inpcb *inp;
1300         struct tcpcb *tp = NULL;
1301
1302         TCPDEBUG0;
1303         inp = sotoinpcb(so);
1304         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1305         INP_WLOCK(inp);
1306         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1307                 error = ECONNRESET;
1308                 goto out;
1309         }
1310         tp = intotcpcb(inp);
1311         TCPDEBUG1();
1312         if ((so->so_oobmark == 0 &&
1313              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1314             so->so_options & SO_OOBINLINE ||
1315             tp->t_oobflags & TCPOOB_HADDATA) {
1316                 error = EINVAL;
1317                 goto out;
1318         }
1319         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1320                 error = EWOULDBLOCK;
1321                 goto out;
1322         }
1323         m->m_len = 1;
1324         *mtod(m, caddr_t) = tp->t_iobc;
1325         if ((flags & MSG_PEEK) == 0)
1326                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1327
1328 out:
1329         TCPDEBUG2(PRU_RCVOOB);
1330         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1331         INP_WUNLOCK(inp);
1332         return (error);
1333 }
1334
1335 #ifdef INET
1336 struct pr_usrreqs tcp_usrreqs = {
1337         .pru_abort =            tcp_usr_abort,
1338         .pru_accept =           tcp_usr_accept,
1339         .pru_attach =           tcp_usr_attach,
1340         .pru_bind =             tcp_usr_bind,
1341         .pru_connect =          tcp_usr_connect,
1342         .pru_control =          in_control,
1343         .pru_detach =           tcp_usr_detach,
1344         .pru_disconnect =       tcp_usr_disconnect,
1345         .pru_listen =           tcp_usr_listen,
1346         .pru_peeraddr =         in_getpeeraddr,
1347         .pru_rcvd =             tcp_usr_rcvd,
1348         .pru_rcvoob =           tcp_usr_rcvoob,
1349         .pru_send =             tcp_usr_send,
1350         .pru_ready =            tcp_usr_ready,
1351         .pru_shutdown =         tcp_usr_shutdown,
1352         .pru_sockaddr =         in_getsockaddr,
1353         .pru_sosetlabel =       in_pcbsosetlabel,
1354         .pru_close =            tcp_usr_close,
1355 };
1356 #endif /* INET */
1357
1358 #ifdef INET6
1359 struct pr_usrreqs tcp6_usrreqs = {
1360         .pru_abort =            tcp_usr_abort,
1361         .pru_accept =           tcp6_usr_accept,
1362         .pru_attach =           tcp_usr_attach,
1363         .pru_bind =             tcp6_usr_bind,
1364         .pru_connect =          tcp6_usr_connect,
1365         .pru_control =          in6_control,
1366         .pru_detach =           tcp_usr_detach,
1367         .pru_disconnect =       tcp_usr_disconnect,
1368         .pru_listen =           tcp6_usr_listen,
1369         .pru_peeraddr =         in6_mapped_peeraddr,
1370         .pru_rcvd =             tcp_usr_rcvd,
1371         .pru_rcvoob =           tcp_usr_rcvoob,
1372         .pru_send =             tcp_usr_send,
1373         .pru_ready =            tcp_usr_ready,
1374         .pru_shutdown =         tcp_usr_shutdown,
1375         .pru_sockaddr =         in6_mapped_sockaddr,
1376         .pru_sosetlabel =       in_pcbsosetlabel,
1377         .pru_close =            tcp_usr_close,
1378 };
1379 #endif /* INET6 */
1380
1381 #ifdef INET
1382 /*
1383  * Common subroutine to open a TCP connection to remote host specified
1384  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1385  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1386  * to choose a local host address (interface).  If there is an existing
1387  * incarnation of the same connection in TIME-WAIT state and if the remote
1388  * host was sending CC options and if the connection duration was < MSL, then
1389  * truncate the previous TIME-WAIT state and proceed.
1390  * Initialize connection parameters and enter SYN-SENT state.
1391  */
1392 static int
1393 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1394 {
1395         struct inpcb *inp = tp->t_inpcb, *oinp;
1396         struct socket *so = inp->inp_socket;
1397         struct in_addr laddr;
1398         u_short lport;
1399         int error;
1400
1401         INP_WLOCK_ASSERT(inp);
1402         INP_HASH_WLOCK(&V_tcbinfo);
1403
1404         if (inp->inp_lport == 0) {
1405                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1406                 if (error)
1407                         goto out;
1408         }
1409
1410         /*
1411          * Cannot simply call in_pcbconnect, because there might be an
1412          * earlier incarnation of this same connection still in
1413          * TIME_WAIT state, creating an ADDRINUSE error.
1414          */
1415         laddr = inp->inp_laddr;
1416         lport = inp->inp_lport;
1417         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1418             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1419         if (error && oinp == NULL)
1420                 goto out;
1421         if (oinp) {
1422                 error = EADDRINUSE;
1423                 goto out;
1424         }
1425         inp->inp_laddr = laddr;
1426         in_pcbrehash(inp);
1427         INP_HASH_WUNLOCK(&V_tcbinfo);
1428
1429         /*
1430          * Compute window scaling to request:
1431          * Scale to fit into sweet spot.  See tcp_syncache.c.
1432          * XXX: This should move to tcp_output().
1433          */
1434         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1435             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1436                 tp->request_r_scale++;
1437
1438         soisconnecting(so);
1439         TCPSTAT_INC(tcps_connattempt);
1440         tcp_state_change(tp, TCPS_SYN_SENT);
1441         tp->iss = tcp_new_isn(&inp->inp_inc);
1442         if (tp->t_flags & TF_REQ_TSTMP)
1443                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1444         tcp_sendseqinit(tp);
1445
1446         return 0;
1447
1448 out:
1449         INP_HASH_WUNLOCK(&V_tcbinfo);
1450         return (error);
1451 }
1452 #endif /* INET */
1453
1454 #ifdef INET6
1455 static int
1456 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1457 {
1458         struct inpcb *inp = tp->t_inpcb;
1459         int error;
1460
1461         INP_WLOCK_ASSERT(inp);
1462         INP_HASH_WLOCK(&V_tcbinfo);
1463
1464         if (inp->inp_lport == 0) {
1465                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1466                 if (error)
1467                         goto out;
1468         }
1469         error = in6_pcbconnect(inp, nam, td->td_ucred);
1470         if (error != 0)
1471                 goto out;
1472         INP_HASH_WUNLOCK(&V_tcbinfo);
1473
1474         /* Compute window scaling to request.  */
1475         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1476             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1477                 tp->request_r_scale++;
1478
1479         soisconnecting(inp->inp_socket);
1480         TCPSTAT_INC(tcps_connattempt);
1481         tcp_state_change(tp, TCPS_SYN_SENT);
1482         tp->iss = tcp_new_isn(&inp->inp_inc);
1483         if (tp->t_flags & TF_REQ_TSTMP)
1484                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1485         tcp_sendseqinit(tp);
1486
1487         return 0;
1488
1489 out:
1490         INP_HASH_WUNLOCK(&V_tcbinfo);
1491         return error;
1492 }
1493 #endif /* INET6 */
1494
1495 /*
1496  * Export TCP internal state information via a struct tcp_info, based on the
1497  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1498  * (TCP state machine, etc).  We export all information using FreeBSD-native
1499  * constants -- for example, the numeric values for tcpi_state will differ
1500  * from Linux.
1501  */
1502 static void
1503 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1504 {
1505
1506         INP_WLOCK_ASSERT(tp->t_inpcb);
1507         bzero(ti, sizeof(*ti));
1508
1509         ti->tcpi_state = tp->t_state;
1510         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1511                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1512         if (tp->t_flags & TF_SACK_PERMIT)
1513                 ti->tcpi_options |= TCPI_OPT_SACK;
1514         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1515                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1516                 ti->tcpi_snd_wscale = tp->snd_scale;
1517                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1518         }
1519         if (tp->t_flags & TF_ECN_PERMIT)
1520                 ti->tcpi_options |= TCPI_OPT_ECN;
1521
1522         ti->tcpi_rto = tp->t_rxtcur * tick;
1523         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1524         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1525         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1526
1527         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1528         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1529
1530         /*
1531          * FreeBSD-specific extension fields for tcp_info.
1532          */
1533         ti->tcpi_rcv_space = tp->rcv_wnd;
1534         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1535         ti->tcpi_snd_wnd = tp->snd_wnd;
1536         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1537         ti->tcpi_snd_nxt = tp->snd_nxt;
1538         ti->tcpi_snd_mss = tp->t_maxseg;
1539         ti->tcpi_rcv_mss = tp->t_maxseg;
1540         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1541         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1542         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1543 #ifdef TCP_OFFLOAD
1544         if (tp->t_flags & TF_TOE) {
1545                 ti->tcpi_options |= TCPI_OPT_TOE;
1546                 tcp_offload_tcp_info(tp, ti);
1547         }
1548 #endif
1549 }
1550
1551 /*
1552  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1553  * socket option arguments.  When it re-acquires the lock after the copy, it
1554  * has to revalidate that the connection is still valid for the socket
1555  * option.
1556  */
1557 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1558         INP_WLOCK(inp);                                                 \
1559         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1560                 INP_WUNLOCK(inp);                                       \
1561                 cleanup;                                                \
1562                 return (ECONNRESET);                                    \
1563         }                                                               \
1564         tp = intotcpcb(inp);                                            \
1565 } while(0)
1566 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1567
1568 int
1569 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1570 {
1571         int     error;
1572         struct  inpcb *inp;
1573         struct  tcpcb *tp;
1574         struct tcp_function_block *blk;
1575         struct tcp_function_set fsn;
1576
1577         error = 0;
1578         inp = sotoinpcb(so);
1579         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1580         if (sopt->sopt_level != IPPROTO_TCP) {
1581 #ifdef INET6
1582                 if (inp->inp_vflag & INP_IPV6PROTO) {
1583                         error = ip6_ctloutput(so, sopt);
1584                         /*
1585                          * In case of the IPV6_USE_MIN_MTU socket option,
1586                          * the INC_IPV6MINMTU flag to announce a corresponding
1587                          * MSS during the initial handshake.
1588                          * If the TCP connection is not in the front states,
1589                          * just reduce the MSS being used.
1590                          * This avoids the sending of TCP segments which will
1591                          * be fragmented at the IPv6 layer.
1592                          */
1593                         if ((error == 0) &&
1594                             (sopt->sopt_dir == SOPT_SET) &&
1595                             (sopt->sopt_level == IPPROTO_IPV6) &&
1596                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1597                                 INP_WLOCK(inp);
1598                                 if ((inp->inp_flags &
1599                                     (INP_TIMEWAIT | INP_DROPPED))) {
1600                                         INP_WUNLOCK(inp);
1601                                         return (ECONNRESET);
1602                                 }
1603                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1604                                 tp = intotcpcb(inp);
1605                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1606                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1607                                         struct ip6_pktopts *opt;
1608
1609                                         opt = inp->in6p_outputopts;
1610                                         if ((opt != NULL) &&
1611                                             (opt->ip6po_minmtu ==
1612                                             IP6PO_MINMTU_ALL)) {
1613                                                 if (tp->t_maxseg > TCP6_MSS) {
1614                                                         tp->t_maxseg = TCP6_MSS;
1615                                                 }
1616                                         }
1617                                 }
1618                                 INP_WUNLOCK(inp);
1619                         }
1620                 }
1621 #endif /* INET6 */
1622 #if defined(INET6) && defined(INET)
1623                 else
1624 #endif
1625 #ifdef INET
1626                 {
1627                         error = ip_ctloutput(so, sopt);
1628                 }
1629 #endif
1630                 return (error);
1631         }
1632         INP_WLOCK(inp);
1633         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1634                 INP_WUNLOCK(inp);
1635                 return (ECONNRESET);
1636         }
1637         tp = intotcpcb(inp);
1638         /*
1639          * Protect the TCP option TCP_FUNCTION_BLK so
1640          * that a sub-function can *never* overwrite this.
1641          */
1642         if ((sopt->sopt_dir == SOPT_SET) &&
1643             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1644                 INP_WUNLOCK(inp);
1645                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1646                     sizeof fsn);
1647                 if (error)
1648                         return (error);
1649                 INP_WLOCK_RECHECK(inp);
1650                 blk = find_and_ref_tcp_functions(&fsn);
1651                 if (blk == NULL) {
1652                         INP_WUNLOCK(inp);
1653                         return (ENOENT);
1654                 }
1655                 if (tp->t_fb == blk) {
1656                         /* You already have this */
1657                         refcount_release(&blk->tfb_refcnt);
1658                         INP_WUNLOCK(inp);
1659                         return (0);
1660                 }
1661                 if (tp->t_state != TCPS_CLOSED) {
1662                         /*
1663                          * The user has advanced the state
1664                          * past the initial point, we may not
1665                          * be able to switch.
1666                          */
1667                         if (blk->tfb_tcp_handoff_ok != NULL) {
1668                                 /*
1669                                  * Does the stack provide a
1670                                  * query mechanism, if so it may
1671                                  * still be possible?
1672                                  */
1673                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1674                         } else
1675                                 error = EINVAL;
1676                         if (error) {
1677                                 refcount_release(&blk->tfb_refcnt);
1678                                 INP_WUNLOCK(inp);
1679                                 return(error);
1680                         }
1681                 }
1682                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1683                         refcount_release(&blk->tfb_refcnt);
1684                         INP_WUNLOCK(inp);
1685                         return (ENOENT);
1686                 }
1687                 /*
1688                  * Release the old refcnt, the
1689                  * lookup acquired a ref on the
1690                  * new one already.
1691                  */
1692                 if (tp->t_fb->tfb_tcp_fb_fini) {
1693                         /*
1694                          * Tell the stack to cleanup with 0 i.e.
1695                          * the tcb is not going away.
1696                          */
1697                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1698                 }
1699 #ifdef TCPHPTS
1700                 /* Assure that we are not on any hpts */
1701                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1702 #endif
1703                 if (blk->tfb_tcp_fb_init) {
1704                         error = (*blk->tfb_tcp_fb_init)(tp);
1705                         if (error) {
1706                                 refcount_release(&blk->tfb_refcnt);
1707                                 if (tp->t_fb->tfb_tcp_fb_init) {
1708                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1709                                                 /* Fall back failed, drop the connection */
1710                                                 INP_WUNLOCK(inp);
1711                                                 soabort(so);
1712                                                 return(error);
1713                                         }
1714                                 }
1715                                 goto err_out;
1716                         }
1717                 }
1718                 refcount_release(&tp->t_fb->tfb_refcnt);
1719                 tp->t_fb = blk;
1720 #ifdef TCP_OFFLOAD
1721                 if (tp->t_flags & TF_TOE) {
1722                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1723                              sopt->sopt_name);
1724                 }
1725 #endif
1726 err_out:
1727                 INP_WUNLOCK(inp);
1728                 return (error);
1729         } else if ((sopt->sopt_dir == SOPT_GET) &&
1730             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1731                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1732                     TCP_FUNCTION_NAME_LEN_MAX);
1733                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1734                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1735                 INP_WUNLOCK(inp);
1736                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1737                 return (error);
1738         }
1739         /* Pass in the INP locked, called must unlock it */
1740         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1741 }
1742
1743 /*
1744  * If this assert becomes untrue, we need to change the size of the buf
1745  * variable in tcp_default_ctloutput().
1746  */
1747 #ifdef CTASSERT
1748 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1749 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1750 #endif
1751
1752 int
1753 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1754 {
1755         int     error, opt, optval;
1756         u_int   ui;
1757         struct  tcp_info ti;
1758         struct cc_algo *algo;
1759         char    *pbuf, buf[TCP_LOG_ID_LEN];
1760         size_t  len;
1761
1762         /*
1763          * For TCP_CCALGOOPT forward the control to CC module, for both
1764          * SOPT_SET and SOPT_GET.
1765          */
1766         switch (sopt->sopt_name) {
1767         case TCP_CCALGOOPT:
1768                 INP_WUNLOCK(inp);
1769                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1770                         return (EINVAL);
1771                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1772                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1773                     sopt->sopt_valsize);
1774                 if (error) {
1775                         free(pbuf, M_TEMP);
1776                         return (error);
1777                 }
1778                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1779                 if (CC_ALGO(tp)->ctl_output != NULL)
1780                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1781                 else
1782                         error = ENOENT;
1783                 INP_WUNLOCK(inp);
1784                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1785                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1786                 free(pbuf, M_TEMP);
1787                 return (error);
1788         }
1789
1790         switch (sopt->sopt_dir) {
1791         case SOPT_SET:
1792                 switch (sopt->sopt_name) {
1793 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1794                 case TCP_MD5SIG:
1795                         if (!TCPMD5_ENABLED()) {
1796                                 INP_WUNLOCK(inp);
1797                                 return (ENOPROTOOPT);
1798                         }
1799                         error = TCPMD5_PCBCTL(inp, sopt);
1800                         if (error)
1801                                 return (error);
1802                         goto unlock_and_done;
1803 #endif /* IPSEC */
1804
1805                 case TCP_NODELAY:
1806                 case TCP_NOOPT:
1807                         INP_WUNLOCK(inp);
1808                         error = sooptcopyin(sopt, &optval, sizeof optval,
1809                             sizeof optval);
1810                         if (error)
1811                                 return (error);
1812
1813                         INP_WLOCK_RECHECK(inp);
1814                         switch (sopt->sopt_name) {
1815                         case TCP_NODELAY:
1816                                 opt = TF_NODELAY;
1817                                 break;
1818                         case TCP_NOOPT:
1819                                 opt = TF_NOOPT;
1820                                 break;
1821                         default:
1822                                 opt = 0; /* dead code to fool gcc */
1823                                 break;
1824                         }
1825
1826                         if (optval)
1827                                 tp->t_flags |= opt;
1828                         else
1829                                 tp->t_flags &= ~opt;
1830 unlock_and_done:
1831 #ifdef TCP_OFFLOAD
1832                         if (tp->t_flags & TF_TOE) {
1833                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1834                                     sopt->sopt_name);
1835                         }
1836 #endif
1837                         INP_WUNLOCK(inp);
1838                         break;
1839
1840                 case TCP_NOPUSH:
1841                         INP_WUNLOCK(inp);
1842                         error = sooptcopyin(sopt, &optval, sizeof optval,
1843                             sizeof optval);
1844                         if (error)
1845                                 return (error);
1846
1847                         INP_WLOCK_RECHECK(inp);
1848                         if (optval)
1849                                 tp->t_flags |= TF_NOPUSH;
1850                         else if (tp->t_flags & TF_NOPUSH) {
1851                                 tp->t_flags &= ~TF_NOPUSH;
1852                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1853                                         error = tp->t_fb->tfb_tcp_output(tp);
1854                         }
1855                         goto unlock_and_done;
1856
1857                 case TCP_MAXSEG:
1858                         INP_WUNLOCK(inp);
1859                         error = sooptcopyin(sopt, &optval, sizeof optval,
1860                             sizeof optval);
1861                         if (error)
1862                                 return (error);
1863
1864                         INP_WLOCK_RECHECK(inp);
1865                         if (optval > 0 && optval <= tp->t_maxseg &&
1866                             optval + 40 >= V_tcp_minmss)
1867                                 tp->t_maxseg = optval;
1868                         else
1869                                 error = EINVAL;
1870                         goto unlock_and_done;
1871
1872                 case TCP_INFO:
1873                         INP_WUNLOCK(inp);
1874                         error = EINVAL;
1875                         break;
1876
1877                 case TCP_CONGESTION:
1878                         INP_WUNLOCK(inp);
1879                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1880                         if (error)
1881                                 break;
1882                         buf[sopt->sopt_valsize] = '\0';
1883                         INP_WLOCK_RECHECK(inp);
1884                         CC_LIST_RLOCK();
1885                         STAILQ_FOREACH(algo, &cc_list, entries)
1886                                 if (strncmp(buf, algo->name,
1887                                     TCP_CA_NAME_MAX) == 0)
1888                                         break;
1889                         CC_LIST_RUNLOCK();
1890                         if (algo == NULL) {
1891                                 INP_WUNLOCK(inp);
1892                                 error = EINVAL;
1893                                 break;
1894                         }
1895                         /*
1896                          * We hold a write lock over the tcb so it's safe to
1897                          * do these things without ordering concerns.
1898                          */
1899                         if (CC_ALGO(tp)->cb_destroy != NULL)
1900                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1901                         CC_DATA(tp) = NULL;
1902                         CC_ALGO(tp) = algo;
1903                         /*
1904                          * If something goes pear shaped initialising the new
1905                          * algo, fall back to newreno (which does not
1906                          * require initialisation).
1907                          */
1908                         if (algo->cb_init != NULL &&
1909                             algo->cb_init(tp->ccv) != 0) {
1910                                 CC_ALGO(tp) = &newreno_cc_algo;
1911                                 /*
1912                                  * The only reason init should fail is
1913                                  * because of malloc.
1914                                  */
1915                                 error = ENOMEM;
1916                         }
1917                         INP_WUNLOCK(inp);
1918                         break;
1919
1920                 case TCP_KEEPIDLE:
1921                 case TCP_KEEPINTVL:
1922                 case TCP_KEEPINIT:
1923                         INP_WUNLOCK(inp);
1924                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1925                         if (error)
1926                                 return (error);
1927
1928                         if (ui > (UINT_MAX / hz)) {
1929                                 error = EINVAL;
1930                                 break;
1931                         }
1932                         ui *= hz;
1933
1934                         INP_WLOCK_RECHECK(inp);
1935                         switch (sopt->sopt_name) {
1936                         case TCP_KEEPIDLE:
1937                                 tp->t_keepidle = ui;
1938                                 /*
1939                                  * XXX: better check current remaining
1940                                  * timeout and "merge" it with new value.
1941                                  */
1942                                 if ((tp->t_state > TCPS_LISTEN) &&
1943                                     (tp->t_state <= TCPS_CLOSING))
1944                                         tcp_timer_activate(tp, TT_KEEP,
1945                                             TP_KEEPIDLE(tp));
1946                                 break;
1947                         case TCP_KEEPINTVL:
1948                                 tp->t_keepintvl = ui;
1949                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1950                                     (TP_MAXIDLE(tp) > 0))
1951                                         tcp_timer_activate(tp, TT_2MSL,
1952                                             TP_MAXIDLE(tp));
1953                                 break;
1954                         case TCP_KEEPINIT:
1955                                 tp->t_keepinit = ui;
1956                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1957                                     tp->t_state == TCPS_SYN_SENT)
1958                                         tcp_timer_activate(tp, TT_KEEP,
1959                                             TP_KEEPINIT(tp));
1960                                 break;
1961                         }
1962                         goto unlock_and_done;
1963
1964                 case TCP_KEEPCNT:
1965                         INP_WUNLOCK(inp);
1966                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1967                         if (error)
1968                                 return (error);
1969
1970                         INP_WLOCK_RECHECK(inp);
1971                         tp->t_keepcnt = ui;
1972                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1973                             (TP_MAXIDLE(tp) > 0))
1974                                 tcp_timer_activate(tp, TT_2MSL,
1975                                     TP_MAXIDLE(tp));
1976                         goto unlock_and_done;
1977
1978 #ifdef TCPPCAP
1979                 case TCP_PCAP_OUT:
1980                 case TCP_PCAP_IN:
1981                         INP_WUNLOCK(inp);
1982                         error = sooptcopyin(sopt, &optval, sizeof optval,
1983                             sizeof optval);
1984                         if (error)
1985                                 return (error);
1986
1987                         INP_WLOCK_RECHECK(inp);
1988                         if (optval >= 0)
1989                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1990                                         &(tp->t_outpkts) : &(tp->t_inpkts),
1991                                         optval);
1992                         else
1993                                 error = EINVAL;
1994                         goto unlock_and_done;
1995 #endif
1996
1997                 case TCP_FASTOPEN: {
1998                         struct tcp_fastopen tfo_optval;
1999
2000                         INP_WUNLOCK(inp);
2001                         if (!V_tcp_fastopen_client_enable &&
2002                             !V_tcp_fastopen_server_enable)
2003                                 return (EPERM);
2004
2005                         error = sooptcopyin(sopt, &tfo_optval,
2006                                     sizeof(tfo_optval), sizeof(int));
2007                         if (error)
2008                                 return (error);
2009
2010                         INP_WLOCK_RECHECK(inp);
2011                         if (tfo_optval.enable) {
2012                                 if (tp->t_state == TCPS_LISTEN) {
2013                                         if (!V_tcp_fastopen_server_enable) {
2014                                                 error = EPERM;
2015                                                 goto unlock_and_done;
2016                                         }
2017
2018                                         tp->t_flags |= TF_FASTOPEN;
2019                                         if (tp->t_tfo_pending == NULL)
2020                                                 tp->t_tfo_pending =
2021                                                     tcp_fastopen_alloc_counter();
2022                                 } else {
2023                                         /*
2024                                          * If a pre-shared key was provided,
2025                                          * stash it in the client cookie
2026                                          * field of the tcpcb for use during
2027                                          * connect.
2028                                          */
2029                                         if (sopt->sopt_valsize ==
2030                                             sizeof(tfo_optval)) {
2031                                                 memcpy(tp->t_tfo_cookie.client,
2032                                                        tfo_optval.psk,
2033                                                        TCP_FASTOPEN_PSK_LEN);
2034                                                 tp->t_tfo_client_cookie_len =
2035                                                     TCP_FASTOPEN_PSK_LEN;
2036                                         }
2037                                         tp->t_flags |= TF_FASTOPEN;
2038                                 }
2039                         } else
2040                                 tp->t_flags &= ~TF_FASTOPEN;
2041                         goto unlock_and_done;
2042                 }
2043
2044 #ifdef TCP_BLACKBOX
2045                 case TCP_LOG:
2046                         INP_WUNLOCK(inp);
2047                         error = sooptcopyin(sopt, &optval, sizeof optval,
2048                             sizeof optval);
2049                         if (error)
2050                                 return (error);
2051
2052                         INP_WLOCK_RECHECK(inp);
2053                         error = tcp_log_state_change(tp, optval);
2054                         goto unlock_and_done;
2055
2056                 case TCP_LOGBUF:
2057                         INP_WUNLOCK(inp);
2058                         error = EINVAL;
2059                         break;
2060
2061                 case TCP_LOGID:
2062                         INP_WUNLOCK(inp);
2063                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2064                         if (error)
2065                                 break;
2066                         buf[sopt->sopt_valsize] = '\0';
2067                         INP_WLOCK_RECHECK(inp);
2068                         error = tcp_log_set_id(tp, buf);
2069                         /* tcp_log_set_id() unlocks the INP. */
2070                         break;
2071
2072                 case TCP_LOGDUMP:
2073                 case TCP_LOGDUMPID:
2074                         INP_WUNLOCK(inp);
2075                         error =
2076                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2077                         if (error)
2078                                 break;
2079                         buf[sopt->sopt_valsize] = '\0';
2080                         INP_WLOCK_RECHECK(inp);
2081                         if (sopt->sopt_name == TCP_LOGDUMP) {
2082                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2083                                     M_WAITOK, true);
2084                                 INP_WUNLOCK(inp);
2085                         } else {
2086                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2087                                 /*
2088                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2089                                  * INP lock.
2090                                  */
2091                         }
2092                         break;
2093 #endif
2094
2095                 default:
2096                         INP_WUNLOCK(inp);
2097                         error = ENOPROTOOPT;
2098                         break;
2099                 }
2100                 break;
2101
2102         case SOPT_GET:
2103                 tp = intotcpcb(inp);
2104                 switch (sopt->sopt_name) {
2105 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2106                 case TCP_MD5SIG:
2107                         if (!TCPMD5_ENABLED()) {
2108                                 INP_WUNLOCK(inp);
2109                                 return (ENOPROTOOPT);
2110                         }
2111                         error = TCPMD5_PCBCTL(inp, sopt);
2112                         break;
2113 #endif
2114
2115                 case TCP_NODELAY:
2116                         optval = tp->t_flags & TF_NODELAY;
2117                         INP_WUNLOCK(inp);
2118                         error = sooptcopyout(sopt, &optval, sizeof optval);
2119                         break;
2120                 case TCP_MAXSEG:
2121                         optval = tp->t_maxseg;
2122                         INP_WUNLOCK(inp);
2123                         error = sooptcopyout(sopt, &optval, sizeof optval);
2124                         break;
2125                 case TCP_NOOPT:
2126                         optval = tp->t_flags & TF_NOOPT;
2127                         INP_WUNLOCK(inp);
2128                         error = sooptcopyout(sopt, &optval, sizeof optval);
2129                         break;
2130                 case TCP_NOPUSH:
2131                         optval = tp->t_flags & TF_NOPUSH;
2132                         INP_WUNLOCK(inp);
2133                         error = sooptcopyout(sopt, &optval, sizeof optval);
2134                         break;
2135                 case TCP_INFO:
2136                         tcp_fill_info(tp, &ti);
2137                         INP_WUNLOCK(inp);
2138                         error = sooptcopyout(sopt, &ti, sizeof ti);
2139                         break;
2140                 case TCP_CONGESTION:
2141                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2142                         INP_WUNLOCK(inp);
2143                         error = sooptcopyout(sopt, buf, len + 1);
2144                         break;
2145                 case TCP_KEEPIDLE:
2146                 case TCP_KEEPINTVL:
2147                 case TCP_KEEPINIT:
2148                 case TCP_KEEPCNT:
2149                         switch (sopt->sopt_name) {
2150                         case TCP_KEEPIDLE:
2151                                 ui = TP_KEEPIDLE(tp) / hz;
2152                                 break;
2153                         case TCP_KEEPINTVL:
2154                                 ui = TP_KEEPINTVL(tp) / hz;
2155                                 break;
2156                         case TCP_KEEPINIT:
2157                                 ui = TP_KEEPINIT(tp) / hz;
2158                                 break;
2159                         case TCP_KEEPCNT:
2160                                 ui = TP_KEEPCNT(tp);
2161                                 break;
2162                         }
2163                         INP_WUNLOCK(inp);
2164                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2165                         break;
2166 #ifdef TCPPCAP
2167                 case TCP_PCAP_OUT:
2168                 case TCP_PCAP_IN:
2169                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2170                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2171                         INP_WUNLOCK(inp);
2172                         error = sooptcopyout(sopt, &optval, sizeof optval);
2173                         break;
2174 #endif
2175                 case TCP_FASTOPEN:
2176                         optval = tp->t_flags & TF_FASTOPEN;
2177                         INP_WUNLOCK(inp);
2178                         error = sooptcopyout(sopt, &optval, sizeof optval);
2179                         break;
2180 #ifdef TCP_BLACKBOX
2181                 case TCP_LOG:
2182                         optval = tp->t_logstate;
2183                         INP_WUNLOCK(inp);
2184                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2185                         break;
2186                 case TCP_LOGBUF:
2187                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2188                         error = tcp_log_getlogbuf(sopt, tp);
2189                         break;
2190                 case TCP_LOGID:
2191                         len = tcp_log_get_id(tp, buf);
2192                         INP_WUNLOCK(inp);
2193                         error = sooptcopyout(sopt, buf, len + 1);
2194                         break;
2195                 case TCP_LOGDUMP:
2196                 case TCP_LOGDUMPID:
2197                         INP_WUNLOCK(inp);
2198                         error = EINVAL;
2199                         break;
2200 #endif
2201                 default:
2202                         INP_WUNLOCK(inp);
2203                         error = ENOPROTOOPT;
2204                         break;
2205                 }
2206                 break;
2207         }
2208         return (error);
2209 }
2210 #undef INP_WLOCK_RECHECK
2211 #undef INP_WLOCK_RECHECK_CLEANUP
2212
2213 /*
2214  * Attach TCP protocol to socket, allocating
2215  * internet protocol control block, tcp control block,
2216  * bufer space, and entering LISTEN state if to accept connections.
2217  */
2218 static int
2219 tcp_attach(struct socket *so)
2220 {
2221         struct tcpcb *tp;
2222         struct inpcb *inp;
2223         struct epoch_tracker et;
2224         int error;
2225
2226         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2227                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
2228                 if (error)
2229                         return (error);
2230         }
2231         so->so_rcv.sb_flags |= SB_AUTOSIZE;
2232         so->so_snd.sb_flags |= SB_AUTOSIZE;
2233         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2234         error = in_pcballoc(so, &V_tcbinfo);
2235         if (error) {
2236                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2237                 return (error);
2238         }
2239         inp = sotoinpcb(so);
2240 #ifdef INET6
2241         if (inp->inp_vflag & INP_IPV6PROTO) {
2242                 inp->inp_vflag |= INP_IPV6;
2243                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
2244                         inp->inp_vflag |= INP_IPV4;
2245                 inp->in6p_hops = -1;    /* use kernel default */
2246         }
2247         else
2248 #endif
2249         inp->inp_vflag |= INP_IPV4;
2250         tp = tcp_newtcpcb(inp);
2251         if (tp == NULL) {
2252                 in_pcbdetach(inp);
2253                 in_pcbfree(inp);
2254                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2255                 return (ENOBUFS);
2256         }
2257         tp->t_state = TCPS_CLOSED;
2258         INP_WUNLOCK(inp);
2259         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2260         TCPSTATES_INC(TCPS_CLOSED);
2261         return (0);
2262 }
2263
2264 /*
2265  * Initiate (or continue) disconnect.
2266  * If embryonic state, just send reset (once).
2267  * If in ``let data drain'' option and linger null, just drop.
2268  * Otherwise (hard), mark socket disconnecting and drop
2269  * current input data; switch states based on user close, and
2270  * send segment to peer (with FIN).
2271  */
2272 static void
2273 tcp_disconnect(struct tcpcb *tp)
2274 {
2275         struct inpcb *inp = tp->t_inpcb;
2276         struct socket *so = inp->inp_socket;
2277
2278         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2279         INP_WLOCK_ASSERT(inp);
2280
2281         /*
2282          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2283          * socket is still open.
2284          */
2285         if (tp->t_state < TCPS_ESTABLISHED &&
2286             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2287                 tp = tcp_close(tp);
2288                 KASSERT(tp != NULL,
2289                     ("tcp_disconnect: tcp_close() returned NULL"));
2290         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2291                 tp = tcp_drop(tp, 0);
2292                 KASSERT(tp != NULL,
2293                     ("tcp_disconnect: tcp_drop() returned NULL"));
2294         } else {
2295                 soisdisconnecting(so);
2296                 sbflush(&so->so_rcv);
2297                 tcp_usrclosed(tp);
2298                 if (!(inp->inp_flags & INP_DROPPED))
2299                         tp->t_fb->tfb_tcp_output(tp);
2300         }
2301 }
2302
2303 /*
2304  * User issued close, and wish to trail through shutdown states:
2305  * if never received SYN, just forget it.  If got a SYN from peer,
2306  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2307  * If already got a FIN from peer, then almost done; go to LAST_ACK
2308  * state.  In all other cases, have already sent FIN to peer (e.g.
2309  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2310  * for peer to send FIN or not respond to keep-alives, etc.
2311  * We can let the user exit from the close as soon as the FIN is acked.
2312  */
2313 static void
2314 tcp_usrclosed(struct tcpcb *tp)
2315 {
2316
2317         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2318         INP_WLOCK_ASSERT(tp->t_inpcb);
2319
2320         switch (tp->t_state) {
2321         case TCPS_LISTEN:
2322 #ifdef TCP_OFFLOAD
2323                 tcp_offload_listen_stop(tp);
2324 #endif
2325                 tcp_state_change(tp, TCPS_CLOSED);
2326                 /* FALLTHROUGH */
2327         case TCPS_CLOSED:
2328                 tp = tcp_close(tp);
2329                 /*
2330                  * tcp_close() should never return NULL here as the socket is
2331                  * still open.
2332                  */
2333                 KASSERT(tp != NULL,
2334                     ("tcp_usrclosed: tcp_close() returned NULL"));
2335                 break;
2336
2337         case TCPS_SYN_SENT:
2338         case TCPS_SYN_RECEIVED:
2339                 tp->t_flags |= TF_NEEDFIN;
2340                 break;
2341
2342         case TCPS_ESTABLISHED:
2343                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2344                 break;
2345
2346         case TCPS_CLOSE_WAIT:
2347                 tcp_state_change(tp, TCPS_LAST_ACK);
2348                 break;
2349         }
2350         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2351                 soisdisconnected(tp->t_inpcb->inp_socket);
2352                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2353                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2354                         int timeout;
2355
2356                         timeout = (tcp_fast_finwait2_recycle) ?
2357                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2358                         tcp_timer_activate(tp, TT_2MSL, timeout);
2359                 }
2360         }
2361 }
2362
2363 #ifdef DDB
2364 static void
2365 db_print_indent(int indent)
2366 {
2367         int i;
2368
2369         for (i = 0; i < indent; i++)
2370                 db_printf(" ");
2371 }
2372
2373 static void
2374 db_print_tstate(int t_state)
2375 {
2376
2377         switch (t_state) {
2378         case TCPS_CLOSED:
2379                 db_printf("TCPS_CLOSED");
2380                 return;
2381
2382         case TCPS_LISTEN:
2383                 db_printf("TCPS_LISTEN");
2384                 return;
2385
2386         case TCPS_SYN_SENT:
2387                 db_printf("TCPS_SYN_SENT");
2388                 return;
2389
2390         case TCPS_SYN_RECEIVED:
2391                 db_printf("TCPS_SYN_RECEIVED");
2392                 return;
2393
2394         case TCPS_ESTABLISHED:
2395                 db_printf("TCPS_ESTABLISHED");
2396                 return;
2397
2398         case TCPS_CLOSE_WAIT:
2399                 db_printf("TCPS_CLOSE_WAIT");
2400                 return;
2401
2402         case TCPS_FIN_WAIT_1:
2403                 db_printf("TCPS_FIN_WAIT_1");
2404                 return;
2405
2406         case TCPS_CLOSING:
2407                 db_printf("TCPS_CLOSING");
2408                 return;
2409
2410         case TCPS_LAST_ACK:
2411                 db_printf("TCPS_LAST_ACK");
2412                 return;
2413
2414         case TCPS_FIN_WAIT_2:
2415                 db_printf("TCPS_FIN_WAIT_2");
2416                 return;
2417
2418         case TCPS_TIME_WAIT:
2419                 db_printf("TCPS_TIME_WAIT");
2420                 return;
2421
2422         default:
2423                 db_printf("unknown");
2424                 return;
2425         }
2426 }
2427
2428 static void
2429 db_print_tflags(u_int t_flags)
2430 {
2431         int comma;
2432
2433         comma = 0;
2434         if (t_flags & TF_ACKNOW) {
2435                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2436                 comma = 1;
2437         }
2438         if (t_flags & TF_DELACK) {
2439                 db_printf("%sTF_DELACK", comma ? ", " : "");
2440                 comma = 1;
2441         }
2442         if (t_flags & TF_NODELAY) {
2443                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2444                 comma = 1;
2445         }
2446         if (t_flags & TF_NOOPT) {
2447                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2448                 comma = 1;
2449         }
2450         if (t_flags & TF_SENTFIN) {
2451                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2452                 comma = 1;
2453         }
2454         if (t_flags & TF_REQ_SCALE) {
2455                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2456                 comma = 1;
2457         }
2458         if (t_flags & TF_RCVD_SCALE) {
2459                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2460                 comma = 1;
2461         }
2462         if (t_flags & TF_REQ_TSTMP) {
2463                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2464                 comma = 1;
2465         }
2466         if (t_flags & TF_RCVD_TSTMP) {
2467                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2468                 comma = 1;
2469         }
2470         if (t_flags & TF_SACK_PERMIT) {
2471                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2472                 comma = 1;
2473         }
2474         if (t_flags & TF_NEEDSYN) {
2475                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2476                 comma = 1;
2477         }
2478         if (t_flags & TF_NEEDFIN) {
2479                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2480                 comma = 1;
2481         }
2482         if (t_flags & TF_NOPUSH) {
2483                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2484                 comma = 1;
2485         }
2486         if (t_flags & TF_MORETOCOME) {
2487                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2488                 comma = 1;
2489         }
2490         if (t_flags & TF_LQ_OVERFLOW) {
2491                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2492                 comma = 1;
2493         }
2494         if (t_flags & TF_LASTIDLE) {
2495                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2496                 comma = 1;
2497         }
2498         if (t_flags & TF_RXWIN0SENT) {
2499                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2500                 comma = 1;
2501         }
2502         if (t_flags & TF_FASTRECOVERY) {
2503                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2504                 comma = 1;
2505         }
2506         if (t_flags & TF_CONGRECOVERY) {
2507                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2508                 comma = 1;
2509         }
2510         if (t_flags & TF_WASFRECOVERY) {
2511                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2512                 comma = 1;
2513         }
2514         if (t_flags & TF_SIGNATURE) {
2515                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2516                 comma = 1;
2517         }
2518         if (t_flags & TF_FORCEDATA) {
2519                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2520                 comma = 1;
2521         }
2522         if (t_flags & TF_TSO) {
2523                 db_printf("%sTF_TSO", comma ? ", " : "");
2524                 comma = 1;
2525         }
2526         if (t_flags & TF_ECN_PERMIT) {
2527                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2528                 comma = 1;
2529         }
2530         if (t_flags & TF_FASTOPEN) {
2531                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2532                 comma = 1;
2533         }
2534 }
2535
2536 static void
2537 db_print_toobflags(char t_oobflags)
2538 {
2539         int comma;
2540
2541         comma = 0;
2542         if (t_oobflags & TCPOOB_HAVEDATA) {
2543                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2544                 comma = 1;
2545         }
2546         if (t_oobflags & TCPOOB_HADDATA) {
2547                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2548                 comma = 1;
2549         }
2550 }
2551
2552 static void
2553 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2554 {
2555
2556         db_print_indent(indent);
2557         db_printf("%s at %p\n", name, tp);
2558
2559         indent += 2;
2560
2561         db_print_indent(indent);
2562         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2563            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2564
2565         db_print_indent(indent);
2566         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2567             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2568
2569         db_print_indent(indent);
2570         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2571             &tp->t_timers->tt_delack, tp->t_inpcb);
2572
2573         db_print_indent(indent);
2574         db_printf("t_state: %d (", tp->t_state);
2575         db_print_tstate(tp->t_state);
2576         db_printf(")\n");
2577
2578         db_print_indent(indent);
2579         db_printf("t_flags: 0x%x (", tp->t_flags);
2580         db_print_tflags(tp->t_flags);
2581         db_printf(")\n");
2582
2583         db_print_indent(indent);
2584         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2585             tp->snd_una, tp->snd_max, tp->snd_nxt);
2586
2587         db_print_indent(indent);
2588         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2589            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2590
2591         db_print_indent(indent);
2592         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2593             tp->iss, tp->irs, tp->rcv_nxt);
2594
2595         db_print_indent(indent);
2596         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2597             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2598
2599         db_print_indent(indent);
2600         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2601            tp->snd_wnd, tp->snd_cwnd);
2602
2603         db_print_indent(indent);
2604         db_printf("snd_ssthresh: %u   snd_recover: "
2605             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2606
2607         db_print_indent(indent);
2608         db_printf("t_rcvtime: %u   t_startime: %u\n",
2609             tp->t_rcvtime, tp->t_starttime);
2610
2611         db_print_indent(indent);
2612         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2613             tp->t_rtttime, tp->t_rtseq);
2614
2615         db_print_indent(indent);
2616         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2617             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2618
2619         db_print_indent(indent);
2620         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2621             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2622             tp->t_rttbest);
2623
2624         db_print_indent(indent);
2625         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2626             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2627
2628         db_print_indent(indent);
2629         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2630         db_print_toobflags(tp->t_oobflags);
2631         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2632
2633         db_print_indent(indent);
2634         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2635             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2636
2637         db_print_indent(indent);
2638         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2639             tp->ts_recent, tp->ts_recent_age);
2640
2641         db_print_indent(indent);
2642         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2643             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2644
2645         db_print_indent(indent);
2646         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2647             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2648             tp->snd_recover_prev, tp->t_badrxtwin);
2649
2650         db_print_indent(indent);
2651         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2652             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2653
2654         db_print_indent(indent);
2655         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2656             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2657
2658         /* Skip sackblks, sackhint. */
2659
2660         db_print_indent(indent);
2661         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2662             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2663 }
2664
2665 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2666 {
2667         struct tcpcb *tp;
2668
2669         if (!have_addr) {
2670                 db_printf("usage: show tcpcb <addr>\n");
2671                 return;
2672         }
2673         tp = (struct tcpcb *)addr;
2674
2675         db_print_tcpcb(tp, "tcpcb", 0);
2676 }
2677 #endif