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_kern_tls.h"
  48 #include "opt_tcpdebug.h"
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/limits.h>
  53 #include <sys/malloc.h>
  54 #include <sys/refcount.h>
  55 #include <sys/kernel.h>
  56 #include <sys/ktls.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/mbuf.h>
  59 #ifdef INET6
  60 #include <sys/domain.h>
  61 #endif /* INET6 */
  62 #include <sys/socket.h>
  63 #include <sys/socketvar.h>
  64 #include <sys/protosw.h>
  65 #include <sys/proc.h>
  66 #include <sys/jail.h>
  67 #include <sys/syslog.h>
  68
  69 #ifdef DDB
  70 #include <ddb/ddb.h>
  71 #endif
  72
  73 #include <net/if.h>
  74 #include <net/if_var.h>
  75 #include <net/route.h>
  76 #include <net/vnet.h>
  77
  78 #include <netinet/in.h>
  79 #include <netinet/in_kdtrace.h>
  80 #include <netinet/in_pcb.h>
  81 #include <netinet/in_systm.h>
  82 #include <netinet/in_var.h>
  83 #include <netinet/ip_var.h>
  84 #ifdef INET6
  85 #include <netinet/ip6.h>
  86 #include <netinet6/in6_pcb.h>
  87 #include <netinet6/ip6_var.h>
  88 #include <netinet6/scope6_var.h>
  89 #endif
  90 #include <netinet/tcp.h>
  91 #include <netinet/tcp_fsm.h>
  92 #include <netinet/tcp_seq.h>
  93 #include <netinet/tcp_timer.h>
  94 #include <netinet/tcp_var.h>
  95 #include <netinet/tcp_log_buf.h>
  96 #include <netinet/tcpip.h>
  97 #include <netinet/cc/cc.h>
  98 #include <netinet/tcp_fastopen.h>
  99 #include <netinet/tcp_hpts.h>
 100 #ifdef TCPPCAP
 101 #include <netinet/tcp_pcap.h>
 102 #endif
 103 #ifdef TCPDEBUG
 104 #include <netinet/tcp_debug.h>
 105 #endif
 106 #ifdef TCP_OFFLOAD
 107 #include <netinet/tcp_offload.h>
 108 #endif
 109 #include <netipsec/ipsec_support.h>
 110
 111 /*
 112  * TCP protocol interface to socket abstraction.
 113  */
 114 static int      tcp_attach(struct socket *);
 115 #ifdef INET
 116 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 117                     struct thread *td);
 118 #endif /* INET */
 119 #ifdef INET6
 120 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 121                     struct thread *td);
 122 #endif /* INET6 */
 123 static void     tcp_disconnect(struct tcpcb *);
 124 static void     tcp_usrclosed(struct tcpcb *);
 125 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 126
 127 #ifdef TCPDEBUG
 128 #define TCPDEBUG0       int ostate = 0
 129 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 130 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 131                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 132 #else
 133 #define TCPDEBUG0
 134 #define TCPDEBUG1()
 135 #define TCPDEBUG2(req)
 136 #endif
 137
 138 /*
 139  * TCP attaches to socket via pru_attach(), reserving space,
 140  * and an internet control block.
 141  */
 142 static int
 143 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 144 {
 145         struct inpcb *inp;
 146         struct tcpcb *tp = NULL;
 147         int error;
 148         TCPDEBUG0;
 149
 150         inp = sotoinpcb(so);
 151         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 152         TCPDEBUG1();
 153
 154         error = tcp_attach(so);
 155         if (error)
 156                 goto out;
 157
 158         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 159                 so->so_linger = TCP_LINGERTIME;
 160
 161         inp = sotoinpcb(so);
 162         tp = intotcpcb(inp);
 163 out:
 164         TCPDEBUG2(PRU_ATTACH);
 165         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 166         return error;
 167 }
 168
 169 /*
 170  * tcp_detach is called when the socket layer loses its final reference
 171  * to the socket, be it a file descriptor reference, a reference from TCP,
 172  * etc.  At this point, there is only one case in which we will keep around
 173  * inpcb state: time wait.
 174  *
 175  * This function can probably be re-absorbed back into tcp_usr_detach() now
 176  * that there is a single detach path.
 177  */
 178 static void
 179 tcp_detach(struct socket *so, struct inpcb *inp)
 180 {
 181         struct tcpcb *tp;
 182
 183         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
 184         INP_WLOCK_ASSERT(inp);
 185
 186         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 187         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 188
 189         tp = intotcpcb(inp);
 190
 191         if (inp->inp_flags & INP_TIMEWAIT) {
 192                 /*
 193                  * There are two cases to handle: one in which the time wait
 194                  * state is being discarded (INP_DROPPED), and one in which
 195                  * this connection will remain in timewait.  In the former,
 196                  * it is time to discard all state (except tcptw, which has
 197                  * already been discarded by the timewait close code, which
 198                  * should be further up the call stack somewhere).  In the
 199                  * latter case, we detach from the socket, but leave the pcb
 200                  * present until timewait ends.
 201                  *
 202                  * XXXRW: Would it be cleaner to free the tcptw here?
 203                  *
 204                  * Astute question indeed, from twtcp perspective there are
 205                  * four cases to consider:
 206                  *
 207                  * #1 tcp_detach is called at tcptw creation time by
 208                  *  tcp_twstart, then do not discard the newly created tcptw
 209                  *  and leave inpcb present until timewait ends
 210                  * #2 tcp_detach is called at tcptw creation time by
 211                  *  tcp_twstart, but connection is local and tw will be
 212                  *  discarded immediately
 213                  * #3 tcp_detach is called at timewait end (or reuse) by
 214                  *  tcp_twclose, then the tcptw has already been discarded
 215                  *  (or reused) and inpcb is freed here
 216                  * #4 tcp_detach is called() after timewait ends (or reuse)
 217                  *  (e.g. by soclose), then tcptw has already been discarded
 218                  *  (or reused) and inpcb is freed here
 219                  *
 220                  *  In all three cases the tcptw should not be freed here.
 221                  */
 222                 if (inp->inp_flags & INP_DROPPED) {
 223                         in_pcbdetach(inp);
 224                         if (__predict_true(tp == NULL)) {
 225                                 in_pcbfree(inp);
 226                         } else {
 227                                 /*
 228                                  * This case should not happen as in TIMEWAIT
 229                                  * state the inp should not be destroyed before
 230                                  * its tcptw.  If INVARIANTS is defined, panic.
 231                                  */
 232 #ifdef INVARIANTS
 233                                 panic("%s: Panic before an inp double-free: "
 234                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 235                                     , __func__);
 236 #else
 237                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 238                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 239                                     , __func__);
 240 #endif
 241                                 INP_WUNLOCK(inp);
 242                         }
 243                 } else {
 244                         in_pcbdetach(inp);
 245                         INP_WUNLOCK(inp);
 246                 }
 247         } else {
 248                 /*
 249                  * If the connection is not in timewait, we consider two
 250                  * two conditions: one in which no further processing is
 251                  * necessary (dropped || embryonic), and one in which TCP is
 252                  * not yet done, but no longer requires the socket, so the
 253                  * pcb will persist for the time being.
 254                  *
 255                  * XXXRW: Does the second case still occur?
 256                  */
 257                 if (inp->inp_flags & INP_DROPPED ||
 258                     tp->t_state < TCPS_SYN_SENT) {
 259                         tcp_discardcb(tp);
 260                         in_pcbdetach(inp);
 261                         in_pcbfree(inp);
 262                 } else {
 263                         in_pcbdetach(inp);
 264                         INP_WUNLOCK(inp);
 265                 }
 266         }
 267 }
 268
 269 /*
 270  * pru_detach() detaches the TCP protocol from the socket.
 271  * If the protocol state is non-embryonic, then can't
 272  * do this directly: have to initiate a pru_disconnect(),
 273  * which may finish later; embryonic TCB's can just
 274  * be discarded here.
 275  */
 276 static void
 277 tcp_usr_detach(struct socket *so)
 278 {
 279         struct inpcb *inp;
 280         int rlock = 0;
 281         struct epoch_tracker et;
 282
 283         inp = sotoinpcb(so);
 284         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 285         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 286                 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 287                 rlock = 1;
 288         }
 289         INP_WLOCK(inp);
 290         KASSERT(inp->inp_socket != NULL,
 291             ("tcp_usr_detach: inp_socket == NULL"));
 292         tcp_detach(so, inp);
 293         if (rlock)
 294                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 295 }
 296
 297 #ifdef INET
 298 /*
 299  * Give the socket an address.
 300  */
 301 static int
 302 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 303 {
 304         int error = 0;
 305         struct inpcb *inp;
 306         struct tcpcb *tp = NULL;
 307         struct sockaddr_in *sinp;
 308
 309         sinp = (struct sockaddr_in *)nam;
 310         if (nam->sa_len != sizeof (*sinp))
 311                 return (EINVAL);
 312         /*
 313          * Must check for multicast addresses and disallow binding
 314          * to them.
 315          */
 316         if (sinp->sin_family == AF_INET &&
 317             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 318                 return (EAFNOSUPPORT);
 319
 320         TCPDEBUG0;
 321         inp = sotoinpcb(so);
 322         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 323         INP_WLOCK(inp);
 324         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 325                 error = EINVAL;
 326                 goto out;
 327         }
 328         tp = intotcpcb(inp);
 329         TCPDEBUG1();
 330         INP_HASH_WLOCK(&V_tcbinfo);
 331         error = in_pcbbind(inp, nam, td->td_ucred);
 332         INP_HASH_WUNLOCK(&V_tcbinfo);
 333 out:
 334         TCPDEBUG2(PRU_BIND);
 335         TCP_PROBE2(debug__user, tp, PRU_BIND);
 336         INP_WUNLOCK(inp);
 337
 338         return (error);
 339 }
 340 #endif /* INET */
 341
 342 #ifdef INET6
 343 static int
 344 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 345 {
 346         int error = 0;
 347         struct inpcb *inp;
 348         struct tcpcb *tp = NULL;
 349         struct sockaddr_in6 *sin6;
 350
 351         sin6 = (struct sockaddr_in6 *)nam;
 352         if (nam->sa_len != sizeof (*sin6))
 353                 return (EINVAL);
 354         /*
 355          * Must check for multicast addresses and disallow binding
 356          * to them.
 357          */
 358         if (sin6->sin6_family == AF_INET6 &&
 359             IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 360                 return (EAFNOSUPPORT);
 361
 362         TCPDEBUG0;
 363         inp = sotoinpcb(so);
 364         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 365         INP_WLOCK(inp);
 366         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 367                 error = EINVAL;
 368                 goto out;
 369         }
 370         tp = intotcpcb(inp);
 371         TCPDEBUG1();
 372         INP_HASH_WLOCK(&V_tcbinfo);
 373         inp->inp_vflag &= ~INP_IPV4;
 374         inp->inp_vflag |= INP_IPV6;
 375 #ifdef INET
 376         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 377                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
 378                         inp->inp_vflag |= INP_IPV4;
 379                 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 380                         struct sockaddr_in sin;
 381
 382                         in6_sin6_2_sin(&sin, sin6);
 383                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 384                                 error = EAFNOSUPPORT;
 385                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 386                                 goto out;
 387                         }
 388                         inp->inp_vflag |= INP_IPV4;
 389                         inp->inp_vflag &= ~INP_IPV6;
 390                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 391                             td->td_ucred);
 392                         INP_HASH_WUNLOCK(&V_tcbinfo);
 393                         goto out;
 394                 }
 395         }
 396 #endif
 397         error = in6_pcbbind(inp, nam, td->td_ucred);
 398         INP_HASH_WUNLOCK(&V_tcbinfo);
 399 out:
 400         TCPDEBUG2(PRU_BIND);
 401         TCP_PROBE2(debug__user, tp, PRU_BIND);
 402         INP_WUNLOCK(inp);
 403         return (error);
 404 }
 405 #endif /* INET6 */
 406
 407 #ifdef INET
 408 /*
 409  * Prepare to accept connections.
 410  */
 411 static int
 412 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 413 {
 414         int error = 0;
 415         struct inpcb *inp;
 416         struct tcpcb *tp = NULL;
 417
 418         TCPDEBUG0;
 419         inp = sotoinpcb(so);
 420         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 421         INP_WLOCK(inp);
 422         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 423                 error = EINVAL;
 424                 goto out;
 425         }
 426         tp = intotcpcb(inp);
 427         TCPDEBUG1();
 428         SOCK_LOCK(so);
 429         error = solisten_proto_check(so);
 430         INP_HASH_WLOCK(&V_tcbinfo);
 431         if (error == 0 && inp->inp_lport == 0)
 432                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 433         INP_HASH_WUNLOCK(&V_tcbinfo);
 434         if (error == 0) {
 435                 tcp_state_change(tp, TCPS_LISTEN);
 436                 solisten_proto(so, backlog);
 437 #ifdef TCP_OFFLOAD
 438                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 439                         tcp_offload_listen_start(tp);
 440 #endif
 441         }
 442         SOCK_UNLOCK(so);
 443
 444         if (IS_FASTOPEN(tp->t_flags))
 445                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 446
 447 out:
 448         TCPDEBUG2(PRU_LISTEN);
 449         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 450         INP_WUNLOCK(inp);
 451         return (error);
 452 }
 453 #endif /* INET */
 454
 455 #ifdef INET6
 456 static int
 457 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 458 {
 459         int error = 0;
 460         struct inpcb *inp;
 461         struct tcpcb *tp = NULL;
 462
 463         TCPDEBUG0;
 464         inp = sotoinpcb(so);
 465         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 466         INP_WLOCK(inp);
 467         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 468                 error = EINVAL;
 469                 goto out;
 470         }
 471         tp = intotcpcb(inp);
 472         TCPDEBUG1();
 473         SOCK_LOCK(so);
 474         error = solisten_proto_check(so);
 475         INP_HASH_WLOCK(&V_tcbinfo);
 476         if (error == 0 && inp->inp_lport == 0) {
 477                 inp->inp_vflag &= ~INP_IPV4;
 478                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 479                         inp->inp_vflag |= INP_IPV4;
 480                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 481         }
 482         INP_HASH_WUNLOCK(&V_tcbinfo);
 483         if (error == 0) {
 484                 tcp_state_change(tp, TCPS_LISTEN);
 485                 solisten_proto(so, backlog);
 486 #ifdef TCP_OFFLOAD
 487                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 488                         tcp_offload_listen_start(tp);
 489 #endif
 490         }
 491         SOCK_UNLOCK(so);
 492
 493         if (IS_FASTOPEN(tp->t_flags))
 494                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 495
 496 out:
 497         TCPDEBUG2(PRU_LISTEN);
 498         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 499         INP_WUNLOCK(inp);
 500         return (error);
 501 }
 502 #endif /* INET6 */
 503
 504 #ifdef INET
 505 /*
 506  * Initiate connection to peer.
 507  * Create a template for use in transmissions on this connection.
 508  * Enter SYN_SENT state, and mark socket as connecting.
 509  * Start keep-alive timer, and seed output sequence space.
 510  * Send initial segment on connection.
 511  */
 512 static int
 513 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 514 {
 515         int error = 0;
 516         struct inpcb *inp;
 517         struct tcpcb *tp = NULL;
 518         struct sockaddr_in *sinp;
 519
 520         sinp = (struct sockaddr_in *)nam;
 521         if (nam->sa_len != sizeof (*sinp))
 522                 return (EINVAL);
 523         /*
 524          * Must disallow TCP ``connections'' to multicast addresses.
 525          */
 526         if (sinp->sin_family == AF_INET
 527             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 528                 return (EAFNOSUPPORT);
 529         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 530                 return (error);
 531
 532         TCPDEBUG0;
 533         inp = sotoinpcb(so);
 534         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 535         INP_WLOCK(inp);
 536         if (inp->inp_flags & INP_TIMEWAIT) {
 537                 error = EADDRINUSE;
 538                 goto out;
 539         }
 540         if (inp->inp_flags & INP_DROPPED) {
 541                 error = ECONNREFUSED;
 542                 goto out;
 543         }
 544         tp = intotcpcb(inp);
 545         TCPDEBUG1();
 546         if ((error = tcp_connect(tp, nam, td)) != 0)
 547                 goto out;
 548 #ifdef TCP_OFFLOAD
 549         if (registered_toedevs > 0 &&
 550             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 551             (error = tcp_offload_connect(so, nam)) == 0)
 552                 goto out;
 553 #endif
 554         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 555         error = tp->t_fb->tfb_tcp_output(tp);
 556 out:
 557         TCPDEBUG2(PRU_CONNECT);
 558         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 559         INP_WUNLOCK(inp);
 560         return (error);
 561 }
 562 #endif /* INET */
 563
 564 #ifdef INET6
 565 static int
 566 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 567 {
 568         int error = 0;
 569         struct inpcb *inp;
 570         struct tcpcb *tp = NULL;
 571         struct sockaddr_in6 *sin6;
 572
 573         TCPDEBUG0;
 574
 575         sin6 = (struct sockaddr_in6 *)nam;
 576         if (nam->sa_len != sizeof (*sin6))
 577                 return (EINVAL);
 578         /*
 579          * Must disallow TCP ``connections'' to multicast addresses.
 580          */
 581         if (sin6->sin6_family == AF_INET6
 582             && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 583                 return (EAFNOSUPPORT);
 584
 585         inp = sotoinpcb(so);
 586         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 587         INP_WLOCK(inp);
 588         if (inp->inp_flags & INP_TIMEWAIT) {
 589                 error = EADDRINUSE;
 590                 goto out;
 591         }
 592         if (inp->inp_flags & INP_DROPPED) {
 593                 error = ECONNREFUSED;
 594                 goto out;
 595         }
 596         tp = intotcpcb(inp);
 597         TCPDEBUG1();
 598 #ifdef INET
 599         /*
 600          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 601          * therefore probably require the hash lock, which isn't held here.
 602          * Is this a significant problem?
 603          */
 604         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 605                 struct sockaddr_in sin;
 606
 607                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 608                         error = EINVAL;
 609                         goto out;
 610                 }
 611                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 612                         error = EAFNOSUPPORT;
 613                         goto out;
 614                 }
 615
 616                 in6_sin6_2_sin(&sin, sin6);
 617                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 618                         error = EAFNOSUPPORT;
 619                         goto out;
 620                 }
 621                 inp->inp_vflag |= INP_IPV4;
 622                 inp->inp_vflag &= ~INP_IPV6;
 623                 if ((error = prison_remote_ip4(td->td_ucred,
 624                     &sin.sin_addr)) != 0)
 625                         goto out;
 626                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 627                         goto out;
 628 #ifdef TCP_OFFLOAD
 629                 if (registered_toedevs > 0 &&
 630                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 631                     (error = tcp_offload_connect(so, nam)) == 0)
 632                         goto out;
 633 #endif
 634                 error = tp->t_fb->tfb_tcp_output(tp);
 635                 goto out;
 636         } else {
 637                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 638                         error = EAFNOSUPPORT;
 639                         goto out;
 640                 }
 641         }
 642 #endif
 643         inp->inp_vflag &= ~INP_IPV4;
 644         inp->inp_vflag |= INP_IPV6;
 645         inp->inp_inc.inc_flags |= INC_ISIPV6;
 646         if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
 647                 goto out;
 648         if ((error = tcp6_connect(tp, nam, td)) != 0)
 649                 goto out;
 650 #ifdef TCP_OFFLOAD
 651         if (registered_toedevs > 0 &&
 652             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 653             (error = tcp_offload_connect(so, nam)) == 0)
 654                 goto out;
 655 #endif
 656         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 657         error = tp->t_fb->tfb_tcp_output(tp);
 658
 659 out:
 660         TCPDEBUG2(PRU_CONNECT);
 661         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 662         INP_WUNLOCK(inp);
 663         return (error);
 664 }
 665 #endif /* INET6 */
 666
 667 /*
 668  * Initiate disconnect from peer.
 669  * If connection never passed embryonic stage, just drop;
 670  * else if don't need to let data drain, then can just drop anyways,
 671  * else have to begin TCP shutdown process: mark socket disconnecting,
 672  * drain unread data, state switch to reflect user close, and
 673  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 674  * when peer sends FIN and acks ours.
 675  *
 676  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 677  */
 678 static int
 679 tcp_usr_disconnect(struct socket *so)
 680 {
 681         struct inpcb *inp;
 682         struct tcpcb *tp = NULL;
 683         struct epoch_tracker et;
 684         int error = 0;
 685
 686         TCPDEBUG0;
 687         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 688         inp = sotoinpcb(so);
 689         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 690         INP_WLOCK(inp);
 691         if (inp->inp_flags & INP_TIMEWAIT)
 692                 goto out;
 693         if (inp->inp_flags & INP_DROPPED) {
 694                 error = ECONNRESET;
 695                 goto out;
 696         }
 697         tp = intotcpcb(inp);
 698         TCPDEBUG1();
 699         tcp_disconnect(tp);
 700 out:
 701         TCPDEBUG2(PRU_DISCONNECT);
 702         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 703         INP_WUNLOCK(inp);
 704         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 705         return (error);
 706 }
 707
 708 #ifdef INET
 709 /*
 710  * Accept a connection.  Essentially all the work is done at higher levels;
 711  * just return the address of the peer, storing through addr.
 712  */
 713 static int
 714 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 715 {
 716         int error = 0;
 717         struct inpcb *inp = NULL;
 718         struct tcpcb *tp = NULL;
 719         struct in_addr addr;
 720         in_port_t port = 0;
 721         TCPDEBUG0;
 722
 723         if (so->so_state & SS_ISDISCONNECTED)
 724                 return (ECONNABORTED);
 725
 726         inp = sotoinpcb(so);
 727         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 728         INP_WLOCK(inp);
 729         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 730                 error = ECONNABORTED;
 731                 goto out;
 732         }
 733         tp = intotcpcb(inp);
 734         TCPDEBUG1();
 735
 736         /*
 737          * We inline in_getpeeraddr and COMMON_END here, so that we can
 738          * copy the data of interest and defer the malloc until after we
 739          * release the lock.
 740          */
 741         port = inp->inp_fport;
 742         addr = inp->inp_faddr;
 743
 744 out:
 745         TCPDEBUG2(PRU_ACCEPT);
 746         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 747         INP_WUNLOCK(inp);
 748         if (error == 0)
 749                 *nam = in_sockaddr(port, &addr);
 750         return error;
 751 }
 752 #endif /* INET */
 753
 754 #ifdef INET6
 755 static int
 756 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 757 {
 758         struct inpcb *inp = NULL;
 759         int error = 0;
 760         struct tcpcb *tp = NULL;
 761         struct in_addr addr;
 762         struct in6_addr addr6;
 763         struct epoch_tracker et;
 764         in_port_t port = 0;
 765         int v4 = 0;
 766         TCPDEBUG0;
 767
 768         if (so->so_state & SS_ISDISCONNECTED)
 769                 return (ECONNABORTED);
 770
 771         inp = sotoinpcb(so);
 772         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 773         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 774         INP_WLOCK(inp);
 775         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 776                 error = ECONNABORTED;
 777                 goto out;
 778         }
 779         tp = intotcpcb(inp);
 780         TCPDEBUG1();
 781
 782         /*
 783          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 784          * copy the data of interest and defer the malloc until after we
 785          * release the lock.
 786          */
 787         if (inp->inp_vflag & INP_IPV4) {
 788                 v4 = 1;
 789                 port = inp->inp_fport;
 790                 addr = inp->inp_faddr;
 791         } else {
 792                 port = inp->inp_fport;
 793                 addr6 = inp->in6p_faddr;
 794         }
 795
 796 out:
 797         TCPDEBUG2(PRU_ACCEPT);
 798         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 799         INP_WUNLOCK(inp);
 800         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 801         if (error == 0) {
 802                 if (v4)
 803                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 804                 else
 805                         *nam = in6_sockaddr(port, &addr6);
 806         }
 807         return error;
 808 }
 809 #endif /* INET6 */
 810
 811 /*
 812  * Mark the connection as being incapable of further output.
 813  */
 814 static int
 815 tcp_usr_shutdown(struct socket *so)
 816 {
 817         int error = 0;
 818         struct inpcb *inp;
 819         struct tcpcb *tp = NULL;
 820         struct epoch_tracker et;
 821
 822         TCPDEBUG0;
 823         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
 824         inp = sotoinpcb(so);
 825         KASSERT(inp != NULL, ("inp == NULL"));
 826         INP_WLOCK(inp);
 827         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 828                 error = ECONNRESET;
 829                 goto out;
 830         }
 831         tp = intotcpcb(inp);
 832         TCPDEBUG1();
 833         socantsendmore(so);
 834         tcp_usrclosed(tp);
 835         if (!(inp->inp_flags & INP_DROPPED))
 836                 error = tp->t_fb->tfb_tcp_output(tp);
 837
 838 out:
 839         TCPDEBUG2(PRU_SHUTDOWN);
 840         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 841         INP_WUNLOCK(inp);
 842         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
 843
 844         return (error);
 845 }
 846
 847 /*
 848  * After a receive, possibly send window update to peer.
 849  */
 850 static int
 851 tcp_usr_rcvd(struct socket *so, int flags)
 852 {
 853         struct inpcb *inp;
 854         struct tcpcb *tp = NULL;
 855         int error = 0;
 856
 857         TCPDEBUG0;
 858         inp = sotoinpcb(so);
 859         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 860         INP_WLOCK(inp);
 861         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 862                 error = ECONNRESET;
 863                 goto out;
 864         }
 865         tp = intotcpcb(inp);
 866         TCPDEBUG1();
 867         /*
 868          * For passively-created TFO connections, don't attempt a window
 869          * update while still in SYN_RECEIVED as this may trigger an early
 870          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 871          * application response data, or failing that, when the DELACK timer
 872          * expires.
 873          */
 874         if (IS_FASTOPEN(tp->t_flags) &&
 875             (tp->t_state == TCPS_SYN_RECEIVED))
 876                 goto out;
 877 #ifdef TCP_OFFLOAD
 878         if (tp->t_flags & TF_TOE)
 879                 tcp_offload_rcvd(tp);
 880         else
 881 #endif
 882         tp->t_fb->tfb_tcp_output(tp);
 883
 884 out:
 885         TCPDEBUG2(PRU_RCVD);
 886         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 887         INP_WUNLOCK(inp);
 888         return (error);
 889 }
 890
 891 /*
 892  * Do a send by putting data in output queue and updating urgent
 893  * marker if URG set.  Possibly send more data.  Unlike the other
 894  * pru_*() routines, the mbuf chains are our responsibility.  We
 895  * must either enqueue them or free them.  The other pru_* routines
 896  * generally are caller-frees.
 897  */
 898 static int
 899 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 900     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 901 {
 902         int error = 0;
 903         struct inpcb *inp;
 904         struct tcpcb *tp = NULL;
 905         struct epoch_tracker net_et;
 906 #ifdef INET
 907 #ifdef INET6
 908         struct sockaddr_in sin;
 909 #endif
 910         struct sockaddr_in *sinp;
 911 #endif
 912 #ifdef INET6
 913         int isipv6;
 914 #endif
 915         TCPDEBUG0;
 916
 917         /*
 918          * We require the pcbinfo lock if we will close the socket as part of
 919          * this call.
 920          */
 921         if (flags & PRUS_EOF)
 922                 INP_INFO_RLOCK_ET(&V_tcbinfo, net_et);
 923         inp = sotoinpcb(so);
 924         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 925         INP_WLOCK(inp);
 926         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 927                 if (control)
 928                         m_freem(control);
 929                 /*
 930                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 931                  * for freeing memory.
 932                  */
 933                 if (m && (flags & PRUS_NOTREADY) == 0)
 934                         m_freem(m);
 935                 error = ECONNRESET;
 936                 goto out;
 937         }
 938         tp = intotcpcb(inp);
 939         TCPDEBUG1();
 940         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
 941                 switch (nam->sa_family) {
 942 #ifdef INET
 943                 case AF_INET:
 944                         sinp = (struct sockaddr_in *)nam;
 945                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
 946                                 if (m)
 947                                         m_freem(m);
 948                                 error = EINVAL;
 949                                 goto out;
 950                         }
 951                         if ((inp->inp_vflag & INP_IPV6) != 0) {
 952                                 if (m)
 953                                         m_freem(m);
 954                                 error = EAFNOSUPPORT;
 955                                 goto out;
 956                         }
 957                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 958                                 if (m)
 959                                         m_freem(m);
 960                                 error = EAFNOSUPPORT;
 961                                 goto out;
 962                         }
 963                         if ((error = prison_remote_ip4(td->td_ucred,
 964                             &sinp->sin_addr))) {
 965                                 if (m)
 966                                         m_freem(m);
 967                                 goto out;
 968                         }
 969 #ifdef INET6
 970                         isipv6 = 0;
 971 #endif
 972                         break;
 973 #endif /* INET */
 974 #ifdef INET6
 975                 case AF_INET6:
 976                 {
 977                         struct sockaddr_in6 *sin6;
 978
 979                         sin6 = (struct sockaddr_in6 *)nam;
 980                         if (sin6->sin6_len != sizeof(*sin6)) {
 981                                 if (m)
 982                                         m_freem(m);
 983                                 error = EINVAL;
 984                                 goto out;
 985                         }
 986                         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
 987                                 if (m)
 988                                         m_freem(m);
 989                                 error = EAFNOSUPPORT;
 990                                 goto out;
 991                         }
 992                         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 993 #ifdef INET
 994                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 995                                         error = EINVAL;
 996                                         if (m)
 997                                                 m_freem(m);
 998                                         goto out;
 999                                 }
1000                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1001                                         error = EAFNOSUPPORT;
1002                                         if (m)
1003                                                 m_freem(m);
1004                                         goto out;
1005                                 }
1006                                 inp->inp_vflag &= ~INP_IPV6;
1007                                 sinp = &sin;
1008                                 in6_sin6_2_sin(sinp, sin6);
1009                                 if (IN_MULTICAST(
1010                                     ntohl(sinp->sin_addr.s_addr))) {
1011                                         error = EAFNOSUPPORT;
1012                                         if (m)
1013                                                 m_freem(m);
1014                                         goto out;
1015                                 }
1016                                 if ((error = prison_remote_ip4(td->td_ucred,
1017                                     &sinp->sin_addr))) {
1018                                         if (m)
1019                                                 m_freem(m);
1020                                         goto out;
1021                                 }
1022                                 isipv6 = 0;
1023 #else /* !INET */
1024                                 error = EAFNOSUPPORT;
1025                                 if (m)
1026                                         m_freem(m);
1027                                 goto out;
1028 #endif /* INET */
1029                         } else {
1030                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1031                                         if (m)
1032                                                 m_freem(m);
1033                                         error = EAFNOSUPPORT;
1034                                         goto out;
1035                                 }
1036                                 inp->inp_vflag &= ~INP_IPV4;
1037                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1038                                 if ((error = prison_remote_ip6(td->td_ucred,
1039                                     &sin6->sin6_addr))) {
1040                                         if (m)
1041                                                 m_freem(m);
1042                                         goto out;
1043                                 }
1044                                 isipv6 = 1;
1045                         }
1046                         break;
1047                 }
1048 #endif /* INET6 */
1049                 default:
1050                         if (m)
1051                                 m_freem(m);
1052                         error = EAFNOSUPPORT;
1053                         goto out;
1054                 }
1055         }
1056         if (control) {
1057                 /* TCP doesn't do control messages (rights, creds, etc) */
1058                 if (control->m_len) {
1059                         m_freem(control);
1060                         if (m)
1061                                 m_freem(m);
1062                         error = EINVAL;
1063                         goto out;
1064                 }
1065                 m_freem(control);       /* empty control, just free it */
1066         }
1067         if (!(flags & PRUS_OOB)) {
1068                 sbappendstream(&so->so_snd, m, flags);
1069                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1070                         /*
1071                          * Do implied connect if not yet connected,
1072                          * initialize window to default value, and
1073                          * initialize maxseg using peer's cached MSS.
1074                          */
1075 #ifdef INET6
1076                         if (isipv6)
1077                                 error = tcp6_connect(tp, nam, td);
1078 #endif /* INET6 */
1079 #if defined(INET6) && defined(INET)
1080                         else
1081 #endif
1082 #ifdef INET
1083                                 error = tcp_connect(tp,
1084                                     (struct sockaddr *)sinp, td);
1085 #endif
1086                         if (error)
1087                                 goto out;
1088                         if (IS_FASTOPEN(tp->t_flags))
1089                                 tcp_fastopen_connect(tp);
1090                         else {
1091                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1092                                 tcp_mss(tp, -1);
1093                         }
1094                 }
1095                 if (flags & PRUS_EOF) {
1096                         /*
1097                          * Close the send side of the connection after
1098                          * the data is sent.
1099                          */
1100                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1101                         socantsendmore(so);
1102                         tcp_usrclosed(tp);
1103                 }
1104                 if (!(inp->inp_flags & INP_DROPPED) &&
1105                     !(flags & PRUS_NOTREADY)) {
1106                         if (flags & PRUS_MORETOCOME)
1107                                 tp->t_flags |= TF_MORETOCOME;
1108                         error = tp->t_fb->tfb_tcp_output(tp);
1109                         if (flags & PRUS_MORETOCOME)
1110                                 tp->t_flags &= ~TF_MORETOCOME;
1111                 }
1112         } else {
1113                 /*
1114                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1115                  */
1116                 SOCKBUF_LOCK(&so->so_snd);
1117                 if (sbspace(&so->so_snd) < -512) {
1118                         SOCKBUF_UNLOCK(&so->so_snd);
1119                         m_freem(m);
1120                         error = ENOBUFS;
1121                         goto out;
1122                 }
1123                 /*
1124                  * According to RFC961 (Assigned Protocols),
1125                  * the urgent pointer points to the last octet
1126                  * of urgent data.  We continue, however,
1127                  * to consider it to indicate the first octet
1128                  * of data past the urgent section.
1129                  * Otherwise, snd_up should be one lower.
1130                  */
1131                 sbappendstream_locked(&so->so_snd, m, flags);
1132                 SOCKBUF_UNLOCK(&so->so_snd);
1133                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1134                         /*
1135                          * Do implied connect if not yet connected,
1136                          * initialize window to default value, and
1137                          * initialize maxseg using peer's cached MSS.
1138                          */
1139
1140                         /*
1141                          * Not going to contemplate SYN|URG
1142                          */
1143                         if (IS_FASTOPEN(tp->t_flags))
1144                                 tp->t_flags &= ~TF_FASTOPEN;
1145 #ifdef INET6
1146                         if (isipv6)
1147                                 error = tcp6_connect(tp, nam, td);
1148 #endif /* INET6 */
1149 #if defined(INET6) && defined(INET)
1150                         else
1151 #endif
1152 #ifdef INET
1153                                 error = tcp_connect(tp,
1154                                     (struct sockaddr *)sinp, td);
1155 #endif
1156                         if (error)
1157                                 goto out;
1158                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1159                         tcp_mss(tp, -1);
1160                 }
1161                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1162                 if (!(flags & PRUS_NOTREADY)) {
1163                         tp->t_flags |= TF_FORCEDATA;
1164                         error = tp->t_fb->tfb_tcp_output(tp);
1165                         tp->t_flags &= ~TF_FORCEDATA;
1166                 }
1167         }
1168         TCP_LOG_EVENT(tp, NULL,
1169             &inp->inp_socket->so_rcv,
1170             &inp->inp_socket->so_snd,
1171             TCP_LOG_USERSEND, error,
1172             0, NULL, false);
1173 out:
1174         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1175                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1176         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1177                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1178         INP_WUNLOCK(inp);
1179         if (flags & PRUS_EOF)
1180                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, net_et);
1181         return (error);
1182 }
1183
1184 static int
1185 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1186 {
1187         struct inpcb *inp;
1188         struct tcpcb *tp;
1189         int error;
1190
1191         inp = sotoinpcb(so);
1192         INP_WLOCK(inp);
1193         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1194                 INP_WUNLOCK(inp);
1195                 mb_free_notready(m, count);
1196                 return (ECONNRESET);
1197         }
1198         tp = intotcpcb(inp);
1199
1200         SOCKBUF_LOCK(&so->so_snd);
1201         error = sbready(&so->so_snd, m, count);
1202         SOCKBUF_UNLOCK(&so->so_snd);
1203         if (error == 0)
1204                 error = tp->t_fb->tfb_tcp_output(tp);
1205         INP_WUNLOCK(inp);
1206
1207         return (error);
1208 }
1209
1210 /*
1211  * Abort the TCP.  Drop the connection abruptly.
1212  */
1213 static void
1214 tcp_usr_abort(struct socket *so)
1215 {
1216         struct inpcb *inp;
1217         struct tcpcb *tp = NULL;
1218         struct epoch_tracker et;
1219         TCPDEBUG0;
1220
1221         inp = sotoinpcb(so);
1222         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1223
1224         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1225         INP_WLOCK(inp);
1226         KASSERT(inp->inp_socket != NULL,
1227             ("tcp_usr_abort: inp_socket == NULL"));
1228
1229         /*
1230          * If we still have full TCP state, and we're not dropped, drop.
1231          */
1232         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1233             !(inp->inp_flags & INP_DROPPED)) {
1234                 tp = intotcpcb(inp);
1235                 TCPDEBUG1();
1236                 tp = tcp_drop(tp, ECONNABORTED);
1237                 if (tp == NULL)
1238                         goto dropped;
1239                 TCPDEBUG2(PRU_ABORT);
1240                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1241         }
1242         if (!(inp->inp_flags & INP_DROPPED)) {
1243                 SOCK_LOCK(so);
1244                 so->so_state |= SS_PROTOREF;
1245                 SOCK_UNLOCK(so);
1246                 inp->inp_flags |= INP_SOCKREF;
1247         }
1248         INP_WUNLOCK(inp);
1249 dropped:
1250         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1251 }
1252
1253 /*
1254  * TCP socket is closed.  Start friendly disconnect.
1255  */
1256 static void
1257 tcp_usr_close(struct socket *so)
1258 {
1259         struct inpcb *inp;
1260         struct tcpcb *tp = NULL;
1261         struct epoch_tracker et;
1262         TCPDEBUG0;
1263
1264         inp = sotoinpcb(so);
1265         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1266
1267         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1268         INP_WLOCK(inp);
1269         KASSERT(inp->inp_socket != NULL,
1270             ("tcp_usr_close: inp_socket == NULL"));
1271
1272         /*
1273          * If we still have full TCP state, and we're not dropped, initiate
1274          * a disconnect.
1275          */
1276         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1277             !(inp->inp_flags & INP_DROPPED)) {
1278                 tp = intotcpcb(inp);
1279                 TCPDEBUG1();
1280                 tcp_disconnect(tp);
1281                 TCPDEBUG2(PRU_CLOSE);
1282                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1283         }
1284         if (!(inp->inp_flags & INP_DROPPED)) {
1285                 SOCK_LOCK(so);
1286                 so->so_state |= SS_PROTOREF;
1287                 SOCK_UNLOCK(so);
1288                 inp->inp_flags |= INP_SOCKREF;
1289         }
1290         INP_WUNLOCK(inp);
1291         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1292 }
1293
1294 /*
1295  * Receive out-of-band data.
1296  */
1297 static int
1298 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1299 {
1300         int error = 0;
1301         struct inpcb *inp;
1302         struct tcpcb *tp = NULL;
1303
1304         TCPDEBUG0;
1305         inp = sotoinpcb(so);
1306         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1307         INP_WLOCK(inp);
1308         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1309                 error = ECONNRESET;
1310                 goto out;
1311         }
1312         tp = intotcpcb(inp);
1313         TCPDEBUG1();
1314         if ((so->so_oobmark == 0 &&
1315              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1316             so->so_options & SO_OOBINLINE ||
1317             tp->t_oobflags & TCPOOB_HADDATA) {
1318                 error = EINVAL;
1319                 goto out;
1320         }
1321         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1322                 error = EWOULDBLOCK;
1323                 goto out;
1324         }
1325         m->m_len = 1;
1326         *mtod(m, caddr_t) = tp->t_iobc;
1327         if ((flags & MSG_PEEK) == 0)
1328                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1329
1330 out:
1331         TCPDEBUG2(PRU_RCVOOB);
1332         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1333         INP_WUNLOCK(inp);
1334         return (error);
1335 }
1336
1337 #ifdef INET
1338 struct pr_usrreqs tcp_usrreqs = {
1339         .pru_abort =            tcp_usr_abort,
1340         .pru_accept =           tcp_usr_accept,
1341         .pru_attach =           tcp_usr_attach,
1342         .pru_bind =             tcp_usr_bind,
1343         .pru_connect =          tcp_usr_connect,
1344         .pru_control =          in_control,
1345         .pru_detach =           tcp_usr_detach,
1346         .pru_disconnect =       tcp_usr_disconnect,
1347         .pru_listen =           tcp_usr_listen,
1348         .pru_peeraddr =         in_getpeeraddr,
1349         .pru_rcvd =             tcp_usr_rcvd,
1350         .pru_rcvoob =           tcp_usr_rcvoob,
1351         .pru_send =             tcp_usr_send,
1352         .pru_ready =            tcp_usr_ready,
1353         .pru_shutdown =         tcp_usr_shutdown,
1354         .pru_sockaddr =         in_getsockaddr,
1355         .pru_sosetlabel =       in_pcbsosetlabel,
1356         .pru_close =            tcp_usr_close,
1357 };
1358 #endif /* INET */
1359
1360 #ifdef INET6
1361 struct pr_usrreqs tcp6_usrreqs = {
1362         .pru_abort =            tcp_usr_abort,
1363         .pru_accept =           tcp6_usr_accept,
1364         .pru_attach =           tcp_usr_attach,
1365         .pru_bind =             tcp6_usr_bind,
1366         .pru_connect =          tcp6_usr_connect,
1367         .pru_control =          in6_control,
1368         .pru_detach =           tcp_usr_detach,
1369         .pru_disconnect =       tcp_usr_disconnect,
1370         .pru_listen =           tcp6_usr_listen,
1371         .pru_peeraddr =         in6_mapped_peeraddr,
1372         .pru_rcvd =             tcp_usr_rcvd,
1373         .pru_rcvoob =           tcp_usr_rcvoob,
1374         .pru_send =             tcp_usr_send,
1375         .pru_ready =            tcp_usr_ready,
1376         .pru_shutdown =         tcp_usr_shutdown,
1377         .pru_sockaddr =         in6_mapped_sockaddr,
1378         .pru_sosetlabel =       in_pcbsosetlabel,
1379         .pru_close =            tcp_usr_close,
1380 };
1381 #endif /* INET6 */
1382
1383 #ifdef INET
1384 /*
1385  * Common subroutine to open a TCP connection to remote host specified
1386  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1387  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1388  * to choose a local host address (interface).  If there is an existing
1389  * incarnation of the same connection in TIME-WAIT state and if the remote
1390  * host was sending CC options and if the connection duration was < MSL, then
1391  * truncate the previous TIME-WAIT state and proceed.
1392  * Initialize connection parameters and enter SYN-SENT state.
1393  */
1394 static int
1395 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1396 {
1397         struct inpcb *inp = tp->t_inpcb, *oinp;
1398         struct socket *so = inp->inp_socket;
1399         struct in_addr laddr;
1400         u_short lport;
1401         int error;
1402
1403         INP_WLOCK_ASSERT(inp);
1404         INP_HASH_WLOCK(&V_tcbinfo);
1405
1406         if (inp->inp_lport == 0) {
1407                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1408                 if (error)
1409                         goto out;
1410         }
1411
1412         /*
1413          * Cannot simply call in_pcbconnect, because there might be an
1414          * earlier incarnation of this same connection still in
1415          * TIME_WAIT state, creating an ADDRINUSE error.
1416          */
1417         laddr = inp->inp_laddr;
1418         lport = inp->inp_lport;
1419         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1420             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1421         if (error && oinp == NULL)
1422                 goto out;
1423         if (oinp) {
1424                 error = EADDRINUSE;
1425                 goto out;
1426         }
1427         inp->inp_laddr = laddr;
1428         in_pcbrehash(inp);
1429         INP_HASH_WUNLOCK(&V_tcbinfo);
1430
1431         /*
1432          * Compute window scaling to request:
1433          * Scale to fit into sweet spot.  See tcp_syncache.c.
1434          * XXX: This should move to tcp_output().
1435          */
1436         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1437             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1438                 tp->request_r_scale++;
1439
1440         soisconnecting(so);
1441         TCPSTAT_INC(tcps_connattempt);
1442         tcp_state_change(tp, TCPS_SYN_SENT);
1443         tp->iss = tcp_new_isn(&inp->inp_inc);
1444         if (tp->t_flags & TF_REQ_TSTMP)
1445                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1446         tcp_sendseqinit(tp);
1447
1448         return 0;
1449
1450 out:
1451         INP_HASH_WUNLOCK(&V_tcbinfo);
1452         return (error);
1453 }
1454 #endif /* INET */
1455
1456 #ifdef INET6
1457 static int
1458 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1459 {
1460         struct inpcb *inp = tp->t_inpcb;
1461         int error;
1462
1463         INP_WLOCK_ASSERT(inp);
1464         INP_HASH_WLOCK(&V_tcbinfo);
1465
1466         if (inp->inp_lport == 0) {
1467                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1468                 if (error)
1469                         goto out;
1470         }
1471         error = in6_pcbconnect(inp, nam, td->td_ucred);
1472         if (error != 0)
1473                 goto out;
1474         INP_HASH_WUNLOCK(&V_tcbinfo);
1475
1476         /* Compute window scaling to request.  */
1477         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1478             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1479                 tp->request_r_scale++;
1480
1481         soisconnecting(inp->inp_socket);
1482         TCPSTAT_INC(tcps_connattempt);
1483         tcp_state_change(tp, TCPS_SYN_SENT);
1484         tp->iss = tcp_new_isn(&inp->inp_inc);
1485         if (tp->t_flags & TF_REQ_TSTMP)
1486                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1487         tcp_sendseqinit(tp);
1488
1489         return 0;
1490
1491 out:
1492         INP_HASH_WUNLOCK(&V_tcbinfo);
1493         return error;
1494 }
1495 #endif /* INET6 */
1496
1497 /*
1498  * Export TCP internal state information via a struct tcp_info, based on the
1499  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1500  * (TCP state machine, etc).  We export all information using FreeBSD-native
1501  * constants -- for example, the numeric values for tcpi_state will differ
1502  * from Linux.
1503  */
1504 static void
1505 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1506 {
1507
1508         INP_WLOCK_ASSERT(tp->t_inpcb);
1509         bzero(ti, sizeof(*ti));
1510
1511         ti->tcpi_state = tp->t_state;
1512         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1513                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1514         if (tp->t_flags & TF_SACK_PERMIT)
1515                 ti->tcpi_options |= TCPI_OPT_SACK;
1516         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1517                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1518                 ti->tcpi_snd_wscale = tp->snd_scale;
1519                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1520         }
1521         if (tp->t_flags & TF_ECN_PERMIT)
1522                 ti->tcpi_options |= TCPI_OPT_ECN;
1523
1524         ti->tcpi_rto = tp->t_rxtcur * tick;
1525         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1526         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1527         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1528
1529         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1530         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1531
1532         /*
1533          * FreeBSD-specific extension fields for tcp_info.
1534          */
1535         ti->tcpi_rcv_space = tp->rcv_wnd;
1536         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1537         ti->tcpi_snd_wnd = tp->snd_wnd;
1538         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1539         ti->tcpi_snd_nxt = tp->snd_nxt;
1540         ti->tcpi_snd_mss = tp->t_maxseg;
1541         ti->tcpi_rcv_mss = tp->t_maxseg;
1542         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1543         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1544         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1545 #ifdef TCP_OFFLOAD
1546         if (tp->t_flags & TF_TOE) {
1547                 ti->tcpi_options |= TCPI_OPT_TOE;
1548                 tcp_offload_tcp_info(tp, ti);
1549         }
1550 #endif
1551 }
1552
1553 /*
1554  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1555  * socket option arguments.  When it re-acquires the lock after the copy, it
1556  * has to revalidate that the connection is still valid for the socket
1557  * option.
1558  */
1559 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1560         INP_WLOCK(inp);                                                 \
1561         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1562                 INP_WUNLOCK(inp);                                       \
1563                 cleanup;                                                \
1564                 return (ECONNRESET);                                    \
1565         }                                                               \
1566         tp = intotcpcb(inp);                                            \
1567 } while(0)
1568 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1569
1570 int
1571 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1572 {
1573         int     error;
1574         struct  inpcb *inp;
1575         struct  tcpcb *tp;
1576         struct tcp_function_block *blk;
1577         struct tcp_function_set fsn;
1578
1579         error = 0;
1580         inp = sotoinpcb(so);
1581         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1582         if (sopt->sopt_level != IPPROTO_TCP) {
1583 #ifdef INET6
1584                 if (inp->inp_vflag & INP_IPV6PROTO) {
1585                         error = ip6_ctloutput(so, sopt);
1586                         /*
1587                          * In case of the IPV6_USE_MIN_MTU socket option,
1588                          * the INC_IPV6MINMTU flag to announce a corresponding
1589                          * MSS during the initial handshake.
1590                          * If the TCP connection is not in the front states,
1591                          * just reduce the MSS being used.
1592                          * This avoids the sending of TCP segments which will
1593                          * be fragmented at the IPv6 layer.
1594                          */
1595                         if ((error == 0) &&
1596                             (sopt->sopt_dir == SOPT_SET) &&
1597                             (sopt->sopt_level == IPPROTO_IPV6) &&
1598                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1599                                 INP_WLOCK(inp);
1600                                 if ((inp->inp_flags &
1601                                     (INP_TIMEWAIT | INP_DROPPED))) {
1602                                         INP_WUNLOCK(inp);
1603                                         return (ECONNRESET);
1604                                 }
1605                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1606                                 tp = intotcpcb(inp);
1607                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1608                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1609                                         struct ip6_pktopts *opt;
1610
1611                                         opt = inp->in6p_outputopts;
1612                                         if ((opt != NULL) &&
1613                                             (opt->ip6po_minmtu ==
1614                                             IP6PO_MINMTU_ALL)) {
1615                                                 if (tp->t_maxseg > TCP6_MSS) {
1616                                                         tp->t_maxseg = TCP6_MSS;
1617                                                 }
1618                                         }
1619                                 }
1620                                 INP_WUNLOCK(inp);
1621                         }
1622                 }
1623 #endif /* INET6 */
1624 #if defined(INET6) && defined(INET)
1625                 else
1626 #endif
1627 #ifdef INET
1628                 {
1629                         error = ip_ctloutput(so, sopt);
1630                 }
1631 #endif
1632                 return (error);
1633         }
1634         INP_WLOCK(inp);
1635         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1636                 INP_WUNLOCK(inp);
1637                 return (ECONNRESET);
1638         }
1639         tp = intotcpcb(inp);
1640         /*
1641          * Protect the TCP option TCP_FUNCTION_BLK so
1642          * that a sub-function can *never* overwrite this.
1643          */
1644         if ((sopt->sopt_dir == SOPT_SET) &&
1645             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1646                 INP_WUNLOCK(inp);
1647                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1648                     sizeof fsn);
1649                 if (error)
1650                         return (error);
1651                 INP_WLOCK_RECHECK(inp);
1652                 blk = find_and_ref_tcp_functions(&fsn);
1653                 if (blk == NULL) {
1654                         INP_WUNLOCK(inp);
1655                         return (ENOENT);
1656                 }
1657                 if (tp->t_fb == blk) {
1658                         /* You already have this */
1659                         refcount_release(&blk->tfb_refcnt);
1660                         INP_WUNLOCK(inp);
1661                         return (0);
1662                 }
1663                 if (tp->t_state != TCPS_CLOSED) {
1664                         /*
1665                          * The user has advanced the state
1666                          * past the initial point, we may not
1667                          * be able to switch.
1668                          */
1669                         if (blk->tfb_tcp_handoff_ok != NULL) {
1670                                 /*
1671                                  * Does the stack provide a
1672                                  * query mechanism, if so it may
1673                                  * still be possible?
1674                                  */
1675                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1676                         } else
1677                                 error = EINVAL;
1678                         if (error) {
1679                                 refcount_release(&blk->tfb_refcnt);
1680                                 INP_WUNLOCK(inp);
1681                                 return(error);
1682                         }
1683                 }
1684                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1685                         refcount_release(&blk->tfb_refcnt);
1686                         INP_WUNLOCK(inp);
1687                         return (ENOENT);
1688                 }
1689                 /*
1690                  * Release the old refcnt, the
1691                  * lookup acquired a ref on the
1692                  * new one already.
1693                  */
1694                 if (tp->t_fb->tfb_tcp_fb_fini) {
1695                         /*
1696                          * Tell the stack to cleanup with 0 i.e.
1697                          * the tcb is not going away.
1698                          */
1699                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1700                 }
1701 #ifdef TCPHPTS
1702                 /* Assure that we are not on any hpts */
1703                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1704 #endif
1705                 if (blk->tfb_tcp_fb_init) {
1706                         error = (*blk->tfb_tcp_fb_init)(tp);
1707                         if (error) {
1708                                 refcount_release(&blk->tfb_refcnt);
1709                                 if (tp->t_fb->tfb_tcp_fb_init) {
1710                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1711                                                 /* Fall back failed, drop the connection */
1712                                                 INP_WUNLOCK(inp);
1713                                                 soabort(so);
1714                                                 return(error);
1715                                         }
1716                                 }
1717                                 goto err_out;
1718                         }
1719                 }
1720                 refcount_release(&tp->t_fb->tfb_refcnt);
1721                 tp->t_fb = blk;
1722 #ifdef TCP_OFFLOAD
1723                 if (tp->t_flags & TF_TOE) {
1724                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1725                              sopt->sopt_name);
1726                 }
1727 #endif
1728 err_out:
1729                 INP_WUNLOCK(inp);
1730                 return (error);
1731         } else if ((sopt->sopt_dir == SOPT_GET) &&
1732             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1733                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1734                     TCP_FUNCTION_NAME_LEN_MAX);
1735                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1736                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1737                 INP_WUNLOCK(inp);
1738                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1739                 return (error);
1740         }
1741         /* Pass in the INP locked, called must unlock it */
1742         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1743 }
1744
1745 /*
1746  * If this assert becomes untrue, we need to change the size of the buf
1747  * variable in tcp_default_ctloutput().
1748  */
1749 #ifdef CTASSERT
1750 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1751 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1752 #endif
1753
1754 int
1755 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1756 {
1757         int     error, opt, optval;
1758         u_int   ui;
1759         struct  tcp_info ti;
1760 #ifdef KERN_TLS
1761         struct tls_enable tls;
1762 #endif
1763         struct cc_algo *algo;
1764         char    *pbuf, buf[TCP_LOG_ID_LEN];
1765         size_t  len;
1766
1767         /*
1768          * For TCP_CCALGOOPT forward the control to CC module, for both
1769          * SOPT_SET and SOPT_GET.
1770          */
1771         switch (sopt->sopt_name) {
1772         case TCP_CCALGOOPT:
1773                 INP_WUNLOCK(inp);
1774                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1775                         return (EINVAL);
1776                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1777                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1778                     sopt->sopt_valsize);
1779                 if (error) {
1780                         free(pbuf, M_TEMP);
1781                         return (error);
1782                 }
1783                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1784                 if (CC_ALGO(tp)->ctl_output != NULL)
1785                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1786                 else
1787                         error = ENOENT;
1788                 INP_WUNLOCK(inp);
1789                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1790                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1791                 free(pbuf, M_TEMP);
1792                 return (error);
1793         }
1794
1795         switch (sopt->sopt_dir) {
1796         case SOPT_SET:
1797                 switch (sopt->sopt_name) {
1798 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1799                 case TCP_MD5SIG:
1800                         if (!TCPMD5_ENABLED()) {
1801                                 INP_WUNLOCK(inp);
1802                                 return (ENOPROTOOPT);
1803                         }
1804                         error = TCPMD5_PCBCTL(inp, sopt);
1805                         if (error)
1806                                 return (error);
1807                         goto unlock_and_done;
1808 #endif /* IPSEC */
1809
1810                 case TCP_NODELAY:
1811                 case TCP_NOOPT:
1812                         INP_WUNLOCK(inp);
1813                         error = sooptcopyin(sopt, &optval, sizeof optval,
1814                             sizeof optval);
1815                         if (error)
1816                                 return (error);
1817
1818                         INP_WLOCK_RECHECK(inp);
1819                         switch (sopt->sopt_name) {
1820                         case TCP_NODELAY:
1821                                 opt = TF_NODELAY;
1822                                 break;
1823                         case TCP_NOOPT:
1824                                 opt = TF_NOOPT;
1825                                 break;
1826                         default:
1827                                 opt = 0; /* dead code to fool gcc */
1828                                 break;
1829                         }
1830
1831                         if (optval)
1832                                 tp->t_flags |= opt;
1833                         else
1834                                 tp->t_flags &= ~opt;
1835 unlock_and_done:
1836 #ifdef TCP_OFFLOAD
1837                         if (tp->t_flags & TF_TOE) {
1838                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1839                                     sopt->sopt_name);
1840                         }
1841 #endif
1842                         INP_WUNLOCK(inp);
1843                         break;
1844
1845                 case TCP_NOPUSH:
1846                         INP_WUNLOCK(inp);
1847                         error = sooptcopyin(sopt, &optval, sizeof optval,
1848                             sizeof optval);
1849                         if (error)
1850                                 return (error);
1851
1852                         INP_WLOCK_RECHECK(inp);
1853                         if (optval)
1854                                 tp->t_flags |= TF_NOPUSH;
1855                         else if (tp->t_flags & TF_NOPUSH) {
1856                                 tp->t_flags &= ~TF_NOPUSH;
1857                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1858                                         error = tp->t_fb->tfb_tcp_output(tp);
1859                         }
1860                         goto unlock_and_done;
1861
1862                 case TCP_MAXSEG:
1863                         INP_WUNLOCK(inp);
1864                         error = sooptcopyin(sopt, &optval, sizeof optval,
1865                             sizeof optval);
1866                         if (error)
1867                                 return (error);
1868
1869                         INP_WLOCK_RECHECK(inp);
1870                         if (optval > 0 && optval <= tp->t_maxseg &&
1871                             optval + 40 >= V_tcp_minmss)
1872                                 tp->t_maxseg = optval;
1873                         else
1874                                 error = EINVAL;
1875                         goto unlock_and_done;
1876
1877                 case TCP_INFO:
1878                         INP_WUNLOCK(inp);
1879                         error = EINVAL;
1880                         break;
1881
1882                 case TCP_CONGESTION:
1883                         INP_WUNLOCK(inp);
1884                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1885                         if (error)
1886                                 break;
1887                         buf[sopt->sopt_valsize] = '\0';
1888                         INP_WLOCK_RECHECK(inp);
1889                         CC_LIST_RLOCK();
1890                         STAILQ_FOREACH(algo, &cc_list, entries)
1891                                 if (strncmp(buf, algo->name,
1892                                     TCP_CA_NAME_MAX) == 0)
1893                                         break;
1894                         CC_LIST_RUNLOCK();
1895                         if (algo == NULL) {
1896                                 INP_WUNLOCK(inp);
1897                                 error = EINVAL;
1898                                 break;
1899                         }
1900                         /*
1901                          * We hold a write lock over the tcb so it's safe to
1902                          * do these things without ordering concerns.
1903                          */
1904                         if (CC_ALGO(tp)->cb_destroy != NULL)
1905                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1906                         CC_DATA(tp) = NULL;
1907                         CC_ALGO(tp) = algo;
1908                         /*
1909                          * If something goes pear shaped initialising the new
1910                          * algo, fall back to newreno (which does not
1911                          * require initialisation).
1912                          */
1913                         if (algo->cb_init != NULL &&
1914                             algo->cb_init(tp->ccv) != 0) {
1915                                 CC_ALGO(tp) = &newreno_cc_algo;
1916                                 /*
1917                                  * The only reason init should fail is
1918                                  * because of malloc.
1919                                  */
1920                                 error = ENOMEM;
1921                         }
1922                         INP_WUNLOCK(inp);
1923                         break;
1924
1925 #ifdef KERN_TLS
1926                 case TCP_TXTLS_ENABLE:
1927                         INP_WUNLOCK(inp);
1928                         error = sooptcopyin(sopt, &tls, sizeof(tls),
1929                             sizeof(tls));
1930                         if (error)
1931                                 break;
1932                         error = ktls_enable_tx(so, &tls);
1933                         break;
1934                 case TCP_TXTLS_MODE:
1935                         INP_WUNLOCK(inp);
1936                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1937                         if (error)
1938                                 return (error);
1939                         if (ui != TCP_TLS_MODE_SW && ui != TCP_TLS_MODE_IFNET)
1940                                 return (EINVAL);
1941
1942                         INP_WLOCK_RECHECK(inp);
1943                         error = ktls_set_tx_mode(so, ui);
1944                         INP_WUNLOCK(inp);
1945                         break;
1946 #endif
1947
1948                 case TCP_KEEPIDLE:
1949                 case TCP_KEEPINTVL:
1950                 case TCP_KEEPINIT:
1951                         INP_WUNLOCK(inp);
1952                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1953                         if (error)
1954                                 return (error);
1955
1956                         if (ui > (UINT_MAX / hz)) {
1957                                 error = EINVAL;
1958                                 break;
1959                         }
1960                         ui *= hz;
1961
1962                         INP_WLOCK_RECHECK(inp);
1963                         switch (sopt->sopt_name) {
1964                         case TCP_KEEPIDLE:
1965                                 tp->t_keepidle = ui;
1966                                 /*
1967                                  * XXX: better check current remaining
1968                                  * timeout and "merge" it with new value.
1969                                  */
1970                                 if ((tp->t_state > TCPS_LISTEN) &&
1971                                     (tp->t_state <= TCPS_CLOSING))
1972                                         tcp_timer_activate(tp, TT_KEEP,
1973                                             TP_KEEPIDLE(tp));
1974                                 break;
1975                         case TCP_KEEPINTVL:
1976                                 tp->t_keepintvl = ui;
1977                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1978                                     (TP_MAXIDLE(tp) > 0))
1979                                         tcp_timer_activate(tp, TT_2MSL,
1980                                             TP_MAXIDLE(tp));
1981                                 break;
1982                         case TCP_KEEPINIT:
1983                                 tp->t_keepinit = ui;
1984                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1985                                     tp->t_state == TCPS_SYN_SENT)
1986                                         tcp_timer_activate(tp, TT_KEEP,
1987                                             TP_KEEPINIT(tp));
1988                                 break;
1989                         }
1990                         goto unlock_and_done;
1991
1992                 case TCP_KEEPCNT:
1993                         INP_WUNLOCK(inp);
1994                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1995                         if (error)
1996                                 return (error);
1997
1998                         INP_WLOCK_RECHECK(inp);
1999                         tp->t_keepcnt = ui;
2000                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2001                             (TP_MAXIDLE(tp) > 0))
2002                                 tcp_timer_activate(tp, TT_2MSL,
2003                                     TP_MAXIDLE(tp));
2004                         goto unlock_and_done;
2005
2006 #ifdef TCPPCAP
2007                 case TCP_PCAP_OUT:
2008                 case TCP_PCAP_IN:
2009                         INP_WUNLOCK(inp);
2010                         error = sooptcopyin(sopt, &optval, sizeof optval,
2011                             sizeof optval);
2012                         if (error)
2013                                 return (error);
2014
2015                         INP_WLOCK_RECHECK(inp);
2016                         if (optval >= 0)
2017                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2018                                         &(tp->t_outpkts) : &(tp->t_inpkts),
2019                                         optval);
2020                         else
2021                                 error = EINVAL;
2022                         goto unlock_and_done;
2023 #endif
2024
2025                 case TCP_FASTOPEN: {
2026                         struct tcp_fastopen tfo_optval;
2027
2028                         INP_WUNLOCK(inp);
2029                         if (!V_tcp_fastopen_client_enable &&
2030                             !V_tcp_fastopen_server_enable)
2031                                 return (EPERM);
2032
2033                         error = sooptcopyin(sopt, &tfo_optval,
2034                                     sizeof(tfo_optval), sizeof(int));
2035                         if (error)
2036                                 return (error);
2037
2038                         INP_WLOCK_RECHECK(inp);
2039                         if (tfo_optval.enable) {
2040                                 if (tp->t_state == TCPS_LISTEN) {
2041                                         if (!V_tcp_fastopen_server_enable) {
2042                                                 error = EPERM;
2043                                                 goto unlock_and_done;
2044                                         }
2045
2046                                         tp->t_flags |= TF_FASTOPEN;
2047                                         if (tp->t_tfo_pending == NULL)
2048                                                 tp->t_tfo_pending =
2049                                                     tcp_fastopen_alloc_counter();
2050                                 } else {
2051                                         /*
2052                                          * If a pre-shared key was provided,
2053                                          * stash it in the client cookie
2054                                          * field of the tcpcb for use during
2055                                          * connect.
2056                                          */
2057                                         if (sopt->sopt_valsize ==
2058                                             sizeof(tfo_optval)) {
2059                                                 memcpy(tp->t_tfo_cookie.client,
2060                                                        tfo_optval.psk,
2061                                                        TCP_FASTOPEN_PSK_LEN);
2062                                                 tp->t_tfo_client_cookie_len =
2063                                                     TCP_FASTOPEN_PSK_LEN;
2064                                         }
2065                                         tp->t_flags |= TF_FASTOPEN;
2066                                 }
2067                         } else
2068                                 tp->t_flags &= ~TF_FASTOPEN;
2069                         goto unlock_and_done;
2070                 }
2071
2072 #ifdef TCP_BLACKBOX
2073                 case TCP_LOG:
2074                         INP_WUNLOCK(inp);
2075                         error = sooptcopyin(sopt, &optval, sizeof optval,
2076                             sizeof optval);
2077                         if (error)
2078                                 return (error);
2079
2080                         INP_WLOCK_RECHECK(inp);
2081                         error = tcp_log_state_change(tp, optval);
2082                         goto unlock_and_done;
2083
2084                 case TCP_LOGBUF:
2085                         INP_WUNLOCK(inp);
2086                         error = EINVAL;
2087                         break;
2088
2089                 case TCP_LOGID:
2090                         INP_WUNLOCK(inp);
2091                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2092                         if (error)
2093                                 break;
2094                         buf[sopt->sopt_valsize] = '\0';
2095                         INP_WLOCK_RECHECK(inp);
2096                         error = tcp_log_set_id(tp, buf);
2097                         /* tcp_log_set_id() unlocks the INP. */
2098                         break;
2099
2100                 case TCP_LOGDUMP:
2101                 case TCP_LOGDUMPID:
2102                         INP_WUNLOCK(inp);
2103                         error =
2104                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2105                         if (error)
2106                                 break;
2107                         buf[sopt->sopt_valsize] = '\0';
2108                         INP_WLOCK_RECHECK(inp);
2109                         if (sopt->sopt_name == TCP_LOGDUMP) {
2110                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2111                                     M_WAITOK, true);
2112                                 INP_WUNLOCK(inp);
2113                         } else {
2114                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2115                                 /*
2116                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2117                                  * INP lock.
2118                                  */
2119                         }
2120                         break;
2121 #endif
2122
2123                 default:
2124                         INP_WUNLOCK(inp);
2125                         error = ENOPROTOOPT;
2126                         break;
2127                 }
2128                 break;
2129
2130         case SOPT_GET:
2131                 tp = intotcpcb(inp);
2132                 switch (sopt->sopt_name) {
2133 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2134                 case TCP_MD5SIG:
2135                         if (!TCPMD5_ENABLED()) {
2136                                 INP_WUNLOCK(inp);
2137                                 return (ENOPROTOOPT);
2138                         }
2139                         error = TCPMD5_PCBCTL(inp, sopt);
2140                         break;
2141 #endif
2142
2143                 case TCP_NODELAY:
2144                         optval = tp->t_flags & TF_NODELAY;
2145                         INP_WUNLOCK(inp);
2146                         error = sooptcopyout(sopt, &optval, sizeof optval);
2147                         break;
2148                 case TCP_MAXSEG:
2149                         optval = tp->t_maxseg;
2150                         INP_WUNLOCK(inp);
2151                         error = sooptcopyout(sopt, &optval, sizeof optval);
2152                         break;
2153                 case TCP_NOOPT:
2154                         optval = tp->t_flags & TF_NOOPT;
2155                         INP_WUNLOCK(inp);
2156                         error = sooptcopyout(sopt, &optval, sizeof optval);
2157                         break;
2158                 case TCP_NOPUSH:
2159                         optval = tp->t_flags & TF_NOPUSH;
2160                         INP_WUNLOCK(inp);
2161                         error = sooptcopyout(sopt, &optval, sizeof optval);
2162                         break;
2163                 case TCP_INFO:
2164                         tcp_fill_info(tp, &ti);
2165                         INP_WUNLOCK(inp);
2166                         error = sooptcopyout(sopt, &ti, sizeof ti);
2167                         break;
2168                 case TCP_CONGESTION:
2169                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2170                         INP_WUNLOCK(inp);
2171                         error = sooptcopyout(sopt, buf, len + 1);
2172                         break;
2173                 case TCP_KEEPIDLE:
2174                 case TCP_KEEPINTVL:
2175                 case TCP_KEEPINIT:
2176                 case TCP_KEEPCNT:
2177                         switch (sopt->sopt_name) {
2178                         case TCP_KEEPIDLE:
2179                                 ui = TP_KEEPIDLE(tp) / hz;
2180                                 break;
2181                         case TCP_KEEPINTVL:
2182                                 ui = TP_KEEPINTVL(tp) / hz;
2183                                 break;
2184                         case TCP_KEEPINIT:
2185                                 ui = TP_KEEPINIT(tp) / hz;
2186                                 break;
2187                         case TCP_KEEPCNT:
2188                                 ui = TP_KEEPCNT(tp);
2189                                 break;
2190                         }
2191                         INP_WUNLOCK(inp);
2192                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2193                         break;
2194 #ifdef TCPPCAP
2195                 case TCP_PCAP_OUT:
2196                 case TCP_PCAP_IN:
2197                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2198                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2199                         INP_WUNLOCK(inp);
2200                         error = sooptcopyout(sopt, &optval, sizeof optval);
2201                         break;
2202 #endif
2203                 case TCP_FASTOPEN:
2204                         optval = tp->t_flags & TF_FASTOPEN;
2205                         INP_WUNLOCK(inp);
2206                         error = sooptcopyout(sopt, &optval, sizeof optval);
2207                         break;
2208 #ifdef TCP_BLACKBOX
2209                 case TCP_LOG:
2210                         optval = tp->t_logstate;
2211                         INP_WUNLOCK(inp);
2212                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2213                         break;
2214                 case TCP_LOGBUF:
2215                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2216                         error = tcp_log_getlogbuf(sopt, tp);
2217                         break;
2218                 case TCP_LOGID:
2219                         len = tcp_log_get_id(tp, buf);
2220                         INP_WUNLOCK(inp);
2221                         error = sooptcopyout(sopt, buf, len + 1);
2222                         break;
2223                 case TCP_LOGDUMP:
2224                 case TCP_LOGDUMPID:
2225                         INP_WUNLOCK(inp);
2226                         error = EINVAL;
2227                         break;
2228 #endif
2229 #ifdef KERN_TLS
2230                 case TCP_TXTLS_MODE:
2231                         optval = ktls_get_tx_mode(so);
2232                         INP_WUNLOCK(inp);
2233                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2234                         break;
2235 #endif
2236                 default:
2237                         INP_WUNLOCK(inp);
2238                         error = ENOPROTOOPT;
2239                         break;
2240                 }
2241                 break;
2242         }
2243         return (error);
2244 }
2245 #undef INP_WLOCK_RECHECK
2246 #undef INP_WLOCK_RECHECK_CLEANUP
2247
2248 /*
2249  * Attach TCP protocol to socket, allocating
2250  * internet protocol control block, tcp control block,
2251  * bufer space, and entering LISTEN state if to accept connections.
2252  */
2253 static int
2254 tcp_attach(struct socket *so)
2255 {
2256         struct tcpcb *tp;
2257         struct inpcb *inp;
2258         struct epoch_tracker et;
2259         int error;
2260
2261         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2262                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
2263                 if (error)
2264                         return (error);
2265         }
2266         so->so_rcv.sb_flags |= SB_AUTOSIZE;
2267         so->so_snd.sb_flags |= SB_AUTOSIZE;
2268         INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2269         error = in_pcballoc(so, &V_tcbinfo);
2270         if (error) {
2271                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2272                 return (error);
2273         }
2274         inp = sotoinpcb(so);
2275 #ifdef INET6
2276         if (inp->inp_vflag & INP_IPV6PROTO) {
2277                 inp->inp_vflag |= INP_IPV6;
2278                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
2279                         inp->inp_vflag |= INP_IPV4;
2280                 inp->in6p_hops = -1;    /* use kernel default */
2281         }
2282         else
2283 #endif
2284         inp->inp_vflag |= INP_IPV4;
2285         tp = tcp_newtcpcb(inp);
2286         if (tp == NULL) {
2287                 in_pcbdetach(inp);
2288                 in_pcbfree(inp);
2289                 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2290                 return (ENOBUFS);
2291         }
2292         tp->t_state = TCPS_CLOSED;
2293         INP_WUNLOCK(inp);
2294         INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2295         TCPSTATES_INC(TCPS_CLOSED);
2296         return (0);
2297 }
2298
2299 /*
2300  * Initiate (or continue) disconnect.
2301  * If embryonic state, just send reset (once).
2302  * If in ``let data drain'' option and linger null, just drop.
2303  * Otherwise (hard), mark socket disconnecting and drop
2304  * current input data; switch states based on user close, and
2305  * send segment to peer (with FIN).
2306  */
2307 static void
2308 tcp_disconnect(struct tcpcb *tp)
2309 {
2310         struct inpcb *inp = tp->t_inpcb;
2311         struct socket *so = inp->inp_socket;
2312
2313         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2314         INP_WLOCK_ASSERT(inp);
2315
2316         /*
2317          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2318          * socket is still open.
2319          */
2320         if (tp->t_state < TCPS_ESTABLISHED &&
2321             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2322                 tp = tcp_close(tp);
2323                 KASSERT(tp != NULL,
2324                     ("tcp_disconnect: tcp_close() returned NULL"));
2325         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2326                 tp = tcp_drop(tp, 0);
2327                 KASSERT(tp != NULL,
2328                     ("tcp_disconnect: tcp_drop() returned NULL"));
2329         } else {
2330                 soisdisconnecting(so);
2331                 sbflush(&so->so_rcv);
2332                 tcp_usrclosed(tp);
2333                 if (!(inp->inp_flags & INP_DROPPED))
2334                         tp->t_fb->tfb_tcp_output(tp);
2335         }
2336 }
2337
2338 /*
2339  * User issued close, and wish to trail through shutdown states:
2340  * if never received SYN, just forget it.  If got a SYN from peer,
2341  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2342  * If already got a FIN from peer, then almost done; go to LAST_ACK
2343  * state.  In all other cases, have already sent FIN to peer (e.g.
2344  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2345  * for peer to send FIN or not respond to keep-alives, etc.
2346  * We can let the user exit from the close as soon as the FIN is acked.
2347  */
2348 static void
2349 tcp_usrclosed(struct tcpcb *tp)
2350 {
2351
2352         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2353         INP_WLOCK_ASSERT(tp->t_inpcb);
2354
2355         switch (tp->t_state) {
2356         case TCPS_LISTEN:
2357 #ifdef TCP_OFFLOAD
2358                 tcp_offload_listen_stop(tp);
2359 #endif
2360                 tcp_state_change(tp, TCPS_CLOSED);
2361                 /* FALLTHROUGH */
2362         case TCPS_CLOSED:
2363                 tp = tcp_close(tp);
2364                 /*
2365                  * tcp_close() should never return NULL here as the socket is
2366                  * still open.
2367                  */
2368                 KASSERT(tp != NULL,
2369                     ("tcp_usrclosed: tcp_close() returned NULL"));
2370                 break;
2371
2372         case TCPS_SYN_SENT:
2373         case TCPS_SYN_RECEIVED:
2374                 tp->t_flags |= TF_NEEDFIN;
2375                 break;
2376
2377         case TCPS_ESTABLISHED:
2378                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2379                 break;
2380
2381         case TCPS_CLOSE_WAIT:
2382                 tcp_state_change(tp, TCPS_LAST_ACK);
2383                 break;
2384         }
2385         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2386                 soisdisconnected(tp->t_inpcb->inp_socket);
2387                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2388                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2389                         int timeout;
2390
2391                         timeout = (tcp_fast_finwait2_recycle) ?
2392                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2393                         tcp_timer_activate(tp, TT_2MSL, timeout);
2394                 }
2395         }
2396 }
2397
2398 #ifdef DDB
2399 static void
2400 db_print_indent(int indent)
2401 {
2402         int i;
2403
2404         for (i = 0; i < indent; i++)
2405                 db_printf(" ");
2406 }
2407
2408 static void
2409 db_print_tstate(int t_state)
2410 {
2411
2412         switch (t_state) {
2413         case TCPS_CLOSED:
2414                 db_printf("TCPS_CLOSED");
2415                 return;
2416
2417         case TCPS_LISTEN:
2418                 db_printf("TCPS_LISTEN");
2419                 return;
2420
2421         case TCPS_SYN_SENT:
2422                 db_printf("TCPS_SYN_SENT");
2423                 return;
2424
2425         case TCPS_SYN_RECEIVED:
2426                 db_printf("TCPS_SYN_RECEIVED");
2427                 return;
2428
2429         case TCPS_ESTABLISHED:
2430                 db_printf("TCPS_ESTABLISHED");
2431                 return;
2432
2433         case TCPS_CLOSE_WAIT:
2434                 db_printf("TCPS_CLOSE_WAIT");
2435                 return;
2436
2437         case TCPS_FIN_WAIT_1:
2438                 db_printf("TCPS_FIN_WAIT_1");
2439                 return;
2440
2441         case TCPS_CLOSING:
2442                 db_printf("TCPS_CLOSING");
2443                 return;
2444
2445         case TCPS_LAST_ACK:
2446                 db_printf("TCPS_LAST_ACK");
2447                 return;
2448
2449         case TCPS_FIN_WAIT_2:
2450                 db_printf("TCPS_FIN_WAIT_2");
2451                 return;
2452
2453         case TCPS_TIME_WAIT:
2454                 db_printf("TCPS_TIME_WAIT");
2455                 return;
2456
2457         default:
2458                 db_printf("unknown");
2459                 return;
2460         }
2461 }
2462
2463 static void
2464 db_print_tflags(u_int t_flags)
2465 {
2466         int comma;
2467
2468         comma = 0;
2469         if (t_flags & TF_ACKNOW) {
2470                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2471                 comma = 1;
2472         }
2473         if (t_flags & TF_DELACK) {
2474                 db_printf("%sTF_DELACK", comma ? ", " : "");
2475                 comma = 1;
2476         }
2477         if (t_flags & TF_NODELAY) {
2478                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2479                 comma = 1;
2480         }
2481         if (t_flags & TF_NOOPT) {
2482                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2483                 comma = 1;
2484         }
2485         if (t_flags & TF_SENTFIN) {
2486                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2487                 comma = 1;
2488         }
2489         if (t_flags & TF_REQ_SCALE) {
2490                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2491                 comma = 1;
2492         }
2493         if (t_flags & TF_RCVD_SCALE) {
2494                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2495                 comma = 1;
2496         }
2497         if (t_flags & TF_REQ_TSTMP) {
2498                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2499                 comma = 1;
2500         }
2501         if (t_flags & TF_RCVD_TSTMP) {
2502                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2503                 comma = 1;
2504         }
2505         if (t_flags & TF_SACK_PERMIT) {
2506                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2507                 comma = 1;
2508         }
2509         if (t_flags & TF_NEEDSYN) {
2510                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2511                 comma = 1;
2512         }
2513         if (t_flags & TF_NEEDFIN) {
2514                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2515                 comma = 1;
2516         }
2517         if (t_flags & TF_NOPUSH) {
2518                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2519                 comma = 1;
2520         }
2521         if (t_flags & TF_MORETOCOME) {
2522                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2523                 comma = 1;
2524         }
2525         if (t_flags & TF_LQ_OVERFLOW) {
2526                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2527                 comma = 1;
2528         }
2529         if (t_flags & TF_LASTIDLE) {
2530                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2531                 comma = 1;
2532         }
2533         if (t_flags & TF_RXWIN0SENT) {
2534                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2535                 comma = 1;
2536         }
2537         if (t_flags & TF_FASTRECOVERY) {
2538                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2539                 comma = 1;
2540         }
2541         if (t_flags & TF_CONGRECOVERY) {
2542                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2543                 comma = 1;
2544         }
2545         if (t_flags & TF_WASFRECOVERY) {
2546                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2547                 comma = 1;
2548         }
2549         if (t_flags & TF_SIGNATURE) {
2550                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2551                 comma = 1;
2552         }
2553         if (t_flags & TF_FORCEDATA) {
2554                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2555                 comma = 1;
2556         }
2557         if (t_flags & TF_TSO) {
2558                 db_printf("%sTF_TSO", comma ? ", " : "");
2559                 comma = 1;
2560         }
2561         if (t_flags & TF_ECN_PERMIT) {
2562                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2563                 comma = 1;
2564         }
2565         if (t_flags & TF_FASTOPEN) {
2566                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2567                 comma = 1;
2568         }
2569 }
2570
2571 static void
2572 db_print_toobflags(char t_oobflags)
2573 {
2574         int comma;
2575
2576         comma = 0;
2577         if (t_oobflags & TCPOOB_HAVEDATA) {
2578                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2579                 comma = 1;
2580         }
2581         if (t_oobflags & TCPOOB_HADDATA) {
2582                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2583                 comma = 1;
2584         }
2585 }
2586
2587 static void
2588 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2589 {
2590
2591         db_print_indent(indent);
2592         db_printf("%s at %p\n", name, tp);
2593
2594         indent += 2;
2595
2596         db_print_indent(indent);
2597         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2598            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2599
2600         db_print_indent(indent);
2601         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2602             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2603
2604         db_print_indent(indent);
2605         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2606             &tp->t_timers->tt_delack, tp->t_inpcb);
2607
2608         db_print_indent(indent);
2609         db_printf("t_state: %d (", tp->t_state);
2610         db_print_tstate(tp->t_state);
2611         db_printf(")\n");
2612
2613         db_print_indent(indent);
2614         db_printf("t_flags: 0x%x (", tp->t_flags);
2615         db_print_tflags(tp->t_flags);
2616         db_printf(")\n");
2617
2618         db_print_indent(indent);
2619         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2620             tp->snd_una, tp->snd_max, tp->snd_nxt);
2621
2622         db_print_indent(indent);
2623         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2624            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2625
2626         db_print_indent(indent);
2627         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2628             tp->iss, tp->irs, tp->rcv_nxt);
2629
2630         db_print_indent(indent);
2631         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2632             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2633
2634         db_print_indent(indent);
2635         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2636            tp->snd_wnd, tp->snd_cwnd);
2637
2638         db_print_indent(indent);
2639         db_printf("snd_ssthresh: %u   snd_recover: "
2640             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2641
2642         db_print_indent(indent);
2643         db_printf("t_rcvtime: %u   t_startime: %u\n",
2644             tp->t_rcvtime, tp->t_starttime);
2645
2646         db_print_indent(indent);
2647         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2648             tp->t_rtttime, tp->t_rtseq);
2649
2650         db_print_indent(indent);
2651         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2652             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2653
2654         db_print_indent(indent);
2655         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2656             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2657             tp->t_rttbest);
2658
2659         db_print_indent(indent);
2660         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2661             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2662
2663         db_print_indent(indent);
2664         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2665         db_print_toobflags(tp->t_oobflags);
2666         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2667
2668         db_print_indent(indent);
2669         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2670             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2671
2672         db_print_indent(indent);
2673         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2674             tp->ts_recent, tp->ts_recent_age);
2675
2676         db_print_indent(indent);
2677         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2678             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2679
2680         db_print_indent(indent);
2681         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2682             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2683             tp->snd_recover_prev, tp->t_badrxtwin);
2684
2685         db_print_indent(indent);
2686         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2687             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2688
2689         db_print_indent(indent);
2690         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2691             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2692
2693         /* Skip sackblks, sackhint. */
2694
2695         db_print_indent(indent);
2696         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2697             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2698 }
2699
2700 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2701 {
2702         struct tcpcb *tp;
2703
2704         if (!have_addr) {
2705                 db_printf("usage: show tcpcb <addr>\n");
2706                 return;
2707         }
2708         tp = (struct tcpcb *)addr;
2709
2710         db_print_tcpcb(tp, "tcpcb", 0);
2711 }
2712 #endif