sys/netinet/tcp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2006-2007 Robert N. M. Watson
   5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   6  * All rights reserved.
   7  *
   8  * Portions of this software were developed by Robert N. M. Watson under
   9  * contract to Juniper Networks, Inc.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 4. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_ddb.h"
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44 #include "opt_ipsec.h"
  45 #include "opt_tcpdebug.h"
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/limits.h>
  50 #include <sys/malloc.h>
  51 #include <sys/refcount.h>
  52 #include <sys/kernel.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/mbuf.h>
  55 #ifdef INET6
  56 #include <sys/domain.h>
  57 #endif /* INET6 */
  58 #include <sys/socket.h>
  59 #include <sys/socketvar.h>
  60 #include <sys/protosw.h>
  61 #include <sys/proc.h>
  62 #include <sys/jail.h>
  63 #include <sys/syslog.h>
  64
  65 #ifdef DDB
  66 #include <ddb/ddb.h>
  67 #endif
  68
  69 #include <net/if.h>
  70 #include <net/if_var.h>
  71 #include <net/route.h>
  72 #include <net/vnet.h>
  73
  74 #include <netinet/in.h>
  75 #include <netinet/in_kdtrace.h>
  76 #include <netinet/in_pcb.h>
  77 #include <netinet/in_systm.h>
  78 #include <netinet/in_var.h>
  79 #include <netinet/ip_var.h>
  80 #ifdef INET6
  81 #include <netinet/ip6.h>
  82 #include <netinet6/in6_pcb.h>
  83 #include <netinet6/ip6_var.h>
  84 #include <netinet6/scope6_var.h>
  85 #endif
  86 #ifdef TCP_RFC7413
  87 #include <netinet/tcp_fastopen.h>
  88 #endif
  89 #include <netinet/tcp.h>
  90 #include <netinet/tcp_fsm.h>
  91 #include <netinet/tcp_seq.h>
  92 #include <netinet/tcp_timer.h>
  93 #include <netinet/tcp_var.h>
  94 #include <netinet/tcpip.h>
  95 #include <netinet/cc/cc.h>
  96 #ifdef TCPPCAP
  97 #include <netinet/tcp_pcap.h>
  98 #endif
  99 #ifdef TCPDEBUG
 100 #include <netinet/tcp_debug.h>
 101 #endif
 102 #ifdef TCP_OFFLOAD
 103 #include <netinet/tcp_offload.h>
 104 #endif
 105 #include <netipsec/ipsec_support.h>
 106
 107 /*
 108  * TCP protocol interface to socket abstraction.
 109  */
 110 static int      tcp_attach(struct socket *);
 111 #ifdef INET
 112 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 113                     struct thread *td);
 114 #endif /* INET */
 115 #ifdef INET6
 116 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 117                     struct thread *td);
 118 #endif /* INET6 */
 119 static void     tcp_disconnect(struct tcpcb *);
 120 static void     tcp_usrclosed(struct tcpcb *);
 121 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 122
 123 #ifdef TCPDEBUG
 124 #define TCPDEBUG0       int ostate = 0
 125 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 126 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 127                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 128 #else
 129 #define TCPDEBUG0
 130 #define TCPDEBUG1()
 131 #define TCPDEBUG2(req)
 132 #endif
 133
 134 /*
 135  * TCP attaches to socket via pru_attach(), reserving space,
 136  * and an internet control block.
 137  */
 138 static int
 139 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 140 {
 141         struct inpcb *inp;
 142         struct tcpcb *tp = NULL;
 143         int error;
 144         TCPDEBUG0;
 145
 146         inp = sotoinpcb(so);
 147         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 148         TCPDEBUG1();
 149
 150         error = tcp_attach(so);
 151         if (error)
 152                 goto out;
 153
 154         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 155                 so->so_linger = TCP_LINGERTIME;
 156
 157         inp = sotoinpcb(so);
 158         tp = intotcpcb(inp);
 159 out:
 160         TCPDEBUG2(PRU_ATTACH);
 161         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 162         return error;
 163 }
 164
 165 /*
 166  * tcp_detach is called when the socket layer loses its final reference
 167  * to the socket, be it a file descriptor reference, a reference from TCP,
 168  * etc.  At this point, there is only one case in which we will keep around
 169  * inpcb state: time wait.
 170  *
 171  * This function can probably be re-absorbed back into tcp_usr_detach() now
 172  * that there is a single detach path.
 173  */
 174 static void
 175 tcp_detach(struct socket *so, struct inpcb *inp)
 176 {
 177         struct tcpcb *tp;
 178
 179         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
 180         INP_WLOCK_ASSERT(inp);
 181
 182         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 183         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 184
 185         tp = intotcpcb(inp);
 186
 187         if (inp->inp_flags & INP_TIMEWAIT) {
 188                 /*
 189                  * There are two cases to handle: one in which the time wait
 190                  * state is being discarded (INP_DROPPED), and one in which
 191                  * this connection will remain in timewait.  In the former,
 192                  * it is time to discard all state (except tcptw, which has
 193                  * already been discarded by the timewait close code, which
 194                  * should be further up the call stack somewhere).  In the
 195                  * latter case, we detach from the socket, but leave the pcb
 196                  * present until timewait ends.
 197                  *
 198                  * XXXRW: Would it be cleaner to free the tcptw here?
 199                  *
 200                  * Astute question indeed, from twtcp perspective there are
 201                  * three cases to consider:
 202                  *
 203                  * #1 tcp_detach is called at tcptw creation time by
 204                  *  tcp_twstart, then do not discard the newly created tcptw
 205                  *  and leave inpcb present until timewait ends
 206                  * #2 tcp_detach is called at timewait end (or reuse) by
 207                  *  tcp_twclose, then the tcptw has already been discarded
 208                  *  (or reused) and inpcb is freed here
 209                  * #3 tcp_detach is called() after timewait ends (or reuse)
 210                  *  (e.g. by soclose), then tcptw has already been discarded
 211                  *  (or reused) and inpcb is freed here
 212                  *
 213                  *  In all three cases the tcptw should not be freed here.
 214                  */
 215                 if (inp->inp_flags & INP_DROPPED) {
 216                         in_pcbdetach(inp);
 217                         if (__predict_true(tp == NULL)) {
 218                                 in_pcbfree(inp);
 219                         } else {
 220                                 /*
 221                                  * This case should not happen as in TIMEWAIT
 222                                  * state the inp should not be destroyed before
 223                                  * its tcptw.  If INVARIANTS is defined, panic.
 224                                  */
 225 #ifdef INVARIANTS
 226                                 panic("%s: Panic before an inp double-free: "
 227                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 228                                     , __func__);
 229 #else
 230                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 231                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 232                                     , __func__);
 233 #endif
 234                                 INP_WUNLOCK(inp);
 235                         }
 236                 } else {
 237                         in_pcbdetach(inp);
 238                         INP_WUNLOCK(inp);
 239                 }
 240         } else {
 241                 /*
 242                  * If the connection is not in timewait, we consider two
 243                  * two conditions: one in which no further processing is
 244                  * necessary (dropped || embryonic), and one in which TCP is
 245                  * not yet done, but no longer requires the socket, so the
 246                  * pcb will persist for the time being.
 247                  *
 248                  * XXXRW: Does the second case still occur?
 249                  */
 250                 if (inp->inp_flags & INP_DROPPED ||
 251                     tp->t_state < TCPS_SYN_SENT) {
 252                         tcp_discardcb(tp);
 253                         in_pcbdetach(inp);
 254                         in_pcbfree(inp);
 255                 } else {
 256                         in_pcbdetach(inp);
 257                         INP_WUNLOCK(inp);
 258                 }
 259         }
 260 }
 261
 262 /*
 263  * pru_detach() detaches the TCP protocol from the socket.
 264  * If the protocol state is non-embryonic, then can't
 265  * do this directly: have to initiate a pru_disconnect(),
 266  * which may finish later; embryonic TCB's can just
 267  * be discarded here.
 268  */
 269 static void
 270 tcp_usr_detach(struct socket *so)
 271 {
 272         struct inpcb *inp;
 273         int rlock = 0;
 274
 275         inp = sotoinpcb(so);
 276         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 277         if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
 278                 INP_INFO_RLOCK(&V_tcbinfo);
 279                 rlock = 1;
 280         }
 281         INP_WLOCK(inp);
 282         KASSERT(inp->inp_socket != NULL,
 283             ("tcp_usr_detach: inp_socket == NULL"));
 284         tcp_detach(so, inp);
 285         if (rlock)
 286                 INP_INFO_RUNLOCK(&V_tcbinfo);
 287 }
 288
 289 #ifdef INET
 290 /*
 291  * Give the socket an address.
 292  */
 293 static int
 294 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 295 {
 296         int error = 0;
 297         struct inpcb *inp;
 298         struct tcpcb *tp = NULL;
 299         struct sockaddr_in *sinp;
 300
 301         sinp = (struct sockaddr_in *)nam;
 302         if (nam->sa_len != sizeof (*sinp))
 303                 return (EINVAL);
 304         /*
 305          * Must check for multicast addresses and disallow binding
 306          * to them.
 307          */
 308         if (sinp->sin_family == AF_INET &&
 309             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 310                 return (EAFNOSUPPORT);
 311
 312         TCPDEBUG0;
 313         inp = sotoinpcb(so);
 314         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 315         INP_WLOCK(inp);
 316         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 317                 error = EINVAL;
 318                 goto out;
 319         }
 320         tp = intotcpcb(inp);
 321         TCPDEBUG1();
 322         INP_HASH_WLOCK(&V_tcbinfo);
 323         error = in_pcbbind(inp, nam, td->td_ucred);
 324         INP_HASH_WUNLOCK(&V_tcbinfo);
 325 out:
 326         TCPDEBUG2(PRU_BIND);
 327         TCP_PROBE2(debug__user, tp, PRU_BIND);
 328         INP_WUNLOCK(inp);
 329
 330         return (error);
 331 }
 332 #endif /* INET */
 333
 334 #ifdef INET6
 335 static int
 336 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 337 {
 338         int error = 0;
 339         struct inpcb *inp;
 340         struct tcpcb *tp = NULL;
 341         struct sockaddr_in6 *sin6p;
 342         u_char vflagsav;
 343
 344         sin6p = (struct sockaddr_in6 *)nam;
 345         if (nam->sa_len != sizeof (*sin6p))
 346                 return (EINVAL);
 347         /*
 348          * Must check for multicast addresses and disallow binding
 349          * to them.
 350          */
 351         if (sin6p->sin6_family == AF_INET6 &&
 352             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 353                 return (EAFNOSUPPORT);
 354
 355         TCPDEBUG0;
 356         inp = sotoinpcb(so);
 357         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 358         INP_WLOCK(inp);
 359         vflagsav = inp->inp_vflag;
 360         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 361                 error = EINVAL;
 362                 goto out;
 363         }
 364         tp = intotcpcb(inp);
 365         TCPDEBUG1();
 366         INP_HASH_WLOCK(&V_tcbinfo);
 367         inp->inp_vflag &= ~INP_IPV4;
 368         inp->inp_vflag |= INP_IPV6;
 369 #ifdef INET
 370         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 371                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 372                         inp->inp_vflag |= INP_IPV4;
 373                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 374                         struct sockaddr_in sin;
 375
 376                         in6_sin6_2_sin(&sin, sin6p);
 377                         inp->inp_vflag |= INP_IPV4;
 378                         inp->inp_vflag &= ~INP_IPV6;
 379                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 380                             td->td_ucred);
 381                         INP_HASH_WUNLOCK(&V_tcbinfo);
 382                         goto out;
 383                 }
 384         }
 385 #endif
 386         error = in6_pcbbind(inp, nam, td->td_ucred);
 387         INP_HASH_WUNLOCK(&V_tcbinfo);
 388 out:
 389         if (error != 0)
 390                 inp->inp_vflag = vflagsav;
 391         TCPDEBUG2(PRU_BIND);
 392         TCP_PROBE2(debug__user, tp, PRU_BIND);
 393         INP_WUNLOCK(inp);
 394         return (error);
 395 }
 396 #endif /* INET6 */
 397
 398 #ifdef INET
 399 /*
 400  * Prepare to accept connections.
 401  */
 402 static int
 403 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 404 {
 405         int error = 0;
 406         struct inpcb *inp;
 407         struct tcpcb *tp = NULL;
 408
 409         TCPDEBUG0;
 410         inp = sotoinpcb(so);
 411         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 412         INP_WLOCK(inp);
 413         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 414                 error = EINVAL;
 415                 goto out;
 416         }
 417         tp = intotcpcb(inp);
 418         TCPDEBUG1();
 419         SOCK_LOCK(so);
 420         error = solisten_proto_check(so);
 421         INP_HASH_WLOCK(&V_tcbinfo);
 422         if (error == 0 && inp->inp_lport == 0)
 423                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 424         INP_HASH_WUNLOCK(&V_tcbinfo);
 425         if (error == 0) {
 426                 tcp_state_change(tp, TCPS_LISTEN);
 427                 solisten_proto(so, backlog);
 428 #ifdef TCP_OFFLOAD
 429                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 430                         tcp_offload_listen_start(tp);
 431 #endif
 432         }
 433         SOCK_UNLOCK(so);
 434
 435 #ifdef TCP_RFC7413
 436         if (tp->t_flags & TF_FASTOPEN)
 437                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 438 #endif
 439 out:
 440         TCPDEBUG2(PRU_LISTEN);
 441         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 442         INP_WUNLOCK(inp);
 443         return (error);
 444 }
 445 #endif /* INET */
 446
 447 #ifdef INET6
 448 static int
 449 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 450 {
 451         int error = 0;
 452         struct inpcb *inp;
 453         struct tcpcb *tp = NULL;
 454         u_char vflagsav;
 455
 456         TCPDEBUG0;
 457         inp = sotoinpcb(so);
 458         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 459         INP_WLOCK(inp);
 460         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 461                 error = EINVAL;
 462                 goto out;
 463         }
 464         vflagsav = inp->inp_vflag;
 465         tp = intotcpcb(inp);
 466         TCPDEBUG1();
 467         SOCK_LOCK(so);
 468         error = solisten_proto_check(so);
 469         INP_HASH_WLOCK(&V_tcbinfo);
 470         if (error == 0 && inp->inp_lport == 0) {
 471                 inp->inp_vflag &= ~INP_IPV4;
 472                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 473                         inp->inp_vflag |= INP_IPV4;
 474                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 475         }
 476         INP_HASH_WUNLOCK(&V_tcbinfo);
 477         if (error == 0) {
 478                 tcp_state_change(tp, TCPS_LISTEN);
 479                 solisten_proto(so, backlog);
 480 #ifdef TCP_OFFLOAD
 481                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 482                         tcp_offload_listen_start(tp);
 483 #endif
 484         }
 485         SOCK_UNLOCK(so);
 486
 487 #ifdef TCP_RFC7413
 488         if (tp->t_flags & TF_FASTOPEN)
 489                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 490 #endif
 491         if (error != 0)
 492                 inp->inp_vflag = vflagsav;
 493
 494 out:
 495         TCPDEBUG2(PRU_LISTEN);
 496         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 497         INP_WUNLOCK(inp);
 498         return (error);
 499 }
 500 #endif /* INET6 */
 501
 502 #ifdef INET
 503 /*
 504  * Initiate connection to peer.
 505  * Create a template for use in transmissions on this connection.
 506  * Enter SYN_SENT state, and mark socket as connecting.
 507  * Start keep-alive timer, and seed output sequence space.
 508  * Send initial segment on connection.
 509  */
 510 static int
 511 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 512 {
 513         int error = 0;
 514         struct inpcb *inp;
 515         struct tcpcb *tp = NULL;
 516         struct sockaddr_in *sinp;
 517
 518         sinp = (struct sockaddr_in *)nam;
 519         if (nam->sa_len != sizeof (*sinp))
 520                 return (EINVAL);
 521         /*
 522          * Must disallow TCP ``connections'' to multicast addresses.
 523          */
 524         if (sinp->sin_family == AF_INET
 525             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 526                 return (EAFNOSUPPORT);
 527         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 528                 return (error);
 529
 530         TCPDEBUG0;
 531         inp = sotoinpcb(so);
 532         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 533         INP_WLOCK(inp);
 534         if (inp->inp_flags & INP_TIMEWAIT) {
 535                 error = EADDRINUSE;
 536                 goto out;
 537         }
 538         if (inp->inp_flags & INP_DROPPED) {
 539                 error = ECONNREFUSED;
 540                 goto out;
 541         }
 542         tp = intotcpcb(inp);
 543         TCPDEBUG1();
 544         if ((error = tcp_connect(tp, nam, td)) != 0)
 545                 goto out;
 546 #ifdef TCP_OFFLOAD
 547         if (registered_toedevs > 0 &&
 548             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 549             (error = tcp_offload_connect(so, nam)) == 0)
 550                 goto out;
 551 #endif
 552         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 553         error = tp->t_fb->tfb_tcp_output(tp);
 554 out:
 555         TCPDEBUG2(PRU_CONNECT);
 556         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 557         INP_WUNLOCK(inp);
 558         return (error);
 559 }
 560 #endif /* INET */
 561
 562 #ifdef INET6
 563 static int
 564 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 565 {
 566         int error = 0;
 567         struct inpcb *inp;
 568         struct tcpcb *tp = NULL;
 569         struct sockaddr_in6 *sin6p;
 570         u_int8_t incflagsav;
 571         u_char vflagsav;
 572
 573         TCPDEBUG0;
 574
 575         sin6p = (struct sockaddr_in6 *)nam;
 576         if (nam->sa_len != sizeof (*sin6p))
 577                 return (EINVAL);
 578         /*
 579          * Must disallow TCP ``connections'' to multicast addresses.
 580          */
 581         if (sin6p->sin6_family == AF_INET6
 582             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 583                 return (EAFNOSUPPORT);
 584
 585         inp = sotoinpcb(so);
 586         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 587         INP_WLOCK(inp);
 588         vflagsav = inp->inp_vflag;
 589         incflagsav = inp->inp_inc.inc_flags;
 590         if (inp->inp_flags & INP_TIMEWAIT) {
 591                 error = EADDRINUSE;
 592                 goto out;
 593         }
 594         if (inp->inp_flags & INP_DROPPED) {
 595                 error = ECONNREFUSED;
 596                 goto out;
 597         }
 598         tp = intotcpcb(inp);
 599         TCPDEBUG1();
 600 #ifdef INET
 601         /*
 602          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 603          * therefore probably require the hash lock, which isn't held here.
 604          * Is this a significant problem?
 605          */
 606         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 607                 struct sockaddr_in sin;
 608
 609                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 610                         error = EINVAL;
 611                         goto out;
 612                 }
 613                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 614                         error = EAFNOSUPPORT;
 615                         goto out;
 616                 }
 617
 618                 in6_sin6_2_sin(&sin, sin6p);
 619                 if ((error = prison_remote_ip4(td->td_ucred,
 620                     &sin.sin_addr)) != 0)
 621                         goto out;
 622                 inp->inp_vflag |= INP_IPV4;
 623                 inp->inp_vflag &= ~INP_IPV6;
 624                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 625                         goto out;
 626 #ifdef TCP_OFFLOAD
 627                 if (registered_toedevs > 0 &&
 628                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 629                     (error = tcp_offload_connect(so, nam)) == 0)
 630                         goto out;
 631 #endif
 632                 error = tp->t_fb->tfb_tcp_output(tp);
 633                 goto out;
 634         } else {
 635                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 636                         error = EAFNOSUPPORT;
 637                         goto out;
 638                 }
 639         }
 640 #endif
 641         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 642                 goto out;
 643         inp->inp_vflag &= ~INP_IPV4;
 644         inp->inp_vflag |= INP_IPV6;
 645         inp->inp_inc.inc_flags |= INC_ISIPV6;
 646         if ((error = tcp6_connect(tp, nam, td)) != 0)
 647                 goto out;
 648 #ifdef TCP_OFFLOAD
 649         if (registered_toedevs > 0 &&
 650             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 651             (error = tcp_offload_connect(so, nam)) == 0)
 652                 goto out;
 653 #endif
 654         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 655         error = tp->t_fb->tfb_tcp_output(tp);
 656
 657 out:
 658         /*
 659          * If the implicit bind in the connect call fails, restore
 660          * the flags we modified.
 661          */
 662         if (error != 0 && inp->inp_lport == 0) {
 663                 inp->inp_vflag = vflagsav;
 664                 inp->inp_inc.inc_flags = incflagsav;
 665         }
 666
 667         TCPDEBUG2(PRU_CONNECT);
 668         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 669         INP_WUNLOCK(inp);
 670         return (error);
 671 }
 672 #endif /* INET6 */
 673
 674 /*
 675  * Initiate disconnect from peer.
 676  * If connection never passed embryonic stage, just drop;
 677  * else if don't need to let data drain, then can just drop anyways,
 678  * else have to begin TCP shutdown process: mark socket disconnecting,
 679  * drain unread data, state switch to reflect user close, and
 680  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 681  * when peer sends FIN and acks ours.
 682  *
 683  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 684  */
 685 static int
 686 tcp_usr_disconnect(struct socket *so)
 687 {
 688         struct inpcb *inp;
 689         struct tcpcb *tp = NULL;
 690         int error = 0;
 691
 692         TCPDEBUG0;
 693         INP_INFO_RLOCK(&V_tcbinfo);
 694         inp = sotoinpcb(so);
 695         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 696         INP_WLOCK(inp);
 697         if (inp->inp_flags & INP_TIMEWAIT)
 698                 goto out;
 699         if (inp->inp_flags & INP_DROPPED) {
 700                 error = ECONNRESET;
 701                 goto out;
 702         }
 703         tp = intotcpcb(inp);
 704         TCPDEBUG1();
 705         tcp_disconnect(tp);
 706 out:
 707         TCPDEBUG2(PRU_DISCONNECT);
 708         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 709         INP_WUNLOCK(inp);
 710         INP_INFO_RUNLOCK(&V_tcbinfo);
 711         return (error);
 712 }
 713
 714 #ifdef INET
 715 /*
 716  * Accept a connection.  Essentially all the work is done at higher levels;
 717  * just return the address of the peer, storing through addr.
 718  */
 719 static int
 720 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 721 {
 722         int error = 0;
 723         struct inpcb *inp = NULL;
 724         struct tcpcb *tp = NULL;
 725         struct in_addr addr;
 726         in_port_t port = 0;
 727         TCPDEBUG0;
 728
 729         if (so->so_state & SS_ISDISCONNECTED)
 730                 return (ECONNABORTED);
 731
 732         inp = sotoinpcb(so);
 733         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 734         INP_WLOCK(inp);
 735         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 736                 error = ECONNABORTED;
 737                 goto out;
 738         }
 739         tp = intotcpcb(inp);
 740         TCPDEBUG1();
 741
 742         /*
 743          * We inline in_getpeeraddr and COMMON_END here, so that we can
 744          * copy the data of interest and defer the malloc until after we
 745          * release the lock.
 746          */
 747         port = inp->inp_fport;
 748         addr = inp->inp_faddr;
 749
 750 out:
 751         TCPDEBUG2(PRU_ACCEPT);
 752         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 753         INP_WUNLOCK(inp);
 754         if (error == 0)
 755                 *nam = in_sockaddr(port, &addr);
 756         return error;
 757 }
 758 #endif /* INET */
 759
 760 #ifdef INET6
 761 static int
 762 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 763 {
 764         struct inpcb *inp = NULL;
 765         int error = 0;
 766         struct tcpcb *tp = NULL;
 767         struct in_addr addr;
 768         struct in6_addr addr6;
 769         in_port_t port = 0;
 770         int v4 = 0;
 771         TCPDEBUG0;
 772
 773         if (so->so_state & SS_ISDISCONNECTED)
 774                 return (ECONNABORTED);
 775
 776         inp = sotoinpcb(so);
 777         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 778         INP_INFO_RLOCK(&V_tcbinfo);
 779         INP_WLOCK(inp);
 780         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 781                 error = ECONNABORTED;
 782                 goto out;
 783         }
 784         tp = intotcpcb(inp);
 785         TCPDEBUG1();
 786
 787         /*
 788          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 789          * copy the data of interest and defer the malloc until after we
 790          * release the lock.
 791          */
 792         if (inp->inp_vflag & INP_IPV4) {
 793                 v4 = 1;
 794                 port = inp->inp_fport;
 795                 addr = inp->inp_faddr;
 796         } else {
 797                 port = inp->inp_fport;
 798                 addr6 = inp->in6p_faddr;
 799         }
 800
 801 out:
 802         TCPDEBUG2(PRU_ACCEPT);
 803         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 804         INP_WUNLOCK(inp);
 805         INP_INFO_RUNLOCK(&V_tcbinfo);
 806         if (error == 0) {
 807                 if (v4)
 808                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 809                 else
 810                         *nam = in6_sockaddr(port, &addr6);
 811         }
 812         return error;
 813 }
 814 #endif /* INET6 */
 815
 816 /*
 817  * Mark the connection as being incapable of further output.
 818  */
 819 static int
 820 tcp_usr_shutdown(struct socket *so)
 821 {
 822         int error = 0;
 823         struct inpcb *inp;
 824         struct tcpcb *tp = NULL;
 825
 826         TCPDEBUG0;
 827         INP_INFO_RLOCK(&V_tcbinfo);
 828         inp = sotoinpcb(so);
 829         KASSERT(inp != NULL, ("inp == NULL"));
 830         INP_WLOCK(inp);
 831         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 832                 error = ECONNRESET;
 833                 goto out;
 834         }
 835         tp = intotcpcb(inp);
 836         TCPDEBUG1();
 837         socantsendmore(so);
 838         tcp_usrclosed(tp);
 839         if (!(inp->inp_flags & INP_DROPPED))
 840                 error = tp->t_fb->tfb_tcp_output(tp);
 841
 842 out:
 843         TCPDEBUG2(PRU_SHUTDOWN);
 844         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 845         INP_WUNLOCK(inp);
 846         INP_INFO_RUNLOCK(&V_tcbinfo);
 847
 848         return (error);
 849 }
 850
 851 /*
 852  * After a receive, possibly send window update to peer.
 853  */
 854 static int
 855 tcp_usr_rcvd(struct socket *so, int flags)
 856 {
 857         struct inpcb *inp;
 858         struct tcpcb *tp = NULL;
 859         int error = 0;
 860
 861         TCPDEBUG0;
 862         inp = sotoinpcb(so);
 863         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 864         INP_WLOCK(inp);
 865         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 866                 error = ECONNRESET;
 867                 goto out;
 868         }
 869         tp = intotcpcb(inp);
 870         TCPDEBUG1();
 871 #ifdef TCP_RFC7413
 872         /*
 873          * For passively-created TFO connections, don't attempt a window
 874          * update while still in SYN_RECEIVED as this may trigger an early
 875          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 876          * application response data, or failing that, when the DELACK timer
 877          * expires.
 878          */
 879         if ((tp->t_flags & TF_FASTOPEN) &&
 880             (tp->t_state == TCPS_SYN_RECEIVED))
 881                 goto out;
 882 #endif
 883 #ifdef TCP_OFFLOAD
 884         if (tp->t_flags & TF_TOE)
 885                 tcp_offload_rcvd(tp);
 886         else
 887 #endif
 888         tp->t_fb->tfb_tcp_output(tp);
 889
 890 out:
 891         TCPDEBUG2(PRU_RCVD);
 892         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 893         INP_WUNLOCK(inp);
 894         return (error);
 895 }
 896
 897 /*
 898  * Do a send by putting data in output queue and updating urgent
 899  * marker if URG set.  Possibly send more data.  Unlike the other
 900  * pru_*() routines, the mbuf chains are our responsibility.  We
 901  * must either enqueue them or free them.  The other pru_* routines
 902  * generally are caller-frees.
 903  */
 904 static int
 905 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 906     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 907 {
 908         int error = 0;
 909         struct inpcb *inp;
 910         struct tcpcb *tp = NULL;
 911 #ifdef INET6
 912         int isipv6;
 913 #endif
 914         TCPDEBUG0;
 915
 916         /*
 917          * We require the pcbinfo lock if we will close the socket as part of
 918          * this call.
 919          */
 920         if (flags & PRUS_EOF)
 921                 INP_INFO_RLOCK(&V_tcbinfo);
 922         inp = sotoinpcb(so);
 923         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 924         INP_WLOCK(inp);
 925         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 926                 if (control)
 927                         m_freem(control);
 928                 /*
 929                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 930                  * for freeing memory.
 931                  */
 932                 if (m && (flags & PRUS_NOTREADY) == 0)
 933                         m_freem(m);
 934                 error = ECONNRESET;
 935                 goto out;
 936         }
 937 #ifdef INET6
 938         isipv6 = nam && nam->sa_family == AF_INET6;
 939 #endif /* INET6 */
 940         tp = intotcpcb(inp);
 941         TCPDEBUG1();
 942         if (control) {
 943                 /* TCP doesn't do control messages (rights, creds, etc) */
 944                 if (control->m_len) {
 945                         m_freem(control);
 946                         if (m)
 947                                 m_freem(m);
 948                         error = EINVAL;
 949                         goto out;
 950                 }
 951                 m_freem(control);       /* empty control, just free it */
 952         }
 953         if (!(flags & PRUS_OOB)) {
 954                 sbappendstream(&so->so_snd, m, flags);
 955                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 956                         /*
 957                          * Do implied connect if not yet connected,
 958                          * initialize window to default value, and
 959                          * initialize maxseg using peer's cached MSS.
 960                          */
 961 #ifdef INET6
 962                         if (isipv6)
 963                                 error = tcp6_connect(tp, nam, td);
 964 #endif /* INET6 */
 965 #if defined(INET6) && defined(INET)
 966                         else
 967 #endif
 968 #ifdef INET
 969                                 error = tcp_connect(tp, nam, td);
 970 #endif
 971                         if (error)
 972                                 goto out;
 973                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 974                         tcp_mss(tp, -1);
 975                 }
 976                 if (flags & PRUS_EOF) {
 977                         /*
 978                          * Close the send side of the connection after
 979                          * the data is sent.
 980                          */
 981                         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
 982                         socantsendmore(so);
 983                         tcp_usrclosed(tp);
 984                 }
 985                 if (!(inp->inp_flags & INP_DROPPED) &&
 986                     !(flags & PRUS_NOTREADY)) {
 987                         if (flags & PRUS_MORETOCOME)
 988                                 tp->t_flags |= TF_MORETOCOME;
 989                         error = tp->t_fb->tfb_tcp_output(tp);
 990                         if (flags & PRUS_MORETOCOME)
 991                                 tp->t_flags &= ~TF_MORETOCOME;
 992                 }
 993         } else {
 994                 /*
 995                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 996                  */
 997                 SOCKBUF_LOCK(&so->so_snd);
 998                 if (sbspace(&so->so_snd) < -512) {
 999                         SOCKBUF_UNLOCK(&so->so_snd);
1000                         m_freem(m);
1001                         error = ENOBUFS;
1002                         goto out;
1003                 }
1004                 /*
1005                  * According to RFC961 (Assigned Protocols),
1006                  * the urgent pointer points to the last octet
1007                  * of urgent data.  We continue, however,
1008                  * to consider it to indicate the first octet
1009                  * of data past the urgent section.
1010                  * Otherwise, snd_up should be one lower.
1011                  */
1012                 sbappendstream_locked(&so->so_snd, m, flags);
1013                 SOCKBUF_UNLOCK(&so->so_snd);
1014                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1015                         /*
1016                          * Do implied connect if not yet connected,
1017                          * initialize window to default value, and
1018                          * initialize maxseg using peer's cached MSS.
1019                          */
1020 #ifdef INET6
1021                         if (isipv6)
1022                                 error = tcp6_connect(tp, nam, td);
1023 #endif /* INET6 */
1024 #if defined(INET6) && defined(INET)
1025                         else
1026 #endif
1027 #ifdef INET
1028                                 error = tcp_connect(tp, nam, td);
1029 #endif
1030                         if (error)
1031                                 goto out;
1032                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1033                         tcp_mss(tp, -1);
1034                 }
1035                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1036                 if (!(flags & PRUS_NOTREADY)) {
1037                         tp->t_flags |= TF_FORCEDATA;
1038                         error = tp->t_fb->tfb_tcp_output(tp);
1039                         tp->t_flags &= ~TF_FORCEDATA;
1040                 }
1041         }
1042 out:
1043         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1044                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1045         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1046                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1047         INP_WUNLOCK(inp);
1048         if (flags & PRUS_EOF)
1049                 INP_INFO_RUNLOCK(&V_tcbinfo);
1050         return (error);
1051 }
1052
1053 static int
1054 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1055 {
1056         struct inpcb *inp;
1057         struct tcpcb *tp;
1058         int error;
1059
1060         inp = sotoinpcb(so);
1061         INP_WLOCK(inp);
1062         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1063                 INP_WUNLOCK(inp);
1064                 for (int i = 0; i < count; i++)
1065                         m = m_free(m);
1066                 return (ECONNRESET);
1067         }
1068         tp = intotcpcb(inp);
1069
1070         SOCKBUF_LOCK(&so->so_snd);
1071         error = sbready(&so->so_snd, m, count);
1072         SOCKBUF_UNLOCK(&so->so_snd);
1073         if (error == 0)
1074                 error = tp->t_fb->tfb_tcp_output(tp);
1075         INP_WUNLOCK(inp);
1076
1077         return (error);
1078 }
1079
1080 /*
1081  * Abort the TCP.  Drop the connection abruptly.
1082  */
1083 static void
1084 tcp_usr_abort(struct socket *so)
1085 {
1086         struct inpcb *inp;
1087         struct tcpcb *tp = NULL;
1088         TCPDEBUG0;
1089
1090         inp = sotoinpcb(so);
1091         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1092
1093         INP_INFO_RLOCK(&V_tcbinfo);
1094         INP_WLOCK(inp);
1095         KASSERT(inp->inp_socket != NULL,
1096             ("tcp_usr_abort: inp_socket == NULL"));
1097
1098         /*
1099          * If we still have full TCP state, and we're not dropped, drop.
1100          */
1101         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1102             !(inp->inp_flags & INP_DROPPED)) {
1103                 tp = intotcpcb(inp);
1104                 TCPDEBUG1();
1105                 tp = tcp_drop(tp, ECONNABORTED);
1106                 if (tp == NULL)
1107                         goto dropped;
1108                 TCPDEBUG2(PRU_ABORT);
1109                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1110         }
1111         if (!(inp->inp_flags & INP_DROPPED)) {
1112                 SOCK_LOCK(so);
1113                 so->so_state |= SS_PROTOREF;
1114                 SOCK_UNLOCK(so);
1115                 inp->inp_flags |= INP_SOCKREF;
1116         }
1117         INP_WUNLOCK(inp);
1118 dropped:
1119         INP_INFO_RUNLOCK(&V_tcbinfo);
1120 }
1121
1122 /*
1123  * TCP socket is closed.  Start friendly disconnect.
1124  */
1125 static void
1126 tcp_usr_close(struct socket *so)
1127 {
1128         struct inpcb *inp;
1129         struct tcpcb *tp = NULL;
1130         TCPDEBUG0;
1131
1132         inp = sotoinpcb(so);
1133         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1134
1135         INP_INFO_RLOCK(&V_tcbinfo);
1136         INP_WLOCK(inp);
1137         KASSERT(inp->inp_socket != NULL,
1138             ("tcp_usr_close: inp_socket == NULL"));
1139
1140         /*
1141          * If we still have full TCP state, and we're not dropped, initiate
1142          * a disconnect.
1143          */
1144         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1145             !(inp->inp_flags & INP_DROPPED)) {
1146                 tp = intotcpcb(inp);
1147                 TCPDEBUG1();
1148                 tcp_disconnect(tp);
1149                 TCPDEBUG2(PRU_CLOSE);
1150                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1151         }
1152         if (!(inp->inp_flags & INP_DROPPED)) {
1153                 SOCK_LOCK(so);
1154                 so->so_state |= SS_PROTOREF;
1155                 SOCK_UNLOCK(so);
1156                 inp->inp_flags |= INP_SOCKREF;
1157         }
1158         INP_WUNLOCK(inp);
1159         INP_INFO_RUNLOCK(&V_tcbinfo);
1160 }
1161
1162 /*
1163  * Receive out-of-band data.
1164  */
1165 static int
1166 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1167 {
1168         int error = 0;
1169         struct inpcb *inp;
1170         struct tcpcb *tp = NULL;
1171
1172         TCPDEBUG0;
1173         inp = sotoinpcb(so);
1174         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1175         INP_WLOCK(inp);
1176         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1177                 error = ECONNRESET;
1178                 goto out;
1179         }
1180         tp = intotcpcb(inp);
1181         TCPDEBUG1();
1182         if ((so->so_oobmark == 0 &&
1183              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1184             so->so_options & SO_OOBINLINE ||
1185             tp->t_oobflags & TCPOOB_HADDATA) {
1186                 error = EINVAL;
1187                 goto out;
1188         }
1189         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1190                 error = EWOULDBLOCK;
1191                 goto out;
1192         }
1193         m->m_len = 1;
1194         *mtod(m, caddr_t) = tp->t_iobc;
1195         if ((flags & MSG_PEEK) == 0)
1196                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1197
1198 out:
1199         TCPDEBUG2(PRU_RCVOOB);
1200         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1201         INP_WUNLOCK(inp);
1202         return (error);
1203 }
1204
1205 #ifdef INET
1206 struct pr_usrreqs tcp_usrreqs = {
1207         .pru_abort =            tcp_usr_abort,
1208         .pru_accept =           tcp_usr_accept,
1209         .pru_attach =           tcp_usr_attach,
1210         .pru_bind =             tcp_usr_bind,
1211         .pru_connect =          tcp_usr_connect,
1212         .pru_control =          in_control,
1213         .pru_detach =           tcp_usr_detach,
1214         .pru_disconnect =       tcp_usr_disconnect,
1215         .pru_listen =           tcp_usr_listen,
1216         .pru_peeraddr =         in_getpeeraddr,
1217         .pru_rcvd =             tcp_usr_rcvd,
1218         .pru_rcvoob =           tcp_usr_rcvoob,
1219         .pru_send =             tcp_usr_send,
1220         .pru_ready =            tcp_usr_ready,
1221         .pru_shutdown =         tcp_usr_shutdown,
1222         .pru_sockaddr =         in_getsockaddr,
1223         .pru_sosetlabel =       in_pcbsosetlabel,
1224         .pru_close =            tcp_usr_close,
1225 };
1226 #endif /* INET */
1227
1228 #ifdef INET6
1229 struct pr_usrreqs tcp6_usrreqs = {
1230         .pru_abort =            tcp_usr_abort,
1231         .pru_accept =           tcp6_usr_accept,
1232         .pru_attach =           tcp_usr_attach,
1233         .pru_bind =             tcp6_usr_bind,
1234         .pru_connect =          tcp6_usr_connect,
1235         .pru_control =          in6_control,
1236         .pru_detach =           tcp_usr_detach,
1237         .pru_disconnect =       tcp_usr_disconnect,
1238         .pru_listen =           tcp6_usr_listen,
1239         .pru_peeraddr =         in6_mapped_peeraddr,
1240         .pru_rcvd =             tcp_usr_rcvd,
1241         .pru_rcvoob =           tcp_usr_rcvoob,
1242         .pru_send =             tcp_usr_send,
1243         .pru_ready =            tcp_usr_ready,
1244         .pru_shutdown =         tcp_usr_shutdown,
1245         .pru_sockaddr =         in6_mapped_sockaddr,
1246         .pru_sosetlabel =       in_pcbsosetlabel,
1247         .pru_close =            tcp_usr_close,
1248 };
1249 #endif /* INET6 */
1250
1251 #ifdef INET
1252 /*
1253  * Common subroutine to open a TCP connection to remote host specified
1254  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1255  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1256  * to choose a local host address (interface).  If there is an existing
1257  * incarnation of the same connection in TIME-WAIT state and if the remote
1258  * host was sending CC options and if the connection duration was < MSL, then
1259  * truncate the previous TIME-WAIT state and proceed.
1260  * Initialize connection parameters and enter SYN-SENT state.
1261  */
1262 static int
1263 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1264 {
1265         struct inpcb *inp = tp->t_inpcb, *oinp;
1266         struct socket *so = inp->inp_socket;
1267         struct in_addr laddr;
1268         u_short lport;
1269         int error;
1270
1271         INP_WLOCK_ASSERT(inp);
1272         INP_HASH_WLOCK(&V_tcbinfo);
1273
1274         if (inp->inp_lport == 0) {
1275                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1276                 if (error)
1277                         goto out;
1278         }
1279
1280         /*
1281          * Cannot simply call in_pcbconnect, because there might be an
1282          * earlier incarnation of this same connection still in
1283          * TIME_WAIT state, creating an ADDRINUSE error.
1284          */
1285         laddr = inp->inp_laddr;
1286         lport = inp->inp_lport;
1287         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1288             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1289         if (error && oinp == NULL)
1290                 goto out;
1291         if (oinp) {
1292                 error = EADDRINUSE;
1293                 goto out;
1294         }
1295         inp->inp_laddr = laddr;
1296         in_pcbrehash(inp);
1297         INP_HASH_WUNLOCK(&V_tcbinfo);
1298
1299         /*
1300          * Compute window scaling to request:
1301          * Scale to fit into sweet spot.  See tcp_syncache.c.
1302          * XXX: This should move to tcp_output().
1303          */
1304         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1305             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1306                 tp->request_r_scale++;
1307
1308         soisconnecting(so);
1309         TCPSTAT_INC(tcps_connattempt);
1310         tcp_state_change(tp, TCPS_SYN_SENT);
1311         tp->iss = tcp_new_isn(tp);
1312         tcp_sendseqinit(tp);
1313
1314         return 0;
1315
1316 out:
1317         INP_HASH_WUNLOCK(&V_tcbinfo);
1318         return (error);
1319 }
1320 #endif /* INET */
1321
1322 #ifdef INET6
1323 static int
1324 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1325 {
1326         struct inpcb *inp = tp->t_inpcb;
1327         int error;
1328
1329         INP_WLOCK_ASSERT(inp);
1330         INP_HASH_WLOCK(&V_tcbinfo);
1331
1332         if (inp->inp_lport == 0) {
1333                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1334                 if (error)
1335                         goto out;
1336         }
1337         error = in6_pcbconnect(inp, nam, td->td_ucred);
1338         if (error != 0)
1339                 goto out;
1340         INP_HASH_WUNLOCK(&V_tcbinfo);
1341
1342         /* Compute window scaling to request.  */
1343         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1344             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1345                 tp->request_r_scale++;
1346
1347         soisconnecting(inp->inp_socket);
1348         TCPSTAT_INC(tcps_connattempt);
1349         tcp_state_change(tp, TCPS_SYN_SENT);
1350         tp->iss = tcp_new_isn(tp);
1351         tcp_sendseqinit(tp);
1352
1353         return 0;
1354
1355 out:
1356         INP_HASH_WUNLOCK(&V_tcbinfo);
1357         return error;
1358 }
1359 #endif /* INET6 */
1360
1361 /*
1362  * Export TCP internal state information via a struct tcp_info, based on the
1363  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1364  * (TCP state machine, etc).  We export all information using FreeBSD-native
1365  * constants -- for example, the numeric values for tcpi_state will differ
1366  * from Linux.
1367  */
1368 static void
1369 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1370 {
1371
1372         INP_WLOCK_ASSERT(tp->t_inpcb);
1373         bzero(ti, sizeof(*ti));
1374
1375         ti->tcpi_state = tp->t_state;
1376         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1377                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1378         if (tp->t_flags & TF_SACK_PERMIT)
1379                 ti->tcpi_options |= TCPI_OPT_SACK;
1380         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1381                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1382                 ti->tcpi_snd_wscale = tp->snd_scale;
1383                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1384         }
1385         if (tp->t_flags & TF_ECN_PERMIT)
1386                 ti->tcpi_options |= TCPI_OPT_ECN;
1387
1388         ti->tcpi_rto = tp->t_rxtcur * tick;
1389         ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1390         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1391         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1392
1393         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1394         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1395
1396         /*
1397          * FreeBSD-specific extension fields for tcp_info.
1398          */
1399         ti->tcpi_rcv_space = tp->rcv_wnd;
1400         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1401         ti->tcpi_snd_wnd = tp->snd_wnd;
1402         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1403         ti->tcpi_snd_nxt = tp->snd_nxt;
1404         ti->tcpi_snd_mss = tp->t_maxseg;
1405         ti->tcpi_rcv_mss = tp->t_maxseg;
1406         if (tp->t_flags & TF_TOE)
1407                 ti->tcpi_options |= TCPI_OPT_TOE;
1408         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1409         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1410         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1411 }
1412
1413 /*
1414  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1415  * socket option arguments.  When it re-acquires the lock after the copy, it
1416  * has to revalidate that the connection is still valid for the socket
1417  * option.
1418  */
1419 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1420         INP_WLOCK(inp);                                                 \
1421         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1422                 INP_WUNLOCK(inp);                                       \
1423                 cleanup;                                                \
1424                 return (ECONNRESET);                                    \
1425         }                                                               \
1426         tp = intotcpcb(inp);                                            \
1427 } while(0)
1428 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1429
1430 int
1431 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1432 {
1433         int     error;
1434         struct  inpcb *inp;
1435         struct  tcpcb *tp;
1436         struct tcp_function_block *blk;
1437         struct tcp_function_set fsn;
1438
1439         error = 0;
1440         inp = sotoinpcb(so);
1441         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1442         INP_WLOCK(inp);
1443         if (sopt->sopt_level != IPPROTO_TCP) {
1444 #ifdef INET6
1445                 if (inp->inp_vflag & INP_IPV6PROTO) {
1446                         INP_WUNLOCK(inp);
1447                         error = ip6_ctloutput(so, sopt);
1448                 }
1449 #endif /* INET6 */
1450 #if defined(INET6) && defined(INET)
1451                 else
1452 #endif
1453 #ifdef INET
1454                 {
1455                         INP_WUNLOCK(inp);
1456                         error = ip_ctloutput(so, sopt);
1457                 }
1458 #endif
1459                 return (error);
1460         }
1461         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1462                 INP_WUNLOCK(inp);
1463                 return (ECONNRESET);
1464         }
1465         tp = intotcpcb(inp);
1466         /*
1467          * Protect the TCP option TCP_FUNCTION_BLK so
1468          * that a sub-function can *never* overwrite this.
1469          */
1470         if ((sopt->sopt_dir == SOPT_SET) &&
1471             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1472                 INP_WUNLOCK(inp);
1473                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1474                     sizeof fsn);
1475                 if (error)
1476                         return (error);
1477                 INP_WLOCK_RECHECK(inp);
1478                 if (tp->t_state != TCPS_CLOSED) {
1479                         /*
1480                          * The user has advanced the state
1481                          * past the initial point, we can't
1482                          * switch since we are down the road
1483                          * and a new set of functions may
1484                          * not be compatibile.
1485                          */
1486                         INP_WUNLOCK(inp);
1487                         return(EINVAL);
1488                 }
1489                 blk = find_and_ref_tcp_functions(&fsn);
1490                 if (blk == NULL) {
1491                         INP_WUNLOCK(inp);
1492                         return (ENOENT);
1493                 }
1494                 if (tp->t_fb != blk) {
1495                         if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1496                                 refcount_release(&blk->tfb_refcnt);
1497                                 INP_WUNLOCK(inp);
1498                                 return (ENOENT);
1499                         }
1500                         /*
1501                          * Release the old refcnt, the
1502                          * lookup acquires a ref on the
1503                          * new one.
1504                          */
1505                         if (tp->t_fb->tfb_tcp_fb_fini)
1506                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1507                         refcount_release(&tp->t_fb->tfb_refcnt);
1508                         tp->t_fb = blk;
1509                         if (tp->t_fb->tfb_tcp_fb_init) {
1510                                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1511                         }
1512                 }
1513 #ifdef TCP_OFFLOAD
1514                 if (tp->t_flags & TF_TOE) {
1515                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1516                              sopt->sopt_name);
1517                 }
1518 #endif
1519                 INP_WUNLOCK(inp);
1520                 return (error);
1521         } else if ((sopt->sopt_dir == SOPT_GET) &&
1522             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1523                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1524                     TCP_FUNCTION_NAME_LEN_MAX);
1525                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1526                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1527                 INP_WUNLOCK(inp);
1528                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1529                 return (error);
1530         }
1531         /* Pass in the INP locked, called must unlock it */
1532         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1533 }
1534
1535 int
1536 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1537 {
1538         int     error, opt, optval;
1539         u_int   ui;
1540         struct  tcp_info ti;
1541         struct cc_algo *algo;
1542         char    *pbuf, buf[TCP_CA_NAME_MAX];
1543         size_t  len;
1544
1545         /*
1546          * For TCP_CCALGOOPT forward the control to CC module, for both
1547          * SOPT_SET and SOPT_GET.
1548          */
1549         switch (sopt->sopt_name) {
1550         case TCP_CCALGOOPT:
1551                 INP_WUNLOCK(inp);
1552                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1553                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1554                     sopt->sopt_valsize);
1555                 if (error) {
1556                         free(pbuf, M_TEMP);
1557                         return (error);
1558                 }
1559                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1560                 if (CC_ALGO(tp)->ctl_output != NULL)
1561                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1562                 else
1563                         error = ENOENT;
1564                 INP_WUNLOCK(inp);
1565                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1566                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1567                 free(pbuf, M_TEMP);
1568                 return (error);
1569         }
1570
1571         switch (sopt->sopt_dir) {
1572         case SOPT_SET:
1573                 switch (sopt->sopt_name) {
1574 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1575                 case TCP_MD5SIG:
1576                         if (!TCPMD5_ENABLED()) {
1577                                 INP_WUNLOCK(inp);
1578                                 return (ENOPROTOOPT);
1579                         }
1580                         error = TCPMD5_PCBCTL(inp, sopt);
1581                         if (error)
1582                                 return (error);
1583                         goto unlock_and_done;
1584 #endif /* IPSEC */
1585
1586                 case TCP_NODELAY:
1587                 case TCP_NOOPT:
1588                         INP_WUNLOCK(inp);
1589                         error = sooptcopyin(sopt, &optval, sizeof optval,
1590                             sizeof optval);
1591                         if (error)
1592                                 return (error);
1593
1594                         INP_WLOCK_RECHECK(inp);
1595                         switch (sopt->sopt_name) {
1596                         case TCP_NODELAY:
1597                                 opt = TF_NODELAY;
1598                                 break;
1599                         case TCP_NOOPT:
1600                                 opt = TF_NOOPT;
1601                                 break;
1602                         default:
1603                                 opt = 0; /* dead code to fool gcc */
1604                                 break;
1605                         }
1606
1607                         if (optval)
1608                                 tp->t_flags |= opt;
1609                         else
1610                                 tp->t_flags &= ~opt;
1611 unlock_and_done:
1612 #ifdef TCP_OFFLOAD
1613                         if (tp->t_flags & TF_TOE) {
1614                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1615                                     sopt->sopt_name);
1616                         }
1617 #endif
1618                         INP_WUNLOCK(inp);
1619                         break;
1620
1621                 case TCP_NOPUSH:
1622                         INP_WUNLOCK(inp);
1623                         error = sooptcopyin(sopt, &optval, sizeof optval,
1624                             sizeof optval);
1625                         if (error)
1626                                 return (error);
1627
1628                         INP_WLOCK_RECHECK(inp);
1629                         if (optval)
1630                                 tp->t_flags |= TF_NOPUSH;
1631                         else if (tp->t_flags & TF_NOPUSH) {
1632                                 tp->t_flags &= ~TF_NOPUSH;
1633                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1634                                         error = tp->t_fb->tfb_tcp_output(tp);
1635                         }
1636                         goto unlock_and_done;
1637
1638                 case TCP_MAXSEG:
1639                         INP_WUNLOCK(inp);
1640                         error = sooptcopyin(sopt, &optval, sizeof optval,
1641                             sizeof optval);
1642                         if (error)
1643                                 return (error);
1644
1645                         INP_WLOCK_RECHECK(inp);
1646                         if (optval > 0 && optval <= tp->t_maxseg &&
1647                             optval + 40 >= V_tcp_minmss)
1648                                 tp->t_maxseg = optval;
1649                         else
1650                                 error = EINVAL;
1651                         goto unlock_and_done;
1652
1653                 case TCP_INFO:
1654                         INP_WUNLOCK(inp);
1655                         error = EINVAL;
1656                         break;
1657
1658                 case TCP_CONGESTION:
1659                         INP_WUNLOCK(inp);
1660                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1661                         if (error)
1662                                 break;
1663                         buf[sopt->sopt_valsize] = '\0';
1664                         INP_WLOCK_RECHECK(inp);
1665                         CC_LIST_RLOCK();
1666                         STAILQ_FOREACH(algo, &cc_list, entries)
1667                                 if (strncmp(buf, algo->name,
1668                                     TCP_CA_NAME_MAX) == 0)
1669                                         break;
1670                         CC_LIST_RUNLOCK();
1671                         if (algo == NULL) {
1672                                 INP_WUNLOCK(inp);
1673                                 error = EINVAL;
1674                                 break;
1675                         }
1676                         /*
1677                          * We hold a write lock over the tcb so it's safe to
1678                          * do these things without ordering concerns.
1679                          */
1680                         if (CC_ALGO(tp)->cb_destroy != NULL)
1681                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1682                         CC_ALGO(tp) = algo;
1683                         /*
1684                          * If something goes pear shaped initialising the new
1685                          * algo, fall back to newreno (which does not
1686                          * require initialisation).
1687                          */
1688                         if (algo->cb_init != NULL &&
1689                             algo->cb_init(tp->ccv) != 0) {
1690                                 CC_ALGO(tp) = &newreno_cc_algo;
1691                                 /*
1692                                  * The only reason init should fail is
1693                                  * because of malloc.
1694                                  */
1695                                 error = ENOMEM;
1696                         }
1697                         INP_WUNLOCK(inp);
1698                         break;
1699
1700                 case TCP_KEEPIDLE:
1701                 case TCP_KEEPINTVL:
1702                 case TCP_KEEPINIT:
1703                         INP_WUNLOCK(inp);
1704                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1705                         if (error)
1706                                 return (error);
1707
1708                         if (ui > (UINT_MAX / hz)) {
1709                                 error = EINVAL;
1710                                 break;
1711                         }
1712                         ui *= hz;
1713
1714                         INP_WLOCK_RECHECK(inp);
1715                         switch (sopt->sopt_name) {
1716                         case TCP_KEEPIDLE:
1717                                 tp->t_keepidle = ui;
1718                                 /*
1719                                  * XXX: better check current remaining
1720                                  * timeout and "merge" it with new value.
1721                                  */
1722                                 if ((tp->t_state > TCPS_LISTEN) &&
1723                                     (tp->t_state <= TCPS_CLOSING))
1724                                         tcp_timer_activate(tp, TT_KEEP,
1725                                             TP_KEEPIDLE(tp));
1726                                 break;
1727                         case TCP_KEEPINTVL:
1728                                 tp->t_keepintvl = ui;
1729                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1730                                     (TP_MAXIDLE(tp) > 0))
1731                                         tcp_timer_activate(tp, TT_2MSL,
1732                                             TP_MAXIDLE(tp));
1733                                 break;
1734                         case TCP_KEEPINIT:
1735                                 tp->t_keepinit = ui;
1736                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1737                                     tp->t_state == TCPS_SYN_SENT)
1738                                         tcp_timer_activate(tp, TT_KEEP,
1739                                             TP_KEEPINIT(tp));
1740                                 break;
1741                         }
1742                         goto unlock_and_done;
1743
1744                 case TCP_KEEPCNT:
1745                         INP_WUNLOCK(inp);
1746                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1747                         if (error)
1748                                 return (error);
1749
1750                         INP_WLOCK_RECHECK(inp);
1751                         tp->t_keepcnt = ui;
1752                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1753                             (TP_MAXIDLE(tp) > 0))
1754                                 tcp_timer_activate(tp, TT_2MSL,
1755                                     TP_MAXIDLE(tp));
1756                         goto unlock_and_done;
1757
1758 #ifdef TCPPCAP
1759                 case TCP_PCAP_OUT:
1760                 case TCP_PCAP_IN:
1761                         INP_WUNLOCK(inp);
1762                         error = sooptcopyin(sopt, &optval, sizeof optval,
1763                             sizeof optval);
1764                         if (error)
1765                                 return (error);
1766
1767                         INP_WLOCK_RECHECK(inp);
1768                         if (optval >= 0)
1769                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1770                                         &(tp->t_outpkts) : &(tp->t_inpkts),
1771                                         optval);
1772                         else
1773                                 error = EINVAL;
1774                         goto unlock_and_done;
1775 #endif
1776
1777 #ifdef TCP_RFC7413
1778                 case TCP_FASTOPEN:
1779                         INP_WUNLOCK(inp);
1780                         if (!V_tcp_fastopen_enabled)
1781                                 return (EPERM);
1782
1783                         error = sooptcopyin(sopt, &optval, sizeof optval,
1784                             sizeof optval);
1785                         if (error)
1786                                 return (error);
1787
1788                         INP_WLOCK_RECHECK(inp);
1789                         if (optval) {
1790                                 tp->t_flags |= TF_FASTOPEN;
1791                                 if ((tp->t_state == TCPS_LISTEN) &&
1792                                     (tp->t_tfo_pending == NULL))
1793                                         tp->t_tfo_pending =
1794                                             tcp_fastopen_alloc_counter();
1795                         } else
1796                                 tp->t_flags &= ~TF_FASTOPEN;
1797                         goto unlock_and_done;
1798 #endif
1799
1800                 default:
1801                         INP_WUNLOCK(inp);
1802                         error = ENOPROTOOPT;
1803                         break;
1804                 }
1805                 break;
1806
1807         case SOPT_GET:
1808                 tp = intotcpcb(inp);
1809                 switch (sopt->sopt_name) {
1810 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1811                 case TCP_MD5SIG:
1812                         if (!TCPMD5_ENABLED()) {
1813                                 INP_WUNLOCK(inp);
1814                                 return (ENOPROTOOPT);
1815                         }
1816                         error = TCPMD5_PCBCTL(inp, sopt);
1817                         break;
1818 #endif
1819
1820                 case TCP_NODELAY:
1821                         optval = tp->t_flags & TF_NODELAY;
1822                         INP_WUNLOCK(inp);
1823                         error = sooptcopyout(sopt, &optval, sizeof optval);
1824                         break;
1825                 case TCP_MAXSEG:
1826                         optval = tp->t_maxseg;
1827                         INP_WUNLOCK(inp);
1828                         error = sooptcopyout(sopt, &optval, sizeof optval);
1829                         break;
1830                 case TCP_NOOPT:
1831                         optval = tp->t_flags & TF_NOOPT;
1832                         INP_WUNLOCK(inp);
1833                         error = sooptcopyout(sopt, &optval, sizeof optval);
1834                         break;
1835                 case TCP_NOPUSH:
1836                         optval = tp->t_flags & TF_NOPUSH;
1837                         INP_WUNLOCK(inp);
1838                         error = sooptcopyout(sopt, &optval, sizeof optval);
1839                         break;
1840                 case TCP_INFO:
1841                         tcp_fill_info(tp, &ti);
1842                         INP_WUNLOCK(inp);
1843                         error = sooptcopyout(sopt, &ti, sizeof ti);
1844                         break;
1845                 case TCP_CONGESTION:
1846                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1847                         INP_WUNLOCK(inp);
1848                         error = sooptcopyout(sopt, buf, len + 1);
1849                         break;
1850                 case TCP_KEEPIDLE:
1851                 case TCP_KEEPINTVL:
1852                 case TCP_KEEPINIT:
1853                 case TCP_KEEPCNT:
1854                         switch (sopt->sopt_name) {
1855                         case TCP_KEEPIDLE:
1856                                 ui = TP_KEEPIDLE(tp) / hz;
1857                                 break;
1858                         case TCP_KEEPINTVL:
1859                                 ui = TP_KEEPINTVL(tp) / hz;
1860                                 break;
1861                         case TCP_KEEPINIT:
1862                                 ui = TP_KEEPINIT(tp) / hz;
1863                                 break;
1864                         case TCP_KEEPCNT:
1865                                 ui = TP_KEEPCNT(tp);
1866                                 break;
1867                         }
1868                         INP_WUNLOCK(inp);
1869                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1870                         break;
1871 #ifdef TCPPCAP
1872                 case TCP_PCAP_OUT:
1873                 case TCP_PCAP_IN:
1874                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1875                                         &(tp->t_outpkts) : &(tp->t_inpkts));
1876                         INP_WUNLOCK(inp);
1877                         error = sooptcopyout(sopt, &optval, sizeof optval);
1878                         break;
1879 #endif
1880
1881 #ifdef TCP_RFC7413
1882                 case TCP_FASTOPEN:
1883                         optval = tp->t_flags & TF_FASTOPEN;
1884                         INP_WUNLOCK(inp);
1885                         error = sooptcopyout(sopt, &optval, sizeof optval);
1886                         break;
1887 #endif
1888                 default:
1889                         INP_WUNLOCK(inp);
1890                         error = ENOPROTOOPT;
1891                         break;
1892                 }
1893                 break;
1894         }
1895         return (error);
1896 }
1897 #undef INP_WLOCK_RECHECK
1898 #undef INP_WLOCK_RECHECK_CLEANUP
1899
1900 /*
1901  * Attach TCP protocol to socket, allocating
1902  * internet protocol control block, tcp control block,
1903  * bufer space, and entering LISTEN state if to accept connections.
1904  */
1905 static int
1906 tcp_attach(struct socket *so)
1907 {
1908         struct tcpcb *tp;
1909         struct inpcb *inp;
1910         int error;
1911
1912         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1913                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1914                 if (error)
1915                         return (error);
1916         }
1917         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1918         so->so_snd.sb_flags |= SB_AUTOSIZE;
1919         INP_INFO_RLOCK(&V_tcbinfo);
1920         error = in_pcballoc(so, &V_tcbinfo);
1921         if (error) {
1922                 INP_INFO_RUNLOCK(&V_tcbinfo);
1923                 return (error);
1924         }
1925         inp = sotoinpcb(so);
1926 #ifdef INET6
1927         if (inp->inp_vflag & INP_IPV6PROTO) {
1928                 inp->inp_vflag |= INP_IPV6;
1929                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1930                         inp->inp_vflag |= INP_IPV4;
1931                 inp->in6p_hops = -1;    /* use kernel default */
1932         }
1933         else
1934 #endif
1935         inp->inp_vflag |= INP_IPV4;
1936         tp = tcp_newtcpcb(inp);
1937         if (tp == NULL) {
1938                 in_pcbdetach(inp);
1939                 in_pcbfree(inp);
1940                 INP_INFO_RUNLOCK(&V_tcbinfo);
1941                 return (ENOBUFS);
1942         }
1943         tp->t_state = TCPS_CLOSED;
1944         INP_WUNLOCK(inp);
1945         INP_INFO_RUNLOCK(&V_tcbinfo);
1946         TCPSTATES_INC(TCPS_CLOSED);
1947         return (0);
1948 }
1949
1950 /*
1951  * Initiate (or continue) disconnect.
1952  * If embryonic state, just send reset (once).
1953  * If in ``let data drain'' option and linger null, just drop.
1954  * Otherwise (hard), mark socket disconnecting and drop
1955  * current input data; switch states based on user close, and
1956  * send segment to peer (with FIN).
1957  */
1958 static void
1959 tcp_disconnect(struct tcpcb *tp)
1960 {
1961         struct inpcb *inp = tp->t_inpcb;
1962         struct socket *so = inp->inp_socket;
1963
1964         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1965         INP_WLOCK_ASSERT(inp);
1966
1967         /*
1968          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1969          * socket is still open.
1970          */
1971         if (tp->t_state < TCPS_ESTABLISHED) {
1972                 tp = tcp_close(tp);
1973                 KASSERT(tp != NULL,
1974                     ("tcp_disconnect: tcp_close() returned NULL"));
1975         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1976                 tp = tcp_drop(tp, 0);
1977                 KASSERT(tp != NULL,
1978                     ("tcp_disconnect: tcp_drop() returned NULL"));
1979         } else {
1980                 soisdisconnecting(so);
1981                 sbflush(&so->so_rcv);
1982                 tcp_usrclosed(tp);
1983                 if (!(inp->inp_flags & INP_DROPPED))
1984                         tp->t_fb->tfb_tcp_output(tp);
1985         }
1986 }
1987
1988 /*
1989  * User issued close, and wish to trail through shutdown states:
1990  * if never received SYN, just forget it.  If got a SYN from peer,
1991  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1992  * If already got a FIN from peer, then almost done; go to LAST_ACK
1993  * state.  In all other cases, have already sent FIN to peer (e.g.
1994  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1995  * for peer to send FIN or not respond to keep-alives, etc.
1996  * We can let the user exit from the close as soon as the FIN is acked.
1997  */
1998 static void
1999 tcp_usrclosed(struct tcpcb *tp)
2000 {
2001
2002         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2003         INP_WLOCK_ASSERT(tp->t_inpcb);
2004
2005         switch (tp->t_state) {
2006         case TCPS_LISTEN:
2007 #ifdef TCP_OFFLOAD
2008                 tcp_offload_listen_stop(tp);
2009 #endif
2010                 tcp_state_change(tp, TCPS_CLOSED);
2011                 /* FALLTHROUGH */
2012         case TCPS_CLOSED:
2013                 tp = tcp_close(tp);
2014                 /*
2015                  * tcp_close() should never return NULL here as the socket is
2016                  * still open.
2017                  */
2018                 KASSERT(tp != NULL,
2019                     ("tcp_usrclosed: tcp_close() returned NULL"));
2020                 break;
2021
2022         case TCPS_SYN_SENT:
2023         case TCPS_SYN_RECEIVED:
2024                 tp->t_flags |= TF_NEEDFIN;
2025                 break;
2026
2027         case TCPS_ESTABLISHED:
2028                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2029                 break;
2030
2031         case TCPS_CLOSE_WAIT:
2032                 tcp_state_change(tp, TCPS_LAST_ACK);
2033                 break;
2034         }
2035         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2036                 soisdisconnected(tp->t_inpcb->inp_socket);
2037                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2038                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2039                         int timeout;
2040
2041                         timeout = (tcp_fast_finwait2_recycle) ?
2042                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2043                         tcp_timer_activate(tp, TT_2MSL, timeout);
2044                 }
2045         }
2046 }
2047
2048 #ifdef DDB
2049 static void
2050 db_print_indent(int indent)
2051 {
2052         int i;
2053
2054         for (i = 0; i < indent; i++)
2055                 db_printf(" ");
2056 }
2057
2058 static void
2059 db_print_tstate(int t_state)
2060 {
2061
2062         switch (t_state) {
2063         case TCPS_CLOSED:
2064                 db_printf("TCPS_CLOSED");
2065                 return;
2066
2067         case TCPS_LISTEN:
2068                 db_printf("TCPS_LISTEN");
2069                 return;
2070
2071         case TCPS_SYN_SENT:
2072                 db_printf("TCPS_SYN_SENT");
2073                 return;
2074
2075         case TCPS_SYN_RECEIVED:
2076                 db_printf("TCPS_SYN_RECEIVED");
2077                 return;
2078
2079         case TCPS_ESTABLISHED:
2080                 db_printf("TCPS_ESTABLISHED");
2081                 return;
2082
2083         case TCPS_CLOSE_WAIT:
2084                 db_printf("TCPS_CLOSE_WAIT");
2085                 return;
2086
2087         case TCPS_FIN_WAIT_1:
2088                 db_printf("TCPS_FIN_WAIT_1");
2089                 return;
2090
2091         case TCPS_CLOSING:
2092                 db_printf("TCPS_CLOSING");
2093                 return;
2094
2095         case TCPS_LAST_ACK:
2096                 db_printf("TCPS_LAST_ACK");
2097                 return;
2098
2099         case TCPS_FIN_WAIT_2:
2100                 db_printf("TCPS_FIN_WAIT_2");
2101                 return;
2102
2103         case TCPS_TIME_WAIT:
2104                 db_printf("TCPS_TIME_WAIT");
2105                 return;
2106
2107         default:
2108                 db_printf("unknown");
2109                 return;
2110         }
2111 }
2112
2113 static void
2114 db_print_tflags(u_int t_flags)
2115 {
2116         int comma;
2117
2118         comma = 0;
2119         if (t_flags & TF_ACKNOW) {
2120                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2121                 comma = 1;
2122         }
2123         if (t_flags & TF_DELACK) {
2124                 db_printf("%sTF_DELACK", comma ? ", " : "");
2125                 comma = 1;
2126         }
2127         if (t_flags & TF_NODELAY) {
2128                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2129                 comma = 1;
2130         }
2131         if (t_flags & TF_NOOPT) {
2132                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2133                 comma = 1;
2134         }
2135         if (t_flags & TF_SENTFIN) {
2136                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2137                 comma = 1;
2138         }
2139         if (t_flags & TF_REQ_SCALE) {
2140                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2141                 comma = 1;
2142         }
2143         if (t_flags & TF_RCVD_SCALE) {
2144                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2145                 comma = 1;
2146         }
2147         if (t_flags & TF_REQ_TSTMP) {
2148                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2149                 comma = 1;
2150         }
2151         if (t_flags & TF_RCVD_TSTMP) {
2152                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2153                 comma = 1;
2154         }
2155         if (t_flags & TF_SACK_PERMIT) {
2156                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2157                 comma = 1;
2158         }
2159         if (t_flags & TF_NEEDSYN) {
2160                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2161                 comma = 1;
2162         }
2163         if (t_flags & TF_NEEDFIN) {
2164                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2165                 comma = 1;
2166         }
2167         if (t_flags & TF_NOPUSH) {
2168                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2169                 comma = 1;
2170         }
2171         if (t_flags & TF_MORETOCOME) {
2172                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2173                 comma = 1;
2174         }
2175         if (t_flags & TF_LQ_OVERFLOW) {
2176                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2177                 comma = 1;
2178         }
2179         if (t_flags & TF_LASTIDLE) {
2180                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2181                 comma = 1;
2182         }
2183         if (t_flags & TF_RXWIN0SENT) {
2184                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2185                 comma = 1;
2186         }
2187         if (t_flags & TF_FASTRECOVERY) {
2188                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2189                 comma = 1;
2190         }
2191         if (t_flags & TF_CONGRECOVERY) {
2192                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2193                 comma = 1;
2194         }
2195         if (t_flags & TF_WASFRECOVERY) {
2196                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2197                 comma = 1;
2198         }
2199         if (t_flags & TF_SIGNATURE) {
2200                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2201                 comma = 1;
2202         }
2203         if (t_flags & TF_FORCEDATA) {
2204                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2205                 comma = 1;
2206         }
2207         if (t_flags & TF_TSO) {
2208                 db_printf("%sTF_TSO", comma ? ", " : "");
2209                 comma = 1;
2210         }
2211         if (t_flags & TF_ECN_PERMIT) {
2212                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2213                 comma = 1;
2214         }
2215         if (t_flags & TF_FASTOPEN) {
2216                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2217                 comma = 1;
2218         }
2219 }
2220
2221 static void
2222 db_print_toobflags(char t_oobflags)
2223 {
2224         int comma;
2225
2226         comma = 0;
2227         if (t_oobflags & TCPOOB_HAVEDATA) {
2228                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2229                 comma = 1;
2230         }
2231         if (t_oobflags & TCPOOB_HADDATA) {
2232                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2233                 comma = 1;
2234         }
2235 }
2236
2237 static void
2238 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2239 {
2240
2241         db_print_indent(indent);
2242         db_printf("%s at %p\n", name, tp);
2243
2244         indent += 2;
2245
2246         db_print_indent(indent);
2247         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2248            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2249
2250         db_print_indent(indent);
2251         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2252             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2253
2254         db_print_indent(indent);
2255         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2256             &tp->t_timers->tt_delack, tp->t_inpcb);
2257
2258         db_print_indent(indent);
2259         db_printf("t_state: %d (", tp->t_state);
2260         db_print_tstate(tp->t_state);
2261         db_printf(")\n");
2262
2263         db_print_indent(indent);
2264         db_printf("t_flags: 0x%x (", tp->t_flags);
2265         db_print_tflags(tp->t_flags);
2266         db_printf(")\n");
2267
2268         db_print_indent(indent);
2269         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2270             tp->snd_una, tp->snd_max, tp->snd_nxt);
2271
2272         db_print_indent(indent);
2273         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2274            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2275
2276         db_print_indent(indent);
2277         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2278             tp->iss, tp->irs, tp->rcv_nxt);
2279
2280         db_print_indent(indent);
2281         db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2282             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2283
2284         db_print_indent(indent);
2285         db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2286            tp->snd_wnd, tp->snd_cwnd);
2287
2288         db_print_indent(indent);
2289         db_printf("snd_ssthresh: %lu   snd_recover: "
2290             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2291
2292         db_print_indent(indent);
2293         db_printf("t_rcvtime: %u   t_startime: %u\n",
2294             tp->t_rcvtime, tp->t_starttime);
2295
2296         db_print_indent(indent);
2297         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2298             tp->t_rtttime, tp->t_rtseq);
2299
2300         db_print_indent(indent);
2301         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2302             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2303
2304         db_print_indent(indent);
2305         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2306             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2307             tp->t_rttbest);
2308
2309         db_print_indent(indent);
2310         db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2311             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2312
2313         db_print_indent(indent);
2314         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2315         db_print_toobflags(tp->t_oobflags);
2316         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2317
2318         db_print_indent(indent);
2319         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2320             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2321
2322         db_print_indent(indent);
2323         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2324             tp->ts_recent, tp->ts_recent_age);
2325
2326         db_print_indent(indent);
2327         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2328             "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2329
2330         db_print_indent(indent);
2331         db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2332             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2333             tp->snd_recover_prev, tp->t_badrxtwin);
2334
2335         db_print_indent(indent);
2336         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2337             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2338
2339         db_print_indent(indent);
2340         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2341             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2342
2343         /* Skip sackblks, sackhint. */
2344
2345         db_print_indent(indent);
2346         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2347             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2348 }
2349
2350 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2351 {
2352         struct tcpcb *tp;
2353
2354         if (!have_addr) {
2355                 db_printf("usage: show tcpcb <addr>\n");
2356                 return;
2357         }
2358         tp = (struct tcpcb *)addr;
2359
2360         db_print_tcpcb(tp, "tcpcb", 0);
2361 }
2362 #endif