sys/netinet/tcp_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1993
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2006-2007 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Robert N. M. Watson under
  11  * contract to Juniper Networks, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_ddb.h"
  44 #include "opt_inet.h"
  45 #include "opt_inet6.h"
  46 #include "opt_ipsec.h"
  47 #include "opt_kern_tls.h"
  48 #include "opt_tcpdebug.h"
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/arb.h>
  53 #include <sys/limits.h>
  54 #include <sys/malloc.h>
  55 #include <sys/refcount.h>
  56 #include <sys/kernel.h>
  57 #include <sys/ktls.h>
  58 #include <sys/qmath.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/mbuf.h>
  61 #ifdef INET6
  62 #include <sys/domain.h>
  63 #endif /* INET6 */
  64 #include <sys/socket.h>
  65 #include <sys/socketvar.h>
  66 #include <sys/protosw.h>
  67 #include <sys/proc.h>
  68 #include <sys/jail.h>
  69 #include <sys/syslog.h>
  70 #include <sys/stats.h>
  71
  72 #ifdef DDB
  73 #include <ddb/ddb.h>
  74 #endif
  75
  76 #include <net/if.h>
  77 #include <net/if_var.h>
  78 #include <net/route.h>
  79 #include <net/vnet.h>
  80
  81 #include <netinet/in.h>
  82 #include <netinet/in_kdtrace.h>
  83 #include <netinet/in_pcb.h>
  84 #include <netinet/in_systm.h>
  85 #include <netinet/in_var.h>
  86 #include <netinet/ip_var.h>
  87 #ifdef INET6
  88 #include <netinet/ip6.h>
  89 #include <netinet6/in6_pcb.h>
  90 #include <netinet6/ip6_var.h>
  91 #include <netinet6/scope6_var.h>
  92 #endif
  93 #include <netinet/tcp.h>
  94 #include <netinet/tcp_fsm.h>
  95 #include <netinet/tcp_seq.h>
  96 #include <netinet/tcp_timer.h>
  97 #include <netinet/tcp_var.h>
  98 #include <netinet/tcp_log_buf.h>
  99 #include <netinet/tcpip.h>
 100 #include <netinet/cc/cc.h>
 101 #include <netinet/tcp_fastopen.h>
 102 #include <netinet/tcp_hpts.h>
 103 #ifdef TCPPCAP
 104 #include <netinet/tcp_pcap.h>
 105 #endif
 106 #ifdef TCPDEBUG
 107 #include <netinet/tcp_debug.h>
 108 #endif
 109 #ifdef TCP_OFFLOAD
 110 #include <netinet/tcp_offload.h>
 111 #endif
 112 #include <netipsec/ipsec_support.h>
 113
 114 #include <vm/vm.h>
 115 #include <vm/vm_param.h>
 116 #include <vm/pmap.h>
 117 #include <vm/vm_extern.h>
 118 #include <vm/vm_map.h>
 119 #include <vm/vm_page.h>
 120
 121 /*
 122  * TCP protocol interface to socket abstraction.
 123  */
 124 #ifdef INET
 125 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 126                     struct thread *td);
 127 #endif /* INET */
 128 #ifdef INET6
 129 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 130                     struct thread *td);
 131 #endif /* INET6 */
 132 static void     tcp_disconnect(struct tcpcb *);
 133 static void     tcp_usrclosed(struct tcpcb *);
 134 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 135
 136 static int      tcp_pru_options_support(struct tcpcb *tp, int flags);
 137
 138 #ifdef TCPDEBUG
 139 #define TCPDEBUG0       int ostate = 0
 140 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 141 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 142                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 143 #else
 144 #define TCPDEBUG0
 145 #define TCPDEBUG1()
 146 #define TCPDEBUG2(req)
 147 #endif
 148
 149 /*
 150  * tcp_require_unique port requires a globally-unique source port for each
 151  * outgoing connection.  The default is to require the 4-tuple to be unique.
 152  */
 153 VNET_DEFINE(int, tcp_require_unique_port) = 0;
 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port,
 155     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0,
 156     "Require globally-unique ephemeral port for outgoing connections");
 157 #define V_tcp_require_unique_port       VNET(tcp_require_unique_port)
 158
 159 /*
 160  * TCP attaches to socket via pru_attach(), reserving space,
 161  * and an internet control block.
 162  */
 163 static int
 164 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 165 {
 166         struct inpcb *inp;
 167         struct tcpcb *tp = NULL;
 168         int error;
 169         TCPDEBUG0;
 170
 171         inp = sotoinpcb(so);
 172         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 173         TCPDEBUG1();
 174
 175         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
 176                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
 177                 if (error)
 178                         goto out;
 179         }
 180
 181         so->so_rcv.sb_flags |= SB_AUTOSIZE;
 182         so->so_snd.sb_flags |= SB_AUTOSIZE;
 183         error = in_pcballoc(so, &V_tcbinfo);
 184         if (error)
 185                 goto out;
 186         inp = sotoinpcb(so);
 187 #ifdef INET6
 188         if (inp->inp_vflag & INP_IPV6PROTO) {
 189                 inp->inp_vflag |= INP_IPV6;
 190                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 191                         inp->inp_vflag |= INP_IPV4;
 192                 inp->in6p_hops = -1;    /* use kernel default */
 193         }
 194         else
 195 #endif
 196                 inp->inp_vflag |= INP_IPV4;
 197         tp = tcp_newtcpcb(inp);
 198         if (tp == NULL) {
 199                 error = ENOBUFS;
 200                 in_pcbdetach(inp);
 201                 in_pcbfree(inp);
 202                 goto out;
 203         }
 204         tp->t_state = TCPS_CLOSED;
 205         INP_WUNLOCK(inp);
 206         TCPSTATES_INC(TCPS_CLOSED);
 207 out:
 208         TCPDEBUG2(PRU_ATTACH);
 209         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 210         return (error);
 211 }
 212
 213 /*
 214  * tcp_usr_detach is called when the socket layer loses its final reference
 215  * to the socket, be it a file descriptor reference, a reference from TCP,
 216  * etc.  At this point, there is only one case in which we will keep around
 217  * inpcb state: time wait.
 218  */
 219 static void
 220 tcp_usr_detach(struct socket *so)
 221 {
 222         struct inpcb *inp;
 223         struct tcpcb *tp;
 224
 225         inp = sotoinpcb(so);
 226         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 227         INP_WLOCK(inp);
 228         KASSERT(so->so_pcb == inp && inp->inp_socket == so,
 229                 ("%s: socket %p inp %p mismatch", __func__, so, inp));
 230
 231         tp = intotcpcb(inp);
 232
 233         if (inp->inp_flags & INP_TIMEWAIT) {
 234                 /*
 235                  * There are two cases to handle: one in which the time wait
 236                  * state is being discarded (INP_DROPPED), and one in which
 237                  * this connection will remain in timewait.  In the former,
 238                  * it is time to discard all state (except tcptw, which has
 239                  * already been discarded by the timewait close code, which
 240                  * should be further up the call stack somewhere).  In the
 241                  * latter case, we detach from the socket, but leave the pcb
 242                  * present until timewait ends.
 243                  *
 244                  * XXXRW: Would it be cleaner to free the tcptw here?
 245                  *
 246                  * Astute question indeed, from twtcp perspective there are
 247                  * four cases to consider:
 248                  *
 249                  * #1 tcp_usr_detach is called at tcptw creation time by
 250                  *  tcp_twstart, then do not discard the newly created tcptw
 251                  *  and leave inpcb present until timewait ends
 252                  * #2 tcp_usr_detach is called at tcptw creation time by
 253                  *  tcp_twstart, but connection is local and tw will be
 254                  *  discarded immediately
 255                  * #3 tcp_usr_detach is called at timewait end (or reuse) by
 256                  *  tcp_twclose, then the tcptw has already been discarded
 257                  *  (or reused) and inpcb is freed here
 258                  * #4 tcp_usr_detach is called() after timewait ends (or reuse)
 259                  *  (e.g. by soclose), then tcptw has already been discarded
 260                  *  (or reused) and inpcb is freed here
 261                  *
 262                  *  In all three cases the tcptw should not be freed here.
 263                  */
 264                 if (inp->inp_flags & INP_DROPPED) {
 265                         in_pcbdetach(inp);
 266                         if (__predict_true(tp == NULL)) {
 267                                 in_pcbfree(inp);
 268                         } else {
 269                                 /*
 270                                  * This case should not happen as in TIMEWAIT
 271                                  * state the inp should not be destroyed before
 272                                  * its tcptw.  If INVARIANTS is defined, panic.
 273                                  */
 274 #ifdef INVARIANTS
 275                                 panic("%s: Panic before an inp double-free: "
 276                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 277                                     , __func__);
 278 #else
 279                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 280                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 281                                     , __func__);
 282 #endif
 283                                 INP_WUNLOCK(inp);
 284                         }
 285                 } else {
 286                         in_pcbdetach(inp);
 287                         INP_WUNLOCK(inp);
 288                 }
 289         } else {
 290                 /*
 291                  * If the connection is not in timewait, we consider two
 292                  * two conditions: one in which no further processing is
 293                  * necessary (dropped || embryonic), and one in which TCP is
 294                  * not yet done, but no longer requires the socket, so the
 295                  * pcb will persist for the time being.
 296                  *
 297                  * XXXRW: Does the second case still occur?
 298                  */
 299                 if (inp->inp_flags & INP_DROPPED ||
 300                     tp->t_state < TCPS_SYN_SENT) {
 301                         tcp_discardcb(tp);
 302                         in_pcbdetach(inp);
 303                         in_pcbfree(inp);
 304                 } else {
 305                         in_pcbdetach(inp);
 306                         INP_WUNLOCK(inp);
 307                 }
 308         }
 309 }
 310
 311 #ifdef INET
 312 /*
 313  * Give the socket an address.
 314  */
 315 static int
 316 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 317 {
 318         int error = 0;
 319         struct inpcb *inp;
 320         struct tcpcb *tp = NULL;
 321         struct sockaddr_in *sinp;
 322
 323         sinp = (struct sockaddr_in *)nam;
 324         if (nam->sa_family != AF_INET) {
 325                 /*
 326                  * Preserve compatibility with old programs.
 327                  */
 328                 if (nam->sa_family != AF_UNSPEC ||
 329                     sinp->sin_addr.s_addr != INADDR_ANY)
 330                         return (EAFNOSUPPORT);
 331                 nam->sa_family = AF_INET;
 332         }
 333         if (nam->sa_len != sizeof(*sinp))
 334                 return (EINVAL);
 335
 336         /*
 337          * Must check for multicast addresses and disallow binding
 338          * to them.
 339          */
 340         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 341                 return (EAFNOSUPPORT);
 342
 343         TCPDEBUG0;
 344         inp = sotoinpcb(so);
 345         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 346         INP_WLOCK(inp);
 347         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 348                 error = EINVAL;
 349                 goto out;
 350         }
 351         tp = intotcpcb(inp);
 352         TCPDEBUG1();
 353         INP_HASH_WLOCK(&V_tcbinfo);
 354         error = in_pcbbind(inp, nam, td->td_ucred);
 355         INP_HASH_WUNLOCK(&V_tcbinfo);
 356 out:
 357         TCPDEBUG2(PRU_BIND);
 358         TCP_PROBE2(debug__user, tp, PRU_BIND);
 359         INP_WUNLOCK(inp);
 360
 361         return (error);
 362 }
 363 #endif /* INET */
 364
 365 #ifdef INET6
 366 static int
 367 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 368 {
 369         int error = 0;
 370         struct inpcb *inp;
 371         struct tcpcb *tp = NULL;
 372         struct sockaddr_in6 *sin6;
 373         u_char vflagsav;
 374
 375         sin6 = (struct sockaddr_in6 *)nam;
 376         if (nam->sa_family != AF_INET6)
 377                 return (EAFNOSUPPORT);
 378         if (nam->sa_len != sizeof(*sin6))
 379                 return (EINVAL);
 380
 381         /*
 382          * Must check for multicast addresses and disallow binding
 383          * to them.
 384          */
 385         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 386                 return (EAFNOSUPPORT);
 387
 388         TCPDEBUG0;
 389         inp = sotoinpcb(so);
 390         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 391         INP_WLOCK(inp);
 392         vflagsav = inp->inp_vflag;
 393         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 394                 error = EINVAL;
 395                 goto out;
 396         }
 397         tp = intotcpcb(inp);
 398         TCPDEBUG1();
 399         INP_HASH_WLOCK(&V_tcbinfo);
 400         inp->inp_vflag &= ~INP_IPV4;
 401         inp->inp_vflag |= INP_IPV6;
 402 #ifdef INET
 403         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 404                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
 405                         inp->inp_vflag |= INP_IPV4;
 406                 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 407                         struct sockaddr_in sin;
 408
 409                         in6_sin6_2_sin(&sin, sin6);
 410                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 411                                 error = EAFNOSUPPORT;
 412                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 413                                 goto out;
 414                         }
 415                         inp->inp_vflag |= INP_IPV4;
 416                         inp->inp_vflag &= ~INP_IPV6;
 417                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 418                             td->td_ucred);
 419                         INP_HASH_WUNLOCK(&V_tcbinfo);
 420                         goto out;
 421                 }
 422         }
 423 #endif
 424         error = in6_pcbbind(inp, nam, td->td_ucred);
 425         INP_HASH_WUNLOCK(&V_tcbinfo);
 426 out:
 427         if (error != 0)
 428                 inp->inp_vflag = vflagsav;
 429         TCPDEBUG2(PRU_BIND);
 430         TCP_PROBE2(debug__user, tp, PRU_BIND);
 431         INP_WUNLOCK(inp);
 432         return (error);
 433 }
 434 #endif /* INET6 */
 435
 436 #ifdef INET
 437 /*
 438  * Prepare to accept connections.
 439  */
 440 static int
 441 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 442 {
 443         int error = 0;
 444         struct inpcb *inp;
 445         struct tcpcb *tp = NULL;
 446
 447         TCPDEBUG0;
 448         inp = sotoinpcb(so);
 449         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 450         INP_WLOCK(inp);
 451         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 452                 error = EINVAL;
 453                 goto out;
 454         }
 455         tp = intotcpcb(inp);
 456         TCPDEBUG1();
 457         SOCK_LOCK(so);
 458         error = solisten_proto_check(so);
 459         INP_HASH_WLOCK(&V_tcbinfo);
 460         if (error == 0 && inp->inp_lport == 0)
 461                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 462         INP_HASH_WUNLOCK(&V_tcbinfo);
 463         if (error == 0) {
 464                 tcp_state_change(tp, TCPS_LISTEN);
 465                 solisten_proto(so, backlog);
 466 #ifdef TCP_OFFLOAD
 467                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 468                         tcp_offload_listen_start(tp);
 469 #endif
 470         }
 471         SOCK_UNLOCK(so);
 472
 473         if (IS_FASTOPEN(tp->t_flags))
 474                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 475
 476 out:
 477         TCPDEBUG2(PRU_LISTEN);
 478         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 479         INP_WUNLOCK(inp);
 480         return (error);
 481 }
 482 #endif /* INET */
 483
 484 #ifdef INET6
 485 static int
 486 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 487 {
 488         int error = 0;
 489         struct inpcb *inp;
 490         struct tcpcb *tp = NULL;
 491         u_char vflagsav;
 492
 493         TCPDEBUG0;
 494         inp = sotoinpcb(so);
 495         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 496         INP_WLOCK(inp);
 497         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 498                 error = EINVAL;
 499                 goto out;
 500         }
 501         vflagsav = inp->inp_vflag;
 502         tp = intotcpcb(inp);
 503         TCPDEBUG1();
 504         SOCK_LOCK(so);
 505         error = solisten_proto_check(so);
 506         INP_HASH_WLOCK(&V_tcbinfo);
 507         if (error == 0 && inp->inp_lport == 0) {
 508                 inp->inp_vflag &= ~INP_IPV4;
 509                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 510                         inp->inp_vflag |= INP_IPV4;
 511                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 512         }
 513         INP_HASH_WUNLOCK(&V_tcbinfo);
 514         if (error == 0) {
 515                 tcp_state_change(tp, TCPS_LISTEN);
 516                 solisten_proto(so, backlog);
 517 #ifdef TCP_OFFLOAD
 518                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 519                         tcp_offload_listen_start(tp);
 520 #endif
 521         }
 522         SOCK_UNLOCK(so);
 523
 524         if (IS_FASTOPEN(tp->t_flags))
 525                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 526
 527         if (error != 0)
 528                 inp->inp_vflag = vflagsav;
 529
 530 out:
 531         TCPDEBUG2(PRU_LISTEN);
 532         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 533         INP_WUNLOCK(inp);
 534         return (error);
 535 }
 536 #endif /* INET6 */
 537
 538 #ifdef INET
 539 /*
 540  * Initiate connection to peer.
 541  * Create a template for use in transmissions on this connection.
 542  * Enter SYN_SENT state, and mark socket as connecting.
 543  * Start keep-alive timer, and seed output sequence space.
 544  * Send initial segment on connection.
 545  */
 546 static int
 547 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 548 {
 549         struct epoch_tracker et;
 550         int error = 0;
 551         struct inpcb *inp;
 552         struct tcpcb *tp = NULL;
 553         struct sockaddr_in *sinp;
 554
 555         sinp = (struct sockaddr_in *)nam;
 556         if (nam->sa_family != AF_INET)
 557                 return (EAFNOSUPPORT);
 558         if (nam->sa_len != sizeof (*sinp))
 559                 return (EINVAL);
 560
 561         /*
 562          * Must disallow TCP ``connections'' to multicast addresses.
 563          */
 564         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 565                 return (EAFNOSUPPORT);
 566         if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST)
 567                 return (EACCES);
 568         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 569                 return (error);
 570
 571         TCPDEBUG0;
 572         inp = sotoinpcb(so);
 573         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 574         INP_WLOCK(inp);
 575         if (inp->inp_flags & INP_TIMEWAIT) {
 576                 error = EADDRINUSE;
 577                 goto out;
 578         }
 579         if (inp->inp_flags & INP_DROPPED) {
 580                 error = ECONNREFUSED;
 581                 goto out;
 582         }
 583         tp = intotcpcb(inp);
 584         TCPDEBUG1();
 585         NET_EPOCH_ENTER(et);
 586         if ((error = tcp_connect(tp, nam, td)) != 0)
 587                 goto out_in_epoch;
 588 #ifdef TCP_OFFLOAD
 589         if (registered_toedevs > 0 &&
 590             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 591             (error = tcp_offload_connect(so, nam)) == 0)
 592                 goto out_in_epoch;
 593 #endif
 594         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 595         error = tp->t_fb->tfb_tcp_output(tp);
 596 out_in_epoch:
 597         NET_EPOCH_EXIT(et);
 598 out:
 599         TCPDEBUG2(PRU_CONNECT);
 600         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 601         INP_WUNLOCK(inp);
 602         return (error);
 603 }
 604 #endif /* INET */
 605
 606 #ifdef INET6
 607 static int
 608 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 609 {
 610         struct epoch_tracker et;
 611         int error = 0;
 612         struct inpcb *inp;
 613         struct tcpcb *tp = NULL;
 614         struct sockaddr_in6 *sin6;
 615         u_int8_t incflagsav;
 616         u_char vflagsav;
 617
 618         TCPDEBUG0;
 619
 620         sin6 = (struct sockaddr_in6 *)nam;
 621         if (nam->sa_family != AF_INET6)
 622                 return (EAFNOSUPPORT);
 623         if (nam->sa_len != sizeof (*sin6))
 624                 return (EINVAL);
 625
 626         /*
 627          * Must disallow TCP ``connections'' to multicast addresses.
 628          */
 629         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 630                 return (EAFNOSUPPORT);
 631
 632         inp = sotoinpcb(so);
 633         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 634         INP_WLOCK(inp);
 635         vflagsav = inp->inp_vflag;
 636         incflagsav = inp->inp_inc.inc_flags;
 637         if (inp->inp_flags & INP_TIMEWAIT) {
 638                 error = EADDRINUSE;
 639                 goto out;
 640         }
 641         if (inp->inp_flags & INP_DROPPED) {
 642                 error = ECONNREFUSED;
 643                 goto out;
 644         }
 645         tp = intotcpcb(inp);
 646         TCPDEBUG1();
 647 #ifdef INET
 648         /*
 649          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 650          * therefore probably require the hash lock, which isn't held here.
 651          * Is this a significant problem?
 652          */
 653         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 654                 struct sockaddr_in sin;
 655
 656                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 657                         error = EINVAL;
 658                         goto out;
 659                 }
 660                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 661                         error = EAFNOSUPPORT;
 662                         goto out;
 663                 }
 664
 665                 in6_sin6_2_sin(&sin, sin6);
 666                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 667                         error = EAFNOSUPPORT;
 668                         goto out;
 669                 }
 670                 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
 671                         error = EACCES;
 672                         goto out;
 673                 }
 674                 if ((error = prison_remote_ip4(td->td_ucred,
 675                     &sin.sin_addr)) != 0)
 676                         goto out;
 677                 inp->inp_vflag |= INP_IPV4;
 678                 inp->inp_vflag &= ~INP_IPV6;
 679                 NET_EPOCH_ENTER(et);
 680                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 681                         goto out_in_epoch;
 682 #ifdef TCP_OFFLOAD
 683                 if (registered_toedevs > 0 &&
 684                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 685                     (error = tcp_offload_connect(so, nam)) == 0)
 686                         goto out_in_epoch;
 687 #endif
 688                 error = tp->t_fb->tfb_tcp_output(tp);
 689                 goto out_in_epoch;
 690         } else {
 691                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 692                         error = EAFNOSUPPORT;
 693                         goto out;
 694                 }
 695         }
 696 #endif
 697         if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
 698                 goto out;
 699         inp->inp_vflag &= ~INP_IPV4;
 700         inp->inp_vflag |= INP_IPV6;
 701         inp->inp_inc.inc_flags |= INC_ISIPV6;
 702         if ((error = tcp6_connect(tp, nam, td)) != 0)
 703                 goto out;
 704 #ifdef TCP_OFFLOAD
 705         if (registered_toedevs > 0 &&
 706             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 707             (error = tcp_offload_connect(so, nam)) == 0)
 708                 goto out;
 709 #endif
 710         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 711         NET_EPOCH_ENTER(et);
 712         error = tp->t_fb->tfb_tcp_output(tp);
 713 #ifdef INET
 714 out_in_epoch:
 715 #endif
 716         NET_EPOCH_EXIT(et);
 717 out:
 718         /*
 719          * If the implicit bind in the connect call fails, restore
 720          * the flags we modified.
 721          */
 722         if (error != 0 && inp->inp_lport == 0) {
 723                 inp->inp_vflag = vflagsav;
 724                 inp->inp_inc.inc_flags = incflagsav;
 725         }
 726
 727         TCPDEBUG2(PRU_CONNECT);
 728         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 729         INP_WUNLOCK(inp);
 730         return (error);
 731 }
 732 #endif /* INET6 */
 733
 734 /*
 735  * Initiate disconnect from peer.
 736  * If connection never passed embryonic stage, just drop;
 737  * else if don't need to let data drain, then can just drop anyways,
 738  * else have to begin TCP shutdown process: mark socket disconnecting,
 739  * drain unread data, state switch to reflect user close, and
 740  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 741  * when peer sends FIN and acks ours.
 742  *
 743  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 744  */
 745 static int
 746 tcp_usr_disconnect(struct socket *so)
 747 {
 748         struct inpcb *inp;
 749         struct tcpcb *tp = NULL;
 750         struct epoch_tracker et;
 751         int error = 0;
 752
 753         TCPDEBUG0;
 754         NET_EPOCH_ENTER(et);
 755         inp = sotoinpcb(so);
 756         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 757         INP_WLOCK(inp);
 758         if (inp->inp_flags & INP_TIMEWAIT)
 759                 goto out;
 760         if (inp->inp_flags & INP_DROPPED) {
 761                 error = ECONNRESET;
 762                 goto out;
 763         }
 764         tp = intotcpcb(inp);
 765         TCPDEBUG1();
 766         tcp_disconnect(tp);
 767 out:
 768         TCPDEBUG2(PRU_DISCONNECT);
 769         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 770         INP_WUNLOCK(inp);
 771         NET_EPOCH_EXIT(et);
 772         return (error);
 773 }
 774
 775 #ifdef INET
 776 /*
 777  * Accept a connection.  Essentially all the work is done at higher levels;
 778  * just return the address of the peer, storing through addr.
 779  */
 780 static int
 781 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 782 {
 783         int error = 0;
 784         struct inpcb *inp = NULL;
 785         struct tcpcb *tp = NULL;
 786         struct in_addr addr;
 787         in_port_t port = 0;
 788         TCPDEBUG0;
 789
 790         if (so->so_state & SS_ISDISCONNECTED)
 791                 return (ECONNABORTED);
 792
 793         inp = sotoinpcb(so);
 794         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 795         INP_WLOCK(inp);
 796         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 797                 error = ECONNABORTED;
 798                 goto out;
 799         }
 800         tp = intotcpcb(inp);
 801         TCPDEBUG1();
 802
 803         /*
 804          * We inline in_getpeeraddr and COMMON_END here, so that we can
 805          * copy the data of interest and defer the malloc until after we
 806          * release the lock.
 807          */
 808         port = inp->inp_fport;
 809         addr = inp->inp_faddr;
 810
 811 out:
 812         TCPDEBUG2(PRU_ACCEPT);
 813         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 814         INP_WUNLOCK(inp);
 815         if (error == 0)
 816                 *nam = in_sockaddr(port, &addr);
 817         return error;
 818 }
 819 #endif /* INET */
 820
 821 #ifdef INET6
 822 static int
 823 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 824 {
 825         struct inpcb *inp = NULL;
 826         int error = 0;
 827         struct tcpcb *tp = NULL;
 828         struct in_addr addr;
 829         struct in6_addr addr6;
 830         struct epoch_tracker et;
 831         in_port_t port = 0;
 832         int v4 = 0;
 833         TCPDEBUG0;
 834
 835         if (so->so_state & SS_ISDISCONNECTED)
 836                 return (ECONNABORTED);
 837
 838         inp = sotoinpcb(so);
 839         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 840         NET_EPOCH_ENTER(et);
 841         INP_WLOCK(inp);
 842         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 843                 error = ECONNABORTED;
 844                 goto out;
 845         }
 846         tp = intotcpcb(inp);
 847         TCPDEBUG1();
 848
 849         /*
 850          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 851          * copy the data of interest and defer the malloc until after we
 852          * release the lock.
 853          */
 854         if (inp->inp_vflag & INP_IPV4) {
 855                 v4 = 1;
 856                 port = inp->inp_fport;
 857                 addr = inp->inp_faddr;
 858         } else {
 859                 port = inp->inp_fport;
 860                 addr6 = inp->in6p_faddr;
 861         }
 862
 863 out:
 864         TCPDEBUG2(PRU_ACCEPT);
 865         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 866         INP_WUNLOCK(inp);
 867         NET_EPOCH_EXIT(et);
 868         if (error == 0) {
 869                 if (v4)
 870                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 871                 else
 872                         *nam = in6_sockaddr(port, &addr6);
 873         }
 874         return error;
 875 }
 876 #endif /* INET6 */
 877
 878 /*
 879  * Mark the connection as being incapable of further output.
 880  */
 881 static int
 882 tcp_usr_shutdown(struct socket *so)
 883 {
 884         int error = 0;
 885         struct inpcb *inp;
 886         struct tcpcb *tp = NULL;
 887         struct epoch_tracker et;
 888
 889         TCPDEBUG0;
 890         NET_EPOCH_ENTER(et);
 891         inp = sotoinpcb(so);
 892         KASSERT(inp != NULL, ("inp == NULL"));
 893         INP_WLOCK(inp);
 894         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 895                 error = ECONNRESET;
 896                 goto out;
 897         }
 898         tp = intotcpcb(inp);
 899         TCPDEBUG1();
 900         socantsendmore(so);
 901         tcp_usrclosed(tp);
 902         if (!(inp->inp_flags & INP_DROPPED))
 903                 error = tp->t_fb->tfb_tcp_output(tp);
 904
 905 out:
 906         TCPDEBUG2(PRU_SHUTDOWN);
 907         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 908         INP_WUNLOCK(inp);
 909         NET_EPOCH_EXIT(et);
 910
 911         return (error);
 912 }
 913
 914 /*
 915  * After a receive, possibly send window update to peer.
 916  */
 917 static int
 918 tcp_usr_rcvd(struct socket *so, int flags)
 919 {
 920         struct epoch_tracker et;
 921         struct inpcb *inp;
 922         struct tcpcb *tp = NULL;
 923         int error = 0;
 924
 925         TCPDEBUG0;
 926         inp = sotoinpcb(so);
 927         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 928         INP_WLOCK(inp);
 929         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 930                 error = ECONNRESET;
 931                 goto out;
 932         }
 933         tp = intotcpcb(inp);
 934         TCPDEBUG1();
 935         /*
 936          * For passively-created TFO connections, don't attempt a window
 937          * update while still in SYN_RECEIVED as this may trigger an early
 938          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 939          * application response data, or failing that, when the DELACK timer
 940          * expires.
 941          */
 942         if (IS_FASTOPEN(tp->t_flags) &&
 943             (tp->t_state == TCPS_SYN_RECEIVED))
 944                 goto out;
 945         NET_EPOCH_ENTER(et);
 946 #ifdef TCP_OFFLOAD
 947         if (tp->t_flags & TF_TOE)
 948                 tcp_offload_rcvd(tp);
 949         else
 950 #endif
 951         tp->t_fb->tfb_tcp_output(tp);
 952         NET_EPOCH_EXIT(et);
 953 out:
 954         TCPDEBUG2(PRU_RCVD);
 955         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 956         INP_WUNLOCK(inp);
 957         return (error);
 958 }
 959
 960 /*
 961  * Do a send by putting data in output queue and updating urgent
 962  * marker if URG set.  Possibly send more data.  Unlike the other
 963  * pru_*() routines, the mbuf chains are our responsibility.  We
 964  * must either enqueue them or free them.  The other pru_* routines
 965  * generally are caller-frees.
 966  */
 967 static int
 968 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 969     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 970 {
 971         struct epoch_tracker et;
 972         int error = 0;
 973         struct inpcb *inp;
 974         struct tcpcb *tp = NULL;
 975 #ifdef INET
 976 #ifdef INET6
 977         struct sockaddr_in sin;
 978 #endif
 979         struct sockaddr_in *sinp;
 980 #endif
 981 #ifdef INET6
 982         int isipv6;
 983 #endif
 984         u_int8_t incflagsav;
 985         u_char vflagsav;
 986         bool restoreflags;
 987         TCPDEBUG0;
 988
 989         /*
 990          * We require the pcbinfo "read lock" if we will close the socket
 991          * as part of this call.
 992          */
 993         NET_EPOCH_ENTER(et);
 994         inp = sotoinpcb(so);
 995         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 996         INP_WLOCK(inp);
 997         vflagsav = inp->inp_vflag;
 998         incflagsav = inp->inp_inc.inc_flags;
 999         restoreflags = false;
1000         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1001                 if (control)
1002                         m_freem(control);
1003                 error = ECONNRESET;
1004                 goto out;
1005         }
1006         if (control != NULL) {
1007                 /* TCP doesn't do control messages (rights, creds, etc) */
1008                 if (control->m_len) {
1009                         m_freem(control);
1010                         error = EINVAL;
1011                         goto out;
1012                 }
1013                 m_freem(control);       /* empty control, just free it */
1014                 control = NULL;
1015         }
1016         tp = intotcpcb(inp);
1017         if ((flags & PRUS_OOB) != 0 &&
1018             (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1019                 goto out;
1020
1021         TCPDEBUG1();
1022         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1023                 switch (nam->sa_family) {
1024 #ifdef INET
1025                 case AF_INET:
1026                         sinp = (struct sockaddr_in *)nam;
1027                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1028                                 error = EINVAL;
1029                                 goto out;
1030                         }
1031                         if ((inp->inp_vflag & INP_IPV6) != 0) {
1032                                 error = EAFNOSUPPORT;
1033                                 goto out;
1034                         }
1035                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1036                                 error = EAFNOSUPPORT;
1037                                 goto out;
1038                         }
1039                         if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1040                                 error = EACCES;
1041                                 goto out;
1042                         }
1043                         if ((error = prison_remote_ip4(td->td_ucred,
1044                             &sinp->sin_addr)))
1045                                 goto out;
1046 #ifdef INET6
1047                         isipv6 = 0;
1048 #endif
1049                         break;
1050 #endif /* INET */
1051 #ifdef INET6
1052                 case AF_INET6:
1053                 {
1054                         struct sockaddr_in6 *sin6;
1055
1056                         sin6 = (struct sockaddr_in6 *)nam;
1057                         if (sin6->sin6_len != sizeof(*sin6)) {
1058                                 error = EINVAL;
1059                                 goto out;
1060                         }
1061                         if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1062                                 error = EAFNOSUPPORT;
1063                                 goto out;
1064                         }
1065                         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1066                                 error = EAFNOSUPPORT;
1067                                 goto out;
1068                         }
1069                         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1070 #ifdef INET
1071                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1072                                         error = EINVAL;
1073                                         goto out;
1074                                 }
1075                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1076                                         error = EAFNOSUPPORT;
1077                                         goto out;
1078                                 }
1079                                 restoreflags = true;
1080                                 inp->inp_vflag &= ~INP_IPV6;
1081                                 sinp = &sin;
1082                                 in6_sin6_2_sin(sinp, sin6);
1083                                 if (IN_MULTICAST(
1084                                     ntohl(sinp->sin_addr.s_addr))) {
1085                                         error = EAFNOSUPPORT;
1086                                         goto out;
1087                                 }
1088                                 if ((error = prison_remote_ip4(td->td_ucred,
1089                                     &sinp->sin_addr)))
1090                                         goto out;
1091                                 isipv6 = 0;
1092 #else /* !INET */
1093                                 error = EAFNOSUPPORT;
1094                                 goto out;
1095 #endif /* INET */
1096                         } else {
1097                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1098                                         error = EAFNOSUPPORT;
1099                                         goto out;
1100                                 }
1101                                 restoreflags = true;
1102                                 inp->inp_vflag &= ~INP_IPV4;
1103                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1104                                 if ((error = prison_remote_ip6(td->td_ucred,
1105                                     &sin6->sin6_addr)))
1106                                         goto out;
1107                                 isipv6 = 1;
1108                         }
1109                         break;
1110                 }
1111 #endif /* INET6 */
1112                 default:
1113                         error = EAFNOSUPPORT;
1114                         goto out;
1115                 }
1116         }
1117         if (!(flags & PRUS_OOB)) {
1118                 sbappendstream(&so->so_snd, m, flags);
1119                 m = NULL;
1120                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1121                         /*
1122                          * Do implied connect if not yet connected,
1123                          * initialize window to default value, and
1124                          * initialize maxseg using peer's cached MSS.
1125                          */
1126 #ifdef INET6
1127                         if (isipv6)
1128                                 error = tcp6_connect(tp, nam, td);
1129 #endif /* INET6 */
1130 #if defined(INET6) && defined(INET)
1131                         else
1132 #endif
1133 #ifdef INET
1134                                 error = tcp_connect(tp,
1135                                     (struct sockaddr *)sinp, td);
1136 #endif
1137                         /*
1138                          * The bind operation in tcp_connect succeeded. We
1139                          * no longer want to restore the flags if later
1140                          * operations fail.
1141                          */
1142                         if (error == 0 || inp->inp_lport != 0)
1143                                 restoreflags = false;
1144
1145                         if (error) {
1146                                 /* m is freed if PRUS_NOTREADY is unset. */
1147                                 sbflush(&so->so_snd);
1148                                 goto out;
1149                         }
1150                         if (IS_FASTOPEN(tp->t_flags))
1151                                 tcp_fastopen_connect(tp);
1152                         else {
1153                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1154                                 tcp_mss(tp, -1);
1155                         }
1156                 }
1157                 if (flags & PRUS_EOF) {
1158                         /*
1159                          * Close the send side of the connection after
1160                          * the data is sent.
1161                          */
1162                         socantsendmore(so);
1163                         tcp_usrclosed(tp);
1164                 }
1165                 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1166                     ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1167                     (tp->t_fbyte_out == 0) &&
1168                     (so->so_snd.sb_ccc > 0)) {
1169                         tp->t_fbyte_out = ticks;
1170                         if (tp->t_fbyte_out == 0)
1171                                 tp->t_fbyte_out = 1;
1172                         if (tp->t_fbyte_out && tp->t_fbyte_in)
1173                                 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1174                 }
1175                 if (!(inp->inp_flags & INP_DROPPED) &&
1176                     !(flags & PRUS_NOTREADY)) {
1177                         if (flags & PRUS_MORETOCOME)
1178                                 tp->t_flags |= TF_MORETOCOME;
1179                         error = tp->t_fb->tfb_tcp_output(tp);
1180                         if (flags & PRUS_MORETOCOME)
1181                                 tp->t_flags &= ~TF_MORETOCOME;
1182                 }
1183         } else {
1184                 /*
1185                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1186                  */
1187                 SOCKBUF_LOCK(&so->so_snd);
1188                 if (sbspace(&so->so_snd) < -512) {
1189                         SOCKBUF_UNLOCK(&so->so_snd);
1190                         error = ENOBUFS;
1191                         goto out;
1192                 }
1193                 /*
1194                  * According to RFC961 (Assigned Protocols),
1195                  * the urgent pointer points to the last octet
1196                  * of urgent data.  We continue, however,
1197                  * to consider it to indicate the first octet
1198                  * of data past the urgent section.
1199                  * Otherwise, snd_up should be one lower.
1200                  */
1201                 sbappendstream_locked(&so->so_snd, m, flags);
1202                 SOCKBUF_UNLOCK(&so->so_snd);
1203                 m = NULL;
1204                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1205                         /*
1206                          * Do implied connect if not yet connected,
1207                          * initialize window to default value, and
1208                          * initialize maxseg using peer's cached MSS.
1209                          */
1210
1211                         /*
1212                          * Not going to contemplate SYN|URG
1213                          */
1214                         if (IS_FASTOPEN(tp->t_flags))
1215                                 tp->t_flags &= ~TF_FASTOPEN;
1216 #ifdef INET6
1217                         if (isipv6)
1218                                 error = tcp6_connect(tp, nam, td);
1219 #endif /* INET6 */
1220 #if defined(INET6) && defined(INET)
1221                         else
1222 #endif
1223 #ifdef INET
1224                                 error = tcp_connect(tp,
1225                                     (struct sockaddr *)sinp, td);
1226 #endif
1227                         /*
1228                          * The bind operation in tcp_connect succeeded. We
1229                          * no longer want to restore the flags if later
1230                          * operations fail.
1231                          */
1232                         if (error == 0 || inp->inp_lport != 0)
1233                                 restoreflags = false;
1234
1235                         if (error != 0) {
1236                                 /* m is freed if PRUS_NOTREADY is unset. */
1237                                 sbflush(&so->so_snd);
1238                                 goto out;
1239                         }
1240                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1241                         tcp_mss(tp, -1);
1242                 }
1243                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1244                 if ((flags & PRUS_NOTREADY) == 0) {
1245                         tp->t_flags |= TF_FORCEDATA;
1246                         error = tp->t_fb->tfb_tcp_output(tp);
1247                         tp->t_flags &= ~TF_FORCEDATA;
1248                 }
1249         }
1250         TCP_LOG_EVENT(tp, NULL,
1251             &inp->inp_socket->so_rcv,
1252             &inp->inp_socket->so_snd,
1253             TCP_LOG_USERSEND, error,
1254             0, NULL, false);
1255
1256 out:
1257         /*
1258          * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1259          * responsible for freeing memory.
1260          */
1261         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1262                 m_freem(m);
1263
1264         /*
1265          * If the request was unsuccessful and we changed flags,
1266          * restore the original flags.
1267          */
1268         if (error != 0 && restoreflags) {
1269                 inp->inp_vflag = vflagsav;
1270                 inp->inp_inc.inc_flags = incflagsav;
1271         }
1272         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1273                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1274         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1275                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1276         INP_WUNLOCK(inp);
1277         NET_EPOCH_EXIT(et);
1278         return (error);
1279 }
1280
1281 static int
1282 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1283 {
1284         struct epoch_tracker et;
1285         struct inpcb *inp;
1286         struct tcpcb *tp;
1287         int error;
1288
1289         inp = sotoinpcb(so);
1290         INP_WLOCK(inp);
1291         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1292                 INP_WUNLOCK(inp);
1293                 mb_free_notready(m, count);
1294                 return (ECONNRESET);
1295         }
1296         tp = intotcpcb(inp);
1297
1298         SOCKBUF_LOCK(&so->so_snd);
1299         error = sbready(&so->so_snd, m, count);
1300         SOCKBUF_UNLOCK(&so->so_snd);
1301         if (error == 0) {
1302                 NET_EPOCH_ENTER(et);
1303                 error = tp->t_fb->tfb_tcp_output(tp);
1304                 NET_EPOCH_EXIT(et);
1305         }
1306         INP_WUNLOCK(inp);
1307
1308         return (error);
1309 }
1310
1311 /*
1312  * Abort the TCP.  Drop the connection abruptly.
1313  */
1314 static void
1315 tcp_usr_abort(struct socket *so)
1316 {
1317         struct inpcb *inp;
1318         struct tcpcb *tp = NULL;
1319         struct epoch_tracker et;
1320         TCPDEBUG0;
1321
1322         inp = sotoinpcb(so);
1323         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1324
1325         NET_EPOCH_ENTER(et);
1326         INP_WLOCK(inp);
1327         KASSERT(inp->inp_socket != NULL,
1328             ("tcp_usr_abort: inp_socket == NULL"));
1329
1330         /*
1331          * If we still have full TCP state, and we're not dropped, drop.
1332          */
1333         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1334             !(inp->inp_flags & INP_DROPPED)) {
1335                 tp = intotcpcb(inp);
1336                 TCPDEBUG1();
1337                 tp = tcp_drop(tp, ECONNABORTED);
1338                 if (tp == NULL)
1339                         goto dropped;
1340                 TCPDEBUG2(PRU_ABORT);
1341                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1342         }
1343         if (!(inp->inp_flags & INP_DROPPED)) {
1344                 SOCK_LOCK(so);
1345                 so->so_state |= SS_PROTOREF;
1346                 SOCK_UNLOCK(so);
1347                 inp->inp_flags |= INP_SOCKREF;
1348         }
1349         INP_WUNLOCK(inp);
1350 dropped:
1351         NET_EPOCH_EXIT(et);
1352 }
1353
1354 /*
1355  * TCP socket is closed.  Start friendly disconnect.
1356  */
1357 static void
1358 tcp_usr_close(struct socket *so)
1359 {
1360         struct inpcb *inp;
1361         struct tcpcb *tp = NULL;
1362         struct epoch_tracker et;
1363         TCPDEBUG0;
1364
1365         inp = sotoinpcb(so);
1366         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1367
1368         NET_EPOCH_ENTER(et);
1369         INP_WLOCK(inp);
1370         KASSERT(inp->inp_socket != NULL,
1371             ("tcp_usr_close: inp_socket == NULL"));
1372
1373         /*
1374          * If we still have full TCP state, and we're not dropped, initiate
1375          * a disconnect.
1376          */
1377         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1378             !(inp->inp_flags & INP_DROPPED)) {
1379                 tp = intotcpcb(inp);
1380                 TCPDEBUG1();
1381                 tcp_disconnect(tp);
1382                 TCPDEBUG2(PRU_CLOSE);
1383                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1384         }
1385         if (!(inp->inp_flags & INP_DROPPED)) {
1386                 SOCK_LOCK(so);
1387                 so->so_state |= SS_PROTOREF;
1388                 SOCK_UNLOCK(so);
1389                 inp->inp_flags |= INP_SOCKREF;
1390         }
1391         INP_WUNLOCK(inp);
1392         NET_EPOCH_EXIT(et);
1393 }
1394
1395 static int
1396 tcp_pru_options_support(struct tcpcb *tp, int flags)
1397 {
1398         /*
1399          * If the specific TCP stack has a pru_options
1400          * specified then it does not always support
1401          * all the PRU_XX options and we must ask it.
1402          * If the function is not specified then all
1403          * of the PRU_XX options are supported.
1404          */
1405         int ret = 0;
1406
1407         if (tp->t_fb->tfb_pru_options) {
1408                 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1409         }
1410         return (ret);
1411 }
1412
1413 /*
1414  * Receive out-of-band data.
1415  */
1416 static int
1417 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1418 {
1419         int error = 0;
1420         struct inpcb *inp;
1421         struct tcpcb *tp = NULL;
1422
1423         TCPDEBUG0;
1424         inp = sotoinpcb(so);
1425         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1426         INP_WLOCK(inp);
1427         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1428                 error = ECONNRESET;
1429                 goto out;
1430         }
1431         tp = intotcpcb(inp);
1432         error = tcp_pru_options_support(tp, PRUS_OOB);
1433         if (error) {
1434                 goto out;
1435         }
1436         TCPDEBUG1();
1437         if ((so->so_oobmark == 0 &&
1438              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1439             so->so_options & SO_OOBINLINE ||
1440             tp->t_oobflags & TCPOOB_HADDATA) {
1441                 error = EINVAL;
1442                 goto out;
1443         }
1444         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1445                 error = EWOULDBLOCK;
1446                 goto out;
1447         }
1448         m->m_len = 1;
1449         *mtod(m, caddr_t) = tp->t_iobc;
1450         if ((flags & MSG_PEEK) == 0)
1451                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1452
1453 out:
1454         TCPDEBUG2(PRU_RCVOOB);
1455         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1456         INP_WUNLOCK(inp);
1457         return (error);
1458 }
1459
1460 #ifdef INET
1461 struct pr_usrreqs tcp_usrreqs = {
1462         .pru_abort =            tcp_usr_abort,
1463         .pru_accept =           tcp_usr_accept,
1464         .pru_attach =           tcp_usr_attach,
1465         .pru_bind =             tcp_usr_bind,
1466         .pru_connect =          tcp_usr_connect,
1467         .pru_control =          in_control,
1468         .pru_detach =           tcp_usr_detach,
1469         .pru_disconnect =       tcp_usr_disconnect,
1470         .pru_listen =           tcp_usr_listen,
1471         .pru_peeraddr =         in_getpeeraddr,
1472         .pru_rcvd =             tcp_usr_rcvd,
1473         .pru_rcvoob =           tcp_usr_rcvoob,
1474         .pru_send =             tcp_usr_send,
1475         .pru_ready =            tcp_usr_ready,
1476         .pru_shutdown =         tcp_usr_shutdown,
1477         .pru_sockaddr =         in_getsockaddr,
1478         .pru_sosetlabel =       in_pcbsosetlabel,
1479         .pru_close =            tcp_usr_close,
1480 };
1481 #endif /* INET */
1482
1483 #ifdef INET6
1484 struct pr_usrreqs tcp6_usrreqs = {
1485         .pru_abort =            tcp_usr_abort,
1486         .pru_accept =           tcp6_usr_accept,
1487         .pru_attach =           tcp_usr_attach,
1488         .pru_bind =             tcp6_usr_bind,
1489         .pru_connect =          tcp6_usr_connect,
1490         .pru_control =          in6_control,
1491         .pru_detach =           tcp_usr_detach,
1492         .pru_disconnect =       tcp_usr_disconnect,
1493         .pru_listen =           tcp6_usr_listen,
1494         .pru_peeraddr =         in6_mapped_peeraddr,
1495         .pru_rcvd =             tcp_usr_rcvd,
1496         .pru_rcvoob =           tcp_usr_rcvoob,
1497         .pru_send =             tcp_usr_send,
1498         .pru_ready =            tcp_usr_ready,
1499         .pru_shutdown =         tcp_usr_shutdown,
1500         .pru_sockaddr =         in6_mapped_sockaddr,
1501         .pru_sosetlabel =       in_pcbsosetlabel,
1502         .pru_close =            tcp_usr_close,
1503 };
1504 #endif /* INET6 */
1505
1506 #ifdef INET
1507 /*
1508  * Common subroutine to open a TCP connection to remote host specified
1509  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1510  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1511  * to choose a local host address (interface).  If there is an existing
1512  * incarnation of the same connection in TIME-WAIT state and if the remote
1513  * host was sending CC options and if the connection duration was < MSL, then
1514  * truncate the previous TIME-WAIT state and proceed.
1515  * Initialize connection parameters and enter SYN-SENT state.
1516  */
1517 static int
1518 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1519 {
1520         struct inpcb *inp = tp->t_inpcb, *oinp;
1521         struct socket *so = inp->inp_socket;
1522         struct in_addr laddr;
1523         u_short lport;
1524         int error;
1525
1526         NET_EPOCH_ASSERT();
1527         INP_WLOCK_ASSERT(inp);
1528         INP_HASH_WLOCK(&V_tcbinfo);
1529
1530         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1531                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1532                 if (error)
1533                         goto out;
1534         }
1535
1536         /*
1537          * Cannot simply call in_pcbconnect, because there might be an
1538          * earlier incarnation of this same connection still in
1539          * TIME_WAIT state, creating an ADDRINUSE error.
1540          */
1541         laddr = inp->inp_laddr;
1542         lport = inp->inp_lport;
1543         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1544             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1545         if (error && oinp == NULL)
1546                 goto out;
1547         if (oinp) {
1548                 error = EADDRINUSE;
1549                 goto out;
1550         }
1551         /* Handle initial bind if it hadn't been done in advance. */
1552         if (inp->inp_lport == 0) {
1553                 inp->inp_lport = lport;
1554                 if (in_pcbinshash(inp) != 0) {
1555                         inp->inp_lport = 0;
1556                         error = EAGAIN;
1557                         goto out;
1558                 }
1559         }
1560         inp->inp_laddr = laddr;
1561         in_pcbrehash(inp);
1562         INP_HASH_WUNLOCK(&V_tcbinfo);
1563
1564         /*
1565          * Compute window scaling to request:
1566          * Scale to fit into sweet spot.  See tcp_syncache.c.
1567          * XXX: This should move to tcp_output().
1568          */
1569         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1570             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1571                 tp->request_r_scale++;
1572
1573         soisconnecting(so);
1574         TCPSTAT_INC(tcps_connattempt);
1575         tcp_state_change(tp, TCPS_SYN_SENT);
1576         tp->iss = tcp_new_isn(&inp->inp_inc);
1577         if (tp->t_flags & TF_REQ_TSTMP)
1578                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1579         tcp_sendseqinit(tp);
1580
1581         return 0;
1582
1583 out:
1584         INP_HASH_WUNLOCK(&V_tcbinfo);
1585         return (error);
1586 }
1587 #endif /* INET */
1588
1589 #ifdef INET6
1590 static int
1591 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1592 {
1593         struct inpcb *inp = tp->t_inpcb;
1594         int error;
1595
1596         INP_WLOCK_ASSERT(inp);
1597         INP_HASH_WLOCK(&V_tcbinfo);
1598
1599         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1600                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1601                 if (error)
1602                         goto out;
1603         }
1604         error = in6_pcbconnect(inp, nam, td->td_ucred);
1605         if (error != 0)
1606                 goto out;
1607         INP_HASH_WUNLOCK(&V_tcbinfo);
1608
1609         /* Compute window scaling to request.  */
1610         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1611             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1612                 tp->request_r_scale++;
1613
1614         soisconnecting(inp->inp_socket);
1615         TCPSTAT_INC(tcps_connattempt);
1616         tcp_state_change(tp, TCPS_SYN_SENT);
1617         tp->iss = tcp_new_isn(&inp->inp_inc);
1618         if (tp->t_flags & TF_REQ_TSTMP)
1619                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1620         tcp_sendseqinit(tp);
1621
1622         return 0;
1623
1624 out:
1625         INP_HASH_WUNLOCK(&V_tcbinfo);
1626         return error;
1627 }
1628 #endif /* INET6 */
1629
1630 /*
1631  * Export TCP internal state information via a struct tcp_info, based on the
1632  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1633  * (TCP state machine, etc).  We export all information using FreeBSD-native
1634  * constants -- for example, the numeric values for tcpi_state will differ
1635  * from Linux.
1636  */
1637 static void
1638 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1639 {
1640
1641         INP_WLOCK_ASSERT(tp->t_inpcb);
1642         bzero(ti, sizeof(*ti));
1643
1644         ti->tcpi_state = tp->t_state;
1645         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1646                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1647         if (tp->t_flags & TF_SACK_PERMIT)
1648                 ti->tcpi_options |= TCPI_OPT_SACK;
1649         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1650                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1651                 ti->tcpi_snd_wscale = tp->snd_scale;
1652                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1653         }
1654         if (tp->t_flags2 & TF2_ECN_PERMIT)
1655                 ti->tcpi_options |= TCPI_OPT_ECN;
1656
1657         ti->tcpi_rto = tp->t_rxtcur * tick;
1658         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1659         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1660         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1661
1662         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1663         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1664
1665         /*
1666          * FreeBSD-specific extension fields for tcp_info.
1667          */
1668         ti->tcpi_rcv_space = tp->rcv_wnd;
1669         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1670         ti->tcpi_snd_wnd = tp->snd_wnd;
1671         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1672         ti->tcpi_snd_nxt = tp->snd_nxt;
1673         ti->tcpi_snd_mss = tp->t_maxseg;
1674         ti->tcpi_rcv_mss = tp->t_maxseg;
1675         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1676         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1677         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1678 #ifdef TCP_OFFLOAD
1679         if (tp->t_flags & TF_TOE) {
1680                 ti->tcpi_options |= TCPI_OPT_TOE;
1681                 tcp_offload_tcp_info(tp, ti);
1682         }
1683 #endif
1684 }
1685
1686 /*
1687  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1688  * socket option arguments.  When it re-acquires the lock after the copy, it
1689  * has to revalidate that the connection is still valid for the socket
1690  * option.
1691  */
1692 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1693         INP_WLOCK(inp);                                                 \
1694         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1695                 INP_WUNLOCK(inp);                                       \
1696                 cleanup;                                                \
1697                 return (ECONNRESET);                                    \
1698         }                                                               \
1699         tp = intotcpcb(inp);                                            \
1700 } while(0)
1701 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1702
1703 int
1704 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1705 {
1706         int     error;
1707         struct  inpcb *inp;
1708         struct  tcpcb *tp;
1709         struct tcp_function_block *blk;
1710         struct tcp_function_set fsn;
1711
1712         error = 0;
1713         inp = sotoinpcb(so);
1714         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1715         if (sopt->sopt_level != IPPROTO_TCP) {
1716 #ifdef INET6
1717                 if (inp->inp_vflag & INP_IPV6PROTO) {
1718                         error = ip6_ctloutput(so, sopt);
1719                         /*
1720                          * In case of the IPV6_USE_MIN_MTU socket option,
1721                          * the INC_IPV6MINMTU flag to announce a corresponding
1722                          * MSS during the initial handshake.
1723                          * If the TCP connection is not in the front states,
1724                          * just reduce the MSS being used.
1725                          * This avoids the sending of TCP segments which will
1726                          * be fragmented at the IPv6 layer.
1727                          */
1728                         if ((error == 0) &&
1729                             (sopt->sopt_dir == SOPT_SET) &&
1730                             (sopt->sopt_level == IPPROTO_IPV6) &&
1731                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1732                                 INP_WLOCK(inp);
1733                                 if ((inp->inp_flags &
1734                                     (INP_TIMEWAIT | INP_DROPPED))) {
1735                                         INP_WUNLOCK(inp);
1736                                         return (ECONNRESET);
1737                                 }
1738                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1739                                 tp = intotcpcb(inp);
1740                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1741                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1742                                         struct ip6_pktopts *opt;
1743
1744                                         opt = inp->in6p_outputopts;
1745                                         if ((opt != NULL) &&
1746                                             (opt->ip6po_minmtu ==
1747                                             IP6PO_MINMTU_ALL)) {
1748                                                 if (tp->t_maxseg > TCP6_MSS) {
1749                                                         tp->t_maxseg = TCP6_MSS;
1750                                                 }
1751                                         }
1752                                 }
1753                                 INP_WUNLOCK(inp);
1754                         }
1755                 }
1756 #endif /* INET6 */
1757 #if defined(INET6) && defined(INET)
1758                 else
1759 #endif
1760 #ifdef INET
1761                 {
1762                         error = ip_ctloutput(so, sopt);
1763                 }
1764 #endif
1765                 return (error);
1766         }
1767         INP_WLOCK(inp);
1768         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1769                 INP_WUNLOCK(inp);
1770                 return (ECONNRESET);
1771         }
1772         tp = intotcpcb(inp);
1773         /*
1774          * Protect the TCP option TCP_FUNCTION_BLK so
1775          * that a sub-function can *never* overwrite this.
1776          */
1777         if ((sopt->sopt_dir == SOPT_SET) &&
1778             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1779                 INP_WUNLOCK(inp);
1780                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1781                     sizeof fsn);
1782                 if (error)
1783                         return (error);
1784                 INP_WLOCK_RECHECK(inp);
1785                 blk = find_and_ref_tcp_functions(&fsn);
1786                 if (blk == NULL) {
1787                         INP_WUNLOCK(inp);
1788                         return (ENOENT);
1789                 }
1790                 if (tp->t_fb == blk) {
1791                         /* You already have this */
1792                         refcount_release(&blk->tfb_refcnt);
1793                         INP_WUNLOCK(inp);
1794                         return (0);
1795                 }
1796                 if (tp->t_state != TCPS_CLOSED) {
1797                         /*
1798                          * The user has advanced the state
1799                          * past the initial point, we may not
1800                          * be able to switch.
1801                          */
1802                         if (blk->tfb_tcp_handoff_ok != NULL) {
1803                                 /*
1804                                  * Does the stack provide a
1805                                  * query mechanism, if so it may
1806                                  * still be possible?
1807                                  */
1808                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1809                         } else
1810                                 error = EINVAL;
1811                         if (error) {
1812                                 refcount_release(&blk->tfb_refcnt);
1813                                 INP_WUNLOCK(inp);
1814                                 return(error);
1815                         }
1816                 }
1817                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1818                         refcount_release(&blk->tfb_refcnt);
1819                         INP_WUNLOCK(inp);
1820                         return (ENOENT);
1821                 }
1822                 /*
1823                  * Release the old refcnt, the
1824                  * lookup acquired a ref on the
1825                  * new one already.
1826                  */
1827                 if (tp->t_fb->tfb_tcp_fb_fini) {
1828                         struct epoch_tracker et;
1829                         /*
1830                          * Tell the stack to cleanup with 0 i.e.
1831                          * the tcb is not going away.
1832                          */
1833                         NET_EPOCH_ENTER(et);
1834                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1835                         NET_EPOCH_EXIT(et);
1836                 }
1837 #ifdef TCPHPTS
1838                 /* Assure that we are not on any hpts */
1839                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1840 #endif
1841                 if (blk->tfb_tcp_fb_init) {
1842                         error = (*blk->tfb_tcp_fb_init)(tp);
1843                         if (error) {
1844                                 refcount_release(&blk->tfb_refcnt);
1845                                 if (tp->t_fb->tfb_tcp_fb_init) {
1846                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1847                                                 /* Fall back failed, drop the connection */
1848                                                 INP_WUNLOCK(inp);
1849                                                 soabort(so);
1850                                                 return(error);
1851                                         }
1852                                 }
1853                                 goto err_out;
1854                         }
1855                 }
1856                 refcount_release(&tp->t_fb->tfb_refcnt);
1857                 tp->t_fb = blk;
1858 #ifdef TCP_OFFLOAD
1859                 if (tp->t_flags & TF_TOE) {
1860                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1861                              sopt->sopt_name);
1862                 }
1863 #endif
1864 err_out:
1865                 INP_WUNLOCK(inp);
1866                 return (error);
1867         } else if ((sopt->sopt_dir == SOPT_GET) &&
1868             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1869                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1870                     TCP_FUNCTION_NAME_LEN_MAX);
1871                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1872                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1873                 INP_WUNLOCK(inp);
1874                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1875                 return (error);
1876         }
1877         /* Pass in the INP locked, called must unlock it */
1878         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1879 }
1880
1881 /*
1882  * If this assert becomes untrue, we need to change the size of the buf
1883  * variable in tcp_default_ctloutput().
1884  */
1885 #ifdef CTASSERT
1886 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1887 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1888 #endif
1889
1890 #ifdef KERN_TLS
1891 static int
1892 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1893 {
1894         struct tls_enable_v0 tls_v0;
1895         int error;
1896
1897         if (sopt->sopt_valsize == sizeof(tls_v0)) {
1898                 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1899                     sizeof(tls_v0));
1900                 if (error)
1901                         return (error);
1902                 memset(tls, 0, sizeof(*tls));
1903                 tls->cipher_key = tls_v0.cipher_key;
1904                 tls->iv = tls_v0.iv;
1905                 tls->auth_key = tls_v0.auth_key;
1906                 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1907                 tls->cipher_key_len = tls_v0.cipher_key_len;
1908                 tls->iv_len = tls_v0.iv_len;
1909                 tls->auth_algorithm = tls_v0.auth_algorithm;
1910                 tls->auth_key_len = tls_v0.auth_key_len;
1911                 tls->flags = tls_v0.flags;
1912                 tls->tls_vmajor = tls_v0.tls_vmajor;
1913                 tls->tls_vminor = tls_v0.tls_vminor;
1914                 return (0);
1915         }
1916
1917         return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
1918 }
1919 #endif
1920
1921 int
1922 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1923 {
1924         int     error, opt, optval;
1925         u_int   ui;
1926         struct  tcp_info ti;
1927 #ifdef KERN_TLS
1928         struct tls_enable tls;
1929 #endif
1930         struct cc_algo *algo;
1931         char    *pbuf, buf[TCP_LOG_ID_LEN];
1932 #ifdef STATS
1933         struct statsblob *sbp;
1934 #endif
1935         size_t  len;
1936
1937         /*
1938          * For TCP_CCALGOOPT forward the control to CC module, for both
1939          * SOPT_SET and SOPT_GET.
1940          */
1941         switch (sopt->sopt_name) {
1942         case TCP_CCALGOOPT:
1943                 INP_WUNLOCK(inp);
1944                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1945                         return (EINVAL);
1946                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1947                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1948                     sopt->sopt_valsize);
1949                 if (error) {
1950                         free(pbuf, M_TEMP);
1951                         return (error);
1952                 }
1953                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1954                 if (CC_ALGO(tp)->ctl_output != NULL)
1955                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1956                 else
1957                         error = ENOENT;
1958                 INP_WUNLOCK(inp);
1959                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1960                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1961                 free(pbuf, M_TEMP);
1962                 return (error);
1963         }
1964
1965         switch (sopt->sopt_dir) {
1966         case SOPT_SET:
1967                 switch (sopt->sopt_name) {
1968 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1969                 case TCP_MD5SIG:
1970                         if (!TCPMD5_ENABLED()) {
1971                                 INP_WUNLOCK(inp);
1972                                 return (ENOPROTOOPT);
1973                         }
1974                         error = TCPMD5_PCBCTL(inp, sopt);
1975                         if (error)
1976                                 return (error);
1977                         goto unlock_and_done;
1978 #endif /* IPSEC */
1979
1980                 case TCP_NODELAY:
1981                 case TCP_NOOPT:
1982                 case TCP_LRD:
1983                         INP_WUNLOCK(inp);
1984                         error = sooptcopyin(sopt, &optval, sizeof optval,
1985                             sizeof optval);
1986                         if (error)
1987                                 return (error);
1988
1989                         INP_WLOCK_RECHECK(inp);
1990                         switch (sopt->sopt_name) {
1991                         case TCP_NODELAY:
1992                                 opt = TF_NODELAY;
1993                                 break;
1994                         case TCP_NOOPT:
1995                                 opt = TF_NOOPT;
1996                                 break;
1997                         case TCP_LRD:
1998                                 opt = TF_LRD;
1999                                 break;
2000                         default:
2001                                 opt = 0; /* dead code to fool gcc */
2002                                 break;
2003                         }
2004
2005                         if (optval)
2006                                 tp->t_flags |= opt;
2007                         else
2008                                 tp->t_flags &= ~opt;
2009 unlock_and_done:
2010 #ifdef TCP_OFFLOAD
2011                         if (tp->t_flags & TF_TOE) {
2012                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2013                                     sopt->sopt_name);
2014                         }
2015 #endif
2016                         INP_WUNLOCK(inp);
2017                         break;
2018
2019                 case TCP_NOPUSH:
2020                         INP_WUNLOCK(inp);
2021                         error = sooptcopyin(sopt, &optval, sizeof optval,
2022                             sizeof optval);
2023                         if (error)
2024                                 return (error);
2025
2026                         INP_WLOCK_RECHECK(inp);
2027                         if (optval)
2028                                 tp->t_flags |= TF_NOPUSH;
2029                         else if (tp->t_flags & TF_NOPUSH) {
2030                                 tp->t_flags &= ~TF_NOPUSH;
2031                                 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2032                                         struct epoch_tracker et;
2033
2034                                         NET_EPOCH_ENTER(et);
2035                                         error = tp->t_fb->tfb_tcp_output(tp);
2036                                         NET_EPOCH_EXIT(et);
2037                                 }
2038                         }
2039                         goto unlock_and_done;
2040
2041                 case TCP_REMOTE_UDP_ENCAPS_PORT:
2042                         INP_WUNLOCK(inp);
2043                         error = sooptcopyin(sopt, &optval, sizeof optval,
2044                             sizeof optval);
2045                         if (error)
2046                                 return (error);
2047                         if ((optval < TCP_TUNNELING_PORT_MIN) ||
2048                             (optval > TCP_TUNNELING_PORT_MAX)) {
2049                                 /* Its got to be in range */
2050                                 return (EINVAL);
2051                         }
2052                         if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2053                                 /* You have to have enabled a UDP tunneling port first */
2054                                 return (EINVAL);
2055                         }
2056                         INP_WLOCK_RECHECK(inp);
2057                         if (tp->t_state != TCPS_CLOSED) {
2058                                 /* You can't change after you are connected */
2059                                 error = EINVAL;
2060                         } else {
2061                                 /* Ok we are all good set the port */
2062                                 tp->t_port = htons(optval);
2063                         }
2064                         goto unlock_and_done;
2065
2066                 case TCP_MAXSEG:
2067                         INP_WUNLOCK(inp);
2068                         error = sooptcopyin(sopt, &optval, sizeof optval,
2069                             sizeof optval);
2070                         if (error)
2071                                 return (error);
2072
2073                         INP_WLOCK_RECHECK(inp);
2074                         if (optval > 0 && optval <= tp->t_maxseg &&
2075                             optval + 40 >= V_tcp_minmss)
2076                                 tp->t_maxseg = optval;
2077                         else
2078                                 error = EINVAL;
2079                         goto unlock_and_done;
2080
2081                 case TCP_INFO:
2082                         INP_WUNLOCK(inp);
2083                         error = EINVAL;
2084                         break;
2085
2086                 case TCP_STATS:
2087                         INP_WUNLOCK(inp);
2088 #ifdef STATS
2089                         error = sooptcopyin(sopt, &optval, sizeof optval,
2090                             sizeof optval);
2091                         if (error)
2092                                 return (error);
2093
2094                         if (optval > 0)
2095                                 sbp = stats_blob_alloc(
2096                                     V_tcp_perconn_stats_dflt_tpl, 0);
2097                         else
2098                                 sbp = NULL;
2099
2100                         INP_WLOCK_RECHECK(inp);
2101                         if ((tp->t_stats != NULL && sbp == NULL) ||
2102                             (tp->t_stats == NULL && sbp != NULL)) {
2103                                 struct statsblob *t = tp->t_stats;
2104                                 tp->t_stats = sbp;
2105                                 sbp = t;
2106                         }
2107                         INP_WUNLOCK(inp);
2108
2109                         stats_blob_destroy(sbp);
2110 #else
2111                         return (EOPNOTSUPP);
2112 #endif /* !STATS */
2113                         break;
2114
2115                 case TCP_CONGESTION:
2116                         INP_WUNLOCK(inp);
2117                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2118                         if (error)
2119                                 break;
2120                         buf[sopt->sopt_valsize] = '\0';
2121                         INP_WLOCK_RECHECK(inp);
2122                         CC_LIST_RLOCK();
2123                         STAILQ_FOREACH(algo, &cc_list, entries)
2124                                 if (strncmp(buf, algo->name,
2125                                     TCP_CA_NAME_MAX) == 0)
2126                                         break;
2127                         CC_LIST_RUNLOCK();
2128                         if (algo == NULL) {
2129                                 INP_WUNLOCK(inp);
2130                                 error = EINVAL;
2131                                 break;
2132                         }
2133                         /*
2134                          * We hold a write lock over the tcb so it's safe to
2135                          * do these things without ordering concerns.
2136                          */
2137                         if (CC_ALGO(tp)->cb_destroy != NULL)
2138                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
2139                         CC_DATA(tp) = NULL;
2140                         CC_ALGO(tp) = algo;
2141                         /*
2142                          * If something goes pear shaped initialising the new
2143                          * algo, fall back to newreno (which does not
2144                          * require initialisation).
2145                          */
2146                         if (algo->cb_init != NULL &&
2147                             algo->cb_init(tp->ccv) != 0) {
2148                                 CC_ALGO(tp) = &newreno_cc_algo;
2149                                 /*
2150                                  * The only reason init should fail is
2151                                  * because of malloc.
2152                                  */
2153                                 error = ENOMEM;
2154                         }
2155                         INP_WUNLOCK(inp);
2156                         break;
2157
2158                 case TCP_REUSPORT_LB_NUMA:
2159                         INP_WUNLOCK(inp);
2160                         error = sooptcopyin(sopt, &optval, sizeof(optval),
2161                             sizeof(optval));
2162                         INP_WLOCK_RECHECK(inp);
2163                         if (!error)
2164                                 error = in_pcblbgroup_numa(inp, optval);
2165                         INP_WUNLOCK(inp);
2166                         break;
2167
2168 #ifdef KERN_TLS
2169                 case TCP_TXTLS_ENABLE:
2170                         INP_WUNLOCK(inp);
2171                         error = copyin_tls_enable(sopt, &tls);
2172                         if (error)
2173                                 break;
2174                         error = ktls_enable_tx(so, &tls);
2175                         break;
2176                 case TCP_TXTLS_MODE:
2177                         INP_WUNLOCK(inp);
2178                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2179                         if (error)
2180                                 return (error);
2181
2182                         INP_WLOCK_RECHECK(inp);
2183                         error = ktls_set_tx_mode(so, ui);
2184                         INP_WUNLOCK(inp);
2185                         break;
2186                 case TCP_RXTLS_ENABLE:
2187                         INP_WUNLOCK(inp);
2188                         error = sooptcopyin(sopt, &tls, sizeof(tls),
2189                             sizeof(tls));
2190                         if (error)
2191                                 break;
2192                         error = ktls_enable_rx(so, &tls);
2193                         break;
2194 #endif
2195
2196                 case TCP_KEEPIDLE:
2197                 case TCP_KEEPINTVL:
2198                 case TCP_KEEPINIT:
2199                         INP_WUNLOCK(inp);
2200                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2201                         if (error)
2202                                 return (error);
2203
2204                         if (ui > (UINT_MAX / hz)) {
2205                                 error = EINVAL;
2206                                 break;
2207                         }
2208                         ui *= hz;
2209
2210                         INP_WLOCK_RECHECK(inp);
2211                         switch (sopt->sopt_name) {
2212                         case TCP_KEEPIDLE:
2213                                 tp->t_keepidle = ui;
2214                                 /*
2215                                  * XXX: better check current remaining
2216                                  * timeout and "merge" it with new value.
2217                                  */
2218                                 if ((tp->t_state > TCPS_LISTEN) &&
2219                                     (tp->t_state <= TCPS_CLOSING))
2220                                         tcp_timer_activate(tp, TT_KEEP,
2221                                             TP_KEEPIDLE(tp));
2222                                 break;
2223                         case TCP_KEEPINTVL:
2224                                 tp->t_keepintvl = ui;
2225                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2226                                     (TP_MAXIDLE(tp) > 0))
2227                                         tcp_timer_activate(tp, TT_2MSL,
2228                                             TP_MAXIDLE(tp));
2229                                 break;
2230                         case TCP_KEEPINIT:
2231                                 tp->t_keepinit = ui;
2232                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
2233                                     tp->t_state == TCPS_SYN_SENT)
2234                                         tcp_timer_activate(tp, TT_KEEP,
2235                                             TP_KEEPINIT(tp));
2236                                 break;
2237                         }
2238                         goto unlock_and_done;
2239
2240                 case TCP_KEEPCNT:
2241                         INP_WUNLOCK(inp);
2242                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2243                         if (error)
2244                                 return (error);
2245
2246                         INP_WLOCK_RECHECK(inp);
2247                         tp->t_keepcnt = ui;
2248                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2249                             (TP_MAXIDLE(tp) > 0))
2250                                 tcp_timer_activate(tp, TT_2MSL,
2251                                     TP_MAXIDLE(tp));
2252                         goto unlock_and_done;
2253
2254 #ifdef TCPPCAP
2255                 case TCP_PCAP_OUT:
2256                 case TCP_PCAP_IN:
2257                         INP_WUNLOCK(inp);
2258                         error = sooptcopyin(sopt, &optval, sizeof optval,
2259                             sizeof optval);
2260                         if (error)
2261                                 return (error);
2262
2263                         INP_WLOCK_RECHECK(inp);
2264                         if (optval >= 0)
2265                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2266                                         &(tp->t_outpkts) : &(tp->t_inpkts),
2267                                         optval);
2268                         else
2269                                 error = EINVAL;
2270                         goto unlock_and_done;
2271 #endif
2272
2273                 case TCP_FASTOPEN: {
2274                         struct tcp_fastopen tfo_optval;
2275
2276                         INP_WUNLOCK(inp);
2277                         if (!V_tcp_fastopen_client_enable &&
2278                             !V_tcp_fastopen_server_enable)
2279                                 return (EPERM);
2280
2281                         error = sooptcopyin(sopt, &tfo_optval,
2282                                     sizeof(tfo_optval), sizeof(int));
2283                         if (error)
2284                                 return (error);
2285
2286                         INP_WLOCK_RECHECK(inp);
2287                         if ((tp->t_state != TCPS_CLOSED) &&
2288                             (tp->t_state != TCPS_LISTEN)) {
2289                                 error = EINVAL;
2290                                 goto unlock_and_done;
2291                         }
2292                         if (tfo_optval.enable) {
2293                                 if (tp->t_state == TCPS_LISTEN) {
2294                                         if (!V_tcp_fastopen_server_enable) {
2295                                                 error = EPERM;
2296                                                 goto unlock_and_done;
2297                                         }
2298
2299                                         if (tp->t_tfo_pending == NULL)
2300                                                 tp->t_tfo_pending =
2301                                                     tcp_fastopen_alloc_counter();
2302                                 } else {
2303                                         /*
2304                                          * If a pre-shared key was provided,
2305                                          * stash it in the client cookie
2306                                          * field of the tcpcb for use during
2307                                          * connect.
2308                                          */
2309                                         if (sopt->sopt_valsize ==
2310                                             sizeof(tfo_optval)) {
2311                                                 memcpy(tp->t_tfo_cookie.client,
2312                                                        tfo_optval.psk,
2313                                                        TCP_FASTOPEN_PSK_LEN);
2314                                                 tp->t_tfo_client_cookie_len =
2315                                                     TCP_FASTOPEN_PSK_LEN;
2316                                         }
2317                                 }
2318                                 tp->t_flags |= TF_FASTOPEN;
2319                         } else
2320                                 tp->t_flags &= ~TF_FASTOPEN;
2321                         goto unlock_and_done;
2322                 }
2323
2324 #ifdef TCP_BLACKBOX
2325                 case TCP_LOG:
2326                         INP_WUNLOCK(inp);
2327                         error = sooptcopyin(sopt, &optval, sizeof optval,
2328                             sizeof optval);
2329                         if (error)
2330                                 return (error);
2331
2332                         INP_WLOCK_RECHECK(inp);
2333                         error = tcp_log_state_change(tp, optval);
2334                         goto unlock_and_done;
2335
2336                 case TCP_LOGBUF:
2337                         INP_WUNLOCK(inp);
2338                         error = EINVAL;
2339                         break;
2340
2341                 case TCP_LOGID:
2342                         INP_WUNLOCK(inp);
2343                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2344                         if (error)
2345                                 break;
2346                         buf[sopt->sopt_valsize] = '\0';
2347                         INP_WLOCK_RECHECK(inp);
2348                         error = tcp_log_set_id(tp, buf);
2349                         /* tcp_log_set_id() unlocks the INP. */
2350                         break;
2351
2352                 case TCP_LOGDUMP:
2353                 case TCP_LOGDUMPID:
2354                         INP_WUNLOCK(inp);
2355                         error =
2356                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2357                         if (error)
2358                                 break;
2359                         buf[sopt->sopt_valsize] = '\0';
2360                         INP_WLOCK_RECHECK(inp);
2361                         if (sopt->sopt_name == TCP_LOGDUMP) {
2362                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2363                                     M_WAITOK, true);
2364                                 INP_WUNLOCK(inp);
2365                         } else {
2366                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2367                                 /*
2368                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2369                                  * INP lock.
2370                                  */
2371                         }
2372                         break;
2373 #endif
2374
2375                 default:
2376                         INP_WUNLOCK(inp);
2377                         error = ENOPROTOOPT;
2378                         break;
2379                 }
2380                 break;
2381
2382         case SOPT_GET:
2383                 tp = intotcpcb(inp);
2384                 switch (sopt->sopt_name) {
2385 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2386                 case TCP_MD5SIG:
2387                         if (!TCPMD5_ENABLED()) {
2388                                 INP_WUNLOCK(inp);
2389                                 return (ENOPROTOOPT);
2390                         }
2391                         error = TCPMD5_PCBCTL(inp, sopt);
2392                         break;
2393 #endif
2394
2395                 case TCP_NODELAY:
2396                         optval = tp->t_flags & TF_NODELAY;
2397                         INP_WUNLOCK(inp);
2398                         error = sooptcopyout(sopt, &optval, sizeof optval);
2399                         break;
2400                 case TCP_MAXSEG:
2401                         optval = tp->t_maxseg;
2402                         INP_WUNLOCK(inp);
2403                         error = sooptcopyout(sopt, &optval, sizeof optval);
2404                         break;
2405                 case TCP_REMOTE_UDP_ENCAPS_PORT:
2406                         optval = ntohs(tp->t_port);
2407                         INP_WUNLOCK(inp);
2408                         error = sooptcopyout(sopt, &optval, sizeof optval);
2409                         break;
2410                 case TCP_NOOPT:
2411                         optval = tp->t_flags & TF_NOOPT;
2412                         INP_WUNLOCK(inp);
2413                         error = sooptcopyout(sopt, &optval, sizeof optval);
2414                         break;
2415                 case TCP_NOPUSH:
2416                         optval = tp->t_flags & TF_NOPUSH;
2417                         INP_WUNLOCK(inp);
2418                         error = sooptcopyout(sopt, &optval, sizeof optval);
2419                         break;
2420                 case TCP_INFO:
2421                         tcp_fill_info(tp, &ti);
2422                         INP_WUNLOCK(inp);
2423                         error = sooptcopyout(sopt, &ti, sizeof ti);
2424                         break;
2425                 case TCP_STATS:
2426                         {
2427 #ifdef STATS
2428                         int nheld;
2429                         TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2430
2431                         error = 0;
2432                         socklen_t outsbsz = sopt->sopt_valsize;
2433                         if (tp->t_stats == NULL)
2434                                 error = ENOENT;
2435                         else if (outsbsz >= tp->t_stats->cursz)
2436                                 outsbsz = tp->t_stats->cursz;
2437                         else if (outsbsz >= sizeof(struct statsblob))
2438                                 outsbsz = sizeof(struct statsblob);
2439                         else
2440                                 error = EINVAL;
2441                         INP_WUNLOCK(inp);
2442                         if (error)
2443                                 break;
2444
2445                         sbp = sopt->sopt_val;
2446                         nheld = atop(round_page(((vm_offset_t)sbp) +
2447                             (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2448                         vm_page_t ma[nheld];
2449                         if (vm_fault_quick_hold_pages(
2450                             &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2451                             outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2452                             nheld) < 0) {
2453                                 error = EFAULT;
2454                                 break;
2455                         }
2456
2457                         if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2458                             SIZEOF_MEMBER(struct statsblob, flags))))
2459                                 goto unhold;
2460
2461                         INP_WLOCK_RECHECK(inp);
2462                         error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2463                             sbflags | SB_CLONE_USRDSTNOFAULT);
2464                         INP_WUNLOCK(inp);
2465                         sopt->sopt_valsize = outsbsz;
2466 unhold:
2467                         vm_page_unhold_pages(ma, nheld);
2468 #else
2469                         INP_WUNLOCK(inp);
2470                         error = EOPNOTSUPP;
2471 #endif /* !STATS */
2472                         break;
2473                         }
2474                 case TCP_CONGESTION:
2475                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2476                         INP_WUNLOCK(inp);
2477                         error = sooptcopyout(sopt, buf, len + 1);
2478                         break;
2479                 case TCP_KEEPIDLE:
2480                 case TCP_KEEPINTVL:
2481                 case TCP_KEEPINIT:
2482                 case TCP_KEEPCNT:
2483                         switch (sopt->sopt_name) {
2484                         case TCP_KEEPIDLE:
2485                                 ui = TP_KEEPIDLE(tp) / hz;
2486                                 break;
2487                         case TCP_KEEPINTVL:
2488                                 ui = TP_KEEPINTVL(tp) / hz;
2489                                 break;
2490                         case TCP_KEEPINIT:
2491                                 ui = TP_KEEPINIT(tp) / hz;
2492                                 break;
2493                         case TCP_KEEPCNT:
2494                                 ui = TP_KEEPCNT(tp);
2495                                 break;
2496                         }
2497                         INP_WUNLOCK(inp);
2498                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2499                         break;
2500 #ifdef TCPPCAP
2501                 case TCP_PCAP_OUT:
2502                 case TCP_PCAP_IN:
2503                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2504                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2505                         INP_WUNLOCK(inp);
2506                         error = sooptcopyout(sopt, &optval, sizeof optval);
2507                         break;
2508 #endif
2509                 case TCP_FASTOPEN:
2510                         optval = tp->t_flags & TF_FASTOPEN;
2511                         INP_WUNLOCK(inp);
2512                         error = sooptcopyout(sopt, &optval, sizeof optval);
2513                         break;
2514 #ifdef TCP_BLACKBOX
2515                 case TCP_LOG:
2516                         optval = tp->t_logstate;
2517                         INP_WUNLOCK(inp);
2518                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2519                         break;
2520                 case TCP_LOGBUF:
2521                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2522                         error = tcp_log_getlogbuf(sopt, tp);
2523                         break;
2524                 case TCP_LOGID:
2525                         len = tcp_log_get_id(tp, buf);
2526                         INP_WUNLOCK(inp);
2527                         error = sooptcopyout(sopt, buf, len + 1);
2528                         break;
2529                 case TCP_LOGDUMP:
2530                 case TCP_LOGDUMPID:
2531                         INP_WUNLOCK(inp);
2532                         error = EINVAL;
2533                         break;
2534 #endif
2535 #ifdef KERN_TLS
2536                 case TCP_TXTLS_MODE:
2537                         optval = ktls_get_tx_mode(so);
2538                         INP_WUNLOCK(inp);
2539                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2540                         break;
2541                 case TCP_RXTLS_MODE:
2542                         optval = ktls_get_rx_mode(so);
2543                         INP_WUNLOCK(inp);
2544                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2545                         break;
2546 #endif
2547                 case TCP_LRD:
2548                         optval = tp->t_flags & TF_LRD;
2549                         INP_WUNLOCK(inp);
2550                         error = sooptcopyout(sopt, &optval, sizeof optval);
2551                         break;
2552                 default:
2553                         INP_WUNLOCK(inp);
2554                         error = ENOPROTOOPT;
2555                         break;
2556                 }
2557                 break;
2558         }
2559         return (error);
2560 }
2561 #undef INP_WLOCK_RECHECK
2562 #undef INP_WLOCK_RECHECK_CLEANUP
2563
2564 /*
2565  * Initiate (or continue) disconnect.
2566  * If embryonic state, just send reset (once).
2567  * If in ``let data drain'' option and linger null, just drop.
2568  * Otherwise (hard), mark socket disconnecting and drop
2569  * current input data; switch states based on user close, and
2570  * send segment to peer (with FIN).
2571  */
2572 static void
2573 tcp_disconnect(struct tcpcb *tp)
2574 {
2575         struct inpcb *inp = tp->t_inpcb;
2576         struct socket *so = inp->inp_socket;
2577
2578         NET_EPOCH_ASSERT();
2579         INP_WLOCK_ASSERT(inp);
2580
2581         /*
2582          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2583          * socket is still open.
2584          */
2585         if (tp->t_state < TCPS_ESTABLISHED &&
2586             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2587                 tp = tcp_close(tp);
2588                 KASSERT(tp != NULL,
2589                     ("tcp_disconnect: tcp_close() returned NULL"));
2590         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2591                 tp = tcp_drop(tp, 0);
2592                 KASSERT(tp != NULL,
2593                     ("tcp_disconnect: tcp_drop() returned NULL"));
2594         } else {
2595                 soisdisconnecting(so);
2596                 sbflush(&so->so_rcv);
2597                 tcp_usrclosed(tp);
2598                 if (!(inp->inp_flags & INP_DROPPED))
2599                         tp->t_fb->tfb_tcp_output(tp);
2600         }
2601 }
2602
2603 /*
2604  * User issued close, and wish to trail through shutdown states:
2605  * if never received SYN, just forget it.  If got a SYN from peer,
2606  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2607  * If already got a FIN from peer, then almost done; go to LAST_ACK
2608  * state.  In all other cases, have already sent FIN to peer (e.g.
2609  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2610  * for peer to send FIN or not respond to keep-alives, etc.
2611  * We can let the user exit from the close as soon as the FIN is acked.
2612  */
2613 static void
2614 tcp_usrclosed(struct tcpcb *tp)
2615 {
2616
2617         NET_EPOCH_ASSERT();
2618         INP_WLOCK_ASSERT(tp->t_inpcb);
2619
2620         switch (tp->t_state) {
2621         case TCPS_LISTEN:
2622 #ifdef TCP_OFFLOAD
2623                 tcp_offload_listen_stop(tp);
2624 #endif
2625                 tcp_state_change(tp, TCPS_CLOSED);
2626                 /* FALLTHROUGH */
2627         case TCPS_CLOSED:
2628                 tp = tcp_close(tp);
2629                 /*
2630                  * tcp_close() should never return NULL here as the socket is
2631                  * still open.
2632                  */
2633                 KASSERT(tp != NULL,
2634                     ("tcp_usrclosed: tcp_close() returned NULL"));
2635                 break;
2636
2637         case TCPS_SYN_SENT:
2638         case TCPS_SYN_RECEIVED:
2639                 tp->t_flags |= TF_NEEDFIN;
2640                 break;
2641
2642         case TCPS_ESTABLISHED:
2643                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2644                 break;
2645
2646         case TCPS_CLOSE_WAIT:
2647                 tcp_state_change(tp, TCPS_LAST_ACK);
2648                 break;
2649         }
2650         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2651                 soisdisconnected(tp->t_inpcb->inp_socket);
2652                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2653                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2654                         int timeout;
2655
2656                         timeout = (tcp_fast_finwait2_recycle) ?
2657                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2658                         tcp_timer_activate(tp, TT_2MSL, timeout);
2659                 }
2660         }
2661 }
2662
2663 #ifdef DDB
2664 static void
2665 db_print_indent(int indent)
2666 {
2667         int i;
2668
2669         for (i = 0; i < indent; i++)
2670                 db_printf(" ");
2671 }
2672
2673 static void
2674 db_print_tstate(int t_state)
2675 {
2676
2677         switch (t_state) {
2678         case TCPS_CLOSED:
2679                 db_printf("TCPS_CLOSED");
2680                 return;
2681
2682         case TCPS_LISTEN:
2683                 db_printf("TCPS_LISTEN");
2684                 return;
2685
2686         case TCPS_SYN_SENT:
2687                 db_printf("TCPS_SYN_SENT");
2688                 return;
2689
2690         case TCPS_SYN_RECEIVED:
2691                 db_printf("TCPS_SYN_RECEIVED");
2692                 return;
2693
2694         case TCPS_ESTABLISHED:
2695                 db_printf("TCPS_ESTABLISHED");
2696                 return;
2697
2698         case TCPS_CLOSE_WAIT:
2699                 db_printf("TCPS_CLOSE_WAIT");
2700                 return;
2701
2702         case TCPS_FIN_WAIT_1:
2703                 db_printf("TCPS_FIN_WAIT_1");
2704                 return;
2705
2706         case TCPS_CLOSING:
2707                 db_printf("TCPS_CLOSING");
2708                 return;
2709
2710         case TCPS_LAST_ACK:
2711                 db_printf("TCPS_LAST_ACK");
2712                 return;
2713
2714         case TCPS_FIN_WAIT_2:
2715                 db_printf("TCPS_FIN_WAIT_2");
2716                 return;
2717
2718         case TCPS_TIME_WAIT:
2719                 db_printf("TCPS_TIME_WAIT");
2720                 return;
2721
2722         default:
2723                 db_printf("unknown");
2724                 return;
2725         }
2726 }
2727
2728 static void
2729 db_print_tflags(u_int t_flags)
2730 {
2731         int comma;
2732
2733         comma = 0;
2734         if (t_flags & TF_ACKNOW) {
2735                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2736                 comma = 1;
2737         }
2738         if (t_flags & TF_DELACK) {
2739                 db_printf("%sTF_DELACK", comma ? ", " : "");
2740                 comma = 1;
2741         }
2742         if (t_flags & TF_NODELAY) {
2743                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2744                 comma = 1;
2745         }
2746         if (t_flags & TF_NOOPT) {
2747                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2748                 comma = 1;
2749         }
2750         if (t_flags & TF_SENTFIN) {
2751                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2752                 comma = 1;
2753         }
2754         if (t_flags & TF_REQ_SCALE) {
2755                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2756                 comma = 1;
2757         }
2758         if (t_flags & TF_RCVD_SCALE) {
2759                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2760                 comma = 1;
2761         }
2762         if (t_flags & TF_REQ_TSTMP) {
2763                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2764                 comma = 1;
2765         }
2766         if (t_flags & TF_RCVD_TSTMP) {
2767                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2768                 comma = 1;
2769         }
2770         if (t_flags & TF_SACK_PERMIT) {
2771                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2772                 comma = 1;
2773         }
2774         if (t_flags & TF_NEEDSYN) {
2775                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2776                 comma = 1;
2777         }
2778         if (t_flags & TF_NEEDFIN) {
2779                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2780                 comma = 1;
2781         }
2782         if (t_flags & TF_NOPUSH) {
2783                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2784                 comma = 1;
2785         }
2786         if (t_flags & TF_MORETOCOME) {
2787                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2788                 comma = 1;
2789         }
2790         if (t_flags & TF_LQ_OVERFLOW) {
2791                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2792                 comma = 1;
2793         }
2794         if (t_flags & TF_LASTIDLE) {
2795                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2796                 comma = 1;
2797         }
2798         if (t_flags & TF_RXWIN0SENT) {
2799                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2800                 comma = 1;
2801         }
2802         if (t_flags & TF_FASTRECOVERY) {
2803                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2804                 comma = 1;
2805         }
2806         if (t_flags & TF_CONGRECOVERY) {
2807                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2808                 comma = 1;
2809         }
2810         if (t_flags & TF_WASFRECOVERY) {
2811                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2812                 comma = 1;
2813         }
2814         if (t_flags & TF_SIGNATURE) {
2815                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2816                 comma = 1;
2817         }
2818         if (t_flags & TF_FORCEDATA) {
2819                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2820                 comma = 1;
2821         }
2822         if (t_flags & TF_TSO) {
2823                 db_printf("%sTF_TSO", comma ? ", " : "");
2824                 comma = 1;
2825         }
2826         if (t_flags & TF_FASTOPEN) {
2827                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2828                 comma = 1;
2829         }
2830 }
2831
2832 static void
2833 db_print_tflags2(u_int t_flags2)
2834 {
2835         int comma;
2836
2837         comma = 0;
2838         if (t_flags2 & TF2_ECN_PERMIT) {
2839                 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2840                 comma = 1;
2841         }
2842 }
2843
2844 static void
2845 db_print_toobflags(char t_oobflags)
2846 {
2847         int comma;
2848
2849         comma = 0;
2850         if (t_oobflags & TCPOOB_HAVEDATA) {
2851                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2852                 comma = 1;
2853         }
2854         if (t_oobflags & TCPOOB_HADDATA) {
2855                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2856                 comma = 1;
2857         }
2858 }
2859
2860 static void
2861 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2862 {
2863
2864         db_print_indent(indent);
2865         db_printf("%s at %p\n", name, tp);
2866
2867         indent += 2;
2868
2869         db_print_indent(indent);
2870         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2871            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2872
2873         db_print_indent(indent);
2874         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2875             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2876
2877         db_print_indent(indent);
2878         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2879             &tp->t_timers->tt_delack, tp->t_inpcb);
2880
2881         db_print_indent(indent);
2882         db_printf("t_state: %d (", tp->t_state);
2883         db_print_tstate(tp->t_state);
2884         db_printf(")\n");
2885
2886         db_print_indent(indent);
2887         db_printf("t_flags: 0x%x (", tp->t_flags);
2888         db_print_tflags(tp->t_flags);
2889         db_printf(")\n");
2890
2891         db_print_indent(indent);
2892         db_printf("t_flags2: 0x%x (", tp->t_flags2);
2893         db_print_tflags2(tp->t_flags2);
2894         db_printf(")\n");
2895
2896         db_print_indent(indent);
2897         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2898             tp->snd_una, tp->snd_max, tp->snd_nxt);
2899
2900         db_print_indent(indent);
2901         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2902            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2903
2904         db_print_indent(indent);
2905         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2906             tp->iss, tp->irs, tp->rcv_nxt);
2907
2908         db_print_indent(indent);
2909         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2910             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2911
2912         db_print_indent(indent);
2913         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2914            tp->snd_wnd, tp->snd_cwnd);
2915
2916         db_print_indent(indent);
2917         db_printf("snd_ssthresh: %u   snd_recover: "
2918             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2919
2920         db_print_indent(indent);
2921         db_printf("t_rcvtime: %u   t_startime: %u\n",
2922             tp->t_rcvtime, tp->t_starttime);
2923
2924         db_print_indent(indent);
2925         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2926             tp->t_rtttime, tp->t_rtseq);
2927
2928         db_print_indent(indent);
2929         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2930             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2931
2932         db_print_indent(indent);
2933         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2934             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2935             tp->t_rttbest);
2936
2937         db_print_indent(indent);
2938         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2939             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2940
2941         db_print_indent(indent);
2942         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2943         db_print_toobflags(tp->t_oobflags);
2944         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2945
2946         db_print_indent(indent);
2947         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2948             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2949
2950         db_print_indent(indent);
2951         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2952             tp->ts_recent, tp->ts_recent_age);
2953
2954         db_print_indent(indent);
2955         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2956             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2957
2958         db_print_indent(indent);
2959         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2960             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2961             tp->snd_recover_prev, tp->t_badrxtwin);
2962
2963         db_print_indent(indent);
2964         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2965             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2966
2967         db_print_indent(indent);
2968         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
2969             tp->snd_fack, tp->rcv_numsacks);
2970
2971         /* Skip sackblks, sackhint. */
2972
2973         db_print_indent(indent);
2974         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2975             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2976 }
2977
2978 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2979 {
2980         struct tcpcb *tp;
2981
2982         if (!have_addr) {
2983                 db_printf("usage: show tcpcb <addr>\n");
2984                 return;
2985         }
2986         tp = (struct tcpcb *)addr;
2987
2988         db_print_tcpcb(tp, "tcpcb", 0);
2989 }
2990 #endif