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                     nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
 330                     sinp->sin_addr.s_addr != INADDR_ANY)
 331                         return (EAFNOSUPPORT);
 332                 nam->sa_family = AF_INET;
 333         }
 334         if (nam->sa_len != sizeof(*sinp))
 335                 return (EINVAL);
 336
 337         /*
 338          * Must check for multicast addresses and disallow binding
 339          * to them.
 340          */
 341         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 342                 return (EAFNOSUPPORT);
 343
 344         TCPDEBUG0;
 345         inp = sotoinpcb(so);
 346         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 347         INP_WLOCK(inp);
 348         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 349                 error = EINVAL;
 350                 goto out;
 351         }
 352         tp = intotcpcb(inp);
 353         TCPDEBUG1();
 354         INP_HASH_WLOCK(&V_tcbinfo);
 355         error = in_pcbbind(inp, nam, td->td_ucred);
 356         INP_HASH_WUNLOCK(&V_tcbinfo);
 357 out:
 358         TCPDEBUG2(PRU_BIND);
 359         TCP_PROBE2(debug__user, tp, PRU_BIND);
 360         INP_WUNLOCK(inp);
 361
 362         return (error);
 363 }
 364 #endif /* INET */
 365
 366 #ifdef INET6
 367 static int
 368 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 369 {
 370         int error = 0;
 371         struct inpcb *inp;
 372         struct tcpcb *tp = NULL;
 373         struct sockaddr_in6 *sin6;
 374         u_char vflagsav;
 375
 376         sin6 = (struct sockaddr_in6 *)nam;
 377         if (nam->sa_family != AF_INET6)
 378                 return (EAFNOSUPPORT);
 379         if (nam->sa_len != sizeof(*sin6))
 380                 return (EINVAL);
 381
 382         /*
 383          * Must check for multicast addresses and disallow binding
 384          * to them.
 385          */
 386         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 387                 return (EAFNOSUPPORT);
 388
 389         TCPDEBUG0;
 390         inp = sotoinpcb(so);
 391         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 392         INP_WLOCK(inp);
 393         vflagsav = inp->inp_vflag;
 394         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 395                 error = EINVAL;
 396                 goto out;
 397         }
 398         tp = intotcpcb(inp);
 399         TCPDEBUG1();
 400         INP_HASH_WLOCK(&V_tcbinfo);
 401         inp->inp_vflag &= ~INP_IPV4;
 402         inp->inp_vflag |= INP_IPV6;
 403 #ifdef INET
 404         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 405                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
 406                         inp->inp_vflag |= INP_IPV4;
 407                 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 408                         struct sockaddr_in sin;
 409
 410                         in6_sin6_2_sin(&sin, sin6);
 411                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 412                                 error = EAFNOSUPPORT;
 413                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 414                                 goto out;
 415                         }
 416                         inp->inp_vflag |= INP_IPV4;
 417                         inp->inp_vflag &= ~INP_IPV6;
 418                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 419                             td->td_ucred);
 420                         INP_HASH_WUNLOCK(&V_tcbinfo);
 421                         goto out;
 422                 }
 423         }
 424 #endif
 425         error = in6_pcbbind(inp, nam, td->td_ucred);
 426         INP_HASH_WUNLOCK(&V_tcbinfo);
 427 out:
 428         if (error != 0)
 429                 inp->inp_vflag = vflagsav;
 430         TCPDEBUG2(PRU_BIND);
 431         TCP_PROBE2(debug__user, tp, PRU_BIND);
 432         INP_WUNLOCK(inp);
 433         return (error);
 434 }
 435 #endif /* INET6 */
 436
 437 #ifdef INET
 438 /*
 439  * Prepare to accept connections.
 440  */
 441 static int
 442 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 443 {
 444         int error = 0;
 445         struct inpcb *inp;
 446         struct tcpcb *tp = NULL;
 447
 448         TCPDEBUG0;
 449         inp = sotoinpcb(so);
 450         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 451         INP_WLOCK(inp);
 452         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 453                 error = EINVAL;
 454                 goto out;
 455         }
 456         tp = intotcpcb(inp);
 457         TCPDEBUG1();
 458         SOCK_LOCK(so);
 459         error = solisten_proto_check(so);
 460         INP_HASH_WLOCK(&V_tcbinfo);
 461         if (error == 0 && inp->inp_lport == 0)
 462                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 463         INP_HASH_WUNLOCK(&V_tcbinfo);
 464         if (error == 0) {
 465                 tcp_state_change(tp, TCPS_LISTEN);
 466                 solisten_proto(so, backlog);
 467 #ifdef TCP_OFFLOAD
 468                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 469                         tcp_offload_listen_start(tp);
 470 #endif
 471         }
 472         SOCK_UNLOCK(so);
 473
 474         if (IS_FASTOPEN(tp->t_flags))
 475                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 476
 477 out:
 478         TCPDEBUG2(PRU_LISTEN);
 479         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 480         INP_WUNLOCK(inp);
 481         return (error);
 482 }
 483 #endif /* INET */
 484
 485 #ifdef INET6
 486 static int
 487 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 488 {
 489         int error = 0;
 490         struct inpcb *inp;
 491         struct tcpcb *tp = NULL;
 492         u_char vflagsav;
 493
 494         TCPDEBUG0;
 495         inp = sotoinpcb(so);
 496         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 497         INP_WLOCK(inp);
 498         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 499                 error = EINVAL;
 500                 goto out;
 501         }
 502         vflagsav = inp->inp_vflag;
 503         tp = intotcpcb(inp);
 504         TCPDEBUG1();
 505         SOCK_LOCK(so);
 506         error = solisten_proto_check(so);
 507         INP_HASH_WLOCK(&V_tcbinfo);
 508         if (error == 0 && inp->inp_lport == 0) {
 509                 inp->inp_vflag &= ~INP_IPV4;
 510                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 511                         inp->inp_vflag |= INP_IPV4;
 512                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 513         }
 514         INP_HASH_WUNLOCK(&V_tcbinfo);
 515         if (error == 0) {
 516                 tcp_state_change(tp, TCPS_LISTEN);
 517                 solisten_proto(so, backlog);
 518 #ifdef TCP_OFFLOAD
 519                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 520                         tcp_offload_listen_start(tp);
 521 #endif
 522         }
 523         SOCK_UNLOCK(so);
 524
 525         if (IS_FASTOPEN(tp->t_flags))
 526                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 527
 528         if (error != 0)
 529                 inp->inp_vflag = vflagsav;
 530
 531 out:
 532         TCPDEBUG2(PRU_LISTEN);
 533         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 534         INP_WUNLOCK(inp);
 535         return (error);
 536 }
 537 #endif /* INET6 */
 538
 539 #ifdef INET
 540 /*
 541  * Initiate connection to peer.
 542  * Create a template for use in transmissions on this connection.
 543  * Enter SYN_SENT state, and mark socket as connecting.
 544  * Start keep-alive timer, and seed output sequence space.
 545  * Send initial segment on connection.
 546  */
 547 static int
 548 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 549 {
 550         struct epoch_tracker et;
 551         int error = 0;
 552         struct inpcb *inp;
 553         struct tcpcb *tp = NULL;
 554         struct sockaddr_in *sinp;
 555
 556         sinp = (struct sockaddr_in *)nam;
 557         if (nam->sa_family != AF_INET)
 558                 return (EAFNOSUPPORT);
 559         if (nam->sa_len != sizeof (*sinp))
 560                 return (EINVAL);
 561
 562         /*
 563          * Must disallow TCP ``connections'' to multicast addresses.
 564          */
 565         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 566                 return (EAFNOSUPPORT);
 567         if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST)
 568                 return (EACCES);
 569         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 570                 return (error);
 571
 572         TCPDEBUG0;
 573         inp = sotoinpcb(so);
 574         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 575         INP_WLOCK(inp);
 576         if (inp->inp_flags & INP_TIMEWAIT) {
 577                 error = EADDRINUSE;
 578                 goto out;
 579         }
 580         if (inp->inp_flags & INP_DROPPED) {
 581                 error = ECONNREFUSED;
 582                 goto out;
 583         }
 584         tp = intotcpcb(inp);
 585         TCPDEBUG1();
 586         NET_EPOCH_ENTER(et);
 587         if ((error = tcp_connect(tp, nam, td)) != 0)
 588                 goto out_in_epoch;
 589 #ifdef TCP_OFFLOAD
 590         if (registered_toedevs > 0 &&
 591             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 592             (error = tcp_offload_connect(so, nam)) == 0)
 593                 goto out_in_epoch;
 594 #endif
 595         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 596         error = tp->t_fb->tfb_tcp_output(tp);
 597 out_in_epoch:
 598         NET_EPOCH_EXIT(et);
 599 out:
 600         TCPDEBUG2(PRU_CONNECT);
 601         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 602         INP_WUNLOCK(inp);
 603         return (error);
 604 }
 605 #endif /* INET */
 606
 607 #ifdef INET6
 608 static int
 609 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 610 {
 611         struct epoch_tracker et;
 612         int error = 0;
 613         struct inpcb *inp;
 614         struct tcpcb *tp = NULL;
 615         struct sockaddr_in6 *sin6;
 616         u_int8_t incflagsav;
 617         u_char vflagsav;
 618
 619         TCPDEBUG0;
 620
 621         sin6 = (struct sockaddr_in6 *)nam;
 622         if (nam->sa_family != AF_INET6)
 623                 return (EAFNOSUPPORT);
 624         if (nam->sa_len != sizeof (*sin6))
 625                 return (EINVAL);
 626
 627         /*
 628          * Must disallow TCP ``connections'' to multicast addresses.
 629          */
 630         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 631                 return (EAFNOSUPPORT);
 632
 633         inp = sotoinpcb(so);
 634         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 635         INP_WLOCK(inp);
 636         vflagsav = inp->inp_vflag;
 637         incflagsav = inp->inp_inc.inc_flags;
 638         if (inp->inp_flags & INP_TIMEWAIT) {
 639                 error = EADDRINUSE;
 640                 goto out;
 641         }
 642         if (inp->inp_flags & INP_DROPPED) {
 643                 error = ECONNREFUSED;
 644                 goto out;
 645         }
 646         tp = intotcpcb(inp);
 647         TCPDEBUG1();
 648 #ifdef INET
 649         /*
 650          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 651          * therefore probably require the hash lock, which isn't held here.
 652          * Is this a significant problem?
 653          */
 654         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 655                 struct sockaddr_in sin;
 656
 657                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 658                         error = EINVAL;
 659                         goto out;
 660                 }
 661                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 662                         error = EAFNOSUPPORT;
 663                         goto out;
 664                 }
 665
 666                 in6_sin6_2_sin(&sin, sin6);
 667                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 668                         error = EAFNOSUPPORT;
 669                         goto out;
 670                 }
 671                 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
 672                         error = EACCES;
 673                         goto out;
 674                 }
 675                 if ((error = prison_remote_ip4(td->td_ucred,
 676                     &sin.sin_addr)) != 0)
 677                         goto out;
 678                 inp->inp_vflag |= INP_IPV4;
 679                 inp->inp_vflag &= ~INP_IPV6;
 680                 NET_EPOCH_ENTER(et);
 681                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 682                         goto out_in_epoch;
 683 #ifdef TCP_OFFLOAD
 684                 if (registered_toedevs > 0 &&
 685                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 686                     (error = tcp_offload_connect(so, nam)) == 0)
 687                         goto out_in_epoch;
 688 #endif
 689                 error = tp->t_fb->tfb_tcp_output(tp);
 690                 goto out_in_epoch;
 691         } else {
 692                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 693                         error = EAFNOSUPPORT;
 694                         goto out;
 695                 }
 696         }
 697 #endif
 698         if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
 699                 goto out;
 700         inp->inp_vflag &= ~INP_IPV4;
 701         inp->inp_vflag |= INP_IPV6;
 702         inp->inp_inc.inc_flags |= INC_ISIPV6;
 703         if ((error = tcp6_connect(tp, nam, td)) != 0)
 704                 goto out;
 705 #ifdef TCP_OFFLOAD
 706         if (registered_toedevs > 0 &&
 707             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 708             (error = tcp_offload_connect(so, nam)) == 0)
 709                 goto out;
 710 #endif
 711         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 712         NET_EPOCH_ENTER(et);
 713         error = tp->t_fb->tfb_tcp_output(tp);
 714 #ifdef INET
 715 out_in_epoch:
 716 #endif
 717         NET_EPOCH_EXIT(et);
 718 out:
 719         /*
 720          * If the implicit bind in the connect call fails, restore
 721          * the flags we modified.
 722          */
 723         if (error != 0 && inp->inp_lport == 0) {
 724                 inp->inp_vflag = vflagsav;
 725                 inp->inp_inc.inc_flags = incflagsav;
 726         }
 727
 728         TCPDEBUG2(PRU_CONNECT);
 729         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 730         INP_WUNLOCK(inp);
 731         return (error);
 732 }
 733 #endif /* INET6 */
 734
 735 /*
 736  * Initiate disconnect from peer.
 737  * If connection never passed embryonic stage, just drop;
 738  * else if don't need to let data drain, then can just drop anyways,
 739  * else have to begin TCP shutdown process: mark socket disconnecting,
 740  * drain unread data, state switch to reflect user close, and
 741  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 742  * when peer sends FIN and acks ours.
 743  *
 744  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 745  */
 746 static int
 747 tcp_usr_disconnect(struct socket *so)
 748 {
 749         struct inpcb *inp;
 750         struct tcpcb *tp = NULL;
 751         struct epoch_tracker et;
 752         int error = 0;
 753
 754         TCPDEBUG0;
 755         NET_EPOCH_ENTER(et);
 756         inp = sotoinpcb(so);
 757         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 758         INP_WLOCK(inp);
 759         if (inp->inp_flags & INP_TIMEWAIT)
 760                 goto out;
 761         if (inp->inp_flags & INP_DROPPED) {
 762                 error = ECONNRESET;
 763                 goto out;
 764         }
 765         tp = intotcpcb(inp);
 766         TCPDEBUG1();
 767         tcp_disconnect(tp);
 768 out:
 769         TCPDEBUG2(PRU_DISCONNECT);
 770         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 771         INP_WUNLOCK(inp);
 772         NET_EPOCH_EXIT(et);
 773         return (error);
 774 }
 775
 776 #ifdef INET
 777 /*
 778  * Accept a connection.  Essentially all the work is done at higher levels;
 779  * just return the address of the peer, storing through addr.
 780  */
 781 static int
 782 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 783 {
 784         int error = 0;
 785         struct inpcb *inp = NULL;
 786         struct tcpcb *tp = NULL;
 787         struct in_addr addr;
 788         in_port_t port = 0;
 789         TCPDEBUG0;
 790
 791         if (so->so_state & SS_ISDISCONNECTED)
 792                 return (ECONNABORTED);
 793
 794         inp = sotoinpcb(so);
 795         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 796         INP_WLOCK(inp);
 797         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 798                 error = ECONNABORTED;
 799                 goto out;
 800         }
 801         tp = intotcpcb(inp);
 802         TCPDEBUG1();
 803
 804         /*
 805          * We inline in_getpeeraddr and COMMON_END here, so that we can
 806          * copy the data of interest and defer the malloc until after we
 807          * release the lock.
 808          */
 809         port = inp->inp_fport;
 810         addr = inp->inp_faddr;
 811
 812 out:
 813         TCPDEBUG2(PRU_ACCEPT);
 814         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 815         INP_WUNLOCK(inp);
 816         if (error == 0)
 817                 *nam = in_sockaddr(port, &addr);
 818         return error;
 819 }
 820 #endif /* INET */
 821
 822 #ifdef INET6
 823 static int
 824 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 825 {
 826         struct inpcb *inp = NULL;
 827         int error = 0;
 828         struct tcpcb *tp = NULL;
 829         struct in_addr addr;
 830         struct in6_addr addr6;
 831         struct epoch_tracker et;
 832         in_port_t port = 0;
 833         int v4 = 0;
 834         TCPDEBUG0;
 835
 836         if (so->so_state & SS_ISDISCONNECTED)
 837                 return (ECONNABORTED);
 838
 839         inp = sotoinpcb(so);
 840         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 841         NET_EPOCH_ENTER(et);
 842         INP_WLOCK(inp);
 843         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 844                 error = ECONNABORTED;
 845                 goto out;
 846         }
 847         tp = intotcpcb(inp);
 848         TCPDEBUG1();
 849
 850         /*
 851          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 852          * copy the data of interest and defer the malloc until after we
 853          * release the lock.
 854          */
 855         if (inp->inp_vflag & INP_IPV4) {
 856                 v4 = 1;
 857                 port = inp->inp_fport;
 858                 addr = inp->inp_faddr;
 859         } else {
 860                 port = inp->inp_fport;
 861                 addr6 = inp->in6p_faddr;
 862         }
 863
 864 out:
 865         TCPDEBUG2(PRU_ACCEPT);
 866         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 867         INP_WUNLOCK(inp);
 868         NET_EPOCH_EXIT(et);
 869         if (error == 0) {
 870                 if (v4)
 871                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 872                 else
 873                         *nam = in6_sockaddr(port, &addr6);
 874         }
 875         return error;
 876 }
 877 #endif /* INET6 */
 878
 879 /*
 880  * Mark the connection as being incapable of further output.
 881  */
 882 static int
 883 tcp_usr_shutdown(struct socket *so)
 884 {
 885         int error = 0;
 886         struct inpcb *inp;
 887         struct tcpcb *tp = NULL;
 888         struct epoch_tracker et;
 889
 890         TCPDEBUG0;
 891         NET_EPOCH_ENTER(et);
 892         inp = sotoinpcb(so);
 893         KASSERT(inp != NULL, ("inp == NULL"));
 894         INP_WLOCK(inp);
 895         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 896                 error = ECONNRESET;
 897                 goto out;
 898         }
 899         tp = intotcpcb(inp);
 900         TCPDEBUG1();
 901         socantsendmore(so);
 902         tcp_usrclosed(tp);
 903         if (!(inp->inp_flags & INP_DROPPED))
 904                 error = tp->t_fb->tfb_tcp_output(tp);
 905
 906 out:
 907         TCPDEBUG2(PRU_SHUTDOWN);
 908         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 909         INP_WUNLOCK(inp);
 910         NET_EPOCH_EXIT(et);
 911
 912         return (error);
 913 }
 914
 915 /*
 916  * After a receive, possibly send window update to peer.
 917  */
 918 static int
 919 tcp_usr_rcvd(struct socket *so, int flags)
 920 {
 921         struct epoch_tracker et;
 922         struct inpcb *inp;
 923         struct tcpcb *tp = NULL;
 924         int error = 0;
 925
 926         TCPDEBUG0;
 927         inp = sotoinpcb(so);
 928         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 929         INP_WLOCK(inp);
 930         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 931                 error = ECONNRESET;
 932                 goto out;
 933         }
 934         tp = intotcpcb(inp);
 935         TCPDEBUG1();
 936         /*
 937          * For passively-created TFO connections, don't attempt a window
 938          * update while still in SYN_RECEIVED as this may trigger an early
 939          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 940          * application response data, or failing that, when the DELACK timer
 941          * expires.
 942          */
 943         if (IS_FASTOPEN(tp->t_flags) &&
 944             (tp->t_state == TCPS_SYN_RECEIVED))
 945                 goto out;
 946         NET_EPOCH_ENTER(et);
 947 #ifdef TCP_OFFLOAD
 948         if (tp->t_flags & TF_TOE)
 949                 tcp_offload_rcvd(tp);
 950         else
 951 #endif
 952         tp->t_fb->tfb_tcp_output(tp);
 953         NET_EPOCH_EXIT(et);
 954 out:
 955         TCPDEBUG2(PRU_RCVD);
 956         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 957         INP_WUNLOCK(inp);
 958         return (error);
 959 }
 960
 961 /*
 962  * Do a send by putting data in output queue and updating urgent
 963  * marker if URG set.  Possibly send more data.  Unlike the other
 964  * pru_*() routines, the mbuf chains are our responsibility.  We
 965  * must either enqueue them or free them.  The other pru_* routines
 966  * generally are caller-frees.
 967  */
 968 static int
 969 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 970     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 971 {
 972         struct epoch_tracker et;
 973         int error = 0;
 974         struct inpcb *inp;
 975         struct tcpcb *tp = NULL;
 976 #ifdef INET
 977 #ifdef INET6
 978         struct sockaddr_in sin;
 979 #endif
 980         struct sockaddr_in *sinp;
 981 #endif
 982 #ifdef INET6
 983         int isipv6;
 984 #endif
 985         u_int8_t incflagsav;
 986         u_char vflagsav;
 987         bool restoreflags;
 988         TCPDEBUG0;
 989
 990         /*
 991          * We require the pcbinfo "read lock" if we will close the socket
 992          * as part of this call.
 993          */
 994         NET_EPOCH_ENTER(et);
 995         inp = sotoinpcb(so);
 996         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 997         INP_WLOCK(inp);
 998         vflagsav = inp->inp_vflag;
 999         incflagsav = inp->inp_inc.inc_flags;
1000         restoreflags = false;
1001         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1002                 if (control)
1003                         m_freem(control);
1004                 error = ECONNRESET;
1005                 goto out;
1006         }
1007         if (control != NULL) {
1008                 /* TCP doesn't do control messages (rights, creds, etc) */
1009                 if (control->m_len) {
1010                         m_freem(control);
1011                         error = EINVAL;
1012                         goto out;
1013                 }
1014                 m_freem(control);       /* empty control, just free it */
1015                 control = NULL;
1016         }
1017         tp = intotcpcb(inp);
1018         if ((flags & PRUS_OOB) != 0 &&
1019             (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1020                 goto out;
1021
1022         TCPDEBUG1();
1023         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1024                 switch (nam->sa_family) {
1025 #ifdef INET
1026                 case AF_INET:
1027                         sinp = (struct sockaddr_in *)nam;
1028                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1029                                 error = EINVAL;
1030                                 goto out;
1031                         }
1032                         if ((inp->inp_vflag & INP_IPV6) != 0) {
1033                                 error = EAFNOSUPPORT;
1034                                 goto out;
1035                         }
1036                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1037                                 error = EAFNOSUPPORT;
1038                                 goto out;
1039                         }
1040                         if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1041                                 error = EACCES;
1042                                 goto out;
1043                         }
1044                         if ((error = prison_remote_ip4(td->td_ucred,
1045                             &sinp->sin_addr)))
1046                                 goto out;
1047 #ifdef INET6
1048                         isipv6 = 0;
1049 #endif
1050                         break;
1051 #endif /* INET */
1052 #ifdef INET6
1053                 case AF_INET6:
1054                 {
1055                         struct sockaddr_in6 *sin6;
1056
1057                         sin6 = (struct sockaddr_in6 *)nam;
1058                         if (sin6->sin6_len != sizeof(*sin6)) {
1059                                 error = EINVAL;
1060                                 goto out;
1061                         }
1062                         if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1063                                 error = EAFNOSUPPORT;
1064                                 goto out;
1065                         }
1066                         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1067                                 error = EAFNOSUPPORT;
1068                                 goto out;
1069                         }
1070                         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1071 #ifdef INET
1072                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1073                                         error = EINVAL;
1074                                         goto out;
1075                                 }
1076                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1077                                         error = EAFNOSUPPORT;
1078                                         goto out;
1079                                 }
1080                                 restoreflags = true;
1081                                 inp->inp_vflag &= ~INP_IPV6;
1082                                 sinp = &sin;
1083                                 in6_sin6_2_sin(sinp, sin6);
1084                                 if (IN_MULTICAST(
1085                                     ntohl(sinp->sin_addr.s_addr))) {
1086                                         error = EAFNOSUPPORT;
1087                                         goto out;
1088                                 }
1089                                 if ((error = prison_remote_ip4(td->td_ucred,
1090                                     &sinp->sin_addr)))
1091                                         goto out;
1092                                 isipv6 = 0;
1093 #else /* !INET */
1094                                 error = EAFNOSUPPORT;
1095                                 goto out;
1096 #endif /* INET */
1097                         } else {
1098                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1099                                         error = EAFNOSUPPORT;
1100                                         goto out;
1101                                 }
1102                                 restoreflags = true;
1103                                 inp->inp_vflag &= ~INP_IPV4;
1104                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1105                                 if ((error = prison_remote_ip6(td->td_ucred,
1106                                     &sin6->sin6_addr)))
1107                                         goto out;
1108                                 isipv6 = 1;
1109                         }
1110                         break;
1111                 }
1112 #endif /* INET6 */
1113                 default:
1114                         error = EAFNOSUPPORT;
1115                         goto out;
1116                 }
1117         }
1118         if (!(flags & PRUS_OOB)) {
1119                 sbappendstream(&so->so_snd, m, flags);
1120                 m = NULL;
1121                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1122                         /*
1123                          * Do implied connect if not yet connected,
1124                          * initialize window to default value, and
1125                          * initialize maxseg using peer's cached MSS.
1126                          */
1127 #ifdef INET6
1128                         if (isipv6)
1129                                 error = tcp6_connect(tp, nam, td);
1130 #endif /* INET6 */
1131 #if defined(INET6) && defined(INET)
1132                         else
1133 #endif
1134 #ifdef INET
1135                                 error = tcp_connect(tp,
1136                                     (struct sockaddr *)sinp, td);
1137 #endif
1138                         /*
1139                          * The bind operation in tcp_connect succeeded. We
1140                          * no longer want to restore the flags if later
1141                          * operations fail.
1142                          */
1143                         if (error == 0 || inp->inp_lport != 0)
1144                                 restoreflags = false;
1145
1146                         if (error) {
1147                                 /* m is freed if PRUS_NOTREADY is unset. */
1148                                 sbflush(&so->so_snd);
1149                                 goto out;
1150                         }
1151                         if (IS_FASTOPEN(tp->t_flags))
1152                                 tcp_fastopen_connect(tp);
1153                         else {
1154                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1155                                 tcp_mss(tp, -1);
1156                         }
1157                 }
1158                 if (flags & PRUS_EOF) {
1159                         /*
1160                          * Close the send side of the connection after
1161                          * the data is sent.
1162                          */
1163                         socantsendmore(so);
1164                         tcp_usrclosed(tp);
1165                 }
1166                 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1167                     ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1168                     (tp->t_fbyte_out == 0) &&
1169                     (so->so_snd.sb_ccc > 0)) {
1170                         tp->t_fbyte_out = ticks;
1171                         if (tp->t_fbyte_out == 0)
1172                                 tp->t_fbyte_out = 1;
1173                         if (tp->t_fbyte_out && tp->t_fbyte_in)
1174                                 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1175                 }
1176                 if (!(inp->inp_flags & INP_DROPPED) &&
1177                     !(flags & PRUS_NOTREADY)) {
1178                         if (flags & PRUS_MORETOCOME)
1179                                 tp->t_flags |= TF_MORETOCOME;
1180                         error = tp->t_fb->tfb_tcp_output(tp);
1181                         if (flags & PRUS_MORETOCOME)
1182                                 tp->t_flags &= ~TF_MORETOCOME;
1183                 }
1184         } else {
1185                 /*
1186                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1187                  */
1188                 SOCKBUF_LOCK(&so->so_snd);
1189                 if (sbspace(&so->so_snd) < -512) {
1190                         SOCKBUF_UNLOCK(&so->so_snd);
1191                         error = ENOBUFS;
1192                         goto out;
1193                 }
1194                 /*
1195                  * According to RFC961 (Assigned Protocols),
1196                  * the urgent pointer points to the last octet
1197                  * of urgent data.  We continue, however,
1198                  * to consider it to indicate the first octet
1199                  * of data past the urgent section.
1200                  * Otherwise, snd_up should be one lower.
1201                  */
1202                 sbappendstream_locked(&so->so_snd, m, flags);
1203                 SOCKBUF_UNLOCK(&so->so_snd);
1204                 m = NULL;
1205                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1206                         /*
1207                          * Do implied connect if not yet connected,
1208                          * initialize window to default value, and
1209                          * initialize maxseg using peer's cached MSS.
1210                          */
1211
1212                         /*
1213                          * Not going to contemplate SYN|URG
1214                          */
1215                         if (IS_FASTOPEN(tp->t_flags))
1216                                 tp->t_flags &= ~TF_FASTOPEN;
1217 #ifdef INET6
1218                         if (isipv6)
1219                                 error = tcp6_connect(tp, nam, td);
1220 #endif /* INET6 */
1221 #if defined(INET6) && defined(INET)
1222                         else
1223 #endif
1224 #ifdef INET
1225                                 error = tcp_connect(tp,
1226                                     (struct sockaddr *)sinp, td);
1227 #endif
1228                         /*
1229                          * The bind operation in tcp_connect succeeded. We
1230                          * no longer want to restore the flags if later
1231                          * operations fail.
1232                          */
1233                         if (error == 0 || inp->inp_lport != 0)
1234                                 restoreflags = false;
1235
1236                         if (error != 0) {
1237                                 /* m is freed if PRUS_NOTREADY is unset. */
1238                                 sbflush(&so->so_snd);
1239                                 goto out;
1240                         }
1241                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1242                         tcp_mss(tp, -1);
1243                 }
1244                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1245                 if ((flags & PRUS_NOTREADY) == 0) {
1246                         tp->t_flags |= TF_FORCEDATA;
1247                         error = tp->t_fb->tfb_tcp_output(tp);
1248                         tp->t_flags &= ~TF_FORCEDATA;
1249                 }
1250         }
1251         TCP_LOG_EVENT(tp, NULL,
1252             &inp->inp_socket->so_rcv,
1253             &inp->inp_socket->so_snd,
1254             TCP_LOG_USERSEND, error,
1255             0, NULL, false);
1256
1257 out:
1258         /*
1259          * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1260          * responsible for freeing memory.
1261          */
1262         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1263                 m_freem(m);
1264
1265         /*
1266          * If the request was unsuccessful and we changed flags,
1267          * restore the original flags.
1268          */
1269         if (error != 0 && restoreflags) {
1270                 inp->inp_vflag = vflagsav;
1271                 inp->inp_inc.inc_flags = incflagsav;
1272         }
1273         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1274                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1275         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1276                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1277         INP_WUNLOCK(inp);
1278         NET_EPOCH_EXIT(et);
1279         return (error);
1280 }
1281
1282 static int
1283 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1284 {
1285         struct epoch_tracker et;
1286         struct inpcb *inp;
1287         struct tcpcb *tp;
1288         int error;
1289
1290         inp = sotoinpcb(so);
1291         INP_WLOCK(inp);
1292         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1293                 INP_WUNLOCK(inp);
1294                 mb_free_notready(m, count);
1295                 return (ECONNRESET);
1296         }
1297         tp = intotcpcb(inp);
1298
1299         SOCKBUF_LOCK(&so->so_snd);
1300         error = sbready(&so->so_snd, m, count);
1301         SOCKBUF_UNLOCK(&so->so_snd);
1302         if (error == 0) {
1303                 NET_EPOCH_ENTER(et);
1304                 error = tp->t_fb->tfb_tcp_output(tp);
1305                 NET_EPOCH_EXIT(et);
1306         }
1307         INP_WUNLOCK(inp);
1308
1309         return (error);
1310 }
1311
1312 /*
1313  * Abort the TCP.  Drop the connection abruptly.
1314  */
1315 static void
1316 tcp_usr_abort(struct socket *so)
1317 {
1318         struct inpcb *inp;
1319         struct tcpcb *tp = NULL;
1320         struct epoch_tracker et;
1321         TCPDEBUG0;
1322
1323         inp = sotoinpcb(so);
1324         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1325
1326         NET_EPOCH_ENTER(et);
1327         INP_WLOCK(inp);
1328         KASSERT(inp->inp_socket != NULL,
1329             ("tcp_usr_abort: inp_socket == NULL"));
1330
1331         /*
1332          * If we still have full TCP state, and we're not dropped, drop.
1333          */
1334         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1335             !(inp->inp_flags & INP_DROPPED)) {
1336                 tp = intotcpcb(inp);
1337                 TCPDEBUG1();
1338                 tp = tcp_drop(tp, ECONNABORTED);
1339                 if (tp == NULL)
1340                         goto dropped;
1341                 TCPDEBUG2(PRU_ABORT);
1342                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1343         }
1344         if (!(inp->inp_flags & INP_DROPPED)) {
1345                 SOCK_LOCK(so);
1346                 so->so_state |= SS_PROTOREF;
1347                 SOCK_UNLOCK(so);
1348                 inp->inp_flags |= INP_SOCKREF;
1349         }
1350         INP_WUNLOCK(inp);
1351 dropped:
1352         NET_EPOCH_EXIT(et);
1353 }
1354
1355 /*
1356  * TCP socket is closed.  Start friendly disconnect.
1357  */
1358 static void
1359 tcp_usr_close(struct socket *so)
1360 {
1361         struct inpcb *inp;
1362         struct tcpcb *tp = NULL;
1363         struct epoch_tracker et;
1364         TCPDEBUG0;
1365
1366         inp = sotoinpcb(so);
1367         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1368
1369         NET_EPOCH_ENTER(et);
1370         INP_WLOCK(inp);
1371         KASSERT(inp->inp_socket != NULL,
1372             ("tcp_usr_close: inp_socket == NULL"));
1373
1374         /*
1375          * If we still have full TCP state, and we're not dropped, initiate
1376          * a disconnect.
1377          */
1378         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1379             !(inp->inp_flags & INP_DROPPED)) {
1380                 tp = intotcpcb(inp);
1381                 TCPDEBUG1();
1382                 tcp_disconnect(tp);
1383                 TCPDEBUG2(PRU_CLOSE);
1384                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1385         }
1386         if (!(inp->inp_flags & INP_DROPPED)) {
1387                 SOCK_LOCK(so);
1388                 so->so_state |= SS_PROTOREF;
1389                 SOCK_UNLOCK(so);
1390                 inp->inp_flags |= INP_SOCKREF;
1391         }
1392         INP_WUNLOCK(inp);
1393         NET_EPOCH_EXIT(et);
1394 }
1395
1396 static int
1397 tcp_pru_options_support(struct tcpcb *tp, int flags)
1398 {
1399         /*
1400          * If the specific TCP stack has a pru_options
1401          * specified then it does not always support
1402          * all the PRU_XX options and we must ask it.
1403          * If the function is not specified then all
1404          * of the PRU_XX options are supported.
1405          */
1406         int ret = 0;
1407
1408         if (tp->t_fb->tfb_pru_options) {
1409                 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1410         }
1411         return (ret);
1412 }
1413
1414 /*
1415  * Receive out-of-band data.
1416  */
1417 static int
1418 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1419 {
1420         int error = 0;
1421         struct inpcb *inp;
1422         struct tcpcb *tp = NULL;
1423
1424         TCPDEBUG0;
1425         inp = sotoinpcb(so);
1426         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1427         INP_WLOCK(inp);
1428         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1429                 error = ECONNRESET;
1430                 goto out;
1431         }
1432         tp = intotcpcb(inp);
1433         error = tcp_pru_options_support(tp, PRUS_OOB);
1434         if (error) {
1435                 goto out;
1436         }
1437         TCPDEBUG1();
1438         if ((so->so_oobmark == 0 &&
1439              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1440             so->so_options & SO_OOBINLINE ||
1441             tp->t_oobflags & TCPOOB_HADDATA) {
1442                 error = EINVAL;
1443                 goto out;
1444         }
1445         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1446                 error = EWOULDBLOCK;
1447                 goto out;
1448         }
1449         m->m_len = 1;
1450         *mtod(m, caddr_t) = tp->t_iobc;
1451         if ((flags & MSG_PEEK) == 0)
1452                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1453
1454 out:
1455         TCPDEBUG2(PRU_RCVOOB);
1456         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1457         INP_WUNLOCK(inp);
1458         return (error);
1459 }
1460
1461 #ifdef INET
1462 struct pr_usrreqs tcp_usrreqs = {
1463         .pru_abort =            tcp_usr_abort,
1464         .pru_accept =           tcp_usr_accept,
1465         .pru_attach =           tcp_usr_attach,
1466         .pru_bind =             tcp_usr_bind,
1467         .pru_connect =          tcp_usr_connect,
1468         .pru_control =          in_control,
1469         .pru_detach =           tcp_usr_detach,
1470         .pru_disconnect =       tcp_usr_disconnect,
1471         .pru_listen =           tcp_usr_listen,
1472         .pru_peeraddr =         in_getpeeraddr,
1473         .pru_rcvd =             tcp_usr_rcvd,
1474         .pru_rcvoob =           tcp_usr_rcvoob,
1475         .pru_send =             tcp_usr_send,
1476         .pru_ready =            tcp_usr_ready,
1477         .pru_shutdown =         tcp_usr_shutdown,
1478         .pru_sockaddr =         in_getsockaddr,
1479         .pru_sosetlabel =       in_pcbsosetlabel,
1480         .pru_close =            tcp_usr_close,
1481 };
1482 #endif /* INET */
1483
1484 #ifdef INET6
1485 struct pr_usrreqs tcp6_usrreqs = {
1486         .pru_abort =            tcp_usr_abort,
1487         .pru_accept =           tcp6_usr_accept,
1488         .pru_attach =           tcp_usr_attach,
1489         .pru_bind =             tcp6_usr_bind,
1490         .pru_connect =          tcp6_usr_connect,
1491         .pru_control =          in6_control,
1492         .pru_detach =           tcp_usr_detach,
1493         .pru_disconnect =       tcp_usr_disconnect,
1494         .pru_listen =           tcp6_usr_listen,
1495         .pru_peeraddr =         in6_mapped_peeraddr,
1496         .pru_rcvd =             tcp_usr_rcvd,
1497         .pru_rcvoob =           tcp_usr_rcvoob,
1498         .pru_send =             tcp_usr_send,
1499         .pru_ready =            tcp_usr_ready,
1500         .pru_shutdown =         tcp_usr_shutdown,
1501         .pru_sockaddr =         in6_mapped_sockaddr,
1502         .pru_sosetlabel =       in_pcbsosetlabel,
1503         .pru_close =            tcp_usr_close,
1504 };
1505 #endif /* INET6 */
1506
1507 #ifdef INET
1508 /*
1509  * Common subroutine to open a TCP connection to remote host specified
1510  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1511  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1512  * to choose a local host address (interface).  If there is an existing
1513  * incarnation of the same connection in TIME-WAIT state and if the remote
1514  * host was sending CC options and if the connection duration was < MSL, then
1515  * truncate the previous TIME-WAIT state and proceed.
1516  * Initialize connection parameters and enter SYN-SENT state.
1517  */
1518 static int
1519 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1520 {
1521         struct inpcb *inp = tp->t_inpcb, *oinp;
1522         struct socket *so = inp->inp_socket;
1523         struct in_addr laddr;
1524         u_short lport;
1525         int error;
1526
1527         NET_EPOCH_ASSERT();
1528         INP_WLOCK_ASSERT(inp);
1529         INP_HASH_WLOCK(&V_tcbinfo);
1530
1531         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1532                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1533                 if (error)
1534                         goto out;
1535         }
1536
1537         /*
1538          * Cannot simply call in_pcbconnect, because there might be an
1539          * earlier incarnation of this same connection still in
1540          * TIME_WAIT state, creating an ADDRINUSE error.
1541          */
1542         laddr = inp->inp_laddr;
1543         lport = inp->inp_lport;
1544         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1545             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1546         if (error && oinp == NULL)
1547                 goto out;
1548         if (oinp) {
1549                 error = EADDRINUSE;
1550                 goto out;
1551         }
1552         /* Handle initial bind if it hadn't been done in advance. */
1553         if (inp->inp_lport == 0) {
1554                 inp->inp_lport = lport;
1555                 if (in_pcbinshash(inp) != 0) {
1556                         inp->inp_lport = 0;
1557                         error = EAGAIN;
1558                         goto out;
1559                 }
1560         }
1561         inp->inp_laddr = laddr;
1562         in_pcbrehash(inp);
1563         INP_HASH_WUNLOCK(&V_tcbinfo);
1564
1565         /*
1566          * Compute window scaling to request:
1567          * Scale to fit into sweet spot.  See tcp_syncache.c.
1568          * XXX: This should move to tcp_output().
1569          */
1570         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1571             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1572                 tp->request_r_scale++;
1573
1574         soisconnecting(so);
1575         TCPSTAT_INC(tcps_connattempt);
1576         tcp_state_change(tp, TCPS_SYN_SENT);
1577         tp->iss = tcp_new_isn(&inp->inp_inc);
1578         if (tp->t_flags & TF_REQ_TSTMP)
1579                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1580         tcp_sendseqinit(tp);
1581
1582         return 0;
1583
1584 out:
1585         INP_HASH_WUNLOCK(&V_tcbinfo);
1586         return (error);
1587 }
1588 #endif /* INET */
1589
1590 #ifdef INET6
1591 static int
1592 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1593 {
1594         struct inpcb *inp = tp->t_inpcb;
1595         int error;
1596
1597         INP_WLOCK_ASSERT(inp);
1598         INP_HASH_WLOCK(&V_tcbinfo);
1599
1600         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1601                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1602                 if (error)
1603                         goto out;
1604         }
1605         error = in6_pcbconnect(inp, nam, td->td_ucred);
1606         if (error != 0)
1607                 goto out;
1608         INP_HASH_WUNLOCK(&V_tcbinfo);
1609
1610         /* Compute window scaling to request.  */
1611         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1612             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1613                 tp->request_r_scale++;
1614
1615         soisconnecting(inp->inp_socket);
1616         TCPSTAT_INC(tcps_connattempt);
1617         tcp_state_change(tp, TCPS_SYN_SENT);
1618         tp->iss = tcp_new_isn(&inp->inp_inc);
1619         if (tp->t_flags & TF_REQ_TSTMP)
1620                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1621         tcp_sendseqinit(tp);
1622
1623         return 0;
1624
1625 out:
1626         INP_HASH_WUNLOCK(&V_tcbinfo);
1627         return error;
1628 }
1629 #endif /* INET6 */
1630
1631 /*
1632  * Export TCP internal state information via a struct tcp_info, based on the
1633  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1634  * (TCP state machine, etc).  We export all information using FreeBSD-native
1635  * constants -- for example, the numeric values for tcpi_state will differ
1636  * from Linux.
1637  */
1638 static void
1639 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1640 {
1641
1642         INP_WLOCK_ASSERT(tp->t_inpcb);
1643         bzero(ti, sizeof(*ti));
1644
1645         ti->tcpi_state = tp->t_state;
1646         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1647                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1648         if (tp->t_flags & TF_SACK_PERMIT)
1649                 ti->tcpi_options |= TCPI_OPT_SACK;
1650         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1651                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1652                 ti->tcpi_snd_wscale = tp->snd_scale;
1653                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1654         }
1655         if (tp->t_flags2 & TF2_ECN_PERMIT)
1656                 ti->tcpi_options |= TCPI_OPT_ECN;
1657
1658         ti->tcpi_rto = tp->t_rxtcur * tick;
1659         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1660         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1661         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1662
1663         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1664         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1665
1666         /*
1667          * FreeBSD-specific extension fields for tcp_info.
1668          */
1669         ti->tcpi_rcv_space = tp->rcv_wnd;
1670         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1671         ti->tcpi_snd_wnd = tp->snd_wnd;
1672         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1673         ti->tcpi_snd_nxt = tp->snd_nxt;
1674         ti->tcpi_snd_mss = tp->t_maxseg;
1675         ti->tcpi_rcv_mss = tp->t_maxseg;
1676         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1677         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1678         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1679 #ifdef TCP_OFFLOAD
1680         if (tp->t_flags & TF_TOE) {
1681                 ti->tcpi_options |= TCPI_OPT_TOE;
1682                 tcp_offload_tcp_info(tp, ti);
1683         }
1684 #endif
1685 }
1686
1687 /*
1688  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1689  * socket option arguments.  When it re-acquires the lock after the copy, it
1690  * has to revalidate that the connection is still valid for the socket
1691  * option.
1692  */
1693 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1694         INP_WLOCK(inp);                                                 \
1695         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1696                 INP_WUNLOCK(inp);                                       \
1697                 cleanup;                                                \
1698                 return (ECONNRESET);                                    \
1699         }                                                               \
1700         tp = intotcpcb(inp);                                            \
1701 } while(0)
1702 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1703
1704 int
1705 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1706 {
1707         int     error;
1708         struct  inpcb *inp;
1709         struct  tcpcb *tp;
1710         struct tcp_function_block *blk;
1711         struct tcp_function_set fsn;
1712
1713         error = 0;
1714         inp = sotoinpcb(so);
1715         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1716         if (sopt->sopt_level != IPPROTO_TCP) {
1717 #ifdef INET6
1718                 if (inp->inp_vflag & INP_IPV6PROTO) {
1719                         error = ip6_ctloutput(so, sopt);
1720                         /*
1721                          * In case of the IPV6_USE_MIN_MTU socket option,
1722                          * the INC_IPV6MINMTU flag to announce a corresponding
1723                          * MSS during the initial handshake.
1724                          * If the TCP connection is not in the front states,
1725                          * just reduce the MSS being used.
1726                          * This avoids the sending of TCP segments which will
1727                          * be fragmented at the IPv6 layer.
1728                          */
1729                         if ((error == 0) &&
1730                             (sopt->sopt_dir == SOPT_SET) &&
1731                             (sopt->sopt_level == IPPROTO_IPV6) &&
1732                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1733                                 INP_WLOCK(inp);
1734                                 if ((inp->inp_flags &
1735                                     (INP_TIMEWAIT | INP_DROPPED))) {
1736                                         INP_WUNLOCK(inp);
1737                                         return (ECONNRESET);
1738                                 }
1739                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1740                                 tp = intotcpcb(inp);
1741                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1742                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1743                                         struct ip6_pktopts *opt;
1744
1745                                         opt = inp->in6p_outputopts;
1746                                         if ((opt != NULL) &&
1747                                             (opt->ip6po_minmtu ==
1748                                             IP6PO_MINMTU_ALL)) {
1749                                                 if (tp->t_maxseg > TCP6_MSS) {
1750                                                         tp->t_maxseg = TCP6_MSS;
1751                                                 }
1752                                         }
1753                                 }
1754                                 INP_WUNLOCK(inp);
1755                         }
1756                 }
1757 #endif /* INET6 */
1758 #if defined(INET6) && defined(INET)
1759                 else
1760 #endif
1761 #ifdef INET
1762                 {
1763                         error = ip_ctloutput(so, sopt);
1764                 }
1765 #endif
1766                 return (error);
1767         }
1768         INP_WLOCK(inp);
1769         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1770                 INP_WUNLOCK(inp);
1771                 return (ECONNRESET);
1772         }
1773         tp = intotcpcb(inp);
1774         /*
1775          * Protect the TCP option TCP_FUNCTION_BLK so
1776          * that a sub-function can *never* overwrite this.
1777          */
1778         if ((sopt->sopt_dir == SOPT_SET) &&
1779             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1780                 INP_WUNLOCK(inp);
1781                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1782                     sizeof fsn);
1783                 if (error)
1784                         return (error);
1785                 INP_WLOCK_RECHECK(inp);
1786                 blk = find_and_ref_tcp_functions(&fsn);
1787                 if (blk == NULL) {
1788                         INP_WUNLOCK(inp);
1789                         return (ENOENT);
1790                 }
1791                 if (tp->t_fb == blk) {
1792                         /* You already have this */
1793                         refcount_release(&blk->tfb_refcnt);
1794                         INP_WUNLOCK(inp);
1795                         return (0);
1796                 }
1797                 if (tp->t_state != TCPS_CLOSED) {
1798                         /*
1799                          * The user has advanced the state
1800                          * past the initial point, we may not
1801                          * be able to switch.
1802                          */
1803                         if (blk->tfb_tcp_handoff_ok != NULL) {
1804                                 /*
1805                                  * Does the stack provide a
1806                                  * query mechanism, if so it may
1807                                  * still be possible?
1808                                  */
1809                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1810                         } else
1811                                 error = EINVAL;
1812                         if (error) {
1813                                 refcount_release(&blk->tfb_refcnt);
1814                                 INP_WUNLOCK(inp);
1815                                 return(error);
1816                         }
1817                 }
1818                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1819                         refcount_release(&blk->tfb_refcnt);
1820                         INP_WUNLOCK(inp);
1821                         return (ENOENT);
1822                 }
1823                 /*
1824                  * Release the old refcnt, the
1825                  * lookup acquired a ref on the
1826                  * new one already.
1827                  */
1828                 if (tp->t_fb->tfb_tcp_fb_fini) {
1829                         struct epoch_tracker et;
1830                         /*
1831                          * Tell the stack to cleanup with 0 i.e.
1832                          * the tcb is not going away.
1833                          */
1834                         NET_EPOCH_ENTER(et);
1835                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1836                         NET_EPOCH_EXIT(et);
1837                 }
1838 #ifdef TCPHPTS
1839                 /* Assure that we are not on any hpts */
1840                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1841 #endif
1842                 if (blk->tfb_tcp_fb_init) {
1843                         error = (*blk->tfb_tcp_fb_init)(tp);
1844                         if (error) {
1845                                 refcount_release(&blk->tfb_refcnt);
1846                                 if (tp->t_fb->tfb_tcp_fb_init) {
1847                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1848                                                 /* Fall back failed, drop the connection */
1849                                                 INP_WUNLOCK(inp);
1850                                                 soabort(so);
1851                                                 return(error);
1852                                         }
1853                                 }
1854                                 goto err_out;
1855                         }
1856                 }
1857                 refcount_release(&tp->t_fb->tfb_refcnt);
1858                 tp->t_fb = blk;
1859 #ifdef TCP_OFFLOAD
1860                 if (tp->t_flags & TF_TOE) {
1861                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1862                              sopt->sopt_name);
1863                 }
1864 #endif
1865 err_out:
1866                 INP_WUNLOCK(inp);
1867                 return (error);
1868         } else if ((sopt->sopt_dir == SOPT_GET) &&
1869             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1870                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1871                     TCP_FUNCTION_NAME_LEN_MAX);
1872                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1873                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1874                 INP_WUNLOCK(inp);
1875                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1876                 return (error);
1877         }
1878         /* Pass in the INP locked, called must unlock it */
1879         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1880 }
1881
1882 /*
1883  * If this assert becomes untrue, we need to change the size of the buf
1884  * variable in tcp_default_ctloutput().
1885  */
1886 #ifdef CTASSERT
1887 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1888 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1889 #endif
1890
1891 #ifdef KERN_TLS
1892 static int
1893 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1894 {
1895         struct tls_enable_v0 tls_v0;
1896         int error;
1897
1898         if (sopt->sopt_valsize == sizeof(tls_v0)) {
1899                 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1900                     sizeof(tls_v0));
1901                 if (error)
1902                         return (error);
1903                 memset(tls, 0, sizeof(*tls));
1904                 tls->cipher_key = tls_v0.cipher_key;
1905                 tls->iv = tls_v0.iv;
1906                 tls->auth_key = tls_v0.auth_key;
1907                 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1908                 tls->cipher_key_len = tls_v0.cipher_key_len;
1909                 tls->iv_len = tls_v0.iv_len;
1910                 tls->auth_algorithm = tls_v0.auth_algorithm;
1911                 tls->auth_key_len = tls_v0.auth_key_len;
1912                 tls->flags = tls_v0.flags;
1913                 tls->tls_vmajor = tls_v0.tls_vmajor;
1914                 tls->tls_vminor = tls_v0.tls_vminor;
1915                 return (0);
1916         }
1917
1918         return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
1919 }
1920 #endif
1921
1922 int
1923 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1924 {
1925         int     error, opt, optval;
1926         u_int   ui;
1927         struct  tcp_info ti;
1928 #ifdef KERN_TLS
1929         struct tls_enable tls;
1930 #endif
1931         struct cc_algo *algo;
1932         char    *pbuf, buf[TCP_LOG_ID_LEN];
1933 #ifdef STATS
1934         struct statsblob *sbp;
1935 #endif
1936         size_t  len;
1937
1938         /*
1939          * For TCP_CCALGOOPT forward the control to CC module, for both
1940          * SOPT_SET and SOPT_GET.
1941          */
1942         switch (sopt->sopt_name) {
1943         case TCP_CCALGOOPT:
1944                 INP_WUNLOCK(inp);
1945                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1946                         return (EINVAL);
1947                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1948                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1949                     sopt->sopt_valsize);
1950                 if (error) {
1951                         free(pbuf, M_TEMP);
1952                         return (error);
1953                 }
1954                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1955                 if (CC_ALGO(tp)->ctl_output != NULL)
1956                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1957                 else
1958                         error = ENOENT;
1959                 INP_WUNLOCK(inp);
1960                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1961                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1962                 free(pbuf, M_TEMP);
1963                 return (error);
1964         }
1965
1966         switch (sopt->sopt_dir) {
1967         case SOPT_SET:
1968                 switch (sopt->sopt_name) {
1969 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1970                 case TCP_MD5SIG:
1971                         if (!TCPMD5_ENABLED()) {
1972                                 INP_WUNLOCK(inp);
1973                                 return (ENOPROTOOPT);
1974                         }
1975                         error = TCPMD5_PCBCTL(inp, sopt);
1976                         if (error)
1977                                 return (error);
1978                         goto unlock_and_done;
1979 #endif /* IPSEC */
1980
1981                 case TCP_NODELAY:
1982                 case TCP_NOOPT:
1983                 case TCP_LRD:
1984                         INP_WUNLOCK(inp);
1985                         error = sooptcopyin(sopt, &optval, sizeof optval,
1986                             sizeof optval);
1987                         if (error)
1988                                 return (error);
1989
1990                         INP_WLOCK_RECHECK(inp);
1991                         switch (sopt->sopt_name) {
1992                         case TCP_NODELAY:
1993                                 opt = TF_NODELAY;
1994                                 break;
1995                         case TCP_NOOPT:
1996                                 opt = TF_NOOPT;
1997                                 break;
1998                         case TCP_LRD:
1999                                 opt = TF_LRD;
2000                                 break;
2001                         default:
2002                                 opt = 0; /* dead code to fool gcc */
2003                                 break;
2004                         }
2005
2006                         if (optval)
2007                                 tp->t_flags |= opt;
2008                         else
2009                                 tp->t_flags &= ~opt;
2010 unlock_and_done:
2011 #ifdef TCP_OFFLOAD
2012                         if (tp->t_flags & TF_TOE) {
2013                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2014                                     sopt->sopt_name);
2015                         }
2016 #endif
2017                         INP_WUNLOCK(inp);
2018                         break;
2019
2020                 case TCP_NOPUSH:
2021                         INP_WUNLOCK(inp);
2022                         error = sooptcopyin(sopt, &optval, sizeof optval,
2023                             sizeof optval);
2024                         if (error)
2025                                 return (error);
2026
2027                         INP_WLOCK_RECHECK(inp);
2028                         if (optval)
2029                                 tp->t_flags |= TF_NOPUSH;
2030                         else if (tp->t_flags & TF_NOPUSH) {
2031                                 tp->t_flags &= ~TF_NOPUSH;
2032                                 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2033                                         struct epoch_tracker et;
2034
2035                                         NET_EPOCH_ENTER(et);
2036                                         error = tp->t_fb->tfb_tcp_output(tp);
2037                                         NET_EPOCH_EXIT(et);
2038                                 }
2039                         }
2040                         goto unlock_and_done;
2041
2042                 case TCP_REMOTE_UDP_ENCAPS_PORT:
2043                         INP_WUNLOCK(inp);
2044                         error = sooptcopyin(sopt, &optval, sizeof optval,
2045                             sizeof optval);
2046                         if (error)
2047                                 return (error);
2048                         if ((optval < TCP_TUNNELING_PORT_MIN) ||
2049                             (optval > TCP_TUNNELING_PORT_MAX)) {
2050                                 /* Its got to be in range */
2051                                 return (EINVAL);
2052                         }
2053                         if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2054                                 /* You have to have enabled a UDP tunneling port first */
2055                                 return (EINVAL);
2056                         }
2057                         INP_WLOCK_RECHECK(inp);
2058                         if (tp->t_state != TCPS_CLOSED) {
2059                                 /* You can't change after you are connected */
2060                                 error = EINVAL;
2061                         } else {
2062                                 /* Ok we are all good set the port */
2063                                 tp->t_port = htons(optval);
2064                         }
2065                         goto unlock_and_done;
2066
2067                 case TCP_MAXSEG:
2068                         INP_WUNLOCK(inp);
2069                         error = sooptcopyin(sopt, &optval, sizeof optval,
2070                             sizeof optval);
2071                         if (error)
2072                                 return (error);
2073
2074                         INP_WLOCK_RECHECK(inp);
2075                         if (optval > 0 && optval <= tp->t_maxseg &&
2076                             optval + 40 >= V_tcp_minmss)
2077                                 tp->t_maxseg = optval;
2078                         else
2079                                 error = EINVAL;
2080                         goto unlock_and_done;
2081
2082                 case TCP_INFO:
2083                         INP_WUNLOCK(inp);
2084                         error = EINVAL;
2085                         break;
2086
2087                 case TCP_STATS:
2088                         INP_WUNLOCK(inp);
2089 #ifdef STATS
2090                         error = sooptcopyin(sopt, &optval, sizeof optval,
2091                             sizeof optval);
2092                         if (error)
2093                                 return (error);
2094
2095                         if (optval > 0)
2096                                 sbp = stats_blob_alloc(
2097                                     V_tcp_perconn_stats_dflt_tpl, 0);
2098                         else
2099                                 sbp = NULL;
2100
2101                         INP_WLOCK_RECHECK(inp);
2102                         if ((tp->t_stats != NULL && sbp == NULL) ||
2103                             (tp->t_stats == NULL && sbp != NULL)) {
2104                                 struct statsblob *t = tp->t_stats;
2105                                 tp->t_stats = sbp;
2106                                 sbp = t;
2107                         }
2108                         INP_WUNLOCK(inp);
2109
2110                         stats_blob_destroy(sbp);
2111 #else
2112                         return (EOPNOTSUPP);
2113 #endif /* !STATS */
2114                         break;
2115
2116                 case TCP_CONGESTION:
2117                         INP_WUNLOCK(inp);
2118                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2119                         if (error)
2120                                 break;
2121                         buf[sopt->sopt_valsize] = '\0';
2122                         INP_WLOCK_RECHECK(inp);
2123                         CC_LIST_RLOCK();
2124                         STAILQ_FOREACH(algo, &cc_list, entries)
2125                                 if (strncmp(buf, algo->name,
2126                                     TCP_CA_NAME_MAX) == 0)
2127                                         break;
2128                         CC_LIST_RUNLOCK();
2129                         if (algo == NULL) {
2130                                 INP_WUNLOCK(inp);
2131                                 error = EINVAL;
2132                                 break;
2133                         }
2134                         /*
2135                          * We hold a write lock over the tcb so it's safe to
2136                          * do these things without ordering concerns.
2137                          */
2138                         if (CC_ALGO(tp)->cb_destroy != NULL)
2139                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
2140                         CC_DATA(tp) = NULL;
2141                         CC_ALGO(tp) = algo;
2142                         /*
2143                          * If something goes pear shaped initialising the new
2144                          * algo, fall back to newreno (which does not
2145                          * require initialisation).
2146                          */
2147                         if (algo->cb_init != NULL &&
2148                             algo->cb_init(tp->ccv) != 0) {
2149                                 CC_ALGO(tp) = &newreno_cc_algo;
2150                                 /*
2151                                  * The only reason init should fail is
2152                                  * because of malloc.
2153                                  */
2154                                 error = ENOMEM;
2155                         }
2156                         INP_WUNLOCK(inp);
2157                         break;
2158
2159                 case TCP_REUSPORT_LB_NUMA:
2160                         INP_WUNLOCK(inp);
2161                         error = sooptcopyin(sopt, &optval, sizeof(optval),
2162                             sizeof(optval));
2163                         INP_WLOCK_RECHECK(inp);
2164                         if (!error)
2165                                 error = in_pcblbgroup_numa(inp, optval);
2166                         INP_WUNLOCK(inp);
2167                         break;
2168
2169 #ifdef KERN_TLS
2170                 case TCP_TXTLS_ENABLE:
2171                         INP_WUNLOCK(inp);
2172                         error = copyin_tls_enable(sopt, &tls);
2173                         if (error)
2174                                 break;
2175                         error = ktls_enable_tx(so, &tls);
2176                         break;
2177                 case TCP_TXTLS_MODE:
2178                         INP_WUNLOCK(inp);
2179                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2180                         if (error)
2181                                 return (error);
2182
2183                         INP_WLOCK_RECHECK(inp);
2184                         error = ktls_set_tx_mode(so, ui);
2185                         INP_WUNLOCK(inp);
2186                         break;
2187                 case TCP_RXTLS_ENABLE:
2188                         INP_WUNLOCK(inp);
2189                         error = sooptcopyin(sopt, &tls, sizeof(tls),
2190                             sizeof(tls));
2191                         if (error)
2192                                 break;
2193                         error = ktls_enable_rx(so, &tls);
2194                         break;
2195 #endif
2196
2197                 case TCP_KEEPIDLE:
2198                 case TCP_KEEPINTVL:
2199                 case TCP_KEEPINIT:
2200                         INP_WUNLOCK(inp);
2201                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2202                         if (error)
2203                                 return (error);
2204
2205                         if (ui > (UINT_MAX / hz)) {
2206                                 error = EINVAL;
2207                                 break;
2208                         }
2209                         ui *= hz;
2210
2211                         INP_WLOCK_RECHECK(inp);
2212                         switch (sopt->sopt_name) {
2213                         case TCP_KEEPIDLE:
2214                                 tp->t_keepidle = ui;
2215                                 /*
2216                                  * XXX: better check current remaining
2217                                  * timeout and "merge" it with new value.
2218                                  */
2219                                 if ((tp->t_state > TCPS_LISTEN) &&
2220                                     (tp->t_state <= TCPS_CLOSING))
2221                                         tcp_timer_activate(tp, TT_KEEP,
2222                                             TP_KEEPIDLE(tp));
2223                                 break;
2224                         case TCP_KEEPINTVL:
2225                                 tp->t_keepintvl = ui;
2226                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2227                                     (TP_MAXIDLE(tp) > 0))
2228                                         tcp_timer_activate(tp, TT_2MSL,
2229                                             TP_MAXIDLE(tp));
2230                                 break;
2231                         case TCP_KEEPINIT:
2232                                 tp->t_keepinit = ui;
2233                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
2234                                     tp->t_state == TCPS_SYN_SENT)
2235                                         tcp_timer_activate(tp, TT_KEEP,
2236                                             TP_KEEPINIT(tp));
2237                                 break;
2238                         }
2239                         goto unlock_and_done;
2240
2241                 case TCP_KEEPCNT:
2242                         INP_WUNLOCK(inp);
2243                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2244                         if (error)
2245                                 return (error);
2246
2247                         INP_WLOCK_RECHECK(inp);
2248                         tp->t_keepcnt = ui;
2249                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2250                             (TP_MAXIDLE(tp) > 0))
2251                                 tcp_timer_activate(tp, TT_2MSL,
2252                                     TP_MAXIDLE(tp));
2253                         goto unlock_and_done;
2254
2255 #ifdef TCPPCAP
2256                 case TCP_PCAP_OUT:
2257                 case TCP_PCAP_IN:
2258                         INP_WUNLOCK(inp);
2259                         error = sooptcopyin(sopt, &optval, sizeof optval,
2260                             sizeof optval);
2261                         if (error)
2262                                 return (error);
2263
2264                         INP_WLOCK_RECHECK(inp);
2265                         if (optval >= 0)
2266                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2267                                         &(tp->t_outpkts) : &(tp->t_inpkts),
2268                                         optval);
2269                         else
2270                                 error = EINVAL;
2271                         goto unlock_and_done;
2272 #endif
2273
2274                 case TCP_FASTOPEN: {
2275                         struct tcp_fastopen tfo_optval;
2276
2277                         INP_WUNLOCK(inp);
2278                         if (!V_tcp_fastopen_client_enable &&
2279                             !V_tcp_fastopen_server_enable)
2280                                 return (EPERM);
2281
2282                         error = sooptcopyin(sopt, &tfo_optval,
2283                                     sizeof(tfo_optval), sizeof(int));
2284                         if (error)
2285                                 return (error);
2286
2287                         INP_WLOCK_RECHECK(inp);
2288                         if ((tp->t_state != TCPS_CLOSED) &&
2289                             (tp->t_state != TCPS_LISTEN)) {
2290                                 error = EINVAL;
2291                                 goto unlock_and_done;
2292                         }
2293                         if (tfo_optval.enable) {
2294                                 if (tp->t_state == TCPS_LISTEN) {
2295                                         if (!V_tcp_fastopen_server_enable) {
2296                                                 error = EPERM;
2297                                                 goto unlock_and_done;
2298                                         }
2299
2300                                         if (tp->t_tfo_pending == NULL)
2301                                                 tp->t_tfo_pending =
2302                                                     tcp_fastopen_alloc_counter();
2303                                 } else {
2304                                         /*
2305                                          * If a pre-shared key was provided,
2306                                          * stash it in the client cookie
2307                                          * field of the tcpcb for use during
2308                                          * connect.
2309                                          */
2310                                         if (sopt->sopt_valsize ==
2311                                             sizeof(tfo_optval)) {
2312                                                 memcpy(tp->t_tfo_cookie.client,
2313                                                        tfo_optval.psk,
2314                                                        TCP_FASTOPEN_PSK_LEN);
2315                                                 tp->t_tfo_client_cookie_len =
2316                                                     TCP_FASTOPEN_PSK_LEN;
2317                                         }
2318                                 }
2319                                 tp->t_flags |= TF_FASTOPEN;
2320                         } else
2321                                 tp->t_flags &= ~TF_FASTOPEN;
2322                         goto unlock_and_done;
2323                 }
2324
2325 #ifdef TCP_BLACKBOX
2326                 case TCP_LOG:
2327                         INP_WUNLOCK(inp);
2328                         error = sooptcopyin(sopt, &optval, sizeof optval,
2329                             sizeof optval);
2330                         if (error)
2331                                 return (error);
2332
2333                         INP_WLOCK_RECHECK(inp);
2334                         error = tcp_log_state_change(tp, optval);
2335                         goto unlock_and_done;
2336
2337                 case TCP_LOGBUF:
2338                         INP_WUNLOCK(inp);
2339                         error = EINVAL;
2340                         break;
2341
2342                 case TCP_LOGID:
2343                         INP_WUNLOCK(inp);
2344                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2345                         if (error)
2346                                 break;
2347                         buf[sopt->sopt_valsize] = '\0';
2348                         INP_WLOCK_RECHECK(inp);
2349                         error = tcp_log_set_id(tp, buf);
2350                         /* tcp_log_set_id() unlocks the INP. */
2351                         break;
2352
2353                 case TCP_LOGDUMP:
2354                 case TCP_LOGDUMPID:
2355                         INP_WUNLOCK(inp);
2356                         error =
2357                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2358                         if (error)
2359                                 break;
2360                         buf[sopt->sopt_valsize] = '\0';
2361                         INP_WLOCK_RECHECK(inp);
2362                         if (sopt->sopt_name == TCP_LOGDUMP) {
2363                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2364                                     M_WAITOK, true);
2365                                 INP_WUNLOCK(inp);
2366                         } else {
2367                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2368                                 /*
2369                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2370                                  * INP lock.
2371                                  */
2372                         }
2373                         break;
2374 #endif
2375
2376                 default:
2377                         INP_WUNLOCK(inp);
2378                         error = ENOPROTOOPT;
2379                         break;
2380                 }
2381                 break;
2382
2383         case SOPT_GET:
2384                 tp = intotcpcb(inp);
2385                 switch (sopt->sopt_name) {
2386 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2387                 case TCP_MD5SIG:
2388                         if (!TCPMD5_ENABLED()) {
2389                                 INP_WUNLOCK(inp);
2390                                 return (ENOPROTOOPT);
2391                         }
2392                         error = TCPMD5_PCBCTL(inp, sopt);
2393                         break;
2394 #endif
2395
2396                 case TCP_NODELAY:
2397                         optval = tp->t_flags & TF_NODELAY;
2398                         INP_WUNLOCK(inp);
2399                         error = sooptcopyout(sopt, &optval, sizeof optval);
2400                         break;
2401                 case TCP_MAXSEG:
2402                         optval = tp->t_maxseg;
2403                         INP_WUNLOCK(inp);
2404                         error = sooptcopyout(sopt, &optval, sizeof optval);
2405                         break;
2406                 case TCP_REMOTE_UDP_ENCAPS_PORT:
2407                         optval = ntohs(tp->t_port);
2408                         INP_WUNLOCK(inp);
2409                         error = sooptcopyout(sopt, &optval, sizeof optval);
2410                         break;
2411                 case TCP_NOOPT:
2412                         optval = tp->t_flags & TF_NOOPT;
2413                         INP_WUNLOCK(inp);
2414                         error = sooptcopyout(sopt, &optval, sizeof optval);
2415                         break;
2416                 case TCP_NOPUSH:
2417                         optval = tp->t_flags & TF_NOPUSH;
2418                         INP_WUNLOCK(inp);
2419                         error = sooptcopyout(sopt, &optval, sizeof optval);
2420                         break;
2421                 case TCP_INFO:
2422                         tcp_fill_info(tp, &ti);
2423                         INP_WUNLOCK(inp);
2424                         error = sooptcopyout(sopt, &ti, sizeof ti);
2425                         break;
2426                 case TCP_STATS:
2427                         {
2428 #ifdef STATS
2429                         int nheld;
2430                         TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2431
2432                         error = 0;
2433                         socklen_t outsbsz = sopt->sopt_valsize;
2434                         if (tp->t_stats == NULL)
2435                                 error = ENOENT;
2436                         else if (outsbsz >= tp->t_stats->cursz)
2437                                 outsbsz = tp->t_stats->cursz;
2438                         else if (outsbsz >= sizeof(struct statsblob))
2439                                 outsbsz = sizeof(struct statsblob);
2440                         else
2441                                 error = EINVAL;
2442                         INP_WUNLOCK(inp);
2443                         if (error)
2444                                 break;
2445
2446                         sbp = sopt->sopt_val;
2447                         nheld = atop(round_page(((vm_offset_t)sbp) +
2448                             (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2449                         vm_page_t ma[nheld];
2450                         if (vm_fault_quick_hold_pages(
2451                             &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2452                             outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2453                             nheld) < 0) {
2454                                 error = EFAULT;
2455                                 break;
2456                         }
2457
2458                         if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2459                             SIZEOF_MEMBER(struct statsblob, flags))))
2460                                 goto unhold;
2461
2462                         INP_WLOCK_RECHECK(inp);
2463                         error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2464                             sbflags | SB_CLONE_USRDSTNOFAULT);
2465                         INP_WUNLOCK(inp);
2466                         sopt->sopt_valsize = outsbsz;
2467 unhold:
2468                         vm_page_unhold_pages(ma, nheld);
2469 #else
2470                         INP_WUNLOCK(inp);
2471                         error = EOPNOTSUPP;
2472 #endif /* !STATS */
2473                         break;
2474                         }
2475                 case TCP_CONGESTION:
2476                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2477                         INP_WUNLOCK(inp);
2478                         error = sooptcopyout(sopt, buf, len + 1);
2479                         break;
2480                 case TCP_KEEPIDLE:
2481                 case TCP_KEEPINTVL:
2482                 case TCP_KEEPINIT:
2483                 case TCP_KEEPCNT:
2484                         switch (sopt->sopt_name) {
2485                         case TCP_KEEPIDLE:
2486                                 ui = TP_KEEPIDLE(tp) / hz;
2487                                 break;
2488                         case TCP_KEEPINTVL:
2489                                 ui = TP_KEEPINTVL(tp) / hz;
2490                                 break;
2491                         case TCP_KEEPINIT:
2492                                 ui = TP_KEEPINIT(tp) / hz;
2493                                 break;
2494                         case TCP_KEEPCNT:
2495                                 ui = TP_KEEPCNT(tp);
2496                                 break;
2497                         }
2498                         INP_WUNLOCK(inp);
2499                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2500                         break;
2501 #ifdef TCPPCAP
2502                 case TCP_PCAP_OUT:
2503                 case TCP_PCAP_IN:
2504                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2505                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2506                         INP_WUNLOCK(inp);
2507                         error = sooptcopyout(sopt, &optval, sizeof optval);
2508                         break;
2509 #endif
2510                 case TCP_FASTOPEN:
2511                         optval = tp->t_flags & TF_FASTOPEN;
2512                         INP_WUNLOCK(inp);
2513                         error = sooptcopyout(sopt, &optval, sizeof optval);
2514                         break;
2515 #ifdef TCP_BLACKBOX
2516                 case TCP_LOG:
2517                         optval = tp->t_logstate;
2518                         INP_WUNLOCK(inp);
2519                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2520                         break;
2521                 case TCP_LOGBUF:
2522                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2523                         error = tcp_log_getlogbuf(sopt, tp);
2524                         break;
2525                 case TCP_LOGID:
2526                         len = tcp_log_get_id(tp, buf);
2527                         INP_WUNLOCK(inp);
2528                         error = sooptcopyout(sopt, buf, len + 1);
2529                         break;
2530                 case TCP_LOGDUMP:
2531                 case TCP_LOGDUMPID:
2532                         INP_WUNLOCK(inp);
2533                         error = EINVAL;
2534                         break;
2535 #endif
2536 #ifdef KERN_TLS
2537                 case TCP_TXTLS_MODE:
2538                         optval = ktls_get_tx_mode(so);
2539                         INP_WUNLOCK(inp);
2540                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2541                         break;
2542                 case TCP_RXTLS_MODE:
2543                         optval = ktls_get_rx_mode(so);
2544                         INP_WUNLOCK(inp);
2545                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2546                         break;
2547 #endif
2548                 case TCP_LRD:
2549                         optval = tp->t_flags & TF_LRD;
2550                         INP_WUNLOCK(inp);
2551                         error = sooptcopyout(sopt, &optval, sizeof optval);
2552                         break;
2553                 default:
2554                         INP_WUNLOCK(inp);
2555                         error = ENOPROTOOPT;
2556                         break;
2557                 }
2558                 break;
2559         }
2560         return (error);
2561 }
2562 #undef INP_WLOCK_RECHECK
2563 #undef INP_WLOCK_RECHECK_CLEANUP
2564
2565 /*
2566  * Initiate (or continue) disconnect.
2567  * If embryonic state, just send reset (once).
2568  * If in ``let data drain'' option and linger null, just drop.
2569  * Otherwise (hard), mark socket disconnecting and drop
2570  * current input data; switch states based on user close, and
2571  * send segment to peer (with FIN).
2572  */
2573 static void
2574 tcp_disconnect(struct tcpcb *tp)
2575 {
2576         struct inpcb *inp = tp->t_inpcb;
2577         struct socket *so = inp->inp_socket;
2578
2579         NET_EPOCH_ASSERT();
2580         INP_WLOCK_ASSERT(inp);
2581
2582         /*
2583          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2584          * socket is still open.
2585          */
2586         if (tp->t_state < TCPS_ESTABLISHED &&
2587             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2588                 tp = tcp_close(tp);
2589                 KASSERT(tp != NULL,
2590                     ("tcp_disconnect: tcp_close() returned NULL"));
2591         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2592                 tp = tcp_drop(tp, 0);
2593                 KASSERT(tp != NULL,
2594                     ("tcp_disconnect: tcp_drop() returned NULL"));
2595         } else {
2596                 soisdisconnecting(so);
2597                 sbflush(&so->so_rcv);
2598                 tcp_usrclosed(tp);
2599                 if (!(inp->inp_flags & INP_DROPPED))
2600                         tp->t_fb->tfb_tcp_output(tp);
2601         }
2602 }
2603
2604 /*
2605  * User issued close, and wish to trail through shutdown states:
2606  * if never received SYN, just forget it.  If got a SYN from peer,
2607  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2608  * If already got a FIN from peer, then almost done; go to LAST_ACK
2609  * state.  In all other cases, have already sent FIN to peer (e.g.
2610  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2611  * for peer to send FIN or not respond to keep-alives, etc.
2612  * We can let the user exit from the close as soon as the FIN is acked.
2613  */
2614 static void
2615 tcp_usrclosed(struct tcpcb *tp)
2616 {
2617
2618         NET_EPOCH_ASSERT();
2619         INP_WLOCK_ASSERT(tp->t_inpcb);
2620
2621         switch (tp->t_state) {
2622         case TCPS_LISTEN:
2623 #ifdef TCP_OFFLOAD
2624                 tcp_offload_listen_stop(tp);
2625 #endif
2626                 tcp_state_change(tp, TCPS_CLOSED);
2627                 /* FALLTHROUGH */
2628         case TCPS_CLOSED:
2629                 tp = tcp_close(tp);
2630                 /*
2631                  * tcp_close() should never return NULL here as the socket is
2632                  * still open.
2633                  */
2634                 KASSERT(tp != NULL,
2635                     ("tcp_usrclosed: tcp_close() returned NULL"));
2636                 break;
2637
2638         case TCPS_SYN_SENT:
2639         case TCPS_SYN_RECEIVED:
2640                 tp->t_flags |= TF_NEEDFIN;
2641                 break;
2642
2643         case TCPS_ESTABLISHED:
2644                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2645                 break;
2646
2647         case TCPS_CLOSE_WAIT:
2648                 tcp_state_change(tp, TCPS_LAST_ACK);
2649                 break;
2650         }
2651         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2652                 soisdisconnected(tp->t_inpcb->inp_socket);
2653                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2654                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2655                         int timeout;
2656
2657                         timeout = (tcp_fast_finwait2_recycle) ?
2658                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2659                         tcp_timer_activate(tp, TT_2MSL, timeout);
2660                 }
2661         }
2662 }
2663
2664 #ifdef DDB
2665 static void
2666 db_print_indent(int indent)
2667 {
2668         int i;
2669
2670         for (i = 0; i < indent; i++)
2671                 db_printf(" ");
2672 }
2673
2674 static void
2675 db_print_tstate(int t_state)
2676 {
2677
2678         switch (t_state) {
2679         case TCPS_CLOSED:
2680                 db_printf("TCPS_CLOSED");
2681                 return;
2682
2683         case TCPS_LISTEN:
2684                 db_printf("TCPS_LISTEN");
2685                 return;
2686
2687         case TCPS_SYN_SENT:
2688                 db_printf("TCPS_SYN_SENT");
2689                 return;
2690
2691         case TCPS_SYN_RECEIVED:
2692                 db_printf("TCPS_SYN_RECEIVED");
2693                 return;
2694
2695         case TCPS_ESTABLISHED:
2696                 db_printf("TCPS_ESTABLISHED");
2697                 return;
2698
2699         case TCPS_CLOSE_WAIT:
2700                 db_printf("TCPS_CLOSE_WAIT");
2701                 return;
2702
2703         case TCPS_FIN_WAIT_1:
2704                 db_printf("TCPS_FIN_WAIT_1");
2705                 return;
2706
2707         case TCPS_CLOSING:
2708                 db_printf("TCPS_CLOSING");
2709                 return;
2710
2711         case TCPS_LAST_ACK:
2712                 db_printf("TCPS_LAST_ACK");
2713                 return;
2714
2715         case TCPS_FIN_WAIT_2:
2716                 db_printf("TCPS_FIN_WAIT_2");
2717                 return;
2718
2719         case TCPS_TIME_WAIT:
2720                 db_printf("TCPS_TIME_WAIT");
2721                 return;
2722
2723         default:
2724                 db_printf("unknown");
2725                 return;
2726         }
2727 }
2728
2729 static void
2730 db_print_tflags(u_int t_flags)
2731 {
2732         int comma;
2733
2734         comma = 0;
2735         if (t_flags & TF_ACKNOW) {
2736                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2737                 comma = 1;
2738         }
2739         if (t_flags & TF_DELACK) {
2740                 db_printf("%sTF_DELACK", comma ? ", " : "");
2741                 comma = 1;
2742         }
2743         if (t_flags & TF_NODELAY) {
2744                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2745                 comma = 1;
2746         }
2747         if (t_flags & TF_NOOPT) {
2748                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2749                 comma = 1;
2750         }
2751         if (t_flags & TF_SENTFIN) {
2752                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2753                 comma = 1;
2754         }
2755         if (t_flags & TF_REQ_SCALE) {
2756                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2757                 comma = 1;
2758         }
2759         if (t_flags & TF_RCVD_SCALE) {
2760                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2761                 comma = 1;
2762         }
2763         if (t_flags & TF_REQ_TSTMP) {
2764                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2765                 comma = 1;
2766         }
2767         if (t_flags & TF_RCVD_TSTMP) {
2768                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2769                 comma = 1;
2770         }
2771         if (t_flags & TF_SACK_PERMIT) {
2772                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2773                 comma = 1;
2774         }
2775         if (t_flags & TF_NEEDSYN) {
2776                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2777                 comma = 1;
2778         }
2779         if (t_flags & TF_NEEDFIN) {
2780                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2781                 comma = 1;
2782         }
2783         if (t_flags & TF_NOPUSH) {
2784                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2785                 comma = 1;
2786         }
2787         if (t_flags & TF_MORETOCOME) {
2788                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2789                 comma = 1;
2790         }
2791         if (t_flags & TF_LQ_OVERFLOW) {
2792                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2793                 comma = 1;
2794         }
2795         if (t_flags & TF_LASTIDLE) {
2796                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2797                 comma = 1;
2798         }
2799         if (t_flags & TF_RXWIN0SENT) {
2800                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2801                 comma = 1;
2802         }
2803         if (t_flags & TF_FASTRECOVERY) {
2804                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2805                 comma = 1;
2806         }
2807         if (t_flags & TF_CONGRECOVERY) {
2808                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2809                 comma = 1;
2810         }
2811         if (t_flags & TF_WASFRECOVERY) {
2812                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2813                 comma = 1;
2814         }
2815         if (t_flags & TF_SIGNATURE) {
2816                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2817                 comma = 1;
2818         }
2819         if (t_flags & TF_FORCEDATA) {
2820                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2821                 comma = 1;
2822         }
2823         if (t_flags & TF_TSO) {
2824                 db_printf("%sTF_TSO", comma ? ", " : "");
2825                 comma = 1;
2826         }
2827         if (t_flags & TF_FASTOPEN) {
2828                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2829                 comma = 1;
2830         }
2831 }
2832
2833 static void
2834 db_print_tflags2(u_int t_flags2)
2835 {
2836         int comma;
2837
2838         comma = 0;
2839         if (t_flags2 & TF2_ECN_PERMIT) {
2840                 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2841                 comma = 1;
2842         }
2843 }
2844
2845 static void
2846 db_print_toobflags(char t_oobflags)
2847 {
2848         int comma;
2849
2850         comma = 0;
2851         if (t_oobflags & TCPOOB_HAVEDATA) {
2852                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2853                 comma = 1;
2854         }
2855         if (t_oobflags & TCPOOB_HADDATA) {
2856                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2857                 comma = 1;
2858         }
2859 }
2860
2861 static void
2862 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2863 {
2864
2865         db_print_indent(indent);
2866         db_printf("%s at %p\n", name, tp);
2867
2868         indent += 2;
2869
2870         db_print_indent(indent);
2871         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2872            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2873
2874         db_print_indent(indent);
2875         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2876             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2877
2878         db_print_indent(indent);
2879         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2880             &tp->t_timers->tt_delack, tp->t_inpcb);
2881
2882         db_print_indent(indent);
2883         db_printf("t_state: %d (", tp->t_state);
2884         db_print_tstate(tp->t_state);
2885         db_printf(")\n");
2886
2887         db_print_indent(indent);
2888         db_printf("t_flags: 0x%x (", tp->t_flags);
2889         db_print_tflags(tp->t_flags);
2890         db_printf(")\n");
2891
2892         db_print_indent(indent);
2893         db_printf("t_flags2: 0x%x (", tp->t_flags2);
2894         db_print_tflags2(tp->t_flags2);
2895         db_printf(")\n");
2896
2897         db_print_indent(indent);
2898         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2899             tp->snd_una, tp->snd_max, tp->snd_nxt);
2900
2901         db_print_indent(indent);
2902         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2903            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2904
2905         db_print_indent(indent);
2906         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2907             tp->iss, tp->irs, tp->rcv_nxt);
2908
2909         db_print_indent(indent);
2910         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2911             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2912
2913         db_print_indent(indent);
2914         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2915            tp->snd_wnd, tp->snd_cwnd);
2916
2917         db_print_indent(indent);
2918         db_printf("snd_ssthresh: %u   snd_recover: "
2919             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2920
2921         db_print_indent(indent);
2922         db_printf("t_rcvtime: %u   t_startime: %u\n",
2923             tp->t_rcvtime, tp->t_starttime);
2924
2925         db_print_indent(indent);
2926         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2927             tp->t_rtttime, tp->t_rtseq);
2928
2929         db_print_indent(indent);
2930         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2931             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2932
2933         db_print_indent(indent);
2934         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2935             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2936             tp->t_rttbest);
2937
2938         db_print_indent(indent);
2939         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2940             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2941
2942         db_print_indent(indent);
2943         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2944         db_print_toobflags(tp->t_oobflags);
2945         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2946
2947         db_print_indent(indent);
2948         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2949             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2950
2951         db_print_indent(indent);
2952         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2953             tp->ts_recent, tp->ts_recent_age);
2954
2955         db_print_indent(indent);
2956         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2957             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2958
2959         db_print_indent(indent);
2960         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2961             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2962             tp->snd_recover_prev, tp->t_badrxtwin);
2963
2964         db_print_indent(indent);
2965         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2966             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2967
2968         db_print_indent(indent);
2969         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
2970             tp->snd_fack, tp->rcv_numsacks);
2971
2972         /* Skip sackblks, sackhint. */
2973
2974         db_print_indent(indent);
2975         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2976             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2977 }
2978
2979 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2980 {
2981         struct tcpcb *tp;
2982
2983         if (!have_addr) {
2984                 db_printf("usage: show tcpcb <addr>\n");
2985                 return;
2986         }
2987         tp = (struct tcpcb *)addr;
2988
2989         db_print_tcpcb(tp, "tcpcb", 0);
2990 }
2991 #endif