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         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 208                 so->so_linger = TCP_LINGERTIME;
 209 out:
 210         TCPDEBUG2(PRU_ATTACH);
 211         TCP_PROBE2(debug__user, tp, PRU_ATTACH);
 212         return (error);
 213 }
 214
 215 /*
 216  * tcp_usr_detach is called when the socket layer loses its final reference
 217  * to the socket, be it a file descriptor reference, a reference from TCP,
 218  * etc.  At this point, there is only one case in which we will keep around
 219  * inpcb state: time wait.
 220  */
 221 static void
 222 tcp_usr_detach(struct socket *so)
 223 {
 224         struct inpcb *inp;
 225         struct tcpcb *tp;
 226
 227         inp = sotoinpcb(so);
 228         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 229         INP_WLOCK(inp);
 230         KASSERT(so->so_pcb == inp && inp->inp_socket == so,
 231                 ("%s: socket %p inp %p mismatch", __func__, so, inp));
 232
 233         tp = intotcpcb(inp);
 234
 235         if (inp->inp_flags & INP_TIMEWAIT) {
 236                 /*
 237                  * There are two cases to handle: one in which the time wait
 238                  * state is being discarded (INP_DROPPED), and one in which
 239                  * this connection will remain in timewait.  In the former,
 240                  * it is time to discard all state (except tcptw, which has
 241                  * already been discarded by the timewait close code, which
 242                  * should be further up the call stack somewhere).  In the
 243                  * latter case, we detach from the socket, but leave the pcb
 244                  * present until timewait ends.
 245                  *
 246                  * XXXRW: Would it be cleaner to free the tcptw here?
 247                  *
 248                  * Astute question indeed, from twtcp perspective there are
 249                  * four cases to consider:
 250                  *
 251                  * #1 tcp_usr_detach is called at tcptw creation time by
 252                  *  tcp_twstart, then do not discard the newly created tcptw
 253                  *  and leave inpcb present until timewait ends
 254                  * #2 tcp_usr_detach is called at tcptw creation time by
 255                  *  tcp_twstart, but connection is local and tw will be
 256                  *  discarded immediately
 257                  * #3 tcp_usr_detach is called at timewait end (or reuse) by
 258                  *  tcp_twclose, then the tcptw has already been discarded
 259                  *  (or reused) and inpcb is freed here
 260                  * #4 tcp_usr_detach is called() after timewait ends (or reuse)
 261                  *  (e.g. by soclose), then tcptw has already been discarded
 262                  *  (or reused) and inpcb is freed here
 263                  *
 264                  *  In all three cases the tcptw should not be freed here.
 265                  */
 266                 if (inp->inp_flags & INP_DROPPED) {
 267                         in_pcbdetach(inp);
 268                         if (__predict_true(tp == NULL)) {
 269                                 in_pcbfree(inp);
 270                         } else {
 271                                 /*
 272                                  * This case should not happen as in TIMEWAIT
 273                                  * state the inp should not be destroyed before
 274                                  * its tcptw.  If INVARIANTS is defined, panic.
 275                                  */
 276 #ifdef INVARIANTS
 277                                 panic("%s: Panic before an inp double-free: "
 278                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 279                                     , __func__);
 280 #else
 281                                 log(LOG_ERR, "%s: Avoid an inp double-free: "
 282                                     "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
 283                                     , __func__);
 284 #endif
 285                                 INP_WUNLOCK(inp);
 286                         }
 287                 } else {
 288                         in_pcbdetach(inp);
 289                         INP_WUNLOCK(inp);
 290                 }
 291         } else {
 292                 /*
 293                  * If the connection is not in timewait, we consider two
 294                  * two conditions: one in which no further processing is
 295                  * necessary (dropped || embryonic), and one in which TCP is
 296                  * not yet done, but no longer requires the socket, so the
 297                  * pcb will persist for the time being.
 298                  *
 299                  * XXXRW: Does the second case still occur?
 300                  */
 301                 if (inp->inp_flags & INP_DROPPED ||
 302                     tp->t_state < TCPS_SYN_SENT) {
 303                         tcp_discardcb(tp);
 304                         in_pcbdetach(inp);
 305                         in_pcbfree(inp);
 306                 } else {
 307                         in_pcbdetach(inp);
 308                         INP_WUNLOCK(inp);
 309                 }
 310         }
 311 }
 312
 313 #ifdef INET
 314 /*
 315  * Give the socket an address.
 316  */
 317 static int
 318 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 319 {
 320         int error = 0;
 321         struct inpcb *inp;
 322         struct tcpcb *tp = NULL;
 323         struct sockaddr_in *sinp;
 324
 325         sinp = (struct sockaddr_in *)nam;
 326         if (nam->sa_len != sizeof (*sinp))
 327                 return (EINVAL);
 328         /*
 329          * Must check for multicast addresses and disallow binding
 330          * to them.
 331          */
 332         if (sinp->sin_family == AF_INET &&
 333             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 334                 return (EAFNOSUPPORT);
 335
 336         TCPDEBUG0;
 337         inp = sotoinpcb(so);
 338         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 339         INP_WLOCK(inp);
 340         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 341                 error = EINVAL;
 342                 goto out;
 343         }
 344         tp = intotcpcb(inp);
 345         TCPDEBUG1();
 346         INP_HASH_WLOCK(&V_tcbinfo);
 347         error = in_pcbbind(inp, nam, td->td_ucred);
 348         INP_HASH_WUNLOCK(&V_tcbinfo);
 349 out:
 350         TCPDEBUG2(PRU_BIND);
 351         TCP_PROBE2(debug__user, tp, PRU_BIND);
 352         INP_WUNLOCK(inp);
 353
 354         return (error);
 355 }
 356 #endif /* INET */
 357
 358 #ifdef INET6
 359 static int
 360 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 361 {
 362         int error = 0;
 363         struct inpcb *inp;
 364         struct tcpcb *tp = NULL;
 365         struct sockaddr_in6 *sin6;
 366         u_char vflagsav;
 367
 368         sin6 = (struct sockaddr_in6 *)nam;
 369         if (nam->sa_len != sizeof (*sin6))
 370                 return (EINVAL);
 371         /*
 372          * Must check for multicast addresses and disallow binding
 373          * to them.
 374          */
 375         if (sin6->sin6_family == AF_INET6 &&
 376             IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 377                 return (EAFNOSUPPORT);
 378
 379         TCPDEBUG0;
 380         inp = sotoinpcb(so);
 381         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 382         INP_WLOCK(inp);
 383         vflagsav = inp->inp_vflag;
 384         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 385                 error = EINVAL;
 386                 goto out;
 387         }
 388         tp = intotcpcb(inp);
 389         TCPDEBUG1();
 390         INP_HASH_WLOCK(&V_tcbinfo);
 391         inp->inp_vflag &= ~INP_IPV4;
 392         inp->inp_vflag |= INP_IPV6;
 393 #ifdef INET
 394         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 395                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
 396                         inp->inp_vflag |= INP_IPV4;
 397                 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 398                         struct sockaddr_in sin;
 399
 400                         in6_sin6_2_sin(&sin, sin6);
 401                         if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 402                                 error = EAFNOSUPPORT;
 403                                 INP_HASH_WUNLOCK(&V_tcbinfo);
 404                                 goto out;
 405                         }
 406                         inp->inp_vflag |= INP_IPV4;
 407                         inp->inp_vflag &= ~INP_IPV6;
 408                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 409                             td->td_ucred);
 410                         INP_HASH_WUNLOCK(&V_tcbinfo);
 411                         goto out;
 412                 }
 413         }
 414 #endif
 415         error = in6_pcbbind(inp, nam, td->td_ucred);
 416         INP_HASH_WUNLOCK(&V_tcbinfo);
 417 out:
 418         if (error != 0)
 419                 inp->inp_vflag = vflagsav;
 420         TCPDEBUG2(PRU_BIND);
 421         TCP_PROBE2(debug__user, tp, PRU_BIND);
 422         INP_WUNLOCK(inp);
 423         return (error);
 424 }
 425 #endif /* INET6 */
 426
 427 #ifdef INET
 428 /*
 429  * Prepare to accept connections.
 430  */
 431 static int
 432 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 433 {
 434         int error = 0;
 435         struct inpcb *inp;
 436         struct tcpcb *tp = NULL;
 437
 438         TCPDEBUG0;
 439         inp = sotoinpcb(so);
 440         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 441         INP_WLOCK(inp);
 442         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 443                 error = EINVAL;
 444                 goto out;
 445         }
 446         tp = intotcpcb(inp);
 447         TCPDEBUG1();
 448         SOCK_LOCK(so);
 449         error = solisten_proto_check(so);
 450         INP_HASH_WLOCK(&V_tcbinfo);
 451         if (error == 0 && inp->inp_lport == 0)
 452                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 453         INP_HASH_WUNLOCK(&V_tcbinfo);
 454         if (error == 0) {
 455                 tcp_state_change(tp, TCPS_LISTEN);
 456                 solisten_proto(so, backlog);
 457 #ifdef TCP_OFFLOAD
 458                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 459                         tcp_offload_listen_start(tp);
 460 #endif
 461         }
 462         SOCK_UNLOCK(so);
 463
 464         if (IS_FASTOPEN(tp->t_flags))
 465                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 466
 467 out:
 468         TCPDEBUG2(PRU_LISTEN);
 469         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 470         INP_WUNLOCK(inp);
 471         return (error);
 472 }
 473 #endif /* INET */
 474
 475 #ifdef INET6
 476 static int
 477 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 478 {
 479         int error = 0;
 480         struct inpcb *inp;
 481         struct tcpcb *tp = NULL;
 482         u_char vflagsav;
 483
 484         TCPDEBUG0;
 485         inp = sotoinpcb(so);
 486         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 487         INP_WLOCK(inp);
 488         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 489                 error = EINVAL;
 490                 goto out;
 491         }
 492         vflagsav = inp->inp_vflag;
 493         tp = intotcpcb(inp);
 494         TCPDEBUG1();
 495         SOCK_LOCK(so);
 496         error = solisten_proto_check(so);
 497         INP_HASH_WLOCK(&V_tcbinfo);
 498         if (error == 0 && inp->inp_lport == 0) {
 499                 inp->inp_vflag &= ~INP_IPV4;
 500                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 501                         inp->inp_vflag |= INP_IPV4;
 502                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 503         }
 504         INP_HASH_WUNLOCK(&V_tcbinfo);
 505         if (error == 0) {
 506                 tcp_state_change(tp, TCPS_LISTEN);
 507                 solisten_proto(so, backlog);
 508 #ifdef TCP_OFFLOAD
 509                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 510                         tcp_offload_listen_start(tp);
 511 #endif
 512         }
 513         SOCK_UNLOCK(so);
 514
 515         if (IS_FASTOPEN(tp->t_flags))
 516                 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
 517
 518         if (error != 0)
 519                 inp->inp_vflag = vflagsav;
 520
 521 out:
 522         TCPDEBUG2(PRU_LISTEN);
 523         TCP_PROBE2(debug__user, tp, PRU_LISTEN);
 524         INP_WUNLOCK(inp);
 525         return (error);
 526 }
 527 #endif /* INET6 */
 528
 529 #ifdef INET
 530 /*
 531  * Initiate connection to peer.
 532  * Create a template for use in transmissions on this connection.
 533  * Enter SYN_SENT state, and mark socket as connecting.
 534  * Start keep-alive timer, and seed output sequence space.
 535  * Send initial segment on connection.
 536  */
 537 static int
 538 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 539 {
 540         struct epoch_tracker et;
 541         int error = 0;
 542         struct inpcb *inp;
 543         struct tcpcb *tp = NULL;
 544         struct sockaddr_in *sinp;
 545
 546         sinp = (struct sockaddr_in *)nam;
 547         if (nam->sa_len != sizeof (*sinp))
 548                 return (EINVAL);
 549         /*
 550          * Must disallow TCP ``connections'' to multicast addresses.
 551          */
 552         if (sinp->sin_family == AF_INET
 553             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 554                 return (EAFNOSUPPORT);
 555         if ((sinp->sin_family == AF_INET) &&
 556             ((ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) ||
 557              (sinp->sin_addr.s_addr == INADDR_ANY)))
 558                 return(EAFNOSUPPORT);
 559         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 560                 return (error);
 561
 562         TCPDEBUG0;
 563         inp = sotoinpcb(so);
 564         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 565         INP_WLOCK(inp);
 566         if (inp->inp_flags & INP_TIMEWAIT) {
 567                 error = EADDRINUSE;
 568                 goto out;
 569         }
 570         if (inp->inp_flags & INP_DROPPED) {
 571                 error = ECONNREFUSED;
 572                 goto out;
 573         }
 574         tp = intotcpcb(inp);
 575         TCPDEBUG1();
 576         NET_EPOCH_ENTER(et);
 577         if ((error = tcp_connect(tp, nam, td)) != 0)
 578                 goto out_in_epoch;
 579 #ifdef TCP_OFFLOAD
 580         if (registered_toedevs > 0 &&
 581             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 582             (error = tcp_offload_connect(so, nam)) == 0)
 583                 goto out_in_epoch;
 584 #endif
 585         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 586         error = tp->t_fb->tfb_tcp_output(tp);
 587 out_in_epoch:
 588         NET_EPOCH_EXIT(et);
 589 out:
 590         TCPDEBUG2(PRU_CONNECT);
 591         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 592         INP_WUNLOCK(inp);
 593         return (error);
 594 }
 595 #endif /* INET */
 596
 597 #ifdef INET6
 598 static int
 599 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 600 {
 601         struct epoch_tracker et;
 602         int error = 0;
 603         struct inpcb *inp;
 604         struct tcpcb *tp = NULL;
 605         struct sockaddr_in6 *sin6;
 606         u_int8_t incflagsav;
 607         u_char vflagsav;
 608
 609         TCPDEBUG0;
 610
 611         sin6 = (struct sockaddr_in6 *)nam;
 612         if (nam->sa_len != sizeof (*sin6))
 613                 return (EINVAL);
 614         /*
 615          * Must disallow TCP ``connections'' to multicast addresses.
 616          */
 617         if (sin6->sin6_family == AF_INET6
 618             && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 619                 return (EAFNOSUPPORT);
 620
 621         inp = sotoinpcb(so);
 622         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 623         INP_WLOCK(inp);
 624         vflagsav = inp->inp_vflag;
 625         incflagsav = inp->inp_inc.inc_flags;
 626         if (inp->inp_flags & INP_TIMEWAIT) {
 627                 error = EADDRINUSE;
 628                 goto out;
 629         }
 630         if (inp->inp_flags & INP_DROPPED) {
 631                 error = ECONNREFUSED;
 632                 goto out;
 633         }
 634         tp = intotcpcb(inp);
 635         TCPDEBUG1();
 636 #ifdef INET
 637         /*
 638          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 639          * therefore probably require the hash lock, which isn't held here.
 640          * Is this a significant problem?
 641          */
 642         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 643                 struct sockaddr_in sin;
 644
 645                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 646                         error = EINVAL;
 647                         goto out;
 648                 }
 649                 if ((inp->inp_vflag & INP_IPV4) == 0) {
 650                         error = EAFNOSUPPORT;
 651                         goto out;
 652                 }
 653
 654                 in6_sin6_2_sin(&sin, sin6);
 655                 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
 656                         error = EAFNOSUPPORT;
 657                         goto out;
 658                 }
 659                 if ((ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) ||
 660                     (sin.sin_addr.s_addr == INADDR_ANY)) {
 661                         error = EAFNOSUPPORT;
 662                         goto out;
 663                 }
 664                 if ((error = prison_remote_ip4(td->td_ucred,
 665                     &sin.sin_addr)) != 0)
 666                         goto out;
 667                 inp->inp_vflag |= INP_IPV4;
 668                 inp->inp_vflag &= ~INP_IPV6;
 669                 NET_EPOCH_ENTER(et);
 670                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 671                         goto out_in_epoch;
 672 #ifdef TCP_OFFLOAD
 673                 if (registered_toedevs > 0 &&
 674                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 675                     (error = tcp_offload_connect(so, nam)) == 0)
 676                         goto out_in_epoch;
 677 #endif
 678                 error = tp->t_fb->tfb_tcp_output(tp);
 679                 goto out_in_epoch;
 680         } else {
 681                 if ((inp->inp_vflag & INP_IPV6) == 0) {
 682                         error = EAFNOSUPPORT;
 683                         goto out;
 684                 }
 685         }
 686 #endif
 687         if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
 688                 goto out;
 689         inp->inp_vflag &= ~INP_IPV4;
 690         inp->inp_vflag |= INP_IPV6;
 691         inp->inp_inc.inc_flags |= INC_ISIPV6;
 692         if ((error = tcp6_connect(tp, nam, td)) != 0)
 693                 goto out;
 694 #ifdef TCP_OFFLOAD
 695         if (registered_toedevs > 0 &&
 696             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 697             (error = tcp_offload_connect(so, nam)) == 0)
 698                 goto out;
 699 #endif
 700         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 701         NET_EPOCH_ENTER(et);
 702         error = tp->t_fb->tfb_tcp_output(tp);
 703 #ifdef INET
 704 out_in_epoch:
 705 #endif
 706         NET_EPOCH_EXIT(et);
 707 out:
 708         /*
 709          * If the implicit bind in the connect call fails, restore
 710          * the flags we modified.
 711          */
 712         if (error != 0 && inp->inp_lport == 0) {
 713                 inp->inp_vflag = vflagsav;
 714                 inp->inp_inc.inc_flags = incflagsav;
 715         }
 716
 717         TCPDEBUG2(PRU_CONNECT);
 718         TCP_PROBE2(debug__user, tp, PRU_CONNECT);
 719         INP_WUNLOCK(inp);
 720         return (error);
 721 }
 722 #endif /* INET6 */
 723
 724 /*
 725  * Initiate disconnect from peer.
 726  * If connection never passed embryonic stage, just drop;
 727  * else if don't need to let data drain, then can just drop anyways,
 728  * else have to begin TCP shutdown process: mark socket disconnecting,
 729  * drain unread data, state switch to reflect user close, and
 730  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 731  * when peer sends FIN and acks ours.
 732  *
 733  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 734  */
 735 static int
 736 tcp_usr_disconnect(struct socket *so)
 737 {
 738         struct inpcb *inp;
 739         struct tcpcb *tp = NULL;
 740         struct epoch_tracker et;
 741         int error = 0;
 742
 743         TCPDEBUG0;
 744         NET_EPOCH_ENTER(et);
 745         inp = sotoinpcb(so);
 746         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 747         INP_WLOCK(inp);
 748         if (inp->inp_flags & INP_TIMEWAIT)
 749                 goto out;
 750         if (inp->inp_flags & INP_DROPPED) {
 751                 error = ECONNRESET;
 752                 goto out;
 753         }
 754         tp = intotcpcb(inp);
 755         TCPDEBUG1();
 756         tcp_disconnect(tp);
 757 out:
 758         TCPDEBUG2(PRU_DISCONNECT);
 759         TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
 760         INP_WUNLOCK(inp);
 761         NET_EPOCH_EXIT(et);
 762         return (error);
 763 }
 764
 765 #ifdef INET
 766 /*
 767  * Accept a connection.  Essentially all the work is done at higher levels;
 768  * just return the address of the peer, storing through addr.
 769  */
 770 static int
 771 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 772 {
 773         int error = 0;
 774         struct inpcb *inp = NULL;
 775         struct tcpcb *tp = NULL;
 776         struct in_addr addr;
 777         in_port_t port = 0;
 778         TCPDEBUG0;
 779
 780         if (so->so_state & SS_ISDISCONNECTED)
 781                 return (ECONNABORTED);
 782
 783         inp = sotoinpcb(so);
 784         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 785         INP_WLOCK(inp);
 786         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 787                 error = ECONNABORTED;
 788                 goto out;
 789         }
 790         tp = intotcpcb(inp);
 791         TCPDEBUG1();
 792
 793         /*
 794          * We inline in_getpeeraddr and COMMON_END here, so that we can
 795          * copy the data of interest and defer the malloc until after we
 796          * release the lock.
 797          */
 798         port = inp->inp_fport;
 799         addr = inp->inp_faddr;
 800
 801 out:
 802         TCPDEBUG2(PRU_ACCEPT);
 803         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 804         INP_WUNLOCK(inp);
 805         if (error == 0)
 806                 *nam = in_sockaddr(port, &addr);
 807         return error;
 808 }
 809 #endif /* INET */
 810
 811 #ifdef INET6
 812 static int
 813 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 814 {
 815         struct inpcb *inp = NULL;
 816         int error = 0;
 817         struct tcpcb *tp = NULL;
 818         struct in_addr addr;
 819         struct in6_addr addr6;
 820         struct epoch_tracker et;
 821         in_port_t port = 0;
 822         int v4 = 0;
 823         TCPDEBUG0;
 824
 825         if (so->so_state & SS_ISDISCONNECTED)
 826                 return (ECONNABORTED);
 827
 828         inp = sotoinpcb(so);
 829         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 830         NET_EPOCH_ENTER(et);
 831         INP_WLOCK(inp);
 832         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 833                 error = ECONNABORTED;
 834                 goto out;
 835         }
 836         tp = intotcpcb(inp);
 837         TCPDEBUG1();
 838
 839         /*
 840          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 841          * copy the data of interest and defer the malloc until after we
 842          * release the lock.
 843          */
 844         if (inp->inp_vflag & INP_IPV4) {
 845                 v4 = 1;
 846                 port = inp->inp_fport;
 847                 addr = inp->inp_faddr;
 848         } else {
 849                 port = inp->inp_fport;
 850                 addr6 = inp->in6p_faddr;
 851         }
 852
 853 out:
 854         TCPDEBUG2(PRU_ACCEPT);
 855         TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
 856         INP_WUNLOCK(inp);
 857         NET_EPOCH_EXIT(et);
 858         if (error == 0) {
 859                 if (v4)
 860                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 861                 else
 862                         *nam = in6_sockaddr(port, &addr6);
 863         }
 864         return error;
 865 }
 866 #endif /* INET6 */
 867
 868 /*
 869  * Mark the connection as being incapable of further output.
 870  */
 871 static int
 872 tcp_usr_shutdown(struct socket *so)
 873 {
 874         int error = 0;
 875         struct inpcb *inp;
 876         struct tcpcb *tp = NULL;
 877         struct epoch_tracker et;
 878
 879         TCPDEBUG0;
 880         NET_EPOCH_ENTER(et);
 881         inp = sotoinpcb(so);
 882         KASSERT(inp != NULL, ("inp == NULL"));
 883         INP_WLOCK(inp);
 884         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 885                 error = ECONNRESET;
 886                 goto out;
 887         }
 888         tp = intotcpcb(inp);
 889         TCPDEBUG1();
 890         socantsendmore(so);
 891         tcp_usrclosed(tp);
 892         if (!(inp->inp_flags & INP_DROPPED))
 893                 error = tp->t_fb->tfb_tcp_output(tp);
 894
 895 out:
 896         TCPDEBUG2(PRU_SHUTDOWN);
 897         TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
 898         INP_WUNLOCK(inp);
 899         NET_EPOCH_EXIT(et);
 900
 901         return (error);
 902 }
 903
 904 /*
 905  * After a receive, possibly send window update to peer.
 906  */
 907 static int
 908 tcp_usr_rcvd(struct socket *so, int flags)
 909 {
 910         struct epoch_tracker et;
 911         struct inpcb *inp;
 912         struct tcpcb *tp = NULL;
 913         int error = 0;
 914
 915         TCPDEBUG0;
 916         inp = sotoinpcb(so);
 917         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 918         INP_WLOCK(inp);
 919         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 920                 error = ECONNRESET;
 921                 goto out;
 922         }
 923         tp = intotcpcb(inp);
 924         TCPDEBUG1();
 925         /*
 926          * For passively-created TFO connections, don't attempt a window
 927          * update while still in SYN_RECEIVED as this may trigger an early
 928          * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
 929          * application response data, or failing that, when the DELACK timer
 930          * expires.
 931          */
 932         if (IS_FASTOPEN(tp->t_flags) &&
 933             (tp->t_state == TCPS_SYN_RECEIVED))
 934                 goto out;
 935         NET_EPOCH_ENTER(et);
 936 #ifdef TCP_OFFLOAD
 937         if (tp->t_flags & TF_TOE)
 938                 tcp_offload_rcvd(tp);
 939         else
 940 #endif
 941         tp->t_fb->tfb_tcp_output(tp);
 942         NET_EPOCH_EXIT(et);
 943 out:
 944         TCPDEBUG2(PRU_RCVD);
 945         TCP_PROBE2(debug__user, tp, PRU_RCVD);
 946         INP_WUNLOCK(inp);
 947         return (error);
 948 }
 949
 950 /*
 951  * Do a send by putting data in output queue and updating urgent
 952  * marker if URG set.  Possibly send more data.  Unlike the other
 953  * pru_*() routines, the mbuf chains are our responsibility.  We
 954  * must either enqueue them or free them.  The other pru_* routines
 955  * generally are caller-frees.
 956  */
 957 static int
 958 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 959     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 960 {
 961         struct epoch_tracker et;
 962         int error = 0;
 963         struct inpcb *inp;
 964         struct tcpcb *tp = NULL;
 965 #ifdef INET
 966 #ifdef INET6
 967         struct sockaddr_in sin;
 968 #endif
 969         struct sockaddr_in *sinp;
 970 #endif
 971 #ifdef INET6
 972         int isipv6;
 973 #endif
 974         u_int8_t incflagsav;
 975         u_char vflagsav;
 976         bool restoreflags;
 977         TCPDEBUG0;
 978
 979         /*
 980          * We require the pcbinfo "read lock" if we will close the socket
 981          * as part of this call.
 982          */
 983         NET_EPOCH_ENTER(et);
 984         inp = sotoinpcb(so);
 985         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 986         INP_WLOCK(inp);
 987         vflagsav = inp->inp_vflag;
 988         incflagsav = inp->inp_inc.inc_flags;
 989         restoreflags = false;
 990         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 991                 if (control)
 992                         m_freem(control);
 993                 /*
 994                  * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
 995                  * for freeing memory.
 996                  */
 997                 if (m && (flags & PRUS_NOTREADY) == 0)
 998                         m_freem(m);
 999                 error = ECONNRESET;
1000                 goto out;
1001         }
1002         tp = intotcpcb(inp);
1003         if (flags & PRUS_OOB) {
1004                 if ((error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) {
1005                         if (control)
1006                                 m_freem(control);
1007                         if (m && (flags & PRUS_NOTREADY) == 0)
1008                                 m_freem(m);
1009                         goto out;
1010                 }
1011         }
1012         TCPDEBUG1();
1013         if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1014                 switch (nam->sa_family) {
1015 #ifdef INET
1016                 case AF_INET:
1017                         sinp = (struct sockaddr_in *)nam;
1018                         if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1019                                 if (m)
1020                                         m_freem(m);
1021                                 error = EINVAL;
1022                                 goto out;
1023                         }
1024                         if ((inp->inp_vflag & INP_IPV6) != 0) {
1025                                 if (m)
1026                                         m_freem(m);
1027                                 error = EAFNOSUPPORT;
1028                                 goto out;
1029                         }
1030                         if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1031                                 if (m)
1032                                         m_freem(m);
1033                                 error = EAFNOSUPPORT;
1034                                 goto out;
1035                         }
1036                         if ((ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) ||
1037                             (sinp->sin_addr.s_addr == INADDR_ANY)) {
1038                                 if (m)
1039                                         m_freem(m);
1040                                 error = EAFNOSUPPORT;
1041                                 goto out;
1042                         }
1043                         if ((error = prison_remote_ip4(td->td_ucred,
1044                             &sinp->sin_addr))) {
1045                                 if (m)
1046                                         m_freem(m);
1047                                 goto out;
1048                         }
1049 #ifdef INET6
1050                         isipv6 = 0;
1051 #endif
1052                         break;
1053 #endif /* INET */
1054 #ifdef INET6
1055                 case AF_INET6:
1056                 {
1057                         struct sockaddr_in6 *sin6;
1058
1059                         sin6 = (struct sockaddr_in6 *)nam;
1060                         if (sin6->sin6_len != sizeof(*sin6)) {
1061                                 if (m)
1062                                         m_freem(m);
1063                                 error = EINVAL;
1064                                 goto out;
1065                         }
1066                         if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1067                                 if (m != NULL)
1068                                         m_freem(m);
1069                                 error = EAFNOSUPPORT;
1070                                 goto out;
1071                         }
1072                         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1073                                 if (m)
1074                                         m_freem(m);
1075                                 error = EAFNOSUPPORT;
1076                                 goto out;
1077                         }
1078                         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1079 #ifdef INET
1080                                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1081                                         error = EINVAL;
1082                                         if (m)
1083                                                 m_freem(m);
1084                                         goto out;
1085                                 }
1086                                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1087                                         error = EAFNOSUPPORT;
1088                                         if (m)
1089                                                 m_freem(m);
1090                                         goto out;
1091                                 }
1092                                 restoreflags = true;
1093                                 inp->inp_vflag &= ~INP_IPV6;
1094                                 sinp = &sin;
1095                                 in6_sin6_2_sin(sinp, sin6);
1096                                 if (IN_MULTICAST(
1097                                     ntohl(sinp->sin_addr.s_addr))) {
1098                                         error = EAFNOSUPPORT;
1099                                         if (m)
1100                                                 m_freem(m);
1101                                         goto out;
1102                                 }
1103                                 if ((error = prison_remote_ip4(td->td_ucred,
1104                                     &sinp->sin_addr))) {
1105                                         if (m)
1106                                                 m_freem(m);
1107                                         goto out;
1108                                 }
1109                                 isipv6 = 0;
1110 #else /* !INET */
1111                                 error = EAFNOSUPPORT;
1112                                 if (m)
1113                                         m_freem(m);
1114                                 goto out;
1115 #endif /* INET */
1116                         } else {
1117                                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1118                                         if (m)
1119                                                 m_freem(m);
1120                                         error = EAFNOSUPPORT;
1121                                         goto out;
1122                                 }
1123                                 restoreflags = true;
1124                                 inp->inp_vflag &= ~INP_IPV4;
1125                                 inp->inp_inc.inc_flags |= INC_ISIPV6;
1126                                 if ((error = prison_remote_ip6(td->td_ucred,
1127                                     &sin6->sin6_addr))) {
1128                                         if (m)
1129                                                 m_freem(m);
1130                                         goto out;
1131                                 }
1132                                 isipv6 = 1;
1133                         }
1134                         break;
1135                 }
1136 #endif /* INET6 */
1137                 default:
1138                         if (m)
1139                                 m_freem(m);
1140                         error = EAFNOSUPPORT;
1141                         goto out;
1142                 }
1143         }
1144         if (control) {
1145                 /* TCP doesn't do control messages (rights, creds, etc) */
1146                 if (control->m_len) {
1147                         m_freem(control);
1148                         if (m)
1149                                 m_freem(m);
1150                         error = EINVAL;
1151                         goto out;
1152                 }
1153                 m_freem(control);       /* empty control, just free it */
1154         }
1155         if (!(flags & PRUS_OOB)) {
1156                 sbappendstream(&so->so_snd, m, flags);
1157                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1158                         /*
1159                          * Do implied connect if not yet connected,
1160                          * initialize window to default value, and
1161                          * initialize maxseg using peer's cached MSS.
1162                          */
1163 #ifdef INET6
1164                         if (isipv6)
1165                                 error = tcp6_connect(tp, nam, td);
1166 #endif /* INET6 */
1167 #if defined(INET6) && defined(INET)
1168                         else
1169 #endif
1170 #ifdef INET
1171                                 error = tcp_connect(tp,
1172                                     (struct sockaddr *)sinp, td);
1173 #endif
1174                         /*
1175                          * The bind operation in tcp_connect succeeded. We
1176                          * no longer want to restore the flags if later
1177                          * operations fail.
1178                          */
1179                         if (error == 0 || inp->inp_lport != 0)
1180                                 restoreflags = false;
1181
1182                         if (error)
1183                                 goto out;
1184                         if (IS_FASTOPEN(tp->t_flags))
1185                                 tcp_fastopen_connect(tp);
1186                         else {
1187                                 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1188                                 tcp_mss(tp, -1);
1189                         }
1190                 }
1191                 if (flags & PRUS_EOF) {
1192                         /*
1193                          * Close the send side of the connection after
1194                          * the data is sent.
1195                          */
1196                         socantsendmore(so);
1197                         tcp_usrclosed(tp);
1198                 }
1199                 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1200                     ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1201                     (tp->t_fbyte_out == 0) &&
1202                     (so->so_snd.sb_ccc > 0)) {
1203                         tp->t_fbyte_out = ticks;
1204                         if (tp->t_fbyte_out == 0)
1205                                 tp->t_fbyte_out = 1;
1206                         if (tp->t_fbyte_out && tp->t_fbyte_in)
1207                                 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1208                 }
1209                 if (!(inp->inp_flags & INP_DROPPED) &&
1210                     !(flags & PRUS_NOTREADY)) {
1211                         if (flags & PRUS_MORETOCOME)
1212                                 tp->t_flags |= TF_MORETOCOME;
1213                         error = tp->t_fb->tfb_tcp_output(tp);
1214                         if (flags & PRUS_MORETOCOME)
1215                                 tp->t_flags &= ~TF_MORETOCOME;
1216                 }
1217         } else {
1218                 /*
1219                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1220                  */
1221                 SOCKBUF_LOCK(&so->so_snd);
1222                 if (sbspace(&so->so_snd) < -512) {
1223                         SOCKBUF_UNLOCK(&so->so_snd);
1224                         m_freem(m);
1225                         error = ENOBUFS;
1226                         goto out;
1227                 }
1228                 /*
1229                  * According to RFC961 (Assigned Protocols),
1230                  * the urgent pointer points to the last octet
1231                  * of urgent data.  We continue, however,
1232                  * to consider it to indicate the first octet
1233                  * of data past the urgent section.
1234                  * Otherwise, snd_up should be one lower.
1235                  */
1236                 sbappendstream_locked(&so->so_snd, m, flags);
1237                 SOCKBUF_UNLOCK(&so->so_snd);
1238                 if (nam && tp->t_state < TCPS_SYN_SENT) {
1239                         /*
1240                          * Do implied connect if not yet connected,
1241                          * initialize window to default value, and
1242                          * initialize maxseg using peer's cached MSS.
1243                          */
1244
1245                         /*
1246                          * Not going to contemplate SYN|URG
1247                          */
1248                         if (IS_FASTOPEN(tp->t_flags))
1249                                 tp->t_flags &= ~TF_FASTOPEN;
1250 #ifdef INET6
1251                         if (isipv6)
1252                                 error = tcp6_connect(tp, nam, td);
1253 #endif /* INET6 */
1254 #if defined(INET6) && defined(INET)
1255                         else
1256 #endif
1257 #ifdef INET
1258                                 error = tcp_connect(tp,
1259                                     (struct sockaddr *)sinp, td);
1260 #endif
1261                         /*
1262                          * The bind operation in tcp_connect succeeded. We
1263                          * no longer want to restore the flags if later
1264                          * operations fail.
1265                          */
1266                         if (error == 0 || inp->inp_lport != 0)
1267                                 restoreflags = false;
1268
1269                         if (error)
1270                                 goto out;
1271                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
1272                         tcp_mss(tp, -1);
1273                 }
1274                 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1275                 if (!(flags & PRUS_NOTREADY)) {
1276                         tp->t_flags |= TF_FORCEDATA;
1277                         error = tp->t_fb->tfb_tcp_output(tp);
1278                         tp->t_flags &= ~TF_FORCEDATA;
1279                 }
1280         }
1281         TCP_LOG_EVENT(tp, NULL,
1282             &inp->inp_socket->so_rcv,
1283             &inp->inp_socket->so_snd,
1284             TCP_LOG_USERSEND, error,
1285             0, NULL, false);
1286 out:
1287         /*
1288          * If the request was unsuccessful and we changed flags,
1289          * restore the original flags.
1290          */
1291         if (error != 0 && restoreflags) {
1292                 inp->inp_vflag = vflagsav;
1293                 inp->inp_inc.inc_flags = incflagsav;
1294         }
1295         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1296                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1297         TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1298                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1299         INP_WUNLOCK(inp);
1300         NET_EPOCH_EXIT(et);
1301         return (error);
1302 }
1303
1304 static int
1305 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1306 {
1307         struct epoch_tracker et;
1308         struct inpcb *inp;
1309         struct tcpcb *tp;
1310         int error;
1311
1312         inp = sotoinpcb(so);
1313         INP_WLOCK(inp);
1314         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1315                 INP_WUNLOCK(inp);
1316                 mb_free_notready(m, count);
1317                 return (ECONNRESET);
1318         }
1319         tp = intotcpcb(inp);
1320
1321         SOCKBUF_LOCK(&so->so_snd);
1322         error = sbready(&so->so_snd, m, count);
1323         SOCKBUF_UNLOCK(&so->so_snd);
1324         if (error == 0) {
1325                 NET_EPOCH_ENTER(et);
1326                 error = tp->t_fb->tfb_tcp_output(tp);
1327                 NET_EPOCH_EXIT(et);
1328         }
1329         INP_WUNLOCK(inp);
1330
1331         return (error);
1332 }
1333
1334 /*
1335  * Abort the TCP.  Drop the connection abruptly.
1336  */
1337 static void
1338 tcp_usr_abort(struct socket *so)
1339 {
1340         struct inpcb *inp;
1341         struct tcpcb *tp = NULL;
1342         struct epoch_tracker et;
1343         TCPDEBUG0;
1344
1345         inp = sotoinpcb(so);
1346         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1347
1348         NET_EPOCH_ENTER(et);
1349         INP_WLOCK(inp);
1350         KASSERT(inp->inp_socket != NULL,
1351             ("tcp_usr_abort: inp_socket == NULL"));
1352
1353         /*
1354          * If we still have full TCP state, and we're not dropped, drop.
1355          */
1356         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1357             !(inp->inp_flags & INP_DROPPED)) {
1358                 tp = intotcpcb(inp);
1359                 TCPDEBUG1();
1360                 tp = tcp_drop(tp, ECONNABORTED);
1361                 if (tp == NULL)
1362                         goto dropped;
1363                 TCPDEBUG2(PRU_ABORT);
1364                 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1365         }
1366         if (!(inp->inp_flags & INP_DROPPED)) {
1367                 SOCK_LOCK(so);
1368                 so->so_state |= SS_PROTOREF;
1369                 SOCK_UNLOCK(so);
1370                 inp->inp_flags |= INP_SOCKREF;
1371         }
1372         INP_WUNLOCK(inp);
1373 dropped:
1374         NET_EPOCH_EXIT(et);
1375 }
1376
1377 /*
1378  * TCP socket is closed.  Start friendly disconnect.
1379  */
1380 static void
1381 tcp_usr_close(struct socket *so)
1382 {
1383         struct inpcb *inp;
1384         struct tcpcb *tp = NULL;
1385         struct epoch_tracker et;
1386         TCPDEBUG0;
1387
1388         inp = sotoinpcb(so);
1389         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1390
1391         NET_EPOCH_ENTER(et);
1392         INP_WLOCK(inp);
1393         KASSERT(inp->inp_socket != NULL,
1394             ("tcp_usr_close: inp_socket == NULL"));
1395
1396         /*
1397          * If we still have full TCP state, and we're not dropped, initiate
1398          * a disconnect.
1399          */
1400         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1401             !(inp->inp_flags & INP_DROPPED)) {
1402                 tp = intotcpcb(inp);
1403                 TCPDEBUG1();
1404                 tcp_disconnect(tp);
1405                 TCPDEBUG2(PRU_CLOSE);
1406                 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1407         }
1408         if (!(inp->inp_flags & INP_DROPPED)) {
1409                 SOCK_LOCK(so);
1410                 so->so_state |= SS_PROTOREF;
1411                 SOCK_UNLOCK(so);
1412                 inp->inp_flags |= INP_SOCKREF;
1413         }
1414         INP_WUNLOCK(inp);
1415         NET_EPOCH_EXIT(et);
1416 }
1417
1418 static int
1419 tcp_pru_options_support(struct tcpcb *tp, int flags)
1420 {
1421         /*
1422          * If the specific TCP stack has a pru_options
1423          * specified then it does not always support
1424          * all the PRU_XX options and we must ask it.
1425          * If the function is not specified then all
1426          * of the PRU_XX options are supported.
1427          */
1428         int ret = 0;
1429
1430         if (tp->t_fb->tfb_pru_options) {
1431                 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1432         }
1433         return (ret);
1434 }
1435
1436 /*
1437  * Receive out-of-band data.
1438  */
1439 static int
1440 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1441 {
1442         int error = 0;
1443         struct inpcb *inp;
1444         struct tcpcb *tp = NULL;
1445
1446         TCPDEBUG0;
1447         inp = sotoinpcb(so);
1448         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1449         INP_WLOCK(inp);
1450         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1451                 error = ECONNRESET;
1452                 goto out;
1453         }
1454         tp = intotcpcb(inp);
1455         error = tcp_pru_options_support(tp, PRUS_OOB);
1456         if (error) {
1457                 goto out;
1458         }
1459         TCPDEBUG1();
1460         if ((so->so_oobmark == 0 &&
1461              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1462             so->so_options & SO_OOBINLINE ||
1463             tp->t_oobflags & TCPOOB_HADDATA) {
1464                 error = EINVAL;
1465                 goto out;
1466         }
1467         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1468                 error = EWOULDBLOCK;
1469                 goto out;
1470         }
1471         m->m_len = 1;
1472         *mtod(m, caddr_t) = tp->t_iobc;
1473         if ((flags & MSG_PEEK) == 0)
1474                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1475
1476 out:
1477         TCPDEBUG2(PRU_RCVOOB);
1478         TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1479         INP_WUNLOCK(inp);
1480         return (error);
1481 }
1482
1483 #ifdef INET
1484 struct pr_usrreqs tcp_usrreqs = {
1485         .pru_abort =            tcp_usr_abort,
1486         .pru_accept =           tcp_usr_accept,
1487         .pru_attach =           tcp_usr_attach,
1488         .pru_bind =             tcp_usr_bind,
1489         .pru_connect =          tcp_usr_connect,
1490         .pru_control =          in_control,
1491         .pru_detach =           tcp_usr_detach,
1492         .pru_disconnect =       tcp_usr_disconnect,
1493         .pru_listen =           tcp_usr_listen,
1494         .pru_peeraddr =         in_getpeeraddr,
1495         .pru_rcvd =             tcp_usr_rcvd,
1496         .pru_rcvoob =           tcp_usr_rcvoob,
1497         .pru_send =             tcp_usr_send,
1498         .pru_ready =            tcp_usr_ready,
1499         .pru_shutdown =         tcp_usr_shutdown,
1500         .pru_sockaddr =         in_getsockaddr,
1501         .pru_sosetlabel =       in_pcbsosetlabel,
1502         .pru_close =            tcp_usr_close,
1503 };
1504 #endif /* INET */
1505
1506 #ifdef INET6
1507 struct pr_usrreqs tcp6_usrreqs = {
1508         .pru_abort =            tcp_usr_abort,
1509         .pru_accept =           tcp6_usr_accept,
1510         .pru_attach =           tcp_usr_attach,
1511         .pru_bind =             tcp6_usr_bind,
1512         .pru_connect =          tcp6_usr_connect,
1513         .pru_control =          in6_control,
1514         .pru_detach =           tcp_usr_detach,
1515         .pru_disconnect =       tcp_usr_disconnect,
1516         .pru_listen =           tcp6_usr_listen,
1517         .pru_peeraddr =         in6_mapped_peeraddr,
1518         .pru_rcvd =             tcp_usr_rcvd,
1519         .pru_rcvoob =           tcp_usr_rcvoob,
1520         .pru_send =             tcp_usr_send,
1521         .pru_ready =            tcp_usr_ready,
1522         .pru_shutdown =         tcp_usr_shutdown,
1523         .pru_sockaddr =         in6_mapped_sockaddr,
1524         .pru_sosetlabel =       in_pcbsosetlabel,
1525         .pru_close =            tcp_usr_close,
1526 };
1527 #endif /* INET6 */
1528
1529 #ifdef INET
1530 /*
1531  * Common subroutine to open a TCP connection to remote host specified
1532  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1533  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1534  * to choose a local host address (interface).  If there is an existing
1535  * incarnation of the same connection in TIME-WAIT state and if the remote
1536  * host was sending CC options and if the connection duration was < MSL, then
1537  * truncate the previous TIME-WAIT state and proceed.
1538  * Initialize connection parameters and enter SYN-SENT state.
1539  */
1540 static int
1541 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1542 {
1543         struct inpcb *inp = tp->t_inpcb, *oinp;
1544         struct socket *so = inp->inp_socket;
1545         struct in_addr laddr;
1546         u_short lport;
1547         int error;
1548
1549         NET_EPOCH_ASSERT();
1550         INP_WLOCK_ASSERT(inp);
1551         INP_HASH_WLOCK(&V_tcbinfo);
1552
1553         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1554                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1555                 if (error)
1556                         goto out;
1557         }
1558
1559         /*
1560          * Cannot simply call in_pcbconnect, because there might be an
1561          * earlier incarnation of this same connection still in
1562          * TIME_WAIT state, creating an ADDRINUSE error.
1563          */
1564         laddr = inp->inp_laddr;
1565         lport = inp->inp_lport;
1566         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1567             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1568         if (error && oinp == NULL)
1569                 goto out;
1570         if (oinp) {
1571                 error = EADDRINUSE;
1572                 goto out;
1573         }
1574         /* Handle initial bind if it hadn't been done in advance. */
1575         if (inp->inp_lport == 0) {
1576                 inp->inp_lport = lport;
1577                 if (in_pcbinshash(inp) != 0) {
1578                         inp->inp_lport = 0;
1579                         error = EAGAIN;
1580                         goto out;
1581                 }
1582         }
1583         inp->inp_laddr = laddr;
1584         in_pcbrehash(inp);
1585         INP_HASH_WUNLOCK(&V_tcbinfo);
1586
1587         /*
1588          * Compute window scaling to request:
1589          * Scale to fit into sweet spot.  See tcp_syncache.c.
1590          * XXX: This should move to tcp_output().
1591          */
1592         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1593             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1594                 tp->request_r_scale++;
1595
1596         soisconnecting(so);
1597         TCPSTAT_INC(tcps_connattempt);
1598         tcp_state_change(tp, TCPS_SYN_SENT);
1599         tp->iss = tcp_new_isn(&inp->inp_inc);
1600         if (tp->t_flags & TF_REQ_TSTMP)
1601                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1602         tcp_sendseqinit(tp);
1603
1604         return 0;
1605
1606 out:
1607         INP_HASH_WUNLOCK(&V_tcbinfo);
1608         return (error);
1609 }
1610 #endif /* INET */
1611
1612 #ifdef INET6
1613 static int
1614 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1615 {
1616         struct inpcb *inp = tp->t_inpcb;
1617         int error;
1618
1619         INP_WLOCK_ASSERT(inp);
1620         INP_HASH_WLOCK(&V_tcbinfo);
1621
1622         if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1623                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1624                 if (error)
1625                         goto out;
1626         }
1627         error = in6_pcbconnect(inp, nam, td->td_ucred);
1628         if (error != 0)
1629                 goto out;
1630         INP_HASH_WUNLOCK(&V_tcbinfo);
1631
1632         /* Compute window scaling to request.  */
1633         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1634             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1635                 tp->request_r_scale++;
1636
1637         soisconnecting(inp->inp_socket);
1638         TCPSTAT_INC(tcps_connattempt);
1639         tcp_state_change(tp, TCPS_SYN_SENT);
1640         tp->iss = tcp_new_isn(&inp->inp_inc);
1641         if (tp->t_flags & TF_REQ_TSTMP)
1642                 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1643         tcp_sendseqinit(tp);
1644
1645         return 0;
1646
1647 out:
1648         INP_HASH_WUNLOCK(&V_tcbinfo);
1649         return error;
1650 }
1651 #endif /* INET6 */
1652
1653 /*
1654  * Export TCP internal state information via a struct tcp_info, based on the
1655  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1656  * (TCP state machine, etc).  We export all information using FreeBSD-native
1657  * constants -- for example, the numeric values for tcpi_state will differ
1658  * from Linux.
1659  */
1660 static void
1661 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1662 {
1663
1664         INP_WLOCK_ASSERT(tp->t_inpcb);
1665         bzero(ti, sizeof(*ti));
1666
1667         ti->tcpi_state = tp->t_state;
1668         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1669                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1670         if (tp->t_flags & TF_SACK_PERMIT)
1671                 ti->tcpi_options |= TCPI_OPT_SACK;
1672         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1673                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1674                 ti->tcpi_snd_wscale = tp->snd_scale;
1675                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1676         }
1677         if (tp->t_flags2 & TF2_ECN_PERMIT)
1678                 ti->tcpi_options |= TCPI_OPT_ECN;
1679
1680         ti->tcpi_rto = tp->t_rxtcur * tick;
1681         ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1682         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1683         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1684
1685         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1686         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1687
1688         /*
1689          * FreeBSD-specific extension fields for tcp_info.
1690          */
1691         ti->tcpi_rcv_space = tp->rcv_wnd;
1692         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1693         ti->tcpi_snd_wnd = tp->snd_wnd;
1694         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1695         ti->tcpi_snd_nxt = tp->snd_nxt;
1696         ti->tcpi_snd_mss = tp->t_maxseg;
1697         ti->tcpi_rcv_mss = tp->t_maxseg;
1698         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1699         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1700         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1701 #ifdef TCP_OFFLOAD
1702         if (tp->t_flags & TF_TOE) {
1703                 ti->tcpi_options |= TCPI_OPT_TOE;
1704                 tcp_offload_tcp_info(tp, ti);
1705         }
1706 #endif
1707 }
1708
1709 /*
1710  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1711  * socket option arguments.  When it re-acquires the lock after the copy, it
1712  * has to revalidate that the connection is still valid for the socket
1713  * option.
1714  */
1715 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {                    \
1716         INP_WLOCK(inp);                                                 \
1717         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1718                 INP_WUNLOCK(inp);                                       \
1719                 cleanup;                                                \
1720                 return (ECONNRESET);                                    \
1721         }                                                               \
1722         tp = intotcpcb(inp);                                            \
1723 } while(0)
1724 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1725
1726 int
1727 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1728 {
1729         int     error;
1730         struct  inpcb *inp;
1731         struct  tcpcb *tp;
1732         struct tcp_function_block *blk;
1733         struct tcp_function_set fsn;
1734
1735         error = 0;
1736         inp = sotoinpcb(so);
1737         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1738         if (sopt->sopt_level != IPPROTO_TCP) {
1739 #ifdef INET6
1740                 if (inp->inp_vflag & INP_IPV6PROTO) {
1741                         error = ip6_ctloutput(so, sopt);
1742                         /*
1743                          * In case of the IPV6_USE_MIN_MTU socket option,
1744                          * the INC_IPV6MINMTU flag to announce a corresponding
1745                          * MSS during the initial handshake.
1746                          * If the TCP connection is not in the front states,
1747                          * just reduce the MSS being used.
1748                          * This avoids the sending of TCP segments which will
1749                          * be fragmented at the IPv6 layer.
1750                          */
1751                         if ((error == 0) &&
1752                             (sopt->sopt_dir == SOPT_SET) &&
1753                             (sopt->sopt_level == IPPROTO_IPV6) &&
1754                             (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1755                                 INP_WLOCK(inp);
1756                                 if ((inp->inp_flags &
1757                                     (INP_TIMEWAIT | INP_DROPPED))) {
1758                                         INP_WUNLOCK(inp);
1759                                         return (ECONNRESET);
1760                                 }
1761                                 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1762                                 tp = intotcpcb(inp);
1763                                 if ((tp->t_state >= TCPS_SYN_SENT) &&
1764                                     (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1765                                         struct ip6_pktopts *opt;
1766
1767                                         opt = inp->in6p_outputopts;
1768                                         if ((opt != NULL) &&
1769                                             (opt->ip6po_minmtu ==
1770                                             IP6PO_MINMTU_ALL)) {
1771                                                 if (tp->t_maxseg > TCP6_MSS) {
1772                                                         tp->t_maxseg = TCP6_MSS;
1773                                                 }
1774                                         }
1775                                 }
1776                                 INP_WUNLOCK(inp);
1777                         }
1778                 }
1779 #endif /* INET6 */
1780 #if defined(INET6) && defined(INET)
1781                 else
1782 #endif
1783 #ifdef INET
1784                 {
1785                         error = ip_ctloutput(so, sopt);
1786                 }
1787 #endif
1788                 return (error);
1789         }
1790         INP_WLOCK(inp);
1791         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1792                 INP_WUNLOCK(inp);
1793                 return (ECONNRESET);
1794         }
1795         tp = intotcpcb(inp);
1796         /*
1797          * Protect the TCP option TCP_FUNCTION_BLK so
1798          * that a sub-function can *never* overwrite this.
1799          */
1800         if ((sopt->sopt_dir == SOPT_SET) &&
1801             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1802                 INP_WUNLOCK(inp);
1803                 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1804                     sizeof fsn);
1805                 if (error)
1806                         return (error);
1807                 INP_WLOCK_RECHECK(inp);
1808                 blk = find_and_ref_tcp_functions(&fsn);
1809                 if (blk == NULL) {
1810                         INP_WUNLOCK(inp);
1811                         return (ENOENT);
1812                 }
1813                 if (tp->t_fb == blk) {
1814                         /* You already have this */
1815                         refcount_release(&blk->tfb_refcnt);
1816                         INP_WUNLOCK(inp);
1817                         return (0);
1818                 }
1819                 if (tp->t_state != TCPS_CLOSED) {
1820                         /*
1821                          * The user has advanced the state
1822                          * past the initial point, we may not
1823                          * be able to switch.
1824                          */
1825                         if (blk->tfb_tcp_handoff_ok != NULL) {
1826                                 /*
1827                                  * Does the stack provide a
1828                                  * query mechanism, if so it may
1829                                  * still be possible?
1830                                  */
1831                                 error = (*blk->tfb_tcp_handoff_ok)(tp);
1832                         } else
1833                                 error = EINVAL;
1834                         if (error) {
1835                                 refcount_release(&blk->tfb_refcnt);
1836                                 INP_WUNLOCK(inp);
1837                                 return(error);
1838                         }
1839                 }
1840                 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1841                         refcount_release(&blk->tfb_refcnt);
1842                         INP_WUNLOCK(inp);
1843                         return (ENOENT);
1844                 }
1845                 /*
1846                  * Release the old refcnt, the
1847                  * lookup acquired a ref on the
1848                  * new one already.
1849                  */
1850                 if (tp->t_fb->tfb_tcp_fb_fini) {
1851                         /*
1852                          * Tell the stack to cleanup with 0 i.e.
1853                          * the tcb is not going away.
1854                          */
1855                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1856                 }
1857 #ifdef TCPHPTS
1858                 /* Assure that we are not on any hpts */
1859                 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1860 #endif
1861                 if (blk->tfb_tcp_fb_init) {
1862                         error = (*blk->tfb_tcp_fb_init)(tp);
1863                         if (error) {
1864                                 refcount_release(&blk->tfb_refcnt);
1865                                 if (tp->t_fb->tfb_tcp_fb_init) {
1866                                         if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1867                                                 /* Fall back failed, drop the connection */
1868                                                 INP_WUNLOCK(inp);
1869                                                 soabort(so);
1870                                                 return(error);
1871                                         }
1872                                 }
1873                                 goto err_out;
1874                         }
1875                 }
1876                 refcount_release(&tp->t_fb->tfb_refcnt);
1877                 tp->t_fb = blk;
1878 #ifdef TCP_OFFLOAD
1879                 if (tp->t_flags & TF_TOE) {
1880                         tcp_offload_ctloutput(tp, sopt->sopt_dir,
1881                              sopt->sopt_name);
1882                 }
1883 #endif
1884 err_out:
1885                 INP_WUNLOCK(inp);
1886                 return (error);
1887         } else if ((sopt->sopt_dir == SOPT_GET) &&
1888             (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1889                 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1890                     TCP_FUNCTION_NAME_LEN_MAX);
1891                 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1892                 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1893                 INP_WUNLOCK(inp);
1894                 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1895                 return (error);
1896         }
1897         /* Pass in the INP locked, called must unlock it */
1898         return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1899 }
1900
1901 /*
1902  * If this assert becomes untrue, we need to change the size of the buf
1903  * variable in tcp_default_ctloutput().
1904  */
1905 #ifdef CTASSERT
1906 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1907 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1908 #endif
1909
1910 #ifdef KERN_TLS
1911 static int
1912 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1913 {
1914         struct tls_enable_v0 tls_v0;
1915         int error;
1916
1917         if (sopt->sopt_valsize == sizeof(tls_v0)) {
1918                 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1919                     sizeof(tls_v0));
1920                 if (error)
1921                         return (error);
1922                 memset(tls, 0, sizeof(*tls));
1923                 tls->cipher_key = tls_v0.cipher_key;
1924                 tls->iv = tls_v0.iv;
1925                 tls->auth_key = tls_v0.auth_key;
1926                 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1927                 tls->cipher_key_len = tls_v0.cipher_key_len;
1928                 tls->iv_len = tls_v0.iv_len;
1929                 tls->auth_algorithm = tls_v0.auth_algorithm;
1930                 tls->auth_key_len = tls_v0.auth_key_len;
1931                 tls->flags = tls_v0.flags;
1932                 tls->tls_vmajor = tls_v0.tls_vmajor;
1933                 tls->tls_vminor = tls_v0.tls_vminor;
1934                 return (0);
1935         }
1936
1937         return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
1938 }
1939 #endif
1940
1941 int
1942 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1943 {
1944         int     error, opt, optval;
1945         u_int   ui;
1946         struct  tcp_info ti;
1947 #ifdef KERN_TLS
1948         struct tls_enable tls;
1949 #endif
1950         struct cc_algo *algo;
1951         char    *pbuf, buf[TCP_LOG_ID_LEN];
1952 #ifdef STATS
1953         struct statsblob *sbp;
1954 #endif
1955         size_t  len;
1956
1957         /*
1958          * For TCP_CCALGOOPT forward the control to CC module, for both
1959          * SOPT_SET and SOPT_GET.
1960          */
1961         switch (sopt->sopt_name) {
1962         case TCP_CCALGOOPT:
1963                 INP_WUNLOCK(inp);
1964                 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1965                         return (EINVAL);
1966                 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1967                 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1968                     sopt->sopt_valsize);
1969                 if (error) {
1970                         free(pbuf, M_TEMP);
1971                         return (error);
1972                 }
1973                 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1974                 if (CC_ALGO(tp)->ctl_output != NULL)
1975                         error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1976                 else
1977                         error = ENOENT;
1978                 INP_WUNLOCK(inp);
1979                 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1980                         error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1981                 free(pbuf, M_TEMP);
1982                 return (error);
1983         }
1984
1985         switch (sopt->sopt_dir) {
1986         case SOPT_SET:
1987                 switch (sopt->sopt_name) {
1988 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1989                 case TCP_MD5SIG:
1990                         if (!TCPMD5_ENABLED()) {
1991                                 INP_WUNLOCK(inp);
1992                                 return (ENOPROTOOPT);
1993                         }
1994                         error = TCPMD5_PCBCTL(inp, sopt);
1995                         if (error)
1996                                 return (error);
1997                         goto unlock_and_done;
1998 #endif /* IPSEC */
1999
2000                 case TCP_NODELAY:
2001                 case TCP_NOOPT:
2002                         INP_WUNLOCK(inp);
2003                         error = sooptcopyin(sopt, &optval, sizeof optval,
2004                             sizeof optval);
2005                         if (error)
2006                                 return (error);
2007
2008                         INP_WLOCK_RECHECK(inp);
2009                         switch (sopt->sopt_name) {
2010                         case TCP_NODELAY:
2011                                 opt = TF_NODELAY;
2012                                 break;
2013                         case TCP_NOOPT:
2014                                 opt = TF_NOOPT;
2015                                 break;
2016                         default:
2017                                 opt = 0; /* dead code to fool gcc */
2018                                 break;
2019                         }
2020
2021                         if (optval)
2022                                 tp->t_flags |= opt;
2023                         else
2024                                 tp->t_flags &= ~opt;
2025 unlock_and_done:
2026 #ifdef TCP_OFFLOAD
2027                         if (tp->t_flags & TF_TOE) {
2028                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2029                                     sopt->sopt_name);
2030                         }
2031 #endif
2032                         INP_WUNLOCK(inp);
2033                         break;
2034
2035                 case TCP_NOPUSH:
2036                         INP_WUNLOCK(inp);
2037                         error = sooptcopyin(sopt, &optval, sizeof optval,
2038                             sizeof optval);
2039                         if (error)
2040                                 return (error);
2041
2042                         INP_WLOCK_RECHECK(inp);
2043                         if (optval)
2044                                 tp->t_flags |= TF_NOPUSH;
2045                         else if (tp->t_flags & TF_NOPUSH) {
2046                                 tp->t_flags &= ~TF_NOPUSH;
2047                                 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2048                                         struct epoch_tracker et;
2049
2050                                         NET_EPOCH_ENTER(et);
2051                                         error = tp->t_fb->tfb_tcp_output(tp);
2052                                         NET_EPOCH_EXIT(et);
2053                                 }
2054                         }
2055                         goto unlock_and_done;
2056
2057                 case TCP_MAXSEG:
2058                         INP_WUNLOCK(inp);
2059                         error = sooptcopyin(sopt, &optval, sizeof optval,
2060                             sizeof optval);
2061                         if (error)
2062                                 return (error);
2063
2064                         INP_WLOCK_RECHECK(inp);
2065                         if (optval > 0 && optval <= tp->t_maxseg &&
2066                             optval + 40 >= V_tcp_minmss)
2067                                 tp->t_maxseg = optval;
2068                         else
2069                                 error = EINVAL;
2070                         goto unlock_and_done;
2071
2072                 case TCP_INFO:
2073                         INP_WUNLOCK(inp);
2074                         error = EINVAL;
2075                         break;
2076
2077                 case TCP_STATS:
2078                         INP_WUNLOCK(inp);
2079 #ifdef STATS
2080                         error = sooptcopyin(sopt, &optval, sizeof optval,
2081                             sizeof optval);
2082                         if (error)
2083                                 return (error);
2084
2085                         if (optval > 0)
2086                                 sbp = stats_blob_alloc(
2087                                     V_tcp_perconn_stats_dflt_tpl, 0);
2088                         else
2089                                 sbp = NULL;
2090
2091                         INP_WLOCK_RECHECK(inp);
2092                         if ((tp->t_stats != NULL && sbp == NULL) ||
2093                             (tp->t_stats == NULL && sbp != NULL)) {
2094                                 struct statsblob *t = tp->t_stats;
2095                                 tp->t_stats = sbp;
2096                                 sbp = t;
2097                         }
2098                         INP_WUNLOCK(inp);
2099
2100                         stats_blob_destroy(sbp);
2101 #else
2102                         return (EOPNOTSUPP);
2103 #endif /* !STATS */
2104                         break;
2105
2106                 case TCP_CONGESTION:
2107                         INP_WUNLOCK(inp);
2108                         error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2109                         if (error)
2110                                 break;
2111                         buf[sopt->sopt_valsize] = '\0';
2112                         INP_WLOCK_RECHECK(inp);
2113                         CC_LIST_RLOCK();
2114                         STAILQ_FOREACH(algo, &cc_list, entries)
2115                                 if (strncmp(buf, algo->name,
2116                                     TCP_CA_NAME_MAX) == 0)
2117                                         break;
2118                         CC_LIST_RUNLOCK();
2119                         if (algo == NULL) {
2120                                 INP_WUNLOCK(inp);
2121                                 error = EINVAL;
2122                                 break;
2123                         }
2124                         /*
2125                          * We hold a write lock over the tcb so it's safe to
2126                          * do these things without ordering concerns.
2127                          */
2128                         if (CC_ALGO(tp)->cb_destroy != NULL)
2129                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
2130                         CC_DATA(tp) = NULL;
2131                         CC_ALGO(tp) = algo;
2132                         /*
2133                          * If something goes pear shaped initialising the new
2134                          * algo, fall back to newreno (which does not
2135                          * require initialisation).
2136                          */
2137                         if (algo->cb_init != NULL &&
2138                             algo->cb_init(tp->ccv) != 0) {
2139                                 CC_ALGO(tp) = &newreno_cc_algo;
2140                                 /*
2141                                  * The only reason init should fail is
2142                                  * because of malloc.
2143                                  */
2144                                 error = ENOMEM;
2145                         }
2146                         INP_WUNLOCK(inp);
2147                         break;
2148
2149 #ifdef KERN_TLS
2150                 case TCP_TXTLS_ENABLE:
2151                         INP_WUNLOCK(inp);
2152                         error = copyin_tls_enable(sopt, &tls);
2153                         if (error)
2154                                 break;
2155                         error = ktls_enable_tx(so, &tls);
2156                         break;
2157                 case TCP_TXTLS_MODE:
2158                         INP_WUNLOCK(inp);
2159                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2160                         if (error)
2161                                 return (error);
2162
2163                         INP_WLOCK_RECHECK(inp);
2164                         error = ktls_set_tx_mode(so, ui);
2165                         INP_WUNLOCK(inp);
2166                         break;
2167                 case TCP_RXTLS_ENABLE:
2168                         INP_WUNLOCK(inp);
2169                         error = sooptcopyin(sopt, &tls, sizeof(tls),
2170                             sizeof(tls));
2171                         if (error)
2172                                 break;
2173                         error = ktls_enable_rx(so, &tls);
2174                         break;
2175 #endif
2176
2177                 case TCP_KEEPIDLE:
2178                 case TCP_KEEPINTVL:
2179                 case TCP_KEEPINIT:
2180                         INP_WUNLOCK(inp);
2181                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2182                         if (error)
2183                                 return (error);
2184
2185                         if (ui > (UINT_MAX / hz)) {
2186                                 error = EINVAL;
2187                                 break;
2188                         }
2189                         ui *= hz;
2190
2191                         INP_WLOCK_RECHECK(inp);
2192                         switch (sopt->sopt_name) {
2193                         case TCP_KEEPIDLE:
2194                                 tp->t_keepidle = ui;
2195                                 /*
2196                                  * XXX: better check current remaining
2197                                  * timeout and "merge" it with new value.
2198                                  */
2199                                 if ((tp->t_state > TCPS_LISTEN) &&
2200                                     (tp->t_state <= TCPS_CLOSING))
2201                                         tcp_timer_activate(tp, TT_KEEP,
2202                                             TP_KEEPIDLE(tp));
2203                                 break;
2204                         case TCP_KEEPINTVL:
2205                                 tp->t_keepintvl = ui;
2206                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2207                                     (TP_MAXIDLE(tp) > 0))
2208                                         tcp_timer_activate(tp, TT_2MSL,
2209                                             TP_MAXIDLE(tp));
2210                                 break;
2211                         case TCP_KEEPINIT:
2212                                 tp->t_keepinit = ui;
2213                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
2214                                     tp->t_state == TCPS_SYN_SENT)
2215                                         tcp_timer_activate(tp, TT_KEEP,
2216                                             TP_KEEPINIT(tp));
2217                                 break;
2218                         }
2219                         goto unlock_and_done;
2220
2221                 case TCP_KEEPCNT:
2222                         INP_WUNLOCK(inp);
2223                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2224                         if (error)
2225                                 return (error);
2226
2227                         INP_WLOCK_RECHECK(inp);
2228                         tp->t_keepcnt = ui;
2229                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2230                             (TP_MAXIDLE(tp) > 0))
2231                                 tcp_timer_activate(tp, TT_2MSL,
2232                                     TP_MAXIDLE(tp));
2233                         goto unlock_and_done;
2234
2235 #ifdef TCPPCAP
2236                 case TCP_PCAP_OUT:
2237                 case TCP_PCAP_IN:
2238                         INP_WUNLOCK(inp);
2239                         error = sooptcopyin(sopt, &optval, sizeof optval,
2240                             sizeof optval);
2241                         if (error)
2242                                 return (error);
2243
2244                         INP_WLOCK_RECHECK(inp);
2245                         if (optval >= 0)
2246                                 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2247                                         &(tp->t_outpkts) : &(tp->t_inpkts),
2248                                         optval);
2249                         else
2250                                 error = EINVAL;
2251                         goto unlock_and_done;
2252 #endif
2253
2254                 case TCP_FASTOPEN: {
2255                         struct tcp_fastopen tfo_optval;
2256
2257                         INP_WUNLOCK(inp);
2258                         if (!V_tcp_fastopen_client_enable &&
2259                             !V_tcp_fastopen_server_enable)
2260                                 return (EPERM);
2261
2262                         error = sooptcopyin(sopt, &tfo_optval,
2263                                     sizeof(tfo_optval), sizeof(int));
2264                         if (error)
2265                                 return (error);
2266
2267                         INP_WLOCK_RECHECK(inp);
2268                         if ((tp->t_state != TCPS_CLOSED) &&
2269                             (tp->t_state != TCPS_LISTEN)) {
2270                                 error = EINVAL;
2271                                 goto unlock_and_done;
2272                         }
2273                         if (tfo_optval.enable) {
2274                                 if (tp->t_state == TCPS_LISTEN) {
2275                                         if (!V_tcp_fastopen_server_enable) {
2276                                                 error = EPERM;
2277                                                 goto unlock_and_done;
2278                                         }
2279
2280                                         if (tp->t_tfo_pending == NULL)
2281                                                 tp->t_tfo_pending =
2282                                                     tcp_fastopen_alloc_counter();
2283                                 } else {
2284                                         /*
2285                                          * If a pre-shared key was provided,
2286                                          * stash it in the client cookie
2287                                          * field of the tcpcb for use during
2288                                          * connect.
2289                                          */
2290                                         if (sopt->sopt_valsize ==
2291                                             sizeof(tfo_optval)) {
2292                                                 memcpy(tp->t_tfo_cookie.client,
2293                                                        tfo_optval.psk,
2294                                                        TCP_FASTOPEN_PSK_LEN);
2295                                                 tp->t_tfo_client_cookie_len =
2296                                                     TCP_FASTOPEN_PSK_LEN;
2297                                         }
2298                                 }
2299                                 tp->t_flags |= TF_FASTOPEN;
2300                         } else
2301                                 tp->t_flags &= ~TF_FASTOPEN;
2302                         goto unlock_and_done;
2303                 }
2304
2305 #ifdef TCP_BLACKBOX
2306                 case TCP_LOG:
2307                         INP_WUNLOCK(inp);
2308                         error = sooptcopyin(sopt, &optval, sizeof optval,
2309                             sizeof optval);
2310                         if (error)
2311                                 return (error);
2312
2313                         INP_WLOCK_RECHECK(inp);
2314                         error = tcp_log_state_change(tp, optval);
2315                         goto unlock_and_done;
2316
2317                 case TCP_LOGBUF:
2318                         INP_WUNLOCK(inp);
2319                         error = EINVAL;
2320                         break;
2321
2322                 case TCP_LOGID:
2323                         INP_WUNLOCK(inp);
2324                         error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2325                         if (error)
2326                                 break;
2327                         buf[sopt->sopt_valsize] = '\0';
2328                         INP_WLOCK_RECHECK(inp);
2329                         error = tcp_log_set_id(tp, buf);
2330                         /* tcp_log_set_id() unlocks the INP. */
2331                         break;
2332
2333                 case TCP_LOGDUMP:
2334                 case TCP_LOGDUMPID:
2335                         INP_WUNLOCK(inp);
2336                         error =
2337                             sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2338                         if (error)
2339                                 break;
2340                         buf[sopt->sopt_valsize] = '\0';
2341                         INP_WLOCK_RECHECK(inp);
2342                         if (sopt->sopt_name == TCP_LOGDUMP) {
2343                                 error = tcp_log_dump_tp_logbuf(tp, buf,
2344                                     M_WAITOK, true);
2345                                 INP_WUNLOCK(inp);
2346                         } else {
2347                                 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2348                                 /*
2349                                  * tcp_log_dump_tp_bucket_logbufs() drops the
2350                                  * INP lock.
2351                                  */
2352                         }
2353                         break;
2354 #endif
2355
2356                 default:
2357                         INP_WUNLOCK(inp);
2358                         error = ENOPROTOOPT;
2359                         break;
2360                 }
2361                 break;
2362
2363         case SOPT_GET:
2364                 tp = intotcpcb(inp);
2365                 switch (sopt->sopt_name) {
2366 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2367                 case TCP_MD5SIG:
2368                         if (!TCPMD5_ENABLED()) {
2369                                 INP_WUNLOCK(inp);
2370                                 return (ENOPROTOOPT);
2371                         }
2372                         error = TCPMD5_PCBCTL(inp, sopt);
2373                         break;
2374 #endif
2375
2376                 case TCP_NODELAY:
2377                         optval = tp->t_flags & TF_NODELAY;
2378                         INP_WUNLOCK(inp);
2379                         error = sooptcopyout(sopt, &optval, sizeof optval);
2380                         break;
2381                 case TCP_MAXSEG:
2382                         optval = tp->t_maxseg;
2383                         INP_WUNLOCK(inp);
2384                         error = sooptcopyout(sopt, &optval, sizeof optval);
2385                         break;
2386                 case TCP_NOOPT:
2387                         optval = tp->t_flags & TF_NOOPT;
2388                         INP_WUNLOCK(inp);
2389                         error = sooptcopyout(sopt, &optval, sizeof optval);
2390                         break;
2391                 case TCP_NOPUSH:
2392                         optval = tp->t_flags & TF_NOPUSH;
2393                         INP_WUNLOCK(inp);
2394                         error = sooptcopyout(sopt, &optval, sizeof optval);
2395                         break;
2396                 case TCP_INFO:
2397                         tcp_fill_info(tp, &ti);
2398                         INP_WUNLOCK(inp);
2399                         error = sooptcopyout(sopt, &ti, sizeof ti);
2400                         break;
2401                 case TCP_STATS:
2402                         {
2403 #ifdef STATS
2404                         int nheld;
2405                         TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2406
2407                         error = 0;
2408                         socklen_t outsbsz = sopt->sopt_valsize;
2409                         if (tp->t_stats == NULL)
2410                                 error = ENOENT;
2411                         else if (outsbsz >= tp->t_stats->cursz)
2412                                 outsbsz = tp->t_stats->cursz;
2413                         else if (outsbsz >= sizeof(struct statsblob))
2414                                 outsbsz = sizeof(struct statsblob);
2415                         else
2416                                 error = EINVAL;
2417                         INP_WUNLOCK(inp);
2418                         if (error)
2419                                 break;
2420
2421                         sbp = sopt->sopt_val;
2422                         nheld = atop(round_page(((vm_offset_t)sbp) +
2423                             (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2424                         vm_page_t ma[nheld];
2425                         if (vm_fault_quick_hold_pages(
2426                             &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2427                             outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2428                             nheld) < 0) {
2429                                 error = EFAULT;
2430                                 break;
2431                         }
2432
2433                         if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2434                             SIZEOF_MEMBER(struct statsblob, flags))))
2435                                 goto unhold;
2436
2437                         INP_WLOCK_RECHECK(inp);
2438                         error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2439                             sbflags | SB_CLONE_USRDSTNOFAULT);
2440                         INP_WUNLOCK(inp);
2441                         sopt->sopt_valsize = outsbsz;
2442 unhold:
2443                         vm_page_unhold_pages(ma, nheld);
2444 #else
2445                         INP_WUNLOCK(inp);
2446                         error = EOPNOTSUPP;
2447 #endif /* !STATS */
2448                         break;
2449                         }
2450                 case TCP_CONGESTION:
2451                         len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2452                         INP_WUNLOCK(inp);
2453                         error = sooptcopyout(sopt, buf, len + 1);
2454                         break;
2455                 case TCP_KEEPIDLE:
2456                 case TCP_KEEPINTVL:
2457                 case TCP_KEEPINIT:
2458                 case TCP_KEEPCNT:
2459                         switch (sopt->sopt_name) {
2460                         case TCP_KEEPIDLE:
2461                                 ui = TP_KEEPIDLE(tp) / hz;
2462                                 break;
2463                         case TCP_KEEPINTVL:
2464                                 ui = TP_KEEPINTVL(tp) / hz;
2465                                 break;
2466                         case TCP_KEEPINIT:
2467                                 ui = TP_KEEPINIT(tp) / hz;
2468                                 break;
2469                         case TCP_KEEPCNT:
2470                                 ui = TP_KEEPCNT(tp);
2471                                 break;
2472                         }
2473                         INP_WUNLOCK(inp);
2474                         error = sooptcopyout(sopt, &ui, sizeof(ui));
2475                         break;
2476 #ifdef TCPPCAP
2477                 case TCP_PCAP_OUT:
2478                 case TCP_PCAP_IN:
2479                         optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2480                                         &(tp->t_outpkts) : &(tp->t_inpkts));
2481                         INP_WUNLOCK(inp);
2482                         error = sooptcopyout(sopt, &optval, sizeof optval);
2483                         break;
2484 #endif
2485                 case TCP_FASTOPEN:
2486                         optval = tp->t_flags & TF_FASTOPEN;
2487                         INP_WUNLOCK(inp);
2488                         error = sooptcopyout(sopt, &optval, sizeof optval);
2489                         break;
2490 #ifdef TCP_BLACKBOX
2491                 case TCP_LOG:
2492                         optval = tp->t_logstate;
2493                         INP_WUNLOCK(inp);
2494                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2495                         break;
2496                 case TCP_LOGBUF:
2497                         /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2498                         error = tcp_log_getlogbuf(sopt, tp);
2499                         break;
2500                 case TCP_LOGID:
2501                         len = tcp_log_get_id(tp, buf);
2502                         INP_WUNLOCK(inp);
2503                         error = sooptcopyout(sopt, buf, len + 1);
2504                         break;
2505                 case TCP_LOGDUMP:
2506                 case TCP_LOGDUMPID:
2507                         INP_WUNLOCK(inp);
2508                         error = EINVAL;
2509                         break;
2510 #endif
2511 #ifdef KERN_TLS
2512                 case TCP_TXTLS_MODE:
2513                         optval = ktls_get_tx_mode(so);
2514                         INP_WUNLOCK(inp);
2515                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2516                         break;
2517                 case TCP_RXTLS_MODE:
2518                         optval = ktls_get_rx_mode(so);
2519                         INP_WUNLOCK(inp);
2520                         error = sooptcopyout(sopt, &optval, sizeof(optval));
2521                         break;
2522 #endif
2523                 default:
2524                         INP_WUNLOCK(inp);
2525                         error = ENOPROTOOPT;
2526                         break;
2527                 }
2528                 break;
2529         }
2530         return (error);
2531 }
2532 #undef INP_WLOCK_RECHECK
2533 #undef INP_WLOCK_RECHECK_CLEANUP
2534
2535 /*
2536  * Initiate (or continue) disconnect.
2537  * If embryonic state, just send reset (once).
2538  * If in ``let data drain'' option and linger null, just drop.
2539  * Otherwise (hard), mark socket disconnecting and drop
2540  * current input data; switch states based on user close, and
2541  * send segment to peer (with FIN).
2542  */
2543 static void
2544 tcp_disconnect(struct tcpcb *tp)
2545 {
2546         struct inpcb *inp = tp->t_inpcb;
2547         struct socket *so = inp->inp_socket;
2548
2549         NET_EPOCH_ASSERT();
2550         INP_WLOCK_ASSERT(inp);
2551
2552         /*
2553          * Neither tcp_close() nor tcp_drop() should return NULL, as the
2554          * socket is still open.
2555          */
2556         if (tp->t_state < TCPS_ESTABLISHED &&
2557             !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2558                 tp = tcp_close(tp);
2559                 KASSERT(tp != NULL,
2560                     ("tcp_disconnect: tcp_close() returned NULL"));
2561         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2562                 tp = tcp_drop(tp, 0);
2563                 KASSERT(tp != NULL,
2564                     ("tcp_disconnect: tcp_drop() returned NULL"));
2565         } else {
2566                 soisdisconnecting(so);
2567                 sbflush(&so->so_rcv);
2568                 tcp_usrclosed(tp);
2569                 if (!(inp->inp_flags & INP_DROPPED))
2570                         tp->t_fb->tfb_tcp_output(tp);
2571         }
2572 }
2573
2574 /*
2575  * User issued close, and wish to trail through shutdown states:
2576  * if never received SYN, just forget it.  If got a SYN from peer,
2577  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2578  * If already got a FIN from peer, then almost done; go to LAST_ACK
2579  * state.  In all other cases, have already sent FIN to peer (e.g.
2580  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2581  * for peer to send FIN or not respond to keep-alives, etc.
2582  * We can let the user exit from the close as soon as the FIN is acked.
2583  */
2584 static void
2585 tcp_usrclosed(struct tcpcb *tp)
2586 {
2587
2588         NET_EPOCH_ASSERT();
2589         INP_WLOCK_ASSERT(tp->t_inpcb);
2590
2591         switch (tp->t_state) {
2592         case TCPS_LISTEN:
2593 #ifdef TCP_OFFLOAD
2594                 tcp_offload_listen_stop(tp);
2595 #endif
2596                 tcp_state_change(tp, TCPS_CLOSED);
2597                 /* FALLTHROUGH */
2598         case TCPS_CLOSED:
2599                 tp = tcp_close(tp);
2600                 /*
2601                  * tcp_close() should never return NULL here as the socket is
2602                  * still open.
2603                  */
2604                 KASSERT(tp != NULL,
2605                     ("tcp_usrclosed: tcp_close() returned NULL"));
2606                 break;
2607
2608         case TCPS_SYN_SENT:
2609         case TCPS_SYN_RECEIVED:
2610                 tp->t_flags |= TF_NEEDFIN;
2611                 break;
2612
2613         case TCPS_ESTABLISHED:
2614                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2615                 break;
2616
2617         case TCPS_CLOSE_WAIT:
2618                 tcp_state_change(tp, TCPS_LAST_ACK);
2619                 break;
2620         }
2621         if (tp->t_state >= TCPS_FIN_WAIT_2) {
2622                 soisdisconnected(tp->t_inpcb->inp_socket);
2623                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2624                 if (tp->t_state == TCPS_FIN_WAIT_2) {
2625                         int timeout;
2626
2627                         timeout = (tcp_fast_finwait2_recycle) ?
2628                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
2629                         tcp_timer_activate(tp, TT_2MSL, timeout);
2630                 }
2631         }
2632 }
2633
2634 #ifdef DDB
2635 static void
2636 db_print_indent(int indent)
2637 {
2638         int i;
2639
2640         for (i = 0; i < indent; i++)
2641                 db_printf(" ");
2642 }
2643
2644 static void
2645 db_print_tstate(int t_state)
2646 {
2647
2648         switch (t_state) {
2649         case TCPS_CLOSED:
2650                 db_printf("TCPS_CLOSED");
2651                 return;
2652
2653         case TCPS_LISTEN:
2654                 db_printf("TCPS_LISTEN");
2655                 return;
2656
2657         case TCPS_SYN_SENT:
2658                 db_printf("TCPS_SYN_SENT");
2659                 return;
2660
2661         case TCPS_SYN_RECEIVED:
2662                 db_printf("TCPS_SYN_RECEIVED");
2663                 return;
2664
2665         case TCPS_ESTABLISHED:
2666                 db_printf("TCPS_ESTABLISHED");
2667                 return;
2668
2669         case TCPS_CLOSE_WAIT:
2670                 db_printf("TCPS_CLOSE_WAIT");
2671                 return;
2672
2673         case TCPS_FIN_WAIT_1:
2674                 db_printf("TCPS_FIN_WAIT_1");
2675                 return;
2676
2677         case TCPS_CLOSING:
2678                 db_printf("TCPS_CLOSING");
2679                 return;
2680
2681         case TCPS_LAST_ACK:
2682                 db_printf("TCPS_LAST_ACK");
2683                 return;
2684
2685         case TCPS_FIN_WAIT_2:
2686                 db_printf("TCPS_FIN_WAIT_2");
2687                 return;
2688
2689         case TCPS_TIME_WAIT:
2690                 db_printf("TCPS_TIME_WAIT");
2691                 return;
2692
2693         default:
2694                 db_printf("unknown");
2695                 return;
2696         }
2697 }
2698
2699 static void
2700 db_print_tflags(u_int t_flags)
2701 {
2702         int comma;
2703
2704         comma = 0;
2705         if (t_flags & TF_ACKNOW) {
2706                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2707                 comma = 1;
2708         }
2709         if (t_flags & TF_DELACK) {
2710                 db_printf("%sTF_DELACK", comma ? ", " : "");
2711                 comma = 1;
2712         }
2713         if (t_flags & TF_NODELAY) {
2714                 db_printf("%sTF_NODELAY", comma ? ", " : "");
2715                 comma = 1;
2716         }
2717         if (t_flags & TF_NOOPT) {
2718                 db_printf("%sTF_NOOPT", comma ? ", " : "");
2719                 comma = 1;
2720         }
2721         if (t_flags & TF_SENTFIN) {
2722                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2723                 comma = 1;
2724         }
2725         if (t_flags & TF_REQ_SCALE) {
2726                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2727                 comma = 1;
2728         }
2729         if (t_flags & TF_RCVD_SCALE) {
2730                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2731                 comma = 1;
2732         }
2733         if (t_flags & TF_REQ_TSTMP) {
2734                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2735                 comma = 1;
2736         }
2737         if (t_flags & TF_RCVD_TSTMP) {
2738                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2739                 comma = 1;
2740         }
2741         if (t_flags & TF_SACK_PERMIT) {
2742                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2743                 comma = 1;
2744         }
2745         if (t_flags & TF_NEEDSYN) {
2746                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2747                 comma = 1;
2748         }
2749         if (t_flags & TF_NEEDFIN) {
2750                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2751                 comma = 1;
2752         }
2753         if (t_flags & TF_NOPUSH) {
2754                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2755                 comma = 1;
2756         }
2757         if (t_flags & TF_MORETOCOME) {
2758                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2759                 comma = 1;
2760         }
2761         if (t_flags & TF_LQ_OVERFLOW) {
2762                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2763                 comma = 1;
2764         }
2765         if (t_flags & TF_LASTIDLE) {
2766                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2767                 comma = 1;
2768         }
2769         if (t_flags & TF_RXWIN0SENT) {
2770                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2771                 comma = 1;
2772         }
2773         if (t_flags & TF_FASTRECOVERY) {
2774                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2775                 comma = 1;
2776         }
2777         if (t_flags & TF_CONGRECOVERY) {
2778                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2779                 comma = 1;
2780         }
2781         if (t_flags & TF_WASFRECOVERY) {
2782                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2783                 comma = 1;
2784         }
2785         if (t_flags & TF_SIGNATURE) {
2786                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2787                 comma = 1;
2788         }
2789         if (t_flags & TF_FORCEDATA) {
2790                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2791                 comma = 1;
2792         }
2793         if (t_flags & TF_TSO) {
2794                 db_printf("%sTF_TSO", comma ? ", " : "");
2795                 comma = 1;
2796         }
2797         if (t_flags & TF_FASTOPEN) {
2798                 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2799                 comma = 1;
2800         }
2801 }
2802
2803 static void
2804 db_print_tflags2(u_int t_flags2)
2805 {
2806         int comma;
2807
2808         comma = 0;
2809         if (t_flags2 & TF2_ECN_PERMIT) {
2810                 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2811                 comma = 1;
2812         }
2813 }
2814
2815
2816 static void
2817 db_print_toobflags(char t_oobflags)
2818 {
2819         int comma;
2820
2821         comma = 0;
2822         if (t_oobflags & TCPOOB_HAVEDATA) {
2823                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2824                 comma = 1;
2825         }
2826         if (t_oobflags & TCPOOB_HADDATA) {
2827                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2828                 comma = 1;
2829         }
2830 }
2831
2832 static void
2833 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2834 {
2835
2836         db_print_indent(indent);
2837         db_printf("%s at %p\n", name, tp);
2838
2839         indent += 2;
2840
2841         db_print_indent(indent);
2842         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2843            TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2844
2845         db_print_indent(indent);
2846         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2847             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2848
2849         db_print_indent(indent);
2850         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2851             &tp->t_timers->tt_delack, tp->t_inpcb);
2852
2853         db_print_indent(indent);
2854         db_printf("t_state: %d (", tp->t_state);
2855         db_print_tstate(tp->t_state);
2856         db_printf(")\n");
2857
2858         db_print_indent(indent);
2859         db_printf("t_flags: 0x%x (", tp->t_flags);
2860         db_print_tflags(tp->t_flags);
2861         db_printf(")\n");
2862
2863         db_print_indent(indent);
2864         db_printf("t_flags2: 0x%x (", tp->t_flags2);
2865         db_print_tflags2(tp->t_flags2);
2866         db_printf(")\n");
2867
2868         db_print_indent(indent);
2869         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2870             tp->snd_una, tp->snd_max, tp->snd_nxt);
2871
2872         db_print_indent(indent);
2873         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2874            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2875
2876         db_print_indent(indent);
2877         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2878             tp->iss, tp->irs, tp->rcv_nxt);
2879
2880         db_print_indent(indent);
2881         db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2882             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2883
2884         db_print_indent(indent);
2885         db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2886            tp->snd_wnd, tp->snd_cwnd);
2887
2888         db_print_indent(indent);
2889         db_printf("snd_ssthresh: %u   snd_recover: "
2890             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2891
2892         db_print_indent(indent);
2893         db_printf("t_rcvtime: %u   t_startime: %u\n",
2894             tp->t_rcvtime, tp->t_starttime);
2895
2896         db_print_indent(indent);
2897         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2898             tp->t_rtttime, tp->t_rtseq);
2899
2900         db_print_indent(indent);
2901         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2902             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2903
2904         db_print_indent(indent);
2905         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2906             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2907             tp->t_rttbest);
2908
2909         db_print_indent(indent);
2910         db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2911             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2912
2913         db_print_indent(indent);
2914         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2915         db_print_toobflags(tp->t_oobflags);
2916         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2917
2918         db_print_indent(indent);
2919         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2920             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2921
2922         db_print_indent(indent);
2923         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2924             tp->ts_recent, tp->ts_recent_age);
2925
2926         db_print_indent(indent);
2927         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2928             "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2929
2930         db_print_indent(indent);
2931         db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2932             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2933             tp->snd_recover_prev, tp->t_badrxtwin);
2934
2935         db_print_indent(indent);
2936         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2937             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2938
2939         db_print_indent(indent);
2940         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
2941             tp->snd_fack, tp->rcv_numsacks);
2942
2943         /* Skip sackblks, sackhint. */
2944
2945         db_print_indent(indent);
2946         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2947             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2948 }
2949
2950 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2951 {
2952         struct tcpcb *tp;
2953
2954         if (!have_addr) {
2955                 db_printf("usage: show tcpcb <addr>\n");
2956                 return;
2957         }
2958         tp = (struct tcpcb *)addr;
2959
2960         db_print_tcpcb(tp, "tcpcb", 0);
2961 }
2962 #endif