sys/netinet/tcp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2006-2007 Robert N. M. Watson
   5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   6  * All rights reserved.
   7  *
   8  * Portions of this software were developed by Robert N. M. Watson under
   9  * contract to Juniper Networks, Inc.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 4. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_ddb.h"
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44 #include "opt_tcpdebug.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/limits.h>
  49 #include <sys/malloc.h>
  50 #include <sys/kernel.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/mbuf.h>
  53 #ifdef INET6
  54 #include <sys/domain.h>
  55 #endif /* INET6 */
  56 #include <sys/socket.h>
  57 #include <sys/socketvar.h>
  58 #include <sys/protosw.h>
  59 #include <sys/proc.h>
  60 #include <sys/jail.h>
  61
  62 #ifdef DDB
  63 #include <ddb/ddb.h>
  64 #endif
  65
  66 #include <net/if.h>
  67 #include <net/route.h>
  68 #include <net/vnet.h>
  69
  70 #include <netinet/cc.h>
  71 #include <netinet/in.h>
  72 #include <netinet/in_pcb.h>
  73 #include <netinet/in_systm.h>
  74 #include <netinet/in_var.h>
  75 #include <netinet/ip_var.h>
  76 #ifdef INET6
  77 #include <netinet/ip6.h>
  78 #include <netinet6/in6_pcb.h>
  79 #include <netinet6/ip6_var.h>
  80 #include <netinet6/scope6_var.h>
  81 #endif
  82 #include <netinet/tcp_fsm.h>
  83 #include <netinet/tcp_seq.h>
  84 #include <netinet/tcp_timer.h>
  85 #include <netinet/tcp_var.h>
  86 #include <netinet/tcpip.h>
  87 #ifdef TCPDEBUG
  88 #include <netinet/tcp_debug.h>
  89 #endif
  90 #ifdef TCP_OFFLOAD
  91 #include <netinet/tcp_offload.h>
  92 #endif
  93
  94 /*
  95  * TCP protocol interface to socket abstraction.
  96  */
  97 static int      tcp_attach(struct socket *);
  98 #ifdef INET
  99 static int      tcp_connect(struct tcpcb *, struct sockaddr *,
 100                     struct thread *td);
 101 #endif /* INET */
 102 #ifdef INET6
 103 static int      tcp6_connect(struct tcpcb *, struct sockaddr *,
 104                     struct thread *td);
 105 #endif /* INET6 */
 106 static void     tcp_disconnect(struct tcpcb *);
 107 static void     tcp_usrclosed(struct tcpcb *);
 108 static void     tcp_fill_info(struct tcpcb *, struct tcp_info *);
 109
 110 #ifdef TCPDEBUG
 111 #define TCPDEBUG0       int ostate = 0
 112 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 113 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 114                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 115 #else
 116 #define TCPDEBUG0
 117 #define TCPDEBUG1()
 118 #define TCPDEBUG2(req)
 119 #endif
 120
 121 /*
 122  * TCP attaches to socket via pru_attach(), reserving space,
 123  * and an internet control block.
 124  */
 125 static int
 126 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
 127 {
 128         struct inpcb *inp;
 129         struct tcpcb *tp = NULL;
 130         int error;
 131         TCPDEBUG0;
 132
 133         inp = sotoinpcb(so);
 134         KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
 135         TCPDEBUG1();
 136
 137         error = tcp_attach(so);
 138         if (error)
 139                 goto out;
 140
 141         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 142                 so->so_linger = TCP_LINGERTIME;
 143
 144         inp = sotoinpcb(so);
 145         tp = intotcpcb(inp);
 146 out:
 147         TCPDEBUG2(PRU_ATTACH);
 148         return error;
 149 }
 150
 151 /*
 152  * tcp_detach is called when the socket layer loses its final reference
 153  * to the socket, be it a file descriptor reference, a reference from TCP,
 154  * etc.  At this point, there is only one case in which we will keep around
 155  * inpcb state: time wait.
 156  *
 157  * This function can probably be re-absorbed back into tcp_usr_detach() now
 158  * that there is a single detach path.
 159  */
 160 static void
 161 tcp_detach(struct socket *so, struct inpcb *inp)
 162 {
 163         struct tcpcb *tp;
 164
 165         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 166         INP_WLOCK_ASSERT(inp);
 167
 168         KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
 169         KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
 170
 171         tp = intotcpcb(inp);
 172
 173         if (inp->inp_flags & INP_TIMEWAIT) {
 174                 /*
 175                  * There are two cases to handle: one in which the time wait
 176                  * state is being discarded (INP_DROPPED), and one in which
 177                  * this connection will remain in timewait.  In the former,
 178                  * it is time to discard all state (except tcptw, which has
 179                  * already been discarded by the timewait close code, which
 180                  * should be further up the call stack somewhere).  In the
 181                  * latter case, we detach from the socket, but leave the pcb
 182                  * present until timewait ends.
 183                  *
 184                  * XXXRW: Would it be cleaner to free the tcptw here?
 185                  *
 186                  * Astute question indeed, from twtcp perspective there are
 187                  * three cases to consider:
 188                  *
 189                  * #1 tcp_detach is called at tcptw creation time by
 190                  *  tcp_twstart, then do not discard the newly created tcptw
 191                  *  and leave inpcb present until timewait ends
 192                  * #2 tcp_detach is called at timewait end (or reuse) by
 193                  *  tcp_twclose, then the tcptw has already been discarded
 194                  *  and inpcb is freed here
 195                  * #3 tcp_detach is called() after timewait ends (or reuse)
 196                  *  (e.g. by soclose), then tcptw has already been discarded
 197                  *  and inpcb is freed here
 198                  *
 199                  *  In all three cases the tcptw should not be freed here.
 200                  */
 201                 if (inp->inp_flags & INP_DROPPED) {
 202                         KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
 203                             "INP_DROPPED && tp != NULL"));
 204                         in_pcbdetach(inp);
 205                         in_pcbfree(inp);
 206                 } else {
 207                         in_pcbdetach(inp);
 208                         INP_WUNLOCK(inp);
 209                 }
 210         } else {
 211                 /*
 212                  * If the connection is not in timewait, we consider two
 213                  * two conditions: one in which no further processing is
 214                  * necessary (dropped || embryonic), and one in which TCP is
 215                  * not yet done, but no longer requires the socket, so the
 216                  * pcb will persist for the time being.
 217                  *
 218                  * XXXRW: Does the second case still occur?
 219                  */
 220                 if (inp->inp_flags & INP_DROPPED ||
 221                     tp->t_state < TCPS_SYN_SENT) {
 222                         tcp_discardcb(tp);
 223                         in_pcbdetach(inp);
 224                         in_pcbfree(inp);
 225                 } else {
 226                         in_pcbdetach(inp);
 227                         INP_WUNLOCK(inp);
 228                 }
 229         }
 230 }
 231
 232 /*
 233  * pru_detach() detaches the TCP protocol from the socket.
 234  * If the protocol state is non-embryonic, then can't
 235  * do this directly: have to initiate a pru_disconnect(),
 236  * which may finish later; embryonic TCB's can just
 237  * be discarded here.
 238  */
 239 static void
 240 tcp_usr_detach(struct socket *so)
 241 {
 242         struct inpcb *inp;
 243
 244         inp = sotoinpcb(so);
 245         KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
 246         INP_INFO_WLOCK(&V_tcbinfo);
 247         INP_WLOCK(inp);
 248         KASSERT(inp->inp_socket != NULL,
 249             ("tcp_usr_detach: inp_socket == NULL"));
 250         tcp_detach(so, inp);
 251         INP_INFO_WUNLOCK(&V_tcbinfo);
 252 }
 253
 254 #ifdef INET
 255 /*
 256  * Give the socket an address.
 257  */
 258 static int
 259 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 260 {
 261         int error = 0;
 262         struct inpcb *inp;
 263         struct tcpcb *tp = NULL;
 264         struct sockaddr_in *sinp;
 265
 266         sinp = (struct sockaddr_in *)nam;
 267         if (nam->sa_len != sizeof (*sinp))
 268                 return (EINVAL);
 269         /*
 270          * Must check for multicast addresses and disallow binding
 271          * to them.
 272          */
 273         if (sinp->sin_family == AF_INET &&
 274             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 275                 return (EAFNOSUPPORT);
 276
 277         TCPDEBUG0;
 278         inp = sotoinpcb(so);
 279         KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
 280         INP_WLOCK(inp);
 281         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 282                 error = EINVAL;
 283                 goto out;
 284         }
 285         tp = intotcpcb(inp);
 286         TCPDEBUG1();
 287         INP_HASH_WLOCK(&V_tcbinfo);
 288         error = in_pcbbind(inp, nam, td->td_ucred);
 289         INP_HASH_WUNLOCK(&V_tcbinfo);
 290 out:
 291         TCPDEBUG2(PRU_BIND);
 292         INP_WUNLOCK(inp);
 293
 294         return (error);
 295 }
 296 #endif /* INET */
 297
 298 #ifdef INET6
 299 static int
 300 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 301 {
 302         int error = 0;
 303         struct inpcb *inp;
 304         struct tcpcb *tp = NULL;
 305         struct sockaddr_in6 *sin6p;
 306
 307         sin6p = (struct sockaddr_in6 *)nam;
 308         if (nam->sa_len != sizeof (*sin6p))
 309                 return (EINVAL);
 310         /*
 311          * Must check for multicast addresses and disallow binding
 312          * to them.
 313          */
 314         if (sin6p->sin6_family == AF_INET6 &&
 315             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 316                 return (EAFNOSUPPORT);
 317
 318         TCPDEBUG0;
 319         inp = sotoinpcb(so);
 320         KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
 321         INP_WLOCK(inp);
 322         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 323                 error = EINVAL;
 324                 goto out;
 325         }
 326         tp = intotcpcb(inp);
 327         TCPDEBUG1();
 328         INP_HASH_WLOCK(&V_tcbinfo);
 329         inp->inp_vflag &= ~INP_IPV4;
 330         inp->inp_vflag |= INP_IPV6;
 331 #ifdef INET
 332         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 333                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 334                         inp->inp_vflag |= INP_IPV4;
 335                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 336                         struct sockaddr_in sin;
 337
 338                         in6_sin6_2_sin(&sin, sin6p);
 339                         inp->inp_vflag |= INP_IPV4;
 340                         inp->inp_vflag &= ~INP_IPV6;
 341                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
 342                             td->td_ucred);
 343                         INP_HASH_WUNLOCK(&V_tcbinfo);
 344                         goto out;
 345                 }
 346         }
 347 #endif
 348         error = in6_pcbbind(inp, nam, td->td_ucred);
 349         INP_HASH_WUNLOCK(&V_tcbinfo);
 350 out:
 351         TCPDEBUG2(PRU_BIND);
 352         INP_WUNLOCK(inp);
 353         return (error);
 354 }
 355 #endif /* INET6 */
 356
 357 #ifdef INET
 358 /*
 359  * Prepare to accept connections.
 360  */
 361 static int
 362 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
 363 {
 364         int error = 0;
 365         struct inpcb *inp;
 366         struct tcpcb *tp = NULL;
 367
 368         TCPDEBUG0;
 369         inp = sotoinpcb(so);
 370         KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
 371         INP_WLOCK(inp);
 372         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 373                 error = EINVAL;
 374                 goto out;
 375         }
 376         tp = intotcpcb(inp);
 377         TCPDEBUG1();
 378         SOCK_LOCK(so);
 379         error = solisten_proto_check(so);
 380         INP_HASH_WLOCK(&V_tcbinfo);
 381         if (error == 0 && inp->inp_lport == 0)
 382                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 383         INP_HASH_WUNLOCK(&V_tcbinfo);
 384         if (error == 0) {
 385                 tcp_state_change(tp, TCPS_LISTEN);
 386                 solisten_proto(so, backlog);
 387 #ifdef TCP_OFFLOAD
 388                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 389                         tcp_offload_listen_start(tp);
 390 #endif
 391         }
 392         SOCK_UNLOCK(so);
 393
 394 out:
 395         TCPDEBUG2(PRU_LISTEN);
 396         INP_WUNLOCK(inp);
 397         return (error);
 398 }
 399 #endif /* INET */
 400
 401 #ifdef INET6
 402 static int
 403 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
 404 {
 405         int error = 0;
 406         struct inpcb *inp;
 407         struct tcpcb *tp = NULL;
 408
 409         TCPDEBUG0;
 410         inp = sotoinpcb(so);
 411         KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
 412         INP_WLOCK(inp);
 413         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 414                 error = EINVAL;
 415                 goto out;
 416         }
 417         tp = intotcpcb(inp);
 418         TCPDEBUG1();
 419         SOCK_LOCK(so);
 420         error = solisten_proto_check(so);
 421         INP_HASH_WLOCK(&V_tcbinfo);
 422         if (error == 0 && inp->inp_lport == 0) {
 423                 inp->inp_vflag &= ~INP_IPV4;
 424                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
 425                         inp->inp_vflag |= INP_IPV4;
 426                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
 427         }
 428         INP_HASH_WUNLOCK(&V_tcbinfo);
 429         if (error == 0) {
 430                 tcp_state_change(tp, TCPS_LISTEN);
 431                 solisten_proto(so, backlog);
 432 #ifdef TCP_OFFLOAD
 433                 if ((so->so_options & SO_NO_OFFLOAD) == 0)
 434                         tcp_offload_listen_start(tp);
 435 #endif
 436         }
 437         SOCK_UNLOCK(so);
 438
 439 out:
 440         TCPDEBUG2(PRU_LISTEN);
 441         INP_WUNLOCK(inp);
 442         return (error);
 443 }
 444 #endif /* INET6 */
 445
 446 #ifdef INET
 447 /*
 448  * Initiate connection to peer.
 449  * Create a template for use in transmissions on this connection.
 450  * Enter SYN_SENT state, and mark socket as connecting.
 451  * Start keep-alive timer, and seed output sequence space.
 452  * Send initial segment on connection.
 453  */
 454 static int
 455 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 456 {
 457         int error = 0;
 458         struct inpcb *inp;
 459         struct tcpcb *tp = NULL;
 460         struct sockaddr_in *sinp;
 461
 462         sinp = (struct sockaddr_in *)nam;
 463         if (nam->sa_len != sizeof (*sinp))
 464                 return (EINVAL);
 465         /*
 466          * Must disallow TCP ``connections'' to multicast addresses.
 467          */
 468         if (sinp->sin_family == AF_INET
 469             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 470                 return (EAFNOSUPPORT);
 471         if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
 472                 return (error);
 473
 474         TCPDEBUG0;
 475         inp = sotoinpcb(so);
 476         KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
 477         INP_WLOCK(inp);
 478         if (inp->inp_flags & INP_TIMEWAIT) {
 479                 error = EADDRINUSE;
 480                 goto out;
 481         }
 482         if (inp->inp_flags & INP_DROPPED) {
 483                 error = ECONNREFUSED;
 484                 goto out;
 485         }
 486         tp = intotcpcb(inp);
 487         TCPDEBUG1();
 488         if ((error = tcp_connect(tp, nam, td)) != 0)
 489                 goto out;
 490 #ifdef TCP_OFFLOAD
 491         if (registered_toedevs > 0 &&
 492             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 493             (error = tcp_offload_connect(so, nam)) == 0)
 494                 goto out;
 495 #endif
 496         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 497         error = tcp_output(tp);
 498 out:
 499         TCPDEBUG2(PRU_CONNECT);
 500         INP_WUNLOCK(inp);
 501         return (error);
 502 }
 503 #endif /* INET */
 504
 505 #ifdef INET6
 506 static int
 507 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 508 {
 509         int error = 0;
 510         struct inpcb *inp;
 511         struct tcpcb *tp = NULL;
 512         struct sockaddr_in6 *sin6p;
 513
 514         TCPDEBUG0;
 515
 516         sin6p = (struct sockaddr_in6 *)nam;
 517         if (nam->sa_len != sizeof (*sin6p))
 518                 return (EINVAL);
 519         /*
 520          * Must disallow TCP ``connections'' to multicast addresses.
 521          */
 522         if (sin6p->sin6_family == AF_INET6
 523             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
 524                 return (EAFNOSUPPORT);
 525
 526         inp = sotoinpcb(so);
 527         KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
 528         INP_WLOCK(inp);
 529         if (inp->inp_flags & INP_TIMEWAIT) {
 530                 error = EADDRINUSE;
 531                 goto out;
 532         }
 533         if (inp->inp_flags & INP_DROPPED) {
 534                 error = ECONNREFUSED;
 535                 goto out;
 536         }
 537         tp = intotcpcb(inp);
 538         TCPDEBUG1();
 539 #ifdef INET
 540         /*
 541          * XXXRW: Some confusion: V4/V6 flags relate to binding, and
 542          * therefore probably require the hash lock, which isn't held here.
 543          * Is this a significant problem?
 544          */
 545         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 546                 struct sockaddr_in sin;
 547
 548                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 549                         error = EINVAL;
 550                         goto out;
 551                 }
 552
 553                 in6_sin6_2_sin(&sin, sin6p);
 554                 inp->inp_vflag |= INP_IPV4;
 555                 inp->inp_vflag &= ~INP_IPV6;
 556                 if ((error = prison_remote_ip4(td->td_ucred,
 557                     &sin.sin_addr)) != 0)
 558                         goto out;
 559                 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
 560                         goto out;
 561 #ifdef TCP_OFFLOAD
 562                 if (registered_toedevs > 0 &&
 563                     (so->so_options & SO_NO_OFFLOAD) == 0 &&
 564                     (error = tcp_offload_connect(so, nam)) == 0)
 565                         goto out;
 566 #endif
 567                 error = tcp_output(tp);
 568                 goto out;
 569         }
 570 #endif
 571         inp->inp_vflag &= ~INP_IPV4;
 572         inp->inp_vflag |= INP_IPV6;
 573         inp->inp_inc.inc_flags |= INC_ISIPV6;
 574         if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
 575                 goto out;
 576         if ((error = tcp6_connect(tp, nam, td)) != 0)
 577                 goto out;
 578 #ifdef TCP_OFFLOAD
 579         if (registered_toedevs > 0 &&
 580             (so->so_options & SO_NO_OFFLOAD) == 0 &&
 581             (error = tcp_offload_connect(so, nam)) == 0)
 582                 goto out;
 583 #endif
 584         tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
 585         error = tcp_output(tp);
 586
 587 out:
 588         TCPDEBUG2(PRU_CONNECT);
 589         INP_WUNLOCK(inp);
 590         return (error);
 591 }
 592 #endif /* INET6 */
 593
 594 /*
 595  * Initiate disconnect from peer.
 596  * If connection never passed embryonic stage, just drop;
 597  * else if don't need to let data drain, then can just drop anyways,
 598  * else have to begin TCP shutdown process: mark socket disconnecting,
 599  * drain unread data, state switch to reflect user close, and
 600  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 601  * when peer sends FIN and acks ours.
 602  *
 603  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 604  */
 605 static int
 606 tcp_usr_disconnect(struct socket *so)
 607 {
 608         struct inpcb *inp;
 609         struct tcpcb *tp = NULL;
 610         int error = 0;
 611
 612         TCPDEBUG0;
 613         INP_INFO_WLOCK(&V_tcbinfo);
 614         inp = sotoinpcb(so);
 615         KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
 616         INP_WLOCK(inp);
 617         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 618                 error = ECONNRESET;
 619                 goto out;
 620         }
 621         tp = intotcpcb(inp);
 622         TCPDEBUG1();
 623         tcp_disconnect(tp);
 624 out:
 625         TCPDEBUG2(PRU_DISCONNECT);
 626         INP_WUNLOCK(inp);
 627         INP_INFO_WUNLOCK(&V_tcbinfo);
 628         return (error);
 629 }
 630
 631 #ifdef INET
 632 /*
 633  * Accept a connection.  Essentially all the work is done at higher levels;
 634  * just return the address of the peer, storing through addr.
 635  *
 636  * The rationale for acquiring the tcbinfo lock here is somewhat complicated,
 637  * and is described in detail in the commit log entry for r175612.  Acquiring
 638  * it delays an accept(2) racing with sonewconn(), which inserts the socket
 639  * before the inpcb address/port fields are initialized.  A better fix would
 640  * prevent the socket from being placed in the listen queue until all fields
 641  * are fully initialized.
 642  */
 643 static int
 644 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
 645 {
 646         int error = 0;
 647         struct inpcb *inp = NULL;
 648         struct tcpcb *tp = NULL;
 649         struct in_addr addr;
 650         in_port_t port = 0;
 651         TCPDEBUG0;
 652
 653         if (so->so_state & SS_ISDISCONNECTED)
 654                 return (ECONNABORTED);
 655
 656         inp = sotoinpcb(so);
 657         KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
 658         INP_INFO_RLOCK(&V_tcbinfo);
 659         INP_WLOCK(inp);
 660         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 661                 error = ECONNABORTED;
 662                 goto out;
 663         }
 664         tp = intotcpcb(inp);
 665         TCPDEBUG1();
 666
 667         /*
 668          * We inline in_getpeeraddr and COMMON_END here, so that we can
 669          * copy the data of interest and defer the malloc until after we
 670          * release the lock.
 671          */
 672         port = inp->inp_fport;
 673         addr = inp->inp_faddr;
 674
 675 out:
 676         TCPDEBUG2(PRU_ACCEPT);
 677         INP_WUNLOCK(inp);
 678         INP_INFO_RUNLOCK(&V_tcbinfo);
 679         if (error == 0)
 680                 *nam = in_sockaddr(port, &addr);
 681         return error;
 682 }
 683 #endif /* INET */
 684
 685 #ifdef INET6
 686 static int
 687 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
 688 {
 689         struct inpcb *inp = NULL;
 690         int error = 0;
 691         struct tcpcb *tp = NULL;
 692         struct in_addr addr;
 693         struct in6_addr addr6;
 694         in_port_t port = 0;
 695         int v4 = 0;
 696         TCPDEBUG0;
 697
 698         if (so->so_state & SS_ISDISCONNECTED)
 699                 return (ECONNABORTED);
 700
 701         inp = sotoinpcb(so);
 702         KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
 703         INP_INFO_RLOCK(&V_tcbinfo);
 704         INP_WLOCK(inp);
 705         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 706                 error = ECONNABORTED;
 707                 goto out;
 708         }
 709         tp = intotcpcb(inp);
 710         TCPDEBUG1();
 711
 712         /*
 713          * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
 714          * copy the data of interest and defer the malloc until after we
 715          * release the lock.
 716          */
 717         if (inp->inp_vflag & INP_IPV4) {
 718                 v4 = 1;
 719                 port = inp->inp_fport;
 720                 addr = inp->inp_faddr;
 721         } else {
 722                 port = inp->inp_fport;
 723                 addr6 = inp->in6p_faddr;
 724         }
 725
 726 out:
 727         TCPDEBUG2(PRU_ACCEPT);
 728         INP_WUNLOCK(inp);
 729         INP_INFO_RUNLOCK(&V_tcbinfo);
 730         if (error == 0) {
 731                 if (v4)
 732                         *nam = in6_v4mapsin6_sockaddr(port, &addr);
 733                 else
 734                         *nam = in6_sockaddr(port, &addr6);
 735         }
 736         return error;
 737 }
 738 #endif /* INET6 */
 739
 740 /*
 741  * Mark the connection as being incapable of further output.
 742  */
 743 static int
 744 tcp_usr_shutdown(struct socket *so)
 745 {
 746         int error = 0;
 747         struct inpcb *inp;
 748         struct tcpcb *tp = NULL;
 749
 750         TCPDEBUG0;
 751         INP_INFO_WLOCK(&V_tcbinfo);
 752         inp = sotoinpcb(so);
 753         KASSERT(inp != NULL, ("inp == NULL"));
 754         INP_WLOCK(inp);
 755         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 756                 error = ECONNRESET;
 757                 goto out;
 758         }
 759         tp = intotcpcb(inp);
 760         TCPDEBUG1();
 761         socantsendmore(so);
 762         tcp_usrclosed(tp);
 763         if (!(inp->inp_flags & INP_DROPPED))
 764                 error = tcp_output(tp);
 765
 766 out:
 767         TCPDEBUG2(PRU_SHUTDOWN);
 768         INP_WUNLOCK(inp);
 769         INP_INFO_WUNLOCK(&V_tcbinfo);
 770
 771         return (error);
 772 }
 773
 774 /*
 775  * After a receive, possibly send window update to peer.
 776  */
 777 static int
 778 tcp_usr_rcvd(struct socket *so, int flags)
 779 {
 780         struct inpcb *inp;
 781         struct tcpcb *tp = NULL;
 782         int error = 0;
 783
 784         TCPDEBUG0;
 785         inp = sotoinpcb(so);
 786         KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
 787         INP_WLOCK(inp);
 788         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 789                 error = ECONNRESET;
 790                 goto out;
 791         }
 792         tp = intotcpcb(inp);
 793         TCPDEBUG1();
 794 #ifdef TCP_OFFLOAD
 795         if (tp->t_flags & TF_TOE)
 796                 tcp_offload_rcvd(tp);
 797         else
 798 #endif
 799         tcp_output(tp);
 800
 801 out:
 802         TCPDEBUG2(PRU_RCVD);
 803         INP_WUNLOCK(inp);
 804         return (error);
 805 }
 806
 807 /*
 808  * Do a send by putting data in output queue and updating urgent
 809  * marker if URG set.  Possibly send more data.  Unlike the other
 810  * pru_*() routines, the mbuf chains are our responsibility.  We
 811  * must either enqueue them or free them.  The other pru_* routines
 812  * generally are caller-frees.
 813  */
 814 static int
 815 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
 816     struct sockaddr *nam, struct mbuf *control, struct thread *td)
 817 {
 818         int error = 0;
 819         struct inpcb *inp;
 820         struct tcpcb *tp = NULL;
 821 #ifdef INET6
 822         int isipv6;
 823 #endif
 824         TCPDEBUG0;
 825
 826         /*
 827          * We require the pcbinfo lock if we will close the socket as part of
 828          * this call.
 829          */
 830         if (flags & PRUS_EOF)
 831                 INP_INFO_WLOCK(&V_tcbinfo);
 832         inp = sotoinpcb(so);
 833         KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
 834         INP_WLOCK(inp);
 835         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
 836                 if (control)
 837                         m_freem(control);
 838                 if (m)
 839                         m_freem(m);
 840                 error = ECONNRESET;
 841                 goto out;
 842         }
 843 #ifdef INET6
 844         isipv6 = nam && nam->sa_family == AF_INET6;
 845 #endif /* INET6 */
 846         tp = intotcpcb(inp);
 847         TCPDEBUG1();
 848         if (control) {
 849                 /* TCP doesn't do control messages (rights, creds, etc) */
 850                 if (control->m_len) {
 851                         m_freem(control);
 852                         if (m)
 853                                 m_freem(m);
 854                         error = EINVAL;
 855                         goto out;
 856                 }
 857                 m_freem(control);       /* empty control, just free it */
 858         }
 859         if (!(flags & PRUS_OOB)) {
 860                 sbappendstream(&so->so_snd, m);
 861                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 862                         /*
 863                          * Do implied connect if not yet connected,
 864                          * initialize window to default value, and
 865                          * initialize maxseg/maxopd using peer's cached
 866                          * MSS.
 867                          */
 868 #ifdef INET6
 869                         if (isipv6)
 870                                 error = tcp6_connect(tp, nam, td);
 871 #endif /* INET6 */
 872 #if defined(INET6) && defined(INET)
 873                         else
 874 #endif
 875 #ifdef INET
 876                                 error = tcp_connect(tp, nam, td);
 877 #endif
 878                         if (error)
 879                                 goto out;
 880                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 881                         tcp_mss(tp, -1);
 882                 }
 883                 if (flags & PRUS_EOF) {
 884                         /*
 885                          * Close the send side of the connection after
 886                          * the data is sent.
 887                          */
 888                         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 889                         socantsendmore(so);
 890                         tcp_usrclosed(tp);
 891                 }
 892                 if (!(inp->inp_flags & INP_DROPPED)) {
 893                         if (flags & PRUS_MORETOCOME)
 894                                 tp->t_flags |= TF_MORETOCOME;
 895                         error = tcp_output(tp);
 896                         if (flags & PRUS_MORETOCOME)
 897                                 tp->t_flags &= ~TF_MORETOCOME;
 898                 }
 899         } else {
 900                 /*
 901                  * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
 902                  */
 903                 SOCKBUF_LOCK(&so->so_snd);
 904                 if (sbspace(&so->so_snd) < -512) {
 905                         SOCKBUF_UNLOCK(&so->so_snd);
 906                         m_freem(m);
 907                         error = ENOBUFS;
 908                         goto out;
 909                 }
 910                 /*
 911                  * According to RFC961 (Assigned Protocols),
 912                  * the urgent pointer points to the last octet
 913                  * of urgent data.  We continue, however,
 914                  * to consider it to indicate the first octet
 915                  * of data past the urgent section.
 916                  * Otherwise, snd_up should be one lower.
 917                  */
 918                 sbappendstream_locked(&so->so_snd, m);
 919                 SOCKBUF_UNLOCK(&so->so_snd);
 920                 if (nam && tp->t_state < TCPS_SYN_SENT) {
 921                         /*
 922                          * Do implied connect if not yet connected,
 923                          * initialize window to default value, and
 924                          * initialize maxseg/maxopd using peer's cached
 925                          * MSS.
 926                          */
 927 #ifdef INET6
 928                         if (isipv6)
 929                                 error = tcp6_connect(tp, nam, td);
 930 #endif /* INET6 */
 931 #if defined(INET6) && defined(INET)
 932                         else
 933 #endif
 934 #ifdef INET
 935                                 error = tcp_connect(tp, nam, td);
 936 #endif
 937                         if (error)
 938                                 goto out;
 939                         tp->snd_wnd = TTCP_CLIENT_SND_WND;
 940                         tcp_mss(tp, -1);
 941                 }
 942                 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 943                 tp->t_flags |= TF_FORCEDATA;
 944                 error = tcp_output(tp);
 945                 tp->t_flags &= ~TF_FORCEDATA;
 946         }
 947 out:
 948         TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
 949                   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 950         INP_WUNLOCK(inp);
 951         if (flags & PRUS_EOF)
 952                 INP_INFO_WUNLOCK(&V_tcbinfo);
 953         return (error);
 954 }
 955
 956 /*
 957  * Abort the TCP.  Drop the connection abruptly.
 958  */
 959 static void
 960 tcp_usr_abort(struct socket *so)
 961 {
 962         struct inpcb *inp;
 963         struct tcpcb *tp = NULL;
 964         TCPDEBUG0;
 965
 966         inp = sotoinpcb(so);
 967         KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
 968
 969         INP_INFO_WLOCK(&V_tcbinfo);
 970         INP_WLOCK(inp);
 971         KASSERT(inp->inp_socket != NULL,
 972             ("tcp_usr_abort: inp_socket == NULL"));
 973
 974         /*
 975          * If we still have full TCP state, and we're not dropped, drop.
 976          */
 977         if (!(inp->inp_flags & INP_TIMEWAIT) &&
 978             !(inp->inp_flags & INP_DROPPED)) {
 979                 tp = intotcpcb(inp);
 980                 TCPDEBUG1();
 981                 tcp_drop(tp, ECONNABORTED);
 982                 TCPDEBUG2(PRU_ABORT);
 983         }
 984         if (!(inp->inp_flags & INP_DROPPED)) {
 985                 SOCK_LOCK(so);
 986                 so->so_state |= SS_PROTOREF;
 987                 SOCK_UNLOCK(so);
 988                 inp->inp_flags |= INP_SOCKREF;
 989         }
 990         INP_WUNLOCK(inp);
 991         INP_INFO_WUNLOCK(&V_tcbinfo);
 992 }
 993
 994 /*
 995  * TCP socket is closed.  Start friendly disconnect.
 996  */
 997 static void
 998 tcp_usr_close(struct socket *so)
 999 {
1000         struct inpcb *inp;
1001         struct tcpcb *tp = NULL;
1002         TCPDEBUG0;
1003
1004         inp = sotoinpcb(so);
1005         KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1006
1007         INP_INFO_WLOCK(&V_tcbinfo);
1008         INP_WLOCK(inp);
1009         KASSERT(inp->inp_socket != NULL,
1010             ("tcp_usr_close: inp_socket == NULL"));
1011
1012         /*
1013          * If we still have full TCP state, and we're not dropped, initiate
1014          * a disconnect.
1015          */
1016         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1017             !(inp->inp_flags & INP_DROPPED)) {
1018                 tp = intotcpcb(inp);
1019                 TCPDEBUG1();
1020                 tcp_disconnect(tp);
1021                 TCPDEBUG2(PRU_CLOSE);
1022         }
1023         if (!(inp->inp_flags & INP_DROPPED)) {
1024                 SOCK_LOCK(so);
1025                 so->so_state |= SS_PROTOREF;
1026                 SOCK_UNLOCK(so);
1027                 inp->inp_flags |= INP_SOCKREF;
1028         }
1029         INP_WUNLOCK(inp);
1030         INP_INFO_WUNLOCK(&V_tcbinfo);
1031 }
1032
1033 /*
1034  * Receive out-of-band data.
1035  */
1036 static int
1037 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1038 {
1039         int error = 0;
1040         struct inpcb *inp;
1041         struct tcpcb *tp = NULL;
1042
1043         TCPDEBUG0;
1044         inp = sotoinpcb(so);
1045         KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1046         INP_WLOCK(inp);
1047         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1048                 error = ECONNRESET;
1049                 goto out;
1050         }
1051         tp = intotcpcb(inp);
1052         TCPDEBUG1();
1053         if ((so->so_oobmark == 0 &&
1054              (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1055             so->so_options & SO_OOBINLINE ||
1056             tp->t_oobflags & TCPOOB_HADDATA) {
1057                 error = EINVAL;
1058                 goto out;
1059         }
1060         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1061                 error = EWOULDBLOCK;
1062                 goto out;
1063         }
1064         m->m_len = 1;
1065         *mtod(m, caddr_t) = tp->t_iobc;
1066         if ((flags & MSG_PEEK) == 0)
1067                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1068
1069 out:
1070         TCPDEBUG2(PRU_RCVOOB);
1071         INP_WUNLOCK(inp);
1072         return (error);
1073 }
1074
1075 #ifdef INET
1076 struct pr_usrreqs tcp_usrreqs = {
1077         .pru_abort =            tcp_usr_abort,
1078         .pru_accept =           tcp_usr_accept,
1079         .pru_attach =           tcp_usr_attach,
1080         .pru_bind =             tcp_usr_bind,
1081         .pru_connect =          tcp_usr_connect,
1082         .pru_control =          in_control,
1083         .pru_detach =           tcp_usr_detach,
1084         .pru_disconnect =       tcp_usr_disconnect,
1085         .pru_listen =           tcp_usr_listen,
1086         .pru_peeraddr =         in_getpeeraddr,
1087         .pru_rcvd =             tcp_usr_rcvd,
1088         .pru_rcvoob =           tcp_usr_rcvoob,
1089         .pru_send =             tcp_usr_send,
1090         .pru_shutdown =         tcp_usr_shutdown,
1091         .pru_sockaddr =         in_getsockaddr,
1092         .pru_sosetlabel =       in_pcbsosetlabel,
1093         .pru_close =            tcp_usr_close,
1094 };
1095 #endif /* INET */
1096
1097 #ifdef INET6
1098 struct pr_usrreqs tcp6_usrreqs = {
1099         .pru_abort =            tcp_usr_abort,
1100         .pru_accept =           tcp6_usr_accept,
1101         .pru_attach =           tcp_usr_attach,
1102         .pru_bind =             tcp6_usr_bind,
1103         .pru_connect =          tcp6_usr_connect,
1104         .pru_control =          in6_control,
1105         .pru_detach =           tcp_usr_detach,
1106         .pru_disconnect =       tcp_usr_disconnect,
1107         .pru_listen =           tcp6_usr_listen,
1108         .pru_peeraddr =         in6_mapped_peeraddr,
1109         .pru_rcvd =             tcp_usr_rcvd,
1110         .pru_rcvoob =           tcp_usr_rcvoob,
1111         .pru_send =             tcp_usr_send,
1112         .pru_shutdown =         tcp_usr_shutdown,
1113         .pru_sockaddr =         in6_mapped_sockaddr,
1114         .pru_sosetlabel =       in_pcbsosetlabel,
1115         .pru_close =            tcp_usr_close,
1116 };
1117 #endif /* INET6 */
1118
1119 #ifdef INET
1120 /*
1121  * Common subroutine to open a TCP connection to remote host specified
1122  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1123  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1124  * to choose a local host address (interface).  If there is an existing
1125  * incarnation of the same connection in TIME-WAIT state and if the remote
1126  * host was sending CC options and if the connection duration was < MSL, then
1127  * truncate the previous TIME-WAIT state and proceed.
1128  * Initialize connection parameters and enter SYN-SENT state.
1129  */
1130 static int
1131 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1132 {
1133         struct inpcb *inp = tp->t_inpcb, *oinp;
1134         struct socket *so = inp->inp_socket;
1135         struct in_addr laddr;
1136         u_short lport;
1137         int error;
1138
1139         INP_WLOCK_ASSERT(inp);
1140         INP_HASH_WLOCK(&V_tcbinfo);
1141
1142         if (inp->inp_lport == 0) {
1143                 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1144                 if (error)
1145                         goto out;
1146         }
1147
1148         /*
1149          * Cannot simply call in_pcbconnect, because there might be an
1150          * earlier incarnation of this same connection still in
1151          * TIME_WAIT state, creating an ADDRINUSE error.
1152          */
1153         laddr = inp->inp_laddr;
1154         lport = inp->inp_lport;
1155         error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1156             &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1157         if (error && oinp == NULL)
1158                 goto out;
1159         if (oinp) {
1160                 error = EADDRINUSE;
1161                 goto out;
1162         }
1163         inp->inp_laddr = laddr;
1164         in_pcbrehash(inp);
1165         INP_HASH_WUNLOCK(&V_tcbinfo);
1166
1167         /*
1168          * Compute window scaling to request:
1169          * Scale to fit into sweet spot.  See tcp_syncache.c.
1170          * XXX: This should move to tcp_output().
1171          */
1172         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1173             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1174                 tp->request_r_scale++;
1175
1176         soisconnecting(so);
1177         TCPSTAT_INC(tcps_connattempt);
1178         tcp_state_change(tp, TCPS_SYN_SENT);
1179         tp->iss = tcp_new_isn(tp);
1180         tcp_sendseqinit(tp);
1181
1182         return 0;
1183
1184 out:
1185         INP_HASH_WUNLOCK(&V_tcbinfo);
1186         return (error);
1187 }
1188 #endif /* INET */
1189
1190 #ifdef INET6
1191 static int
1192 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1193 {
1194         struct inpcb *inp = tp->t_inpcb, *oinp;
1195         struct socket *so = inp->inp_socket;
1196         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1197         struct in6_addr addr6;
1198         int error;
1199
1200         INP_WLOCK_ASSERT(inp);
1201         INP_HASH_WLOCK(&V_tcbinfo);
1202
1203         if (inp->inp_lport == 0) {
1204                 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1205                 if (error)
1206                         goto out;
1207         }
1208
1209         /*
1210          * Cannot simply call in_pcbconnect, because there might be an
1211          * earlier incarnation of this same connection still in
1212          * TIME_WAIT state, creating an ADDRINUSE error.
1213          * in6_pcbladdr() also handles scope zone IDs.
1214          *
1215          * XXXRW: We wouldn't need to expose in6_pcblookup_hash_locked()
1216          * outside of in6_pcb.c if there were an in6_pcbconnect_setup().
1217          */
1218         error = in6_pcbladdr(inp, nam, &addr6);
1219         if (error)
1220                 goto out;
1221         oinp = in6_pcblookup_hash_locked(inp->inp_pcbinfo,
1222                                   &sin6->sin6_addr, sin6->sin6_port,
1223                                   IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1224                                   ? &addr6
1225                                   : &inp->in6p_laddr,
1226                                   inp->inp_lport,  0, NULL);
1227         if (oinp) {
1228                 error = EADDRINUSE;
1229                 goto out;
1230         }
1231         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1232                 inp->in6p_laddr = addr6;
1233         inp->in6p_faddr = sin6->sin6_addr;
1234         inp->inp_fport = sin6->sin6_port;
1235         /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1236         inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
1237         if (inp->inp_flags & IN6P_AUTOFLOWLABEL)
1238                 inp->inp_flow |=
1239                     (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1240         in_pcbrehash(inp);
1241         INP_HASH_WUNLOCK(&V_tcbinfo);
1242
1243         /* Compute window scaling to request.  */
1244         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1245             (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1246                 tp->request_r_scale++;
1247
1248         soisconnecting(so);
1249         TCPSTAT_INC(tcps_connattempt);
1250         tcp_state_change(tp, TCPS_SYN_SENT);
1251         tp->iss = tcp_new_isn(tp);
1252         tcp_sendseqinit(tp);
1253
1254         return 0;
1255
1256 out:
1257         INP_HASH_WUNLOCK(&V_tcbinfo);
1258         return error;
1259 }
1260 #endif /* INET6 */
1261
1262 /*
1263  * Export TCP internal state information via a struct tcp_info, based on the
1264  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1265  * (TCP state machine, etc).  We export all information using FreeBSD-native
1266  * constants -- for example, the numeric values for tcpi_state will differ
1267  * from Linux.
1268  */
1269 static void
1270 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1271 {
1272
1273         INP_WLOCK_ASSERT(tp->t_inpcb);
1274         bzero(ti, sizeof(*ti));
1275
1276         ti->tcpi_state = tp->t_state;
1277         if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1278                 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1279         if (tp->t_flags & TF_SACK_PERMIT)
1280                 ti->tcpi_options |= TCPI_OPT_SACK;
1281         if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1282                 ti->tcpi_options |= TCPI_OPT_WSCALE;
1283                 ti->tcpi_snd_wscale = tp->snd_scale;
1284                 ti->tcpi_rcv_wscale = tp->rcv_scale;
1285         }
1286
1287         ti->tcpi_rto = tp->t_rxtcur * tick;
1288         ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1289         ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1290         ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1291
1292         ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1293         ti->tcpi_snd_cwnd = tp->snd_cwnd;
1294
1295         /*
1296          * FreeBSD-specific extension fields for tcp_info.
1297          */
1298         ti->tcpi_rcv_space = tp->rcv_wnd;
1299         ti->tcpi_rcv_nxt = tp->rcv_nxt;
1300         ti->tcpi_snd_wnd = tp->snd_wnd;
1301         ti->tcpi_snd_bwnd = 0;          /* Unused, kept for compat. */
1302         ti->tcpi_snd_nxt = tp->snd_nxt;
1303         ti->tcpi_snd_mss = tp->t_maxseg;
1304         ti->tcpi_rcv_mss = tp->t_maxseg;
1305         if (tp->t_flags & TF_TOE)
1306                 ti->tcpi_options |= TCPI_OPT_TOE;
1307         ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1308         ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1309         ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1310 }
1311
1312 /*
1313  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1314  * socket option arguments.  When it re-acquires the lock after the copy, it
1315  * has to revalidate that the connection is still valid for the socket
1316  * option.
1317  */
1318 #define INP_WLOCK_RECHECK(inp) do {                                     \
1319         INP_WLOCK(inp);                                                 \
1320         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {            \
1321                 INP_WUNLOCK(inp);                                       \
1322                 return (ECONNRESET);                                    \
1323         }                                                               \
1324         tp = intotcpcb(inp);                                            \
1325 } while(0)
1326
1327 int
1328 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1329 {
1330         int     error, opt, optval;
1331         u_int   ui;
1332         struct  inpcb *inp;
1333         struct  tcpcb *tp;
1334         struct  tcp_info ti;
1335         char buf[TCP_CA_NAME_MAX];
1336         struct cc_algo *algo;
1337
1338         error = 0;
1339         inp = sotoinpcb(so);
1340         KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1341         INP_WLOCK(inp);
1342         if (sopt->sopt_level != IPPROTO_TCP) {
1343 #ifdef INET6
1344                 if (inp->inp_vflag & INP_IPV6PROTO) {
1345                         INP_WUNLOCK(inp);
1346                         error = ip6_ctloutput(so, sopt);
1347                 }
1348 #endif /* INET6 */
1349 #if defined(INET6) && defined(INET)
1350                 else
1351 #endif
1352 #ifdef INET
1353                 {
1354                         INP_WUNLOCK(inp);
1355                         error = ip_ctloutput(so, sopt);
1356                 }
1357 #endif
1358                 return (error);
1359         }
1360         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1361                 INP_WUNLOCK(inp);
1362                 return (ECONNRESET);
1363         }
1364
1365         switch (sopt->sopt_dir) {
1366         case SOPT_SET:
1367                 switch (sopt->sopt_name) {
1368 #ifdef TCP_SIGNATURE
1369                 case TCP_MD5SIG:
1370                         INP_WUNLOCK(inp);
1371                         error = sooptcopyin(sopt, &optval, sizeof optval,
1372                             sizeof optval);
1373                         if (error)
1374                                 return (error);
1375
1376                         INP_WLOCK_RECHECK(inp);
1377                         if (optval > 0)
1378                                 tp->t_flags |= TF_SIGNATURE;
1379                         else
1380                                 tp->t_flags &= ~TF_SIGNATURE;
1381                         goto unlock_and_done;
1382 #endif /* TCP_SIGNATURE */
1383
1384                 case TCP_NODELAY:
1385                 case TCP_NOOPT:
1386                         INP_WUNLOCK(inp);
1387                         error = sooptcopyin(sopt, &optval, sizeof optval,
1388                             sizeof optval);
1389                         if (error)
1390                                 return (error);
1391
1392                         INP_WLOCK_RECHECK(inp);
1393                         switch (sopt->sopt_name) {
1394                         case TCP_NODELAY:
1395                                 opt = TF_NODELAY;
1396                                 break;
1397                         case TCP_NOOPT:
1398                                 opt = TF_NOOPT;
1399                                 break;
1400                         default:
1401                                 opt = 0; /* dead code to fool gcc */
1402                                 break;
1403                         }
1404
1405                         if (optval)
1406                                 tp->t_flags |= opt;
1407                         else
1408                                 tp->t_flags &= ~opt;
1409 unlock_and_done:
1410 #ifdef TCP_OFFLOAD
1411                         if (tp->t_flags & TF_TOE) {
1412                                 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1413                                     sopt->sopt_name);
1414                         }
1415 #endif
1416                         INP_WUNLOCK(inp);
1417                         break;
1418
1419                 case TCP_NOPUSH:
1420                         INP_WUNLOCK(inp);
1421                         error = sooptcopyin(sopt, &optval, sizeof optval,
1422                             sizeof optval);
1423                         if (error)
1424                                 return (error);
1425
1426                         INP_WLOCK_RECHECK(inp);
1427                         if (optval)
1428                                 tp->t_flags |= TF_NOPUSH;
1429                         else if (tp->t_flags & TF_NOPUSH) {
1430                                 tp->t_flags &= ~TF_NOPUSH;
1431                                 if (TCPS_HAVEESTABLISHED(tp->t_state))
1432                                         error = tcp_output(tp);
1433                         }
1434                         goto unlock_and_done;
1435
1436                 case TCP_MAXSEG:
1437                         INP_WUNLOCK(inp);
1438                         error = sooptcopyin(sopt, &optval, sizeof optval,
1439                             sizeof optval);
1440                         if (error)
1441                                 return (error);
1442
1443                         INP_WLOCK_RECHECK(inp);
1444                         if (optval > 0 && optval <= tp->t_maxseg &&
1445                             optval + 40 >= V_tcp_minmss)
1446                                 tp->t_maxseg = optval;
1447                         else
1448                                 error = EINVAL;
1449                         goto unlock_and_done;
1450
1451                 case TCP_INFO:
1452                         INP_WUNLOCK(inp);
1453                         error = EINVAL;
1454                         break;
1455
1456                 case TCP_CONGESTION:
1457                         INP_WUNLOCK(inp);
1458                         bzero(buf, sizeof(buf));
1459                         error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1460                         if (error)
1461                                 break;
1462                         INP_WLOCK_RECHECK(inp);
1463                         /*
1464                          * Return EINVAL if we can't find the requested cc algo.
1465                          */
1466                         error = EINVAL;
1467                         CC_LIST_RLOCK();
1468                         STAILQ_FOREACH(algo, &cc_list, entries) {
1469                                 if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1470                                     == 0) {
1471                                         /* We've found the requested algo. */
1472                                         error = 0;
1473                                         /*
1474                                          * We hold a write lock over the tcb
1475                                          * so it's safe to do these things
1476                                          * without ordering concerns.
1477                                          */
1478                                         if (CC_ALGO(tp)->cb_destroy != NULL)
1479                                                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1480                                         CC_ALGO(tp) = algo;
1481                                         /*
1482                                          * If something goes pear shaped
1483                                          * initialising the new algo,
1484                                          * fall back to newreno (which
1485                                          * does not require initialisation).
1486                                          */
1487                                         if (algo->cb_init != NULL)
1488                                                 if (algo->cb_init(tp->ccv) > 0) {
1489                                                         CC_ALGO(tp) = &newreno_cc_algo;
1490                                                         /*
1491                                                          * The only reason init
1492                                                          * should fail is
1493                                                          * because of malloc.
1494                                                          */
1495                                                         error = ENOMEM;
1496                                                 }
1497                                         break; /* Break the STAILQ_FOREACH. */
1498                                 }
1499                         }
1500                         CC_LIST_RUNLOCK();
1501                         goto unlock_and_done;
1502
1503                 case TCP_KEEPIDLE:
1504                 case TCP_KEEPINTVL:
1505                 case TCP_KEEPINIT:
1506                         INP_WUNLOCK(inp);
1507                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1508                         if (error)
1509                                 return (error);
1510
1511                         if (ui > (UINT_MAX / hz)) {
1512                                 error = EINVAL;
1513                                 break;
1514                         }
1515                         ui *= hz;
1516
1517                         INP_WLOCK_RECHECK(inp);
1518                         switch (sopt->sopt_name) {
1519                         case TCP_KEEPIDLE:
1520                                 tp->t_keepidle = ui;
1521                                 /*
1522                                  * XXX: better check current remaining
1523                                  * timeout and "merge" it with new value.
1524                                  */
1525                                 if ((tp->t_state > TCPS_LISTEN) &&
1526                                     (tp->t_state <= TCPS_CLOSING))
1527                                         tcp_timer_activate(tp, TT_KEEP,
1528                                             TP_KEEPIDLE(tp));
1529                                 break;
1530                         case TCP_KEEPINTVL:
1531                                 tp->t_keepintvl = ui;
1532                                 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1533                                     (TP_MAXIDLE(tp) > 0))
1534                                         tcp_timer_activate(tp, TT_2MSL,
1535                                             TP_MAXIDLE(tp));
1536                                 break;
1537                         case TCP_KEEPINIT:
1538                                 tp->t_keepinit = ui;
1539                                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1540                                     tp->t_state == TCPS_SYN_SENT)
1541                                         tcp_timer_activate(tp, TT_KEEP,
1542                                             TP_KEEPINIT(tp));
1543                                 break;
1544                         }
1545                         goto unlock_and_done;
1546
1547                 case TCP_KEEPCNT:
1548                         INP_WUNLOCK(inp);
1549                         error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1550                         if (error)
1551                                 return (error);
1552
1553                         INP_WLOCK_RECHECK(inp);
1554                         tp->t_keepcnt = ui;
1555                         if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1556                             (TP_MAXIDLE(tp) > 0))
1557                                 tcp_timer_activate(tp, TT_2MSL,
1558                                     TP_MAXIDLE(tp));
1559                         goto unlock_and_done;
1560
1561                 default:
1562                         INP_WUNLOCK(inp);
1563                         error = ENOPROTOOPT;
1564                         break;
1565                 }
1566                 break;
1567
1568         case SOPT_GET:
1569                 tp = intotcpcb(inp);
1570                 switch (sopt->sopt_name) {
1571 #ifdef TCP_SIGNATURE
1572                 case TCP_MD5SIG:
1573                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1574                         INP_WUNLOCK(inp);
1575                         error = sooptcopyout(sopt, &optval, sizeof optval);
1576                         break;
1577 #endif
1578
1579                 case TCP_NODELAY:
1580                         optval = tp->t_flags & TF_NODELAY;
1581                         INP_WUNLOCK(inp);
1582                         error = sooptcopyout(sopt, &optval, sizeof optval);
1583                         break;
1584                 case TCP_MAXSEG:
1585                         optval = tp->t_maxseg;
1586                         INP_WUNLOCK(inp);
1587                         error = sooptcopyout(sopt, &optval, sizeof optval);
1588                         break;
1589                 case TCP_NOOPT:
1590                         optval = tp->t_flags & TF_NOOPT;
1591                         INP_WUNLOCK(inp);
1592                         error = sooptcopyout(sopt, &optval, sizeof optval);
1593                         break;
1594                 case TCP_NOPUSH:
1595                         optval = tp->t_flags & TF_NOPUSH;
1596                         INP_WUNLOCK(inp);
1597                         error = sooptcopyout(sopt, &optval, sizeof optval);
1598                         break;
1599                 case TCP_INFO:
1600                         tcp_fill_info(tp, &ti);
1601                         INP_WUNLOCK(inp);
1602                         error = sooptcopyout(sopt, &ti, sizeof ti);
1603                         break;
1604                 case TCP_CONGESTION:
1605                         bzero(buf, sizeof(buf));
1606                         strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1607                         INP_WUNLOCK(inp);
1608                         error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1609                         break;
1610                 case TCP_KEEPIDLE:
1611                 case TCP_KEEPINTVL:
1612                 case TCP_KEEPINIT:
1613                 case TCP_KEEPCNT:
1614                         switch (sopt->sopt_name) {
1615                         case TCP_KEEPIDLE:
1616                                 ui = tp->t_keepidle / hz;
1617                                 break;
1618                         case TCP_KEEPINTVL:
1619                                 ui = tp->t_keepintvl / hz;
1620                                 break;
1621                         case TCP_KEEPINIT:
1622                                 ui = tp->t_keepinit / hz;
1623                                 break;
1624                         case TCP_KEEPCNT:
1625                                 ui = tp->t_keepcnt;
1626                                 break;
1627                         }
1628                         INP_WUNLOCK(inp);
1629                         error = sooptcopyout(sopt, &ui, sizeof(ui));
1630                         break;
1631                 default:
1632                         INP_WUNLOCK(inp);
1633                         error = ENOPROTOOPT;
1634                         break;
1635                 }
1636                 break;
1637         }
1638         return (error);
1639 }
1640 #undef INP_WLOCK_RECHECK
1641
1642 /*
1643  * Attach TCP protocol to socket, allocating
1644  * internet protocol control block, tcp control block,
1645  * bufer space, and entering LISTEN state if to accept connections.
1646  */
1647 static int
1648 tcp_attach(struct socket *so)
1649 {
1650         struct tcpcb *tp;
1651         struct inpcb *inp;
1652         int error;
1653
1654         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1655                 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1656                 if (error)
1657                         return (error);
1658         }
1659         so->so_rcv.sb_flags |= SB_AUTOSIZE;
1660         so->so_snd.sb_flags |= SB_AUTOSIZE;
1661         INP_INFO_WLOCK(&V_tcbinfo);
1662         error = in_pcballoc(so, &V_tcbinfo);
1663         if (error) {
1664                 INP_INFO_WUNLOCK(&V_tcbinfo);
1665                 return (error);
1666         }
1667         inp = sotoinpcb(so);
1668 #ifdef INET6
1669         if (inp->inp_vflag & INP_IPV6PROTO) {
1670                 inp->inp_vflag |= INP_IPV6;
1671                 inp->in6p_hops = -1;    /* use kernel default */
1672         }
1673         else
1674 #endif
1675         inp->inp_vflag |= INP_IPV4;
1676         tp = tcp_newtcpcb(inp);
1677         if (tp == NULL) {
1678                 in_pcbdetach(inp);
1679                 in_pcbfree(inp);
1680                 INP_INFO_WUNLOCK(&V_tcbinfo);
1681                 return (ENOBUFS);
1682         }
1683         tp->t_state = TCPS_CLOSED;
1684         INP_WUNLOCK(inp);
1685         INP_INFO_WUNLOCK(&V_tcbinfo);
1686         return (0);
1687 }
1688
1689 /*
1690  * Initiate (or continue) disconnect.
1691  * If embryonic state, just send reset (once).
1692  * If in ``let data drain'' option and linger null, just drop.
1693  * Otherwise (hard), mark socket disconnecting and drop
1694  * current input data; switch states based on user close, and
1695  * send segment to peer (with FIN).
1696  */
1697 static void
1698 tcp_disconnect(struct tcpcb *tp)
1699 {
1700         struct inpcb *inp = tp->t_inpcb;
1701         struct socket *so = inp->inp_socket;
1702
1703         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1704         INP_WLOCK_ASSERT(inp);
1705
1706         /*
1707          * Neither tcp_close() nor tcp_drop() should return NULL, as the
1708          * socket is still open.
1709          */
1710         if (tp->t_state < TCPS_ESTABLISHED) {
1711                 tp = tcp_close(tp);
1712                 KASSERT(tp != NULL,
1713                     ("tcp_disconnect: tcp_close() returned NULL"));
1714         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1715                 tp = tcp_drop(tp, 0);
1716                 KASSERT(tp != NULL,
1717                     ("tcp_disconnect: tcp_drop() returned NULL"));
1718         } else {
1719                 soisdisconnecting(so);
1720                 sbflush(&so->so_rcv);
1721                 tcp_usrclosed(tp);
1722                 if (!(inp->inp_flags & INP_DROPPED))
1723                         tcp_output(tp);
1724         }
1725 }
1726
1727 /*
1728  * User issued close, and wish to trail through shutdown states:
1729  * if never received SYN, just forget it.  If got a SYN from peer,
1730  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1731  * If already got a FIN from peer, then almost done; go to LAST_ACK
1732  * state.  In all other cases, have already sent FIN to peer (e.g.
1733  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1734  * for peer to send FIN or not respond to keep-alives, etc.
1735  * We can let the user exit from the close as soon as the FIN is acked.
1736  */
1737 static void
1738 tcp_usrclosed(struct tcpcb *tp)
1739 {
1740
1741         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1742         INP_WLOCK_ASSERT(tp->t_inpcb);
1743
1744         switch (tp->t_state) {
1745         case TCPS_LISTEN:
1746 #ifdef TCP_OFFLOAD
1747                 tcp_offload_listen_stop(tp);
1748 #endif
1749                 /* FALLTHROUGH */
1750         case TCPS_CLOSED:
1751                 tcp_state_change(tp, TCPS_CLOSED);
1752                 tp = tcp_close(tp);
1753                 /*
1754                  * tcp_close() should never return NULL here as the socket is
1755                  * still open.
1756                  */
1757                 KASSERT(tp != NULL,
1758                     ("tcp_usrclosed: tcp_close() returned NULL"));
1759                 break;
1760
1761         case TCPS_SYN_SENT:
1762         case TCPS_SYN_RECEIVED:
1763                 tp->t_flags |= TF_NEEDFIN;
1764                 break;
1765
1766         case TCPS_ESTABLISHED:
1767                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1768                 break;
1769
1770         case TCPS_CLOSE_WAIT:
1771                 tcp_state_change(tp, TCPS_LAST_ACK);
1772                 break;
1773         }
1774         if (tp->t_state >= TCPS_FIN_WAIT_2) {
1775                 soisdisconnected(tp->t_inpcb->inp_socket);
1776                 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1777                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1778                         int timeout;
1779
1780                         timeout = (tcp_fast_finwait2_recycle) ?
1781                             tcp_finwait2_timeout : TP_MAXIDLE(tp);
1782                         tcp_timer_activate(tp, TT_2MSL, timeout);
1783                 }
1784         }
1785 }
1786
1787 #ifdef DDB
1788 static void
1789 db_print_indent(int indent)
1790 {
1791         int i;
1792
1793         for (i = 0; i < indent; i++)
1794                 db_printf(" ");
1795 }
1796
1797 static void
1798 db_print_tstate(int t_state)
1799 {
1800
1801         switch (t_state) {
1802         case TCPS_CLOSED:
1803                 db_printf("TCPS_CLOSED");
1804                 return;
1805
1806         case TCPS_LISTEN:
1807                 db_printf("TCPS_LISTEN");
1808                 return;
1809
1810         case TCPS_SYN_SENT:
1811                 db_printf("TCPS_SYN_SENT");
1812                 return;
1813
1814         case TCPS_SYN_RECEIVED:
1815                 db_printf("TCPS_SYN_RECEIVED");
1816                 return;
1817
1818         case TCPS_ESTABLISHED:
1819                 db_printf("TCPS_ESTABLISHED");
1820                 return;
1821
1822         case TCPS_CLOSE_WAIT:
1823                 db_printf("TCPS_CLOSE_WAIT");
1824                 return;
1825
1826         case TCPS_FIN_WAIT_1:
1827                 db_printf("TCPS_FIN_WAIT_1");
1828                 return;
1829
1830         case TCPS_CLOSING:
1831                 db_printf("TCPS_CLOSING");
1832                 return;
1833
1834         case TCPS_LAST_ACK:
1835                 db_printf("TCPS_LAST_ACK");
1836                 return;
1837
1838         case TCPS_FIN_WAIT_2:
1839                 db_printf("TCPS_FIN_WAIT_2");
1840                 return;
1841
1842         case TCPS_TIME_WAIT:
1843                 db_printf("TCPS_TIME_WAIT");
1844                 return;
1845
1846         default:
1847                 db_printf("unknown");
1848                 return;
1849         }
1850 }
1851
1852 static void
1853 db_print_tflags(u_int t_flags)
1854 {
1855         int comma;
1856
1857         comma = 0;
1858         if (t_flags & TF_ACKNOW) {
1859                 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1860                 comma = 1;
1861         }
1862         if (t_flags & TF_DELACK) {
1863                 db_printf("%sTF_DELACK", comma ? ", " : "");
1864                 comma = 1;
1865         }
1866         if (t_flags & TF_NODELAY) {
1867                 db_printf("%sTF_NODELAY", comma ? ", " : "");
1868                 comma = 1;
1869         }
1870         if (t_flags & TF_NOOPT) {
1871                 db_printf("%sTF_NOOPT", comma ? ", " : "");
1872                 comma = 1;
1873         }
1874         if (t_flags & TF_SENTFIN) {
1875                 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1876                 comma = 1;
1877         }
1878         if (t_flags & TF_REQ_SCALE) {
1879                 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1880                 comma = 1;
1881         }
1882         if (t_flags & TF_RCVD_SCALE) {
1883                 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1884                 comma = 1;
1885         }
1886         if (t_flags & TF_REQ_TSTMP) {
1887                 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1888                 comma = 1;
1889         }
1890         if (t_flags & TF_RCVD_TSTMP) {
1891                 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1892                 comma = 1;
1893         }
1894         if (t_flags & TF_SACK_PERMIT) {
1895                 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1896                 comma = 1;
1897         }
1898         if (t_flags & TF_NEEDSYN) {
1899                 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1900                 comma = 1;
1901         }
1902         if (t_flags & TF_NEEDFIN) {
1903                 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1904                 comma = 1;
1905         }
1906         if (t_flags & TF_NOPUSH) {
1907                 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1908                 comma = 1;
1909         }
1910         if (t_flags & TF_MORETOCOME) {
1911                 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1912                 comma = 1;
1913         }
1914         if (t_flags & TF_LQ_OVERFLOW) {
1915                 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1916                 comma = 1;
1917         }
1918         if (t_flags & TF_LASTIDLE) {
1919                 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1920                 comma = 1;
1921         }
1922         if (t_flags & TF_RXWIN0SENT) {
1923                 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1924                 comma = 1;
1925         }
1926         if (t_flags & TF_FASTRECOVERY) {
1927                 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1928                 comma = 1;
1929         }
1930         if (t_flags & TF_CONGRECOVERY) {
1931                 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
1932                 comma = 1;
1933         }
1934         if (t_flags & TF_WASFRECOVERY) {
1935                 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
1936                 comma = 1;
1937         }
1938         if (t_flags & TF_SIGNATURE) {
1939                 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
1940                 comma = 1;
1941         }
1942         if (t_flags & TF_FORCEDATA) {
1943                 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
1944                 comma = 1;
1945         }
1946         if (t_flags & TF_TSO) {
1947                 db_printf("%sTF_TSO", comma ? ", " : "");
1948                 comma = 1;
1949         }
1950         if (t_flags & TF_ECN_PERMIT) {
1951                 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
1952                 comma = 1;
1953         }
1954 }
1955
1956 static void
1957 db_print_toobflags(char t_oobflags)
1958 {
1959         int comma;
1960
1961         comma = 0;
1962         if (t_oobflags & TCPOOB_HAVEDATA) {
1963                 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
1964                 comma = 1;
1965         }
1966         if (t_oobflags & TCPOOB_HADDATA) {
1967                 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
1968                 comma = 1;
1969         }
1970 }
1971
1972 static void
1973 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
1974 {
1975
1976         db_print_indent(indent);
1977         db_printf("%s at %p\n", name, tp);
1978
1979         indent += 2;
1980
1981         db_print_indent(indent);
1982         db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
1983            LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
1984
1985         db_print_indent(indent);
1986         db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
1987             &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
1988
1989         db_print_indent(indent);
1990         db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
1991             &tp->t_timers->tt_delack, tp->t_inpcb);
1992
1993         db_print_indent(indent);
1994         db_printf("t_state: %d (", tp->t_state);
1995         db_print_tstate(tp->t_state);
1996         db_printf(")\n");
1997
1998         db_print_indent(indent);
1999         db_printf("t_flags: 0x%x (", tp->t_flags);
2000         db_print_tflags(tp->t_flags);
2001         db_printf(")\n");
2002
2003         db_print_indent(indent);
2004         db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2005             tp->snd_una, tp->snd_max, tp->snd_nxt);
2006
2007         db_print_indent(indent);
2008         db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2009            tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2010
2011         db_print_indent(indent);
2012         db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2013             tp->iss, tp->irs, tp->rcv_nxt);
2014
2015         db_print_indent(indent);
2016         db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2017             tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2018
2019         db_print_indent(indent);
2020         db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2021            tp->snd_wnd, tp->snd_cwnd);
2022
2023         db_print_indent(indent);
2024         db_printf("snd_ssthresh: %lu   snd_recover: "
2025             "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2026
2027         db_print_indent(indent);
2028         db_printf("t_maxopd: %u   t_rcvtime: %u   t_startime: %u\n",
2029             tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
2030
2031         db_print_indent(indent);
2032         db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2033             tp->t_rtttime, tp->t_rtseq);
2034
2035         db_print_indent(indent);
2036         db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2037             tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2038
2039         db_print_indent(indent);
2040         db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2041             "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2042             tp->t_rttbest);
2043
2044         db_print_indent(indent);
2045         db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2046             tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2047
2048         db_print_indent(indent);
2049         db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2050         db_print_toobflags(tp->t_oobflags);
2051         db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2052
2053         db_print_indent(indent);
2054         db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2055             tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2056
2057         db_print_indent(indent);
2058         db_printf("ts_recent: %u   ts_recent_age: %u\n",
2059             tp->ts_recent, tp->ts_recent_age);
2060
2061         db_print_indent(indent);
2062         db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2063             "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2064
2065         db_print_indent(indent);
2066         db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2067             "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2068             tp->snd_recover_prev, tp->t_badrxtwin);
2069
2070         db_print_indent(indent);
2071         db_printf("snd_numholes: %d  snd_holes first: %p\n",
2072             tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2073
2074         db_print_indent(indent);
2075         db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2076             "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2077
2078         /* Skip sackblks, sackhint. */
2079
2080         db_print_indent(indent);
2081         db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2082             tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2083 }
2084
2085 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2086 {
2087         struct tcpcb *tp;
2088
2089         if (!have_addr) {
2090                 db_printf("usage: show tcpcb <addr>\n");
2091                 return;
2092         }
2093         tp = (struct tcpcb *)addr;
2094
2095         db_print_tcpcb(tp, "tcpcb", 0);
2096 }
2097 #endif