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