sys/netinet/tcp_usrreq.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  64  */
  65
  66 #include "opt_ipsec.h"
  67 #include "opt_inet.h"
  68 #include "opt_inet6.h"
  69 #include "opt_tcpdebug.h"
  70
  71 #include <sys/param.h>
  72 #include <sys/systm.h>
  73 #include <sys/kernel.h>
  74 #include <sys/malloc.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/globaldata.h>
  77 #include <sys/thread.h>
  78
  79 #include <sys/mbuf.h>
  80 #ifdef INET6
  81 #include <sys/domain.h>
  82 #endif /* INET6 */
  83 #include <sys/socket.h>
  84 #include <sys/socketvar.h>
  85 #include <sys/socketops.h>
  86 #include <sys/protosw.h>
  87
  88 #include <sys/thread2.h>
  89 #include <sys/msgport2.h>
  90 #include <sys/socketvar2.h>
  91
  92 #include <net/if.h>
  93 #include <net/netisr.h>
  94 #include <net/route.h>
  95
  96 #include <net/netmsg2.h>
  97 #include <net/netisr2.h>
  98
  99 #include <netinet/in.h>
 100 #include <netinet/in_systm.h>
 101 #ifdef INET6
 102 #include <netinet/ip6.h>
 103 #endif
 104 #include <netinet/in_pcb.h>
 105 #ifdef INET6
 106 #include <netinet6/in6_pcb.h>
 107 #endif
 108 #include <netinet/in_var.h>
 109 #include <netinet/ip_var.h>
 110 #ifdef INET6
 111 #include <netinet6/ip6_var.h>
 112 #include <netinet6/tcp6_var.h>
 113 #endif
 114 #include <netinet/tcp.h>
 115 #include <netinet/tcp_fsm.h>
 116 #include <netinet/tcp_seq.h>
 117 #include <netinet/tcp_timer.h>
 118 #include <netinet/tcp_timer2.h>
 119 #include <netinet/tcp_var.h>
 120 #include <netinet/tcpip.h>
 121 #ifdef TCPDEBUG
 122 #include <netinet/tcp_debug.h>
 123 #endif
 124
 125 #ifdef IPSEC
 126 #include <netinet6/ipsec.h>
 127 #endif /*IPSEC*/
 128
 129 /*
 130  * TCP protocol interface to socket abstraction.
 131  */
 132 extern  char *tcpstates[];      /* XXX ??? */
 133
 134 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 135 static void     tcp_connect (netmsg_t msg);
 136 #ifdef INET6
 137 static void     tcp6_connect (netmsg_t msg);
 138 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 139                                 struct mbuf **mp,
 140                                 struct sockaddr_in6 *sin6,
 141                                 struct in6_addr *addr6);
 142 #endif /* INET6 */
 143 static struct tcpcb *
 144                 tcp_disconnect (struct tcpcb *);
 145 static struct tcpcb *
 146                 tcp_usrclosed (struct tcpcb *);
 147
 148 #ifdef TCPDEBUG
 149 #define TCPDEBUG0       int ostate = 0
 150 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 151 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 152                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 153 #else
 154 #define TCPDEBUG0
 155 #define TCPDEBUG1()
 156 #define TCPDEBUG2(req)
 157 #endif
 158
 159 static int      tcp_lport_extension = 1;
 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 161     &tcp_lport_extension, 0, "");
 162
 163 /*
 164  * For some ill optimized programs, which try to use TCP_NOPUSH
 165  * to improve performance, will have small amount of data sits
 166  * in the sending buffer.  These small amount of data will _not_
 167  * be pushed into the network until more data are written into
 168  * the socket or the socket write side is shutdown.
 169  */
 170 static int      tcp_disable_nopush = 1;
 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
 172     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
 173
 174 /*
 175  * TCP attaches to socket via pru_attach(), reserving space,
 176  * and an internet control block.  This is likely occuring on
 177  * cpu0 and may have to move later when we bind/connect.
 178  */
 179 static void
 180 tcp_usr_attach(netmsg_t msg)
 181 {
 182         struct socket *so = msg->base.nm_so;
 183         struct pru_attach_info *ai = msg->attach.nm_ai;
 184         int error;
 185         struct inpcb *inp;
 186         struct tcpcb *tp = NULL;
 187         TCPDEBUG0;
 188
 189         soreference(so);
 190         inp = so->so_pcb;
 191         TCPDEBUG1();
 192         if (inp) {
 193                 error = EISCONN;
 194                 goto out;
 195         }
 196
 197         error = tcp_attach(so, ai);
 198         if (error)
 199                 goto out;
 200
 201         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 202                 so->so_linger = TCP_LINGERTIME;
 203         tp = sototcpcb(so);
 204 out:
 205         sofree(so);             /* from ref above */
 206         TCPDEBUG2(PRU_ATTACH);
 207         lwkt_replymsg(&msg->lmsg, error);
 208 }
 209
 210 /*
 211  * pru_detach() detaches the TCP protocol from the socket.
 212  * If the protocol state is non-embryonic, then can't
 213  * do this directly: have to initiate a pru_disconnect(),
 214  * which may finish later; embryonic TCB's can just
 215  * be discarded here.
 216  */
 217 static void
 218 tcp_usr_detach(netmsg_t msg)
 219 {
 220         struct socket *so = msg->base.nm_so;
 221         int error = 0;
 222         struct inpcb *inp;
 223         struct tcpcb *tp;
 224         TCPDEBUG0;
 225
 226         inp = so->so_pcb;
 227
 228         /*
 229          * If the inp is already detached it may have been due to an async
 230          * close.  Just return as if no error occured.
 231          *
 232          * It's possible for the tcpcb (tp) to disconnect from the inp due
 233          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 234          * the detach message has been queued so we may find a NULL tp here.
 235          */
 236         if (inp) {
 237                 if ((tp = intotcpcb(inp)) != NULL) {
 238                         TCPDEBUG1();
 239                         tp = tcp_disconnect(tp);
 240                         TCPDEBUG2(PRU_DETACH);
 241                 }
 242         }
 243         lwkt_replymsg(&msg->lmsg, error);
 244 }
 245
 246 /*
 247  * NOTE: ignore_error is non-zero for certain disconnection races
 248  * which we want to silently allow, otherwise close() may return
 249  * an unexpected error.
 250  *
 251  * NOTE: The variables (msg) and (tp) are assumed.
 252  */
 253 #define COMMON_START(so, inp, ignore_error)                     \
 254         TCPDEBUG0;                                              \
 255                                                                 \
 256         inp = so->so_pcb;                                       \
 257         do {                                                    \
 258                  if (inp == NULL) {                             \
 259                         error = ignore_error ? 0 : EINVAL;      \
 260                         tp = NULL;                              \
 261                         goto out;                               \
 262                  }                                              \
 263                  tp = intotcpcb(inp);                           \
 264                  TCPDEBUG1();                                   \
 265         } while(0)
 266
 267 #define COMMON_END1(req, noreply)                               \
 268         out: do {                                               \
 269                 TCPDEBUG2(req);                                 \
 270                 if (!(noreply))                                 \
 271                         lwkt_replymsg(&msg->lmsg, error);       \
 272                 return;                                         \
 273         } while(0)
 274
 275 #define COMMON_END(req)         COMMON_END1((req), 0)
 276
 277 /*
 278  * Give the socket an address.
 279  */
 280 static void
 281 tcp_usr_bind(netmsg_t msg)
 282 {
 283         struct socket *so = msg->bind.base.nm_so;
 284         struct sockaddr *nam = msg->bind.nm_nam;
 285         struct thread *td = msg->bind.nm_td;
 286         int error = 0;
 287         struct inpcb *inp;
 288         struct tcpcb *tp;
 289         struct sockaddr_in *sinp;
 290
 291         COMMON_START(so, inp, 0);
 292
 293         /*
 294          * Must check for multicast addresses and disallow binding
 295          * to them.
 296          */
 297         sinp = (struct sockaddr_in *)nam;
 298         if (sinp->sin_family == AF_INET &&
 299             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 300                 error = EAFNOSUPPORT;
 301                 goto out;
 302         }
 303         error = in_pcbbind(inp, nam, td);
 304         if (error)
 305                 goto out;
 306         COMMON_END(PRU_BIND);
 307
 308 }
 309
 310 #ifdef INET6
 311
 312 static void
 313 tcp6_usr_bind(netmsg_t msg)
 314 {
 315         struct socket *so = msg->bind.base.nm_so;
 316         struct sockaddr *nam = msg->bind.nm_nam;
 317         struct thread *td = msg->bind.nm_td;
 318         int error = 0;
 319         struct inpcb *inp;
 320         struct tcpcb *tp;
 321         struct sockaddr_in6 *sin6p;
 322
 323         COMMON_START(so, inp, 0);
 324
 325         /*
 326          * Must check for multicast addresses and disallow binding
 327          * to them.
 328          */
 329         sin6p = (struct sockaddr_in6 *)nam;
 330         if (sin6p->sin6_family == AF_INET6 &&
 331             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 332                 error = EAFNOSUPPORT;
 333                 goto out;
 334         }
 335         inp->inp_vflag &= ~INP_IPV4;
 336         inp->inp_vflag |= INP_IPV6;
 337         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 338                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 339                         inp->inp_vflag |= INP_IPV4;
 340                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 341                         struct sockaddr_in sin;
 342
 343                         in6_sin6_2_sin(&sin, sin6p);
 344                         inp->inp_vflag |= INP_IPV4;
 345                         inp->inp_vflag &= ~INP_IPV6;
 346                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 347                         goto out;
 348                 }
 349         }
 350         error = in6_pcbbind(inp, nam, td);
 351         if (error)
 352                 goto out;
 353         COMMON_END(PRU_BIND);
 354 }
 355 #endif /* INET6 */
 356
 357 struct netmsg_inswildcard {
 358         struct netmsg_base      base;
 359         struct inpcb            *nm_inp;
 360 };
 361
 362 static void
 363 in_pcbinswildcardhash_handler(netmsg_t msg)
 364 {
 365         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 366         int cpu = mycpuid, nextcpu;
 367
 368         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 369
 370         nextcpu = cpu + 1;
 371         if (nextcpu < ncpus2)
 372                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 373         else
 374                 lwkt_replymsg(&nm->base.lmsg, 0);
 375 }
 376
 377 /*
 378  * Prepare to accept connections.
 379  */
 380 static void
 381 tcp_usr_listen(netmsg_t msg)
 382 {
 383         struct socket *so = msg->listen.base.nm_so;
 384         struct thread *td = msg->listen.nm_td;
 385         int error = 0;
 386         struct inpcb *inp;
 387         struct tcpcb *tp;
 388         struct netmsg_inswildcard nm;
 389
 390         COMMON_START(so, inp, 0);
 391
 392         if (tp->t_flags & TF_LISTEN)
 393                 goto out;
 394
 395         if (inp->inp_lport == 0) {
 396                 error = in_pcbbind(inp, NULL, td);
 397                 if (error)
 398                         goto out;
 399         }
 400
 401         tp->t_state = TCPS_LISTEN;
 402         tp->t_flags |= TF_LISTEN;
 403         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 404
 405         if (ncpus > 1) {
 406                 /*
 407                  * We have to set the flag because we can't have other cpus
 408                  * messing with our inp's flags.
 409                  */
 410                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 411                         ("already on connhash"));
 412                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 413                         ("already on wildcardhash"));
 414                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 415                         ("already on MP wildcardhash"));
 416                 inp->inp_flags |= INP_WILDCARD_MP;
 417
 418                 KKASSERT(so->so_port == netisr_cpuport(0));
 419                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 420                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 421
 422                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 423                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 424                 nm.nm_inp = inp;
 425                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 426         }
 427         in_pcbinswildcardhash(inp);
 428         COMMON_END(PRU_LISTEN);
 429 }
 430
 431 #ifdef INET6
 432
 433 static void
 434 tcp6_usr_listen(netmsg_t msg)
 435 {
 436         struct socket *so = msg->listen.base.nm_so;
 437         struct thread *td = msg->listen.nm_td;
 438         int error = 0;
 439         struct inpcb *inp;
 440         struct tcpcb *tp;
 441         struct netmsg_inswildcard nm;
 442
 443         COMMON_START(so, inp, 0);
 444
 445         if (tp->t_flags & TF_LISTEN)
 446                 goto out;
 447
 448         if (inp->inp_lport == 0) {
 449                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 450                         inp->inp_vflag |= INP_IPV4;
 451                 else
 452                         inp->inp_vflag &= ~INP_IPV4;
 453                 error = in6_pcbbind(inp, NULL, td);
 454                 if (error)
 455                         goto out;
 456         }
 457
 458         tp->t_state = TCPS_LISTEN;
 459         tp->t_flags |= TF_LISTEN;
 460         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 461
 462         if (ncpus > 1) {
 463                 /*
 464                  * We have to set the flag because we can't have other cpus
 465                  * messing with our inp's flags.
 466                  */
 467                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 468                         ("already on connhash"));
 469                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 470                         ("already on wildcardhash"));
 471                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 472                         ("already on MP wildcardhash"));
 473                 inp->inp_flags |= INP_WILDCARD_MP;
 474
 475                 KKASSERT(so->so_port == netisr_cpuport(0));
 476                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 477                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 478
 479                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 480                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 481                 nm.nm_inp = inp;
 482                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 483         }
 484         in_pcbinswildcardhash(inp);
 485         COMMON_END(PRU_LISTEN);
 486 }
 487 #endif /* INET6 */
 488
 489 /*
 490  * Initiate connection to peer.
 491  * Create a template for use in transmissions on this connection.
 492  * Enter SYN_SENT state, and mark socket as connecting.
 493  * Start keep-alive timer, and seed output sequence space.
 494  * Send initial segment on connection.
 495  */
 496 static void
 497 tcp_usr_connect(netmsg_t msg)
 498 {
 499         struct socket *so = msg->connect.base.nm_so;
 500         struct sockaddr *nam = msg->connect.nm_nam;
 501         struct thread *td = msg->connect.nm_td;
 502         int error = 0;
 503         struct inpcb *inp;
 504         struct tcpcb *tp;
 505         struct sockaddr_in *sinp;
 506
 507         COMMON_START(so, inp, 0);
 508
 509         /*
 510          * Must disallow TCP ``connections'' to multicast addresses.
 511          */
 512         sinp = (struct sockaddr_in *)nam;
 513         if (sinp->sin_family == AF_INET
 514             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 515                 error = EAFNOSUPPORT;
 516                 goto out;
 517         }
 518
 519         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 520                 error = EAFNOSUPPORT; /* IPv6 only jail */
 521                 goto out;
 522         }
 523
 524         tcp_connect(msg);
 525         /* msg is invalid now */
 526         return;
 527 out:
 528         if (msg->connect.nm_m) {
 529                 m_freem(msg->connect.nm_m);
 530                 msg->connect.nm_m = NULL;
 531         }
 532         if (msg->connect.nm_flags & PRUC_HELDTD)
 533                 lwkt_rele(td);
 534         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 535                 so->so_error = error;
 536                 soisdisconnected(so);
 537         }
 538         lwkt_replymsg(&msg->lmsg, error);
 539 }
 540
 541 #ifdef INET6
 542
 543 static void
 544 tcp6_usr_connect(netmsg_t msg)
 545 {
 546         struct socket *so = msg->connect.base.nm_so;
 547         struct sockaddr *nam = msg->connect.nm_nam;
 548         struct thread *td = msg->connect.nm_td;
 549         int error = 0;
 550         struct inpcb *inp;
 551         struct tcpcb *tp;
 552         struct sockaddr_in6 *sin6p;
 553
 554         COMMON_START(so, inp, 0);
 555
 556         /*
 557          * Must disallow TCP ``connections'' to multicast addresses.
 558          */
 559         sin6p = (struct sockaddr_in6 *)nam;
 560         if (sin6p->sin6_family == AF_INET6
 561             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 562                 error = EAFNOSUPPORT;
 563                 goto out;
 564         }
 565
 566         if (!prison_remote_ip(td, nam)) {
 567                 error = EAFNOSUPPORT; /* IPv4 only jail */
 568                 goto out;
 569         }
 570
 571         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 572                 struct sockaddr_in *sinp;
 573
 574                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 575                         error = EINVAL;
 576                         goto out;
 577                 }
 578                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 579                 in6_sin6_2_sin(sinp, sin6p);
 580                 inp->inp_vflag |= INP_IPV4;
 581                 inp->inp_vflag &= ~INP_IPV6;
 582                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 583                 msg->connect.nm_flags |= PRUC_NAMALLOC;
 584                 tcp_connect(msg);
 585                 /* msg is invalid now */
 586                 return;
 587         }
 588         inp->inp_vflag &= ~INP_IPV4;
 589         inp->inp_vflag |= INP_IPV6;
 590         inp->inp_inc.inc_isipv6 = 1;
 591
 592         msg->connect.nm_flags |= PRUC_FALLBACK;
 593         tcp6_connect(msg);
 594         /* msg is invalid now */
 595         return;
 596 out:
 597         if (msg->connect.nm_m) {
 598                 m_freem(msg->connect.nm_m);
 599                 msg->connect.nm_m = NULL;
 600         }
 601         lwkt_replymsg(&msg->lmsg, error);
 602 }
 603
 604 #endif /* INET6 */
 605
 606 /*
 607  * Initiate disconnect from peer.
 608  * If connection never passed embryonic stage, just drop;
 609  * else if don't need to let data drain, then can just drop anyways,
 610  * else have to begin TCP shutdown process: mark socket disconnecting,
 611  * drain unread data, state switch to reflect user close, and
 612  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 613  * when peer sends FIN and acks ours.
 614  *
 615  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 616  */
 617 static void
 618 tcp_usr_disconnect(netmsg_t msg)
 619 {
 620         struct socket *so = msg->disconnect.base.nm_so;
 621         int error = 0;
 622         struct inpcb *inp;
 623         struct tcpcb *tp;
 624
 625         COMMON_START(so, inp, 1);
 626         tp = tcp_disconnect(tp);
 627         COMMON_END(PRU_DISCONNECT);
 628 }
 629
 630 /*
 631  * Accept a connection.  Essentially all the work is
 632  * done at higher levels; just return the address
 633  * of the peer, storing through addr.
 634  */
 635 static void
 636 tcp_usr_accept(netmsg_t msg)
 637 {
 638         struct socket *so = msg->accept.base.nm_so;
 639         struct sockaddr **nam = msg->accept.nm_nam;
 640         int error = 0;
 641         struct inpcb *inp;
 642         struct tcpcb *tp = NULL;
 643         TCPDEBUG0;
 644
 645         inp = so->so_pcb;
 646         if (so->so_state & SS_ISDISCONNECTED) {
 647                 error = ECONNABORTED;
 648                 goto out;
 649         }
 650         if (inp == 0) {
 651                 error = EINVAL;
 652                 goto out;
 653         }
 654
 655         tp = intotcpcb(inp);
 656         TCPDEBUG1();
 657         in_setpeeraddr(so, nam);
 658         COMMON_END(PRU_ACCEPT);
 659 }
 660
 661 #ifdef INET6
 662 static void
 663 tcp6_usr_accept(netmsg_t msg)
 664 {
 665         struct socket *so = msg->accept.base.nm_so;
 666         struct sockaddr **nam = msg->accept.nm_nam;
 667         int error = 0;
 668         struct inpcb *inp;
 669         struct tcpcb *tp = NULL;
 670         TCPDEBUG0;
 671
 672         inp = so->so_pcb;
 673
 674         if (so->so_state & SS_ISDISCONNECTED) {
 675                 error = ECONNABORTED;
 676                 goto out;
 677         }
 678         if (inp == 0) {
 679                 error = EINVAL;
 680                 goto out;
 681         }
 682         tp = intotcpcb(inp);
 683         TCPDEBUG1();
 684         in6_mapped_peeraddr(so, nam);
 685         COMMON_END(PRU_ACCEPT);
 686 }
 687 #endif /* INET6 */
 688 /*
 689  * Mark the connection as being incapable of further output.
 690  */
 691 static void
 692 tcp_usr_shutdown(netmsg_t msg)
 693 {
 694         struct socket *so = msg->shutdown.base.nm_so;
 695         int error = 0;
 696         struct inpcb *inp;
 697         struct tcpcb *tp;
 698
 699         COMMON_START(so, inp, 0);
 700         socantsendmore(so);
 701         tp = tcp_usrclosed(tp);
 702         if (tp)
 703                 error = tcp_output(tp);
 704         COMMON_END(PRU_SHUTDOWN);
 705 }
 706
 707 /*
 708  * After a receive, possibly send window update to peer.
 709  */
 710 static void
 711 tcp_usr_rcvd(netmsg_t msg)
 712 {
 713         struct socket *so = msg->rcvd.base.nm_so;
 714         int error = 0, noreply = 0;
 715         struct inpcb *inp;
 716         struct tcpcb *tp;
 717
 718         COMMON_START(so, inp, 0);
 719
 720         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 721                 noreply = 1;
 722                 so_async_rcvd_reply(so);
 723         }
 724         tcp_output(tp);
 725
 726         COMMON_END1(PRU_RCVD, noreply);
 727 }
 728
 729 /*
 730  * Do a send by putting data in output queue and updating urgent
 731  * marker if URG set.  Possibly send more data.  Unlike the other
 732  * pru_*() routines, the mbuf chains are our responsibility.  We
 733  * must either enqueue them or free them.  The other pru_* routines
 734  * generally are caller-frees.
 735  */
 736 static void
 737 tcp_usr_send(netmsg_t msg)
 738 {
 739         struct socket *so = msg->send.base.nm_so;
 740         int flags = msg->send.nm_flags;
 741         struct mbuf *m = msg->send.nm_m;
 742         int error = 0;
 743         struct inpcb *inp;
 744         struct tcpcb *tp;
 745         TCPDEBUG0;
 746
 747         KKASSERT(msg->send.nm_control == NULL);
 748         KKASSERT(msg->send.nm_addr == NULL);
 749         KKASSERT((flags & PRUS_FREEADDR) == 0);
 750
 751         inp = so->so_pcb;
 752
 753         if (inp == NULL) {
 754                 /*
 755                  * OOPS! we lost a race, the TCP session got reset after
 756                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 757                  * network interrupt in the non-critical section of sosend().
 758                  */
 759                 m_freem(m);
 760                 error = ECONNRESET;     /* XXX EPIPE? */
 761                 tp = NULL;
 762                 TCPDEBUG1();
 763                 goto out;
 764         }
 765         tp = intotcpcb(inp);
 766         TCPDEBUG1();
 767
 768 #ifdef foo
 769         /*
 770          * This is no longer necessary, since:
 771          * - sosendtcp() has already checked it for us
 772          * - It does not work with asynchronized send
 773          */
 774
 775         /*
 776          * Don't let too much OOB data build up
 777          */
 778         if (flags & PRUS_OOB) {
 779                 if (ssb_space(&so->so_snd) < -512) {
 780                         m_freem(m);
 781                         error = ENOBUFS;
 782                         goto out;
 783                 }
 784         }
 785 #endif
 786
 787         /*
 788          * Pump the data into the socket.
 789          */
 790         if (m)
 791                 ssb_appendstream(&so->so_snd, m);
 792         if (flags & PRUS_OOB) {
 793                 /*
 794                  * According to RFC961 (Assigned Protocols),
 795                  * the urgent pointer points to the last octet
 796                  * of urgent data.  We continue, however,
 797                  * to consider it to indicate the first octet
 798                  * of data past the urgent section.
 799                  * Otherwise, snd_up should be one lower.
 800                  */
 801                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 802                 tp->t_flags |= TF_FORCE;
 803                 error = tcp_output(tp);
 804                 tp->t_flags &= ~TF_FORCE;
 805         } else {
 806                 if (flags & PRUS_EOF) {
 807                         /*
 808                          * Close the send side of the connection after
 809                          * the data is sent.
 810                          */
 811                         socantsendmore(so);
 812                         tp = tcp_usrclosed(tp);
 813                 }
 814                 if (tp != NULL && !tcp_output_pending(tp)) {
 815                         if (flags & PRUS_MORETOCOME)
 816                                 tp->t_flags |= TF_MORETOCOME;
 817                         error = tcp_output_fair(tp);
 818                         if (flags & PRUS_MORETOCOME)
 819                                 tp->t_flags &= ~TF_MORETOCOME;
 820                 }
 821         }
 822         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 823                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 824                    (flags & PRUS_NOREPLY));
 825 }
 826
 827 /*
 828  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 829  *       will sofree() it when we return.
 830  */
 831 static void
 832 tcp_usr_abort(netmsg_t msg)
 833 {
 834         struct socket *so = msg->abort.base.nm_so;
 835         int error = 0;
 836         struct inpcb *inp;
 837         struct tcpcb *tp;
 838
 839         COMMON_START(so, inp, 1);
 840         tp = tcp_drop(tp, ECONNABORTED);
 841         COMMON_END(PRU_ABORT);
 842 }
 843
 844 /*
 845  * Receive out-of-band data.
 846  */
 847 static void
 848 tcp_usr_rcvoob(netmsg_t msg)
 849 {
 850         struct socket *so = msg->rcvoob.base.nm_so;
 851         struct mbuf *m = msg->rcvoob.nm_m;
 852         int flags = msg->rcvoob.nm_flags;
 853         int error = 0;
 854         struct inpcb *inp;
 855         struct tcpcb *tp;
 856
 857         COMMON_START(so, inp, 0);
 858         if ((so->so_oobmark == 0 &&
 859              (so->so_state & SS_RCVATMARK) == 0) ||
 860             so->so_options & SO_OOBINLINE ||
 861             tp->t_oobflags & TCPOOB_HADDATA) {
 862                 error = EINVAL;
 863                 goto out;
 864         }
 865         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 866                 error = EWOULDBLOCK;
 867                 goto out;
 868         }
 869         m->m_len = 1;
 870         *mtod(m, caddr_t) = tp->t_iobc;
 871         if ((flags & MSG_PEEK) == 0)
 872                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 873         COMMON_END(PRU_RCVOOB);
 874 }
 875
 876 static void
 877 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 878 {
 879         in_savefaddr(so, faddr);
 880 }
 881
 882 #ifdef INET6
 883 static void
 884 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 885 {
 886         in6_mapped_savefaddr(so, faddr);
 887 }
 888 #endif
 889
 890 static int
 891 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 892     struct thread *td __unused)
 893 {
 894         const struct sockaddr_in *sinp;
 895
 896         sinp = (const struct sockaddr_in *)nam;
 897         if (sinp->sin_family == AF_INET &&
 898             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 899                 return EAFNOSUPPORT;
 900
 901         soisconnecting(so);
 902         return 0;
 903 }
 904
 905 /* xxx - should be const */
 906 struct pr_usrreqs tcp_usrreqs = {
 907         .pru_abort = tcp_usr_abort,
 908         .pru_accept = tcp_usr_accept,
 909         .pru_attach = tcp_usr_attach,
 910         .pru_bind = tcp_usr_bind,
 911         .pru_connect = tcp_usr_connect,
 912         .pru_connect2 = pr_generic_notsupp,
 913         .pru_control = in_control_dispatch,
 914         .pru_detach = tcp_usr_detach,
 915         .pru_disconnect = tcp_usr_disconnect,
 916         .pru_listen = tcp_usr_listen,
 917         .pru_peeraddr = in_setpeeraddr_dispatch,
 918         .pru_rcvd = tcp_usr_rcvd,
 919         .pru_rcvoob = tcp_usr_rcvoob,
 920         .pru_send = tcp_usr_send,
 921         .pru_sense = pru_sense_null,
 922         .pru_shutdown = tcp_usr_shutdown,
 923         .pru_sockaddr = in_setsockaddr_dispatch,
 924         .pru_sosend = sosendtcp,
 925         .pru_soreceive = sorecvtcp,
 926         .pru_savefaddr = tcp_usr_savefaddr,
 927         .pru_preconnect = tcp_usr_preconnect
 928 };
 929
 930 #ifdef INET6
 931 struct pr_usrreqs tcp6_usrreqs = {
 932         .pru_abort = tcp_usr_abort,
 933         .pru_accept = tcp6_usr_accept,
 934         .pru_attach = tcp_usr_attach,
 935         .pru_bind = tcp6_usr_bind,
 936         .pru_connect = tcp6_usr_connect,
 937         .pru_connect2 = pr_generic_notsupp,
 938         .pru_control = in6_control_dispatch,
 939         .pru_detach = tcp_usr_detach,
 940         .pru_disconnect = tcp_usr_disconnect,
 941         .pru_listen = tcp6_usr_listen,
 942         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 943         .pru_rcvd = tcp_usr_rcvd,
 944         .pru_rcvoob = tcp_usr_rcvoob,
 945         .pru_send = tcp_usr_send,
 946         .pru_sense = pru_sense_null,
 947         .pru_shutdown = tcp_usr_shutdown,
 948         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 949         .pru_sosend = sosendtcp,
 950         .pru_soreceive = sorecvtcp,
 951         .pru_savefaddr = tcp6_usr_savefaddr
 952 };
 953 #endif /* INET6 */
 954
 955 static int
 956 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 957                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 958 {
 959         struct inpcb *inp = tp->t_inpcb, *oinp;
 960         struct socket *so = inp->inp_socket;
 961         struct route *ro = &inp->inp_route;
 962
 963         KASSERT(inp->inp_cpcbinfo == &tcbinfo[mycpu->gd_cpuid],
 964             ("cpcbinfo mismatch"));
 965
 966         oinp = in_pcblookup_hash(inp->inp_cpcbinfo,
 967                                  sin->sin_addr, sin->sin_port,
 968                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 969                                   inp->inp_laddr : if_sin->sin_addr),
 970                                 inp->inp_lport, 0, NULL);
 971         if (oinp != NULL) {
 972                 m_freem(m);
 973                 return (EADDRINUSE);
 974         }
 975         if (inp->inp_laddr.s_addr == INADDR_ANY)
 976                 inp->inp_laddr = if_sin->sin_addr;
 977         inp->inp_faddr = sin->sin_addr;
 978         inp->inp_fport = sin->sin_port;
 979         in_pcbinsconnhash(inp);
 980
 981         /*
 982          * We are now on the inpcb's owner CPU, if the cached route was
 983          * freed because the rtentry's owner CPU is not the current CPU
 984          * (e.g. in tcp_connect()), then we try to reallocate it here with
 985          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 986          * rtentry.
 987          */
 988         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 989             ro->ro_rt == NULL) {
 990                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 991                 ro->ro_dst.sa_family = AF_INET;
 992                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 993                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 994                         sin->sin_addr;
 995                 rtalloc(ro);
 996         }
 997
 998         /*
 999          * Now that no more errors can occur, change the protocol processing
1000          * port to the current thread (which is the correct thread).
1001          *
1002          * Create TCP timer message now; we are on the tcpcb's owner
1003          * CPU/thread.
1004          */
1005         tcp_create_timermsg(tp, &curthread->td_msgport);
1006
1007         /*
1008          * Compute window scaling to request.  Use a larger scaling then
1009          * needed for the initial receive buffer in case the receive buffer
1010          * gets expanded.
1011          */
1012         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1013                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1014         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1015                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1016         ) {
1017                 tp->request_r_scale++;
1018         }
1019
1020         soisconnecting(so);
1021         tcpstat.tcps_connattempt++;
1022         tp->t_state = TCPS_SYN_SENT;
1023         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1024         tp->iss = tcp_new_isn(tp);
1025         tcp_sendseqinit(tp);
1026         if (m) {
1027                 ssb_appendstream(&so->so_snd, m);
1028                 m = NULL;
1029                 if (flags & PRUS_OOB)
1030                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1031         }
1032
1033         /*
1034          * Close the send side of the connection after
1035          * the data is sent if flagged.
1036          */
1037         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1038                 socantsendmore(so);
1039                 tp = tcp_usrclosed(tp);
1040         }
1041         return (tcp_output(tp));
1042 }
1043
1044 /*
1045  * Common subroutine to open a TCP connection to remote host specified
1046  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1047  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1048  * a local host address (interface).
1049  * Initialize connection parameters and enter SYN-SENT state.
1050  */
1051 static void
1052 tcp_connect(netmsg_t msg)
1053 {
1054         struct socket *so = msg->connect.base.nm_so;
1055         struct sockaddr *nam = msg->connect.nm_nam;
1056         struct thread *td = msg->connect.nm_td;
1057         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1058         struct sockaddr_in *if_sin;
1059         struct inpcb *inp;
1060         struct tcpcb *tp;
1061         int error, calc_laddr = 1;
1062         lwkt_port_t port;
1063
1064         COMMON_START(so, inp, 0);
1065
1066         /*
1067          * Reconnect our pcb if we have to
1068          */
1069         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1070                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1071                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1072         }
1073
1074         /*
1075          * Bind if we have to
1076          */
1077         if (inp->inp_lport == 0) {
1078                 if (tcp_lport_extension) {
1079                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1080
1081                         error = in_pcbladdr(inp, nam, &if_sin, td);
1082                         if (error)
1083                                 goto out;
1084                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1085
1086                         error = in_pcbconn_bind(inp, nam, td);
1087                         if (error)
1088                                 goto out;
1089
1090                         calc_laddr = 0;
1091                 } else {
1092                         error = in_pcbbind(inp, NULL, td);
1093                         if (error)
1094                                 goto out;
1095                 }
1096         }
1097
1098         if (calc_laddr) {
1099                 /*
1100                  * Calculate the correct protocol processing thread.  The
1101                  * connect operation must run there.  Set the forwarding
1102                  * port before we forward the message or it will get bounced
1103                  * right back to us.
1104                  */
1105                 error = in_pcbladdr(inp, nam, &if_sin, td);
1106                 if (error)
1107                         goto out;
1108         }
1109         KKASSERT(inp->inp_socket == so);
1110
1111         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1112                             (inp->inp_laddr.s_addr ?
1113                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1114                             inp->inp_lport);
1115
1116         if (port != &curthread->td_msgport) {
1117                 struct route *ro = &inp->inp_route;
1118
1119                 /*
1120                  * in_pcbladdr() may have allocated a route entry for us
1121                  * on the current CPU, but we need a route entry on the
1122                  * inpcb's owner CPU, so free it here.
1123                  */
1124                 if (ro->ro_rt != NULL)
1125                         RTFREE(ro->ro_rt);
1126                 bzero(ro, sizeof(*ro));
1127
1128                 /*
1129                  * We are moving the protocol processing port the socket
1130                  * is on, we have to unlink here and re-link on the
1131                  * target cpu.
1132                  */
1133                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1134                 sosetport(so, port);
1135                 msg->connect.nm_flags |= PRUC_RECONNECT;
1136                 msg->connect.base.nm_dispatch = tcp_connect;
1137
1138                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1139                 /* msg invalid now */
1140                 return;
1141         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1142                 /*
1143                  * The original thread is no longer needed; release it.
1144                  */
1145                 lwkt_rele(td);
1146                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1147         }
1148         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1149                                   msg->connect.nm_m, sin, if_sin);
1150         msg->connect.nm_m = NULL;
1151 out:
1152         if (msg->connect.nm_m) {
1153                 m_freem(msg->connect.nm_m);
1154                 msg->connect.nm_m = NULL;
1155         }
1156         if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1157                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1158                 msg->connect.nm_nam = NULL;
1159         }
1160         if (msg->connect.nm_flags & PRUC_HELDTD)
1161                 lwkt_rele(td);
1162         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1163                 so->so_error = error;
1164                 soisdisconnected(so);
1165         }
1166         lwkt_replymsg(&msg->connect.base.lmsg, error);
1167         /* msg invalid now */
1168 }
1169
1170 #ifdef INET6
1171
1172 static void
1173 tcp6_connect(netmsg_t msg)
1174 {
1175         struct tcpcb *tp;
1176         struct socket *so = msg->connect.base.nm_so;
1177         struct sockaddr *nam = msg->connect.nm_nam;
1178         struct thread *td = msg->connect.nm_td;
1179         struct inpcb *inp;
1180         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1181         struct in6_addr *addr6;
1182         lwkt_port_t port;
1183         int error;
1184
1185         COMMON_START(so, inp, 0);
1186
1187         /*
1188          * Reconnect our pcb if we have to
1189          */
1190         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1191                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1192                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1193         }
1194
1195         /*
1196          * Bind if we have to
1197          */
1198         if (inp->inp_lport == 0) {
1199                 error = in6_pcbbind(inp, NULL, td);
1200                 if (error)
1201                         goto out;
1202         }
1203
1204         /*
1205          * Cannot simply call in_pcbconnect, because there might be an
1206          * earlier incarnation of this same connection still in
1207          * TIME_WAIT state, creating an ADDRINUSE error.
1208          */
1209         error = in6_pcbladdr(inp, nam, &addr6, td);
1210         if (error)
1211                 goto out;
1212
1213         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1214
1215         if (port != &curthread->td_msgport) {
1216                 struct route *ro = &inp->inp_route;
1217
1218                 /*
1219                  * in_pcbladdr() may have allocated a route entry for us
1220                  * on the current CPU, but we need a route entry on the
1221                  * inpcb's owner CPU, so free it here.
1222                  */
1223                 if (ro->ro_rt != NULL)
1224                         RTFREE(ro->ro_rt);
1225                 bzero(ro, sizeof(*ro));
1226
1227                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1228                 sosetport(so, port);
1229                 msg->connect.nm_flags |= PRUC_RECONNECT;
1230                 msg->connect.base.nm_dispatch = tcp6_connect;
1231
1232                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1233                 /* msg invalid now */
1234                 return;
1235         }
1236         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1237                                    &msg->connect.nm_m, sin6, addr6);
1238         /* nm_m may still be intact */
1239 out:
1240         if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1241                 tcp_connect(msg);
1242                 /* msg invalid now */
1243         } else {
1244                 if (msg->connect.nm_m) {
1245                         m_freem(msg->connect.nm_m);
1246                         msg->connect.nm_m = NULL;
1247                 }
1248                 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1249                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1250                         msg->connect.nm_nam = NULL;
1251                 }
1252                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1253                 /* msg invalid now */
1254         }
1255 }
1256
1257 static int
1258 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1259                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1260 {
1261         struct mbuf *m = *mp;
1262         struct inpcb *inp = tp->t_inpcb;
1263         struct socket *so = inp->inp_socket;
1264         struct inpcb *oinp;
1265
1266         /*
1267          * Cannot simply call in_pcbconnect, because there might be an
1268          * earlier incarnation of this same connection still in
1269          * TIME_WAIT state, creating an ADDRINUSE error.
1270          */
1271         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1272                                   &sin6->sin6_addr, sin6->sin6_port,
1273                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1274                                       addr6 : &inp->in6p_laddr),
1275                                   inp->inp_lport,  0, NULL);
1276         if (oinp)
1277                 return (EADDRINUSE);
1278
1279         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1280                 inp->in6p_laddr = *addr6;
1281         inp->in6p_faddr = sin6->sin6_addr;
1282         inp->inp_fport = sin6->sin6_port;
1283         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1284                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1285         in_pcbinsconnhash(inp);
1286
1287         /*
1288          * Now that no more errors can occur, change the protocol processing
1289          * port to the current thread (which is the correct thread).
1290          *
1291          * Create TCP timer message now; we are on the tcpcb's owner
1292          * CPU/thread.
1293          */
1294         tcp_create_timermsg(tp, &curthread->td_msgport);
1295
1296         /* Compute window scaling to request.  */
1297         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1298                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1299         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1300             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1301                 tp->request_r_scale++;
1302         }
1303
1304         soisconnecting(so);
1305         tcpstat.tcps_connattempt++;
1306         tp->t_state = TCPS_SYN_SENT;
1307         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1308         tp->iss = tcp_new_isn(tp);
1309         tcp_sendseqinit(tp);
1310         if (m) {
1311                 ssb_appendstream(&so->so_snd, m);
1312                 *mp = NULL;
1313                 if (flags & PRUS_OOB)
1314                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1315         }
1316
1317         /*
1318          * Close the send side of the connection after
1319          * the data is sent if flagged.
1320          */
1321         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1322                 socantsendmore(so);
1323                 tp = tcp_usrclosed(tp);
1324         }
1325         return (tcp_output(tp));
1326 }
1327
1328 #endif /* INET6 */
1329
1330 /*
1331  * The new sockopt interface makes it possible for us to block in the
1332  * copyin/out step (if we take a page fault).  Taking a page fault while
1333  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1334  * both now use TSM, there probably isn't any need for this function to
1335  * run in a critical section any more.  This needs more examination.)
1336  */
1337 void
1338 tcp_ctloutput(netmsg_t msg)
1339 {
1340         struct socket *so = msg->base.nm_so;
1341         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1342         int     error, opt, optval, opthz;
1343         struct  inpcb *inp;
1344         struct  tcpcb *tp;
1345
1346         error = 0;
1347         inp = so->so_pcb;
1348         if (inp == NULL) {
1349                 error = ECONNRESET;
1350                 goto done;
1351         }
1352
1353         if (sopt->sopt_level != IPPROTO_TCP) {
1354 #ifdef INET6
1355                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1356                         ip6_ctloutput_dispatch(msg);
1357                 else
1358 #endif /* INET6 */
1359                 ip_ctloutput(msg);
1360                 /* msg invalid now */
1361                 return;
1362         }
1363         tp = intotcpcb(inp);
1364
1365         switch (sopt->sopt_dir) {
1366         case SOPT_SET:
1367                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1368                                       sizeof optval);
1369                 if (error)
1370                         break;
1371                 switch (sopt->sopt_name) {
1372                 case TCP_FASTKEEP:
1373                         if (optval > 0)
1374                                 tp->t_keepidle = tp->t_keepintvl;
1375                         else
1376                                 tp->t_keepidle = tcp_keepidle;
1377                         tcp_timer_keep_activity(tp, 0);
1378                         break;
1379 #ifdef TCP_SIGNATURE
1380                 case TCP_SIGNATURE_ENABLE:
1381                         if (tp->t_state == TCPS_CLOSED) {
1382                                 /*
1383                                  * This is the only safe state that this
1384                                  * option could be changed.  Some segments
1385                                  * could already have been sent in other
1386                                  * states.
1387                                  */
1388                                 if (optval > 0)
1389                                         tp->t_flags |= TF_SIGNATURE;
1390                                 else
1391                                         tp->t_flags &= ~TF_SIGNATURE;
1392                         } else {
1393                                 error = EOPNOTSUPP;
1394                         }
1395                         break;
1396 #endif /* TCP_SIGNATURE */
1397                 case TCP_NODELAY:
1398                 case TCP_NOOPT:
1399                         switch (sopt->sopt_name) {
1400                         case TCP_NODELAY:
1401                                 opt = TF_NODELAY;
1402                                 break;
1403                         case TCP_NOOPT:
1404                                 opt = TF_NOOPT;
1405                                 break;
1406                         default:
1407                                 opt = 0; /* dead code to fool gcc */
1408                                 break;
1409                         }
1410
1411                         if (optval)
1412                                 tp->t_flags |= opt;
1413                         else
1414                                 tp->t_flags &= ~opt;
1415                         break;
1416
1417                 case TCP_NOPUSH:
1418                         if (tcp_disable_nopush)
1419                                 break;
1420                         if (optval)
1421                                 tp->t_flags |= TF_NOPUSH;
1422                         else {
1423                                 tp->t_flags &= ~TF_NOPUSH;
1424                                 error = tcp_output(tp);
1425                         }
1426                         break;
1427
1428                 case TCP_MAXSEG:
1429                         /*
1430                          * Must be between 0 and maxseg.  If the requested
1431                          * maxseg is too small to satisfy the desired minmss,
1432                          * pump it up (silently so sysctl modifications of
1433                          * minmss do not create unexpected program failures).
1434                          * Handle degenerate cases.
1435                          */
1436                         if (optval > 0 && optval <= tp->t_maxseg) {
1437                                 if (optval + 40 < tcp_minmss) {
1438                                         optval = tcp_minmss - 40;
1439                                         if (optval < 0)
1440                                                 optval = 1;
1441                                 }
1442                                 tp->t_maxseg = optval;
1443                         } else {
1444                                 error = EINVAL;
1445                         }
1446                         break;
1447
1448                 case TCP_KEEPINIT:
1449                         opthz = ((int64_t)optval * hz) / 1000;
1450                         if (opthz >= 1)
1451                                 tp->t_keepinit = opthz;
1452                         else
1453                                 error = EINVAL;
1454                         break;
1455
1456                 case TCP_KEEPIDLE:
1457                         opthz = ((int64_t)optval * hz) / 1000;
1458                         if (opthz >= 1) {
1459                                 tp->t_keepidle = opthz;
1460                                 tcp_timer_keep_activity(tp, 0);
1461                         } else {
1462                                 error = EINVAL;
1463                         }
1464                         break;
1465
1466                 case TCP_KEEPINTVL:
1467                         opthz = ((int64_t)optval * hz) / 1000;
1468                         if (opthz >= 1) {
1469                                 tp->t_keepintvl = opthz;
1470                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1471                         } else {
1472                                 error = EINVAL;
1473                         }
1474                         break;
1475
1476                 case TCP_KEEPCNT:
1477                         if (optval > 0) {
1478                                 tp->t_keepcnt = optval;
1479                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1480                         } else {
1481                                 error = EINVAL;
1482                         }
1483                         break;
1484
1485                 default:
1486                         error = ENOPROTOOPT;
1487                         break;
1488                 }
1489                 break;
1490
1491         case SOPT_GET:
1492                 switch (sopt->sopt_name) {
1493 #ifdef TCP_SIGNATURE
1494                 case TCP_SIGNATURE_ENABLE:
1495                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1496                         break;
1497 #endif /* TCP_SIGNATURE */
1498                 case TCP_NODELAY:
1499                         optval = tp->t_flags & TF_NODELAY;
1500                         break;
1501                 case TCP_MAXSEG:
1502                         optval = tp->t_maxseg;
1503                         break;
1504                 case TCP_NOOPT:
1505                         optval = tp->t_flags & TF_NOOPT;
1506                         break;
1507                 case TCP_NOPUSH:
1508                         optval = tp->t_flags & TF_NOPUSH;
1509                         break;
1510                 case TCP_KEEPINIT:
1511                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1512                         break;
1513                 case TCP_KEEPIDLE:
1514                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1515                         break;
1516                 case TCP_KEEPINTVL:
1517                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1518                         break;
1519                 case TCP_KEEPCNT:
1520                         optval = tp->t_keepcnt;
1521                         break;
1522                 default:
1523                         error = ENOPROTOOPT;
1524                         break;
1525                 }
1526                 if (error == 0)
1527                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1528                 break;
1529         }
1530 done:
1531         lwkt_replymsg(&msg->lmsg, error);
1532 }
1533
1534 /*
1535  * tcp_sendspace and tcp_recvspace are the default send and receive window
1536  * sizes, respectively.  These are obsolescent (this information should
1537  * be set by the route).
1538  *
1539  * Use a default that does not require tcp window scaling to be turned
1540  * on.  Individual programs or the administrator can increase the default.
1541  */
1542 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1543 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1544     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1545 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1546 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1547     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1548
1549 /*
1550  * Attach TCP protocol to socket, allocating internet protocol control
1551  * block, tcp control block, buffer space, and entering CLOSED state.
1552  */
1553 static int
1554 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1555 {
1556         struct tcpcb *tp;
1557         struct inpcb *inp;
1558         int error;
1559         int cpu;
1560 #ifdef INET6
1561         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1562 #endif
1563
1564         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1565                 lwkt_gettoken(&so->so_rcv.ssb_token);
1566                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1567                                   ai->sb_rlimit);
1568                 lwkt_reltoken(&so->so_rcv.ssb_token);
1569                 if (error)
1570                         return (error);
1571         }
1572         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1573         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1574         cpu = mycpu->gd_cpuid;
1575
1576         /*
1577          * Set the default port for protocol processing. This will likely
1578          * change when we connect.
1579          */
1580         error = in_pcballoc(so, &tcbinfo[cpu]);
1581         if (error)
1582                 return (error);
1583         inp = so->so_pcb;
1584 #ifdef INET6
1585         if (isipv6) {
1586                 inp->inp_vflag |= INP_IPV6;
1587                 inp->in6p_hops = -1;    /* use kernel default */
1588         }
1589         else
1590 #endif
1591         inp->inp_vflag |= INP_IPV4;
1592         tp = tcp_newtcpcb(inp);
1593         if (tp == NULL) {
1594                 /*
1595                  * Make sure the socket is destroyed by the pcbdetach.
1596                  */
1597                 soreference(so);
1598 #ifdef INET6
1599                 if (isipv6)
1600                         in6_pcbdetach(inp);
1601                 else
1602 #endif
1603                 in_pcbdetach(inp);
1604                 sofree(so);     /* from ref above */
1605                 return (ENOBUFS);
1606         }
1607         tp->t_state = TCPS_CLOSED;
1608         /* Keep a reference for asynchronized pru_rcvd */
1609         soreference(so);
1610         return (0);
1611 }
1612
1613 /*
1614  * Initiate (or continue) disconnect.
1615  * If embryonic state, just send reset (once).
1616  * If in ``let data drain'' option and linger null, just drop.
1617  * Otherwise (hard), mark socket disconnecting and drop
1618  * current input data; switch states based on user close, and
1619  * send segment to peer (with FIN).
1620  */
1621 static struct tcpcb *
1622 tcp_disconnect(struct tcpcb *tp)
1623 {
1624         struct socket *so = tp->t_inpcb->inp_socket;
1625
1626         if (tp->t_state < TCPS_ESTABLISHED) {
1627                 tp = tcp_close(tp);
1628         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1629                 tp = tcp_drop(tp, 0);
1630         } else {
1631                 lwkt_gettoken(&so->so_rcv.ssb_token);
1632                 soisdisconnecting(so);
1633                 sbflush(&so->so_rcv.sb);
1634                 tp = tcp_usrclosed(tp);
1635                 if (tp)
1636                         tcp_output(tp);
1637                 lwkt_reltoken(&so->so_rcv.ssb_token);
1638         }
1639         return (tp);
1640 }
1641
1642 /*
1643  * User issued close, and wish to trail through shutdown states:
1644  * if never received SYN, just forget it.  If got a SYN from peer,
1645  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1646  * If already got a FIN from peer, then almost done; go to LAST_ACK
1647  * state.  In all other cases, have already sent FIN to peer (e.g.
1648  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1649  * for peer to send FIN or not respond to keep-alives, etc.
1650  * We can let the user exit from the close as soon as the FIN is acked.
1651  */
1652 static struct tcpcb *
1653 tcp_usrclosed(struct tcpcb *tp)
1654 {
1655
1656         switch (tp->t_state) {
1657
1658         case TCPS_CLOSED:
1659         case TCPS_LISTEN:
1660                 tp->t_state = TCPS_CLOSED;
1661                 tp = tcp_close(tp);
1662                 break;
1663
1664         case TCPS_SYN_SENT:
1665         case TCPS_SYN_RECEIVED:
1666                 tp->t_flags |= TF_NEEDFIN;
1667                 break;
1668
1669         case TCPS_ESTABLISHED:
1670                 tp->t_state = TCPS_FIN_WAIT_1;
1671                 break;
1672
1673         case TCPS_CLOSE_WAIT:
1674                 tp->t_state = TCPS_LAST_ACK;
1675                 break;
1676         }
1677         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1678                 soisdisconnected(tp->t_inpcb->inp_socket);
1679                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1680                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1681                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1682                             tcp_timer_2msl);
1683                 }
1684         }
1685         return (tp);
1686 }