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