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/socketops.h>
  90 #include <sys/protosw.h>
  91
  92 #include <sys/thread2.h>
  93 #include <sys/msgport2.h>
  94 #include <sys/socketvar2.h>
  95
  96 #include <net/if.h>
  97 #include <net/netisr.h>
  98 #include <net/route.h>
  99
 100 #include <net/netmsg2.h>
 101 #include <net/netisr2.h>
 102
 103 #include <netinet/in.h>
 104 #include <netinet/in_systm.h>
 105 #ifdef INET6
 106 #include <netinet/ip6.h>
 107 #endif
 108 #include <netinet/in_pcb.h>
 109 #ifdef INET6
 110 #include <netinet6/in6_pcb.h>
 111 #endif
 112 #include <netinet/in_var.h>
 113 #include <netinet/ip_var.h>
 114 #ifdef INET6
 115 #include <netinet6/ip6_var.h>
 116 #include <netinet6/tcp6_var.h>
 117 #endif
 118 #include <netinet/tcp.h>
 119 #include <netinet/tcp_fsm.h>
 120 #include <netinet/tcp_seq.h>
 121 #include <netinet/tcp_timer.h>
 122 #include <netinet/tcp_timer2.h>
 123 #include <netinet/tcp_var.h>
 124 #include <netinet/tcpip.h>
 125 #ifdef TCPDEBUG
 126 #include <netinet/tcp_debug.h>
 127 #endif
 128
 129 #ifdef IPSEC
 130 #include <netinet6/ipsec.h>
 131 #endif /*IPSEC*/
 132
 133 /*
 134  * TCP protocol interface to socket abstraction.
 135  */
 136 extern  char *tcpstates[];      /* XXX ??? */
 137
 138 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 139 static void     tcp_connect (netmsg_t msg);
 140 #ifdef INET6
 141 static void     tcp6_connect (netmsg_t msg);
 142 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 143                                 struct mbuf **mp,
 144                                 struct sockaddr_in6 *sin6,
 145                                 struct in6_addr *addr6);
 146 #endif /* INET6 */
 147 static struct tcpcb *
 148                 tcp_disconnect (struct tcpcb *);
 149 static struct tcpcb *
 150                 tcp_usrclosed (struct tcpcb *);
 151
 152 #ifdef TCPDEBUG
 153 #define TCPDEBUG0       int ostate = 0
 154 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 155 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 156                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 157 #else
 158 #define TCPDEBUG0
 159 #define TCPDEBUG1()
 160 #define TCPDEBUG2(req)
 161 #endif
 162
 163 static int      tcp_lport_extension = 1;
 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 165     &tcp_lport_extension, 0, "");
 166
 167 /*
 168  * For some ill optimized programs, which try to use TCP_NOPUSH
 169  * to improve performance, will have small amount of data sits
 170  * in the sending buffer.  These small amount of data will _not_
 171  * be pushed into the network until more data are written into
 172  * the socket or the socket write side is shutdown.
 173  */
 174 static int      tcp_disable_nopush = 1;
 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
 176     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
 177
 178 /*
 179  * TCP attaches to socket via pru_attach(), reserving space,
 180  * and an internet control block.  This is likely occuring on
 181  * cpu0 and may have to move later when we bind/connect.
 182  */
 183 static void
 184 tcp_usr_attach(netmsg_t msg)
 185 {
 186         struct socket *so = msg->base.nm_so;
 187         struct pru_attach_info *ai = msg->attach.nm_ai;
 188         int error;
 189         struct inpcb *inp;
 190         struct tcpcb *tp = NULL;
 191         TCPDEBUG0;
 192
 193         soreference(so);
 194         inp = so->so_pcb;
 195         TCPDEBUG1();
 196         if (inp) {
 197                 error = EISCONN;
 198                 goto out;
 199         }
 200
 201         error = tcp_attach(so, ai);
 202         if (error)
 203                 goto out;
 204
 205         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 206                 so->so_linger = TCP_LINGERTIME;
 207         tp = sototcpcb(so);
 208 out:
 209         sofree(so);             /* from ref above */
 210         TCPDEBUG2(PRU_ATTACH);
 211         lwkt_replymsg(&msg->lmsg, error);
 212 }
 213
 214 /*
 215  * pru_detach() detaches the TCP protocol from the socket.
 216  * If the protocol state is non-embryonic, then can't
 217  * do this directly: have to initiate a pru_disconnect(),
 218  * which may finish later; embryonic TCB's can just
 219  * be discarded here.
 220  */
 221 static void
 222 tcp_usr_detach(netmsg_t msg)
 223 {
 224         struct socket *so = msg->base.nm_so;
 225         int error = 0;
 226         struct inpcb *inp;
 227         struct tcpcb *tp;
 228         TCPDEBUG0;
 229
 230         inp = so->so_pcb;
 231
 232         /*
 233          * If the inp is already detached it may have been due to an async
 234          * close.  Just return as if no error occured.
 235          *
 236          * It's possible for the tcpcb (tp) to disconnect from the inp due
 237          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 238          * the detach message has been queued so we may find a NULL tp here.
 239          */
 240         if (inp) {
 241                 if ((tp = intotcpcb(inp)) != NULL) {
 242                         TCPDEBUG1();
 243                         tp = tcp_disconnect(tp);
 244                         TCPDEBUG2(PRU_DETACH);
 245                 }
 246         }
 247         lwkt_replymsg(&msg->lmsg, error);
 248 }
 249
 250 /*
 251  * NOTE: ignore_error is non-zero for certain disconnection races
 252  * which we want to silently allow, otherwise close() may return
 253  * an unexpected error.
 254  *
 255  * NOTE: The variables (msg) and (tp) are assumed.
 256  */
 257 #define COMMON_START(so, inp, ignore_error)                     \
 258         TCPDEBUG0;                                              \
 259                                                                 \
 260         inp = so->so_pcb;                                       \
 261         do {                                                    \
 262                  if (inp == NULL) {                             \
 263                         error = ignore_error ? 0 : EINVAL;      \
 264                         tp = NULL;                              \
 265                         goto out;                               \
 266                  }                                              \
 267                  tp = intotcpcb(inp);                           \
 268                  TCPDEBUG1();                                   \
 269         } while(0)
 270
 271 #define COMMON_END1(req, noreply)                               \
 272         out: do {                                               \
 273                 TCPDEBUG2(req);                                 \
 274                 if (!(noreply))                                 \
 275                         lwkt_replymsg(&msg->lmsg, error);       \
 276                 return;                                         \
 277         } while(0)
 278
 279 #define COMMON_END(req)         COMMON_END1((req), 0)
 280
 281 /*
 282  * Give the socket an address.
 283  */
 284 static void
 285 tcp_usr_bind(netmsg_t msg)
 286 {
 287         struct socket *so = msg->bind.base.nm_so;
 288         struct sockaddr *nam = msg->bind.nm_nam;
 289         struct thread *td = msg->bind.nm_td;
 290         int error = 0;
 291         struct inpcb *inp;
 292         struct tcpcb *tp;
 293         struct sockaddr_in *sinp;
 294
 295         COMMON_START(so, inp, 0);
 296
 297         /*
 298          * Must check for multicast addresses and disallow binding
 299          * to them.
 300          */
 301         sinp = (struct sockaddr_in *)nam;
 302         if (sinp->sin_family == AF_INET &&
 303             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 304                 error = EAFNOSUPPORT;
 305                 goto out;
 306         }
 307         error = in_pcbbind(inp, nam, td);
 308         if (error)
 309                 goto out;
 310         COMMON_END(PRU_BIND);
 311
 312 }
 313
 314 #ifdef INET6
 315
 316 static void
 317 tcp6_usr_bind(netmsg_t msg)
 318 {
 319         struct socket *so = msg->bind.base.nm_so;
 320         struct sockaddr *nam = msg->bind.nm_nam;
 321         struct thread *td = msg->bind.nm_td;
 322         int error = 0;
 323         struct inpcb *inp;
 324         struct tcpcb *tp;
 325         struct sockaddr_in6 *sin6p;
 326
 327         COMMON_START(so, inp, 0);
 328
 329         /*
 330          * Must check for multicast addresses and disallow binding
 331          * to them.
 332          */
 333         sin6p = (struct sockaddr_in6 *)nam;
 334         if (sin6p->sin6_family == AF_INET6 &&
 335             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 336                 error = EAFNOSUPPORT;
 337                 goto out;
 338         }
 339         inp->inp_vflag &= ~INP_IPV4;
 340         inp->inp_vflag |= INP_IPV6;
 341         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 342                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 343                         inp->inp_vflag |= INP_IPV4;
 344                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 345                         struct sockaddr_in sin;
 346
 347                         in6_sin6_2_sin(&sin, sin6p);
 348                         inp->inp_vflag |= INP_IPV4;
 349                         inp->inp_vflag &= ~INP_IPV6;
 350                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 351                         goto out;
 352                 }
 353         }
 354         error = in6_pcbbind(inp, nam, td);
 355         if (error)
 356                 goto out;
 357         COMMON_END(PRU_BIND);
 358 }
 359 #endif /* INET6 */
 360
 361 struct netmsg_inswildcard {
 362         struct netmsg_base      base;
 363         struct inpcb            *nm_inp;
 364 };
 365
 366 static void
 367 in_pcbinswildcardhash_handler(netmsg_t msg)
 368 {
 369         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 370         int cpu = mycpuid, nextcpu;
 371
 372         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 373
 374         nextcpu = cpu + 1;
 375         if (nextcpu < ncpus2)
 376                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 377         else
 378                 lwkt_replymsg(&nm->base.lmsg, 0);
 379 }
 380
 381 /*
 382  * Prepare to accept connections.
 383  */
 384 static void
 385 tcp_usr_listen(netmsg_t msg)
 386 {
 387         struct socket *so = msg->listen.base.nm_so;
 388         struct thread *td = msg->listen.nm_td;
 389         int error = 0;
 390         struct inpcb *inp;
 391         struct tcpcb *tp;
 392         struct netmsg_inswildcard nm;
 393
 394         COMMON_START(so, inp, 0);
 395
 396         if (tp->t_flags & TF_LISTEN)
 397                 goto out;
 398
 399         if (inp->inp_lport == 0) {
 400                 error = in_pcbbind(inp, NULL, td);
 401                 if (error)
 402                         goto out;
 403         }
 404
 405         tp->t_state = TCPS_LISTEN;
 406         tp->t_flags |= TF_LISTEN;
 407         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 408
 409         if (ncpus > 1) {
 410                 /*
 411                  * We have to set the flag because we can't have other cpus
 412                  * messing with our inp's flags.
 413                  */
 414                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 415                         ("already on connhash"));
 416                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 417                         ("already on wildcardhash"));
 418                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 419                         ("already on MP wildcardhash"));
 420                 inp->inp_flags |= INP_WILDCARD_MP;
 421
 422                 KKASSERT(so->so_port == netisr_cpuport(0));
 423                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 424                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 425
 426                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 427                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 428                 nm.nm_inp = inp;
 429                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 430         }
 431         in_pcbinswildcardhash(inp);
 432         COMMON_END(PRU_LISTEN);
 433 }
 434
 435 #ifdef INET6
 436
 437 static void
 438 tcp6_usr_listen(netmsg_t msg)
 439 {
 440         struct socket *so = msg->listen.base.nm_so;
 441         struct thread *td = msg->listen.nm_td;
 442         int error = 0;
 443         struct inpcb *inp;
 444         struct tcpcb *tp;
 445         struct netmsg_inswildcard nm;
 446
 447         COMMON_START(so, inp, 0);
 448
 449         if (tp->t_flags & TF_LISTEN)
 450                 goto out;
 451
 452         if (inp->inp_lport == 0) {
 453                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 454                         inp->inp_vflag |= INP_IPV4;
 455                 else
 456                         inp->inp_vflag &= ~INP_IPV4;
 457                 error = in6_pcbbind(inp, NULL, td);
 458                 if (error)
 459                         goto out;
 460         }
 461
 462         tp->t_state = TCPS_LISTEN;
 463         tp->t_flags |= TF_LISTEN;
 464         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 465
 466         if (ncpus > 1) {
 467                 /*
 468                  * We have to set the flag because we can't have other cpus
 469                  * messing with our inp's flags.
 470                  */
 471                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
 472                         ("already on connhash"));
 473                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
 474                         ("already on wildcardhash"));
 475                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
 476                         ("already on MP wildcardhash"));
 477                 inp->inp_flags |= INP_WILDCARD_MP;
 478
 479                 KKASSERT(so->so_port == netisr_cpuport(0));
 480                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 481                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 482
 483                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 484                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 485                 nm.nm_inp = inp;
 486                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 487         }
 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, noreply = 0;
 713         struct inpcb *inp;
 714         struct tcpcb *tp;
 715
 716         COMMON_START(so, inp, 0);
 717
 718         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 719                 noreply = 1;
 720                 so_async_rcvd_reply(so);
 721         }
 722         tcp_output(tp);
 723
 724         COMMON_END1(PRU_RCVD, noreply);
 725 }
 726
 727 /*
 728  * Do a send by putting data in output queue and updating urgent
 729  * marker if URG set.  Possibly send more data.  Unlike the other
 730  * pru_*() routines, the mbuf chains are our responsibility.  We
 731  * must either enqueue them or free them.  The other pru_* routines
 732  * generally are caller-frees.
 733  */
 734 static void
 735 tcp_usr_send(netmsg_t msg)
 736 {
 737         struct socket *so = msg->send.base.nm_so;
 738         int flags = msg->send.nm_flags;
 739         struct mbuf *m = msg->send.nm_m;
 740         int error = 0;
 741         struct inpcb *inp;
 742         struct tcpcb *tp;
 743         TCPDEBUG0;
 744
 745         KKASSERT(msg->send.nm_control == NULL);
 746         KKASSERT(msg->send.nm_addr == NULL);
 747         KKASSERT((flags & PRUS_FREEADDR) == 0);
 748
 749         inp = so->so_pcb;
 750
 751         if (inp == NULL) {
 752                 /*
 753                  * OOPS! we lost a race, the TCP session got reset after
 754                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 755                  * network interrupt in the non-critical section of sosend().
 756                  */
 757                 m_freem(m);
 758                 error = ECONNRESET;     /* XXX EPIPE? */
 759                 tp = NULL;
 760                 TCPDEBUG1();
 761                 goto out;
 762         }
 763         tp = intotcpcb(inp);
 764         TCPDEBUG1();
 765
 766 #ifdef foo
 767         /*
 768          * This is no longer necessary, since:
 769          * - sosendtcp() has already checked it for us
 770          * - It does not work with asynchronized send
 771          */
 772
 773         /*
 774          * Don't let too much OOB data build up
 775          */
 776         if (flags & PRUS_OOB) {
 777                 if (ssb_space(&so->so_snd) < -512) {
 778                         m_freem(m);
 779                         error = ENOBUFS;
 780                         goto out;
 781                 }
 782         }
 783 #endif
 784
 785         /*
 786          * Pump the data into the socket.
 787          */
 788         if (m)
 789                 ssb_appendstream(&so->so_snd, m);
 790         if (flags & PRUS_OOB) {
 791                 /*
 792                  * According to RFC961 (Assigned Protocols),
 793                  * the urgent pointer points to the last octet
 794                  * of urgent data.  We continue, however,
 795                  * to consider it to indicate the first octet
 796                  * of data past the urgent section.
 797                  * Otherwise, snd_up should be one lower.
 798                  */
 799                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 800                 tp->t_flags |= TF_FORCE;
 801                 error = tcp_output(tp);
 802                 tp->t_flags &= ~TF_FORCE;
 803         } else {
 804                 if (flags & PRUS_EOF) {
 805                         /*
 806                          * Close the send side of the connection after
 807                          * the data is sent.
 808                          */
 809                         socantsendmore(so);
 810                         tp = tcp_usrclosed(tp);
 811                 }
 812                 if (tp != NULL && !tcp_output_pending(tp)) {
 813                         if (flags & PRUS_MORETOCOME)
 814                                 tp->t_flags |= TF_MORETOCOME;
 815                         error = tcp_output_fair(tp);
 816                         if (flags & PRUS_MORETOCOME)
 817                                 tp->t_flags &= ~TF_MORETOCOME;
 818                 }
 819         }
 820         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 821                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 822                    (flags & PRUS_NOREPLY));
 823 }
 824
 825 /*
 826  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 827  *       will sofree() it when we return.
 828  */
 829 static void
 830 tcp_usr_abort(netmsg_t msg)
 831 {
 832         struct socket *so = msg->abort.base.nm_so;
 833         int error = 0;
 834         struct inpcb *inp;
 835         struct tcpcb *tp;
 836
 837         COMMON_START(so, inp, 1);
 838         tp = tcp_drop(tp, ECONNABORTED);
 839         COMMON_END(PRU_ABORT);
 840 }
 841
 842 /*
 843  * Receive out-of-band data.
 844  */
 845 static void
 846 tcp_usr_rcvoob(netmsg_t msg)
 847 {
 848         struct socket *so = msg->rcvoob.base.nm_so;
 849         struct mbuf *m = msg->rcvoob.nm_m;
 850         int flags = msg->rcvoob.nm_flags;
 851         int error = 0;
 852         struct inpcb *inp;
 853         struct tcpcb *tp;
 854
 855         COMMON_START(so, inp, 0);
 856         if ((so->so_oobmark == 0 &&
 857              (so->so_state & SS_RCVATMARK) == 0) ||
 858             so->so_options & SO_OOBINLINE ||
 859             tp->t_oobflags & TCPOOB_HADDATA) {
 860                 error = EINVAL;
 861                 goto out;
 862         }
 863         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 864                 error = EWOULDBLOCK;
 865                 goto out;
 866         }
 867         m->m_len = 1;
 868         *mtod(m, caddr_t) = tp->t_iobc;
 869         if ((flags & MSG_PEEK) == 0)
 870                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 871         COMMON_END(PRU_RCVOOB);
 872 }
 873
 874 static void
 875 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 876 {
 877         in_savefaddr(so, faddr);
 878 }
 879
 880 #ifdef INET6
 881 static void
 882 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 883 {
 884         in6_mapped_savefaddr(so, faddr);
 885 }
 886 #endif
 887
 888 /* xxx - should be const */
 889 struct pr_usrreqs tcp_usrreqs = {
 890         .pru_abort = tcp_usr_abort,
 891         .pru_accept = tcp_usr_accept,
 892         .pru_attach = tcp_usr_attach,
 893         .pru_bind = tcp_usr_bind,
 894         .pru_connect = tcp_usr_connect,
 895         .pru_connect2 = pr_generic_notsupp,
 896         .pru_control = in_control_dispatch,
 897         .pru_detach = tcp_usr_detach,
 898         .pru_disconnect = tcp_usr_disconnect,
 899         .pru_listen = tcp_usr_listen,
 900         .pru_peeraddr = in_setpeeraddr_dispatch,
 901         .pru_rcvd = tcp_usr_rcvd,
 902         .pru_rcvoob = tcp_usr_rcvoob,
 903         .pru_send = tcp_usr_send,
 904         .pru_sense = pru_sense_null,
 905         .pru_shutdown = tcp_usr_shutdown,
 906         .pru_sockaddr = in_setsockaddr_dispatch,
 907         .pru_sosend = sosendtcp,
 908         .pru_soreceive = sorecvtcp,
 909         .pru_savefaddr = tcp_usr_savefaddr
 910 };
 911
 912 #ifdef INET6
 913 struct pr_usrreqs tcp6_usrreqs = {
 914         .pru_abort = tcp_usr_abort,
 915         .pru_accept = tcp6_usr_accept,
 916         .pru_attach = tcp_usr_attach,
 917         .pru_bind = tcp6_usr_bind,
 918         .pru_connect = tcp6_usr_connect,
 919         .pru_connect2 = pr_generic_notsupp,
 920         .pru_control = in6_control_dispatch,
 921         .pru_detach = tcp_usr_detach,
 922         .pru_disconnect = tcp_usr_disconnect,
 923         .pru_listen = tcp6_usr_listen,
 924         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 925         .pru_rcvd = tcp_usr_rcvd,
 926         .pru_rcvoob = tcp_usr_rcvoob,
 927         .pru_send = tcp_usr_send,
 928         .pru_sense = pru_sense_null,
 929         .pru_shutdown = tcp_usr_shutdown,
 930         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 931         .pru_sosend = sosendtcp,
 932         .pru_soreceive = sorecvtcp,
 933         .pru_savefaddr = tcp6_usr_savefaddr
 934 };
 935 #endif /* INET6 */
 936
 937 static int
 938 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 939                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 940 {
 941         struct inpcb *inp = tp->t_inpcb, *oinp;
 942         struct socket *so = inp->inp_socket;
 943         struct route *ro = &inp->inp_route;
 944
 945         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
 946                                  sin->sin_addr, sin->sin_port,
 947                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 948                                   inp->inp_laddr : if_sin->sin_addr),
 949                                 inp->inp_lport, 0, NULL);
 950         if (oinp != NULL) {
 951                 m_freem(m);
 952                 return (EADDRINUSE);
 953         }
 954         if (inp->inp_laddr.s_addr == INADDR_ANY)
 955                 inp->inp_laddr = if_sin->sin_addr;
 956         inp->inp_faddr = sin->sin_addr;
 957         inp->inp_fport = sin->sin_port;
 958         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
 959         in_pcbinsconnhash(inp);
 960
 961         /*
 962          * We are now on the inpcb's owner CPU, if the cached route was
 963          * freed because the rtentry's owner CPU is not the current CPU
 964          * (e.g. in tcp_connect()), then we try to reallocate it here with
 965          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 966          * rtentry.
 967          */
 968         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 969             ro->ro_rt == NULL) {
 970                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 971                 ro->ro_dst.sa_family = AF_INET;
 972                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 973                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 974                         sin->sin_addr;
 975                 rtalloc(ro);
 976         }
 977
 978         /*
 979          * Now that no more errors can occur, change the protocol processing
 980          * port to the current thread (which is the correct thread).
 981          *
 982          * Create TCP timer message now; we are on the tcpcb's owner
 983          * CPU/thread.
 984          */
 985         tcp_create_timermsg(tp, &curthread->td_msgport);
 986
 987         /*
 988          * Compute window scaling to request.  Use a larger scaling then
 989          * needed for the initial receive buffer in case the receive buffer
 990          * gets expanded.
 991          */
 992         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
 993                 tp->request_r_scale = TCP_MIN_WINSHIFT;
 994         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
 995                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
 996         ) {
 997                 tp->request_r_scale++;
 998         }
 999
1000         soisconnecting(so);
1001         tcpstat.tcps_connattempt++;
1002         tp->t_state = TCPS_SYN_SENT;
1003         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1004         tp->iss = tcp_new_isn(tp);
1005         tcp_sendseqinit(tp);
1006         if (m) {
1007                 ssb_appendstream(&so->so_snd, m);
1008                 m = NULL;
1009                 if (flags & PRUS_OOB)
1010                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1011         }
1012
1013         /*
1014          * Close the send side of the connection after
1015          * the data is sent if flagged.
1016          */
1017         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1018                 socantsendmore(so);
1019                 tp = tcp_usrclosed(tp);
1020         }
1021         return (tcp_output(tp));
1022 }
1023
1024 /*
1025  * Common subroutine to open a TCP connection to remote host specified
1026  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1027  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1028  * a local host address (interface).
1029  * Initialize connection parameters and enter SYN-SENT state.
1030  */
1031 static void
1032 tcp_connect(netmsg_t msg)
1033 {
1034         struct socket *so = msg->connect.base.nm_so;
1035         struct sockaddr *nam = msg->connect.nm_nam;
1036         struct thread *td = msg->connect.nm_td;
1037         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1038         struct sockaddr_in *if_sin;
1039         struct inpcb *inp;
1040         struct tcpcb *tp;
1041         int error, calc_laddr = 1;
1042         lwkt_port_t port;
1043
1044         COMMON_START(so, inp, 0);
1045
1046         /*
1047          * Reconnect our pcb if we have to
1048          */
1049         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1050                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1051                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1052         }
1053
1054         /*
1055          * Bind if we have to
1056          */
1057         if (inp->inp_lport == 0) {
1058                 if (tcp_lport_extension) {
1059                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1060
1061                         error = in_pcbladdr(inp, nam, &if_sin, td);
1062                         if (error)
1063                                 goto out;
1064                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1065
1066                         error = in_pcbconn_bind(inp, nam, td);
1067                         if (error)
1068                                 goto out;
1069
1070                         calc_laddr = 0;
1071                 } else {
1072                         error = in_pcbbind(inp, NULL, td);
1073                         if (error)
1074                                 goto out;
1075                 }
1076         }
1077
1078         if (calc_laddr) {
1079                 /*
1080                  * Calculate the correct protocol processing thread.  The
1081                  * connect operation must run there.  Set the forwarding
1082                  * port before we forward the message or it will get bounced
1083                  * right back to us.
1084                  */
1085                 error = in_pcbladdr(inp, nam, &if_sin, td);
1086                 if (error)
1087                         goto out;
1088         }
1089         KKASSERT(inp->inp_socket == so);
1090
1091         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1092                             (inp->inp_laddr.s_addr ?
1093                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1094                             inp->inp_lport);
1095
1096         if (port != &curthread->td_msgport) {
1097                 struct route *ro = &inp->inp_route;
1098
1099                 /*
1100                  * in_pcbladdr() may have allocated a route entry for us
1101                  * on the current CPU, but we need a route entry on the
1102                  * inpcb's owner CPU, so free it here.
1103                  */
1104                 if (ro->ro_rt != NULL)
1105                         RTFREE(ro->ro_rt);
1106                 bzero(ro, sizeof(*ro));
1107
1108                 /*
1109                  * We are moving the protocol processing port the socket
1110                  * is on, we have to unlink here and re-link on the
1111                  * target cpu.
1112                  */
1113                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1114                 sosetport(so, port);
1115                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1116                 msg->connect.base.nm_dispatch = tcp_connect;
1117
1118                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1119                 /* msg invalid now */
1120                 return;
1121         }
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         lwkt_port_t port;
1151         int error;
1152
1153         COMMON_START(so, inp, 0);
1154
1155         /*
1156          * Reconnect our pcb if we have to
1157          */
1158         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1159                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1160                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1161         }
1162
1163         /*
1164          * Bind if we have to
1165          */
1166         if (inp->inp_lport == 0) {
1167                 error = in6_pcbbind(inp, NULL, td);
1168                 if (error)
1169                         goto out;
1170         }
1171
1172         /*
1173          * Cannot simply call in_pcbconnect, because there might be an
1174          * earlier incarnation of this same connection still in
1175          * TIME_WAIT state, creating an ADDRINUSE error.
1176          */
1177         error = in6_pcbladdr(inp, nam, &addr6, td);
1178         if (error)
1179                 goto out;
1180
1181         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1182
1183         if (port != &curthread->td_msgport) {
1184                 struct route *ro = &inp->inp_route;
1185
1186                 /*
1187                  * in_pcbladdr() may have allocated a route entry for us
1188                  * on the current CPU, but we need a route entry on the
1189                  * inpcb's owner CPU, so free it here.
1190                  */
1191                 if (ro->ro_rt != NULL)
1192                         RTFREE(ro->ro_rt);
1193                 bzero(ro, sizeof(*ro));
1194
1195                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1196                 sosetport(so, port);
1197                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1198                 msg->connect.base.nm_dispatch = tcp6_connect;
1199
1200                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1201                 /* msg invalid now */
1202                 return;
1203         }
1204         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1205                                    &msg->connect.nm_m, sin6, addr6);
1206         /* nm_m may still be intact */
1207 out:
1208         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1209                 tcp_connect(msg);
1210                 /* msg invalid now */
1211         } else {
1212                 if (msg->connect.nm_m) {
1213                         m_freem(msg->connect.nm_m);
1214                         msg->connect.nm_m = NULL;
1215                 }
1216                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1217                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1218                         msg->connect.nm_nam = NULL;
1219                 }
1220                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1221                 /* msg invalid now */
1222         }
1223 }
1224
1225 static int
1226 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1227                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1228 {
1229         struct mbuf *m = *mp;
1230         struct inpcb *inp = tp->t_inpcb;
1231         struct socket *so = inp->inp_socket;
1232         struct inpcb *oinp;
1233
1234         /*
1235          * Cannot simply call in_pcbconnect, because there might be an
1236          * earlier incarnation of this same connection still in
1237          * TIME_WAIT state, creating an ADDRINUSE error.
1238          */
1239         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1240                                   &sin6->sin6_addr, sin6->sin6_port,
1241                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1242                                       addr6 : &inp->in6p_laddr),
1243                                   inp->inp_lport,  0, NULL);
1244         if (oinp)
1245                 return (EADDRINUSE);
1246
1247         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1248                 inp->in6p_laddr = *addr6;
1249         inp->in6p_faddr = sin6->sin6_addr;
1250         inp->inp_fport = sin6->sin6_port;
1251         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1252                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1253         in_pcbinsconnhash(inp);
1254
1255         /*
1256          * Now that no more errors can occur, change the protocol processing
1257          * port to the current thread (which is the correct thread).
1258          *
1259          * Create TCP timer message now; we are on the tcpcb's owner
1260          * CPU/thread.
1261          */
1262         tcp_create_timermsg(tp, &curthread->td_msgport);
1263
1264         /* Compute window scaling to request.  */
1265         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1266                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1267         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1268             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1269                 tp->request_r_scale++;
1270         }
1271
1272         soisconnecting(so);
1273         tcpstat.tcps_connattempt++;
1274         tp->t_state = TCPS_SYN_SENT;
1275         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1276         tp->iss = tcp_new_isn(tp);
1277         tcp_sendseqinit(tp);
1278         if (m) {
1279                 ssb_appendstream(&so->so_snd, m);
1280                 *mp = NULL;
1281                 if (flags & PRUS_OOB)
1282                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1283         }
1284
1285         /*
1286          * Close the send side of the connection after
1287          * the data is sent if flagged.
1288          */
1289         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1290                 socantsendmore(so);
1291                 tp = tcp_usrclosed(tp);
1292         }
1293         return (tcp_output(tp));
1294 }
1295
1296 #endif /* INET6 */
1297
1298 /*
1299  * The new sockopt interface makes it possible for us to block in the
1300  * copyin/out step (if we take a page fault).  Taking a page fault while
1301  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1302  * both now use TSM, there probably isn't any need for this function to
1303  * run in a critical section any more.  This needs more examination.)
1304  */
1305 void
1306 tcp_ctloutput(netmsg_t msg)
1307 {
1308         struct socket *so = msg->base.nm_so;
1309         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1310         int     error, opt, optval, opthz;
1311         struct  inpcb *inp;
1312         struct  tcpcb *tp;
1313
1314         error = 0;
1315         inp = so->so_pcb;
1316         if (inp == NULL) {
1317                 error = ECONNRESET;
1318                 goto done;
1319         }
1320
1321         if (sopt->sopt_level != IPPROTO_TCP) {
1322 #ifdef INET6
1323                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1324                         ip6_ctloutput_dispatch(msg);
1325                 else
1326 #endif /* INET6 */
1327                 ip_ctloutput(msg);
1328                 /* msg invalid now */
1329                 return;
1330         }
1331         tp = intotcpcb(inp);
1332
1333         switch (sopt->sopt_dir) {
1334         case SOPT_SET:
1335                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1336                                       sizeof optval);
1337                 if (error)
1338                         break;
1339                 switch (sopt->sopt_name) {
1340                 case TCP_FASTKEEP:
1341                         if (optval > 0)
1342                                 tp->t_keepidle = tp->t_keepintvl;
1343                         else
1344                                 tp->t_keepidle = tcp_keepidle;
1345                         tcp_timer_keep_activity(tp, 0);
1346                         break;
1347 #ifdef TCP_SIGNATURE
1348                 case TCP_SIGNATURE_ENABLE:
1349                         if (tp->t_state == TCPS_CLOSED) {
1350                                 /*
1351                                  * This is the only safe state that this
1352                                  * option could be changed.  Some segments
1353                                  * could already have been sent in other
1354                                  * states.
1355                                  */
1356                                 if (optval > 0)
1357                                         tp->t_flags |= TF_SIGNATURE;
1358                                 else
1359                                         tp->t_flags &= ~TF_SIGNATURE;
1360                         } else {
1361                                 error = EOPNOTSUPP;
1362                         }
1363                         break;
1364 #endif /* TCP_SIGNATURE */
1365                 case TCP_NODELAY:
1366                 case TCP_NOOPT:
1367                         switch (sopt->sopt_name) {
1368                         case TCP_NODELAY:
1369                                 opt = TF_NODELAY;
1370                                 break;
1371                         case TCP_NOOPT:
1372                                 opt = TF_NOOPT;
1373                                 break;
1374                         default:
1375                                 opt = 0; /* dead code to fool gcc */
1376                                 break;
1377                         }
1378
1379                         if (optval)
1380                                 tp->t_flags |= opt;
1381                         else
1382                                 tp->t_flags &= ~opt;
1383                         break;
1384
1385                 case TCP_NOPUSH:
1386                         if (tcp_disable_nopush)
1387                                 break;
1388                         if (optval)
1389                                 tp->t_flags |= TF_NOPUSH;
1390                         else {
1391                                 tp->t_flags &= ~TF_NOPUSH;
1392                                 error = tcp_output(tp);
1393                         }
1394                         break;
1395
1396                 case TCP_MAXSEG:
1397                         /*
1398                          * Must be between 0 and maxseg.  If the requested
1399                          * maxseg is too small to satisfy the desired minmss,
1400                          * pump it up (silently so sysctl modifications of
1401                          * minmss do not create unexpected program failures).
1402                          * Handle degenerate cases.
1403                          */
1404                         if (optval > 0 && optval <= tp->t_maxseg) {
1405                                 if (optval + 40 < tcp_minmss) {
1406                                         optval = tcp_minmss - 40;
1407                                         if (optval < 0)
1408                                                 optval = 1;
1409                                 }
1410                                 tp->t_maxseg = optval;
1411                         } else {
1412                                 error = EINVAL;
1413                         }
1414                         break;
1415
1416                 case TCP_KEEPINIT:
1417                         opthz = ((int64_t)optval * hz) / 1000;
1418                         if (opthz >= 1)
1419                                 tp->t_keepinit = opthz;
1420                         else
1421                                 error = EINVAL;
1422                         break;
1423
1424                 case TCP_KEEPIDLE:
1425                         opthz = ((int64_t)optval * hz) / 1000;
1426                         if (opthz >= 1) {
1427                                 tp->t_keepidle = opthz;
1428                                 tcp_timer_keep_activity(tp, 0);
1429                         } else {
1430                                 error = EINVAL;
1431                         }
1432                         break;
1433
1434                 case TCP_KEEPINTVL:
1435                         opthz = ((int64_t)optval * hz) / 1000;
1436                         if (opthz >= 1) {
1437                                 tp->t_keepintvl = opthz;
1438                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1439                         } else {
1440                                 error = EINVAL;
1441                         }
1442                         break;
1443
1444                 case TCP_KEEPCNT:
1445                         if (optval > 0) {
1446                                 tp->t_keepcnt = optval;
1447                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1448                         } else {
1449                                 error = EINVAL;
1450                         }
1451                         break;
1452
1453                 default:
1454                         error = ENOPROTOOPT;
1455                         break;
1456                 }
1457                 break;
1458
1459         case SOPT_GET:
1460                 switch (sopt->sopt_name) {
1461 #ifdef TCP_SIGNATURE
1462                 case TCP_SIGNATURE_ENABLE:
1463                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1464                         break;
1465 #endif /* TCP_SIGNATURE */
1466                 case TCP_NODELAY:
1467                         optval = tp->t_flags & TF_NODELAY;
1468                         break;
1469                 case TCP_MAXSEG:
1470                         optval = tp->t_maxseg;
1471                         break;
1472                 case TCP_NOOPT:
1473                         optval = tp->t_flags & TF_NOOPT;
1474                         break;
1475                 case TCP_NOPUSH:
1476                         optval = tp->t_flags & TF_NOPUSH;
1477                         break;
1478                 case TCP_KEEPINIT:
1479                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1480                         break;
1481                 case TCP_KEEPIDLE:
1482                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1483                         break;
1484                 case TCP_KEEPINTVL:
1485                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1486                         break;
1487                 case TCP_KEEPCNT:
1488                         optval = tp->t_keepcnt;
1489                         break;
1490                 default:
1491                         error = ENOPROTOOPT;
1492                         break;
1493                 }
1494                 if (error == 0)
1495                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1496                 break;
1497         }
1498 done:
1499         lwkt_replymsg(&msg->lmsg, error);
1500 }
1501
1502 /*
1503  * tcp_sendspace and tcp_recvspace are the default send and receive window
1504  * sizes, respectively.  These are obsolescent (this information should
1505  * be set by the route).
1506  *
1507  * Use a default that does not require tcp window scaling to be turned
1508  * on.  Individual programs or the administrator can increase the default.
1509  */
1510 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1511 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1512     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1513 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1514 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1515     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1516
1517 /*
1518  * Attach TCP protocol to socket, allocating internet protocol control
1519  * block, tcp control block, bufer space, and entering LISTEN state
1520  * if to accept connections.
1521  */
1522 static int
1523 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1524 {
1525         struct tcpcb *tp;
1526         struct inpcb *inp;
1527         int error;
1528         int cpu;
1529 #ifdef INET6
1530         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1531 #endif
1532
1533         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1534                 lwkt_gettoken(&so->so_rcv.ssb_token);
1535                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1536                                   ai->sb_rlimit);
1537                 lwkt_reltoken(&so->so_rcv.ssb_token);
1538                 if (error)
1539                         return (error);
1540         }
1541         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1542         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1543         cpu = mycpu->gd_cpuid;
1544
1545         /*
1546          * Set the default port for protocol processing. This will likely
1547          * change when we connect.
1548          */
1549         error = in_pcballoc(so, &tcbinfo[cpu]);
1550         if (error)
1551                 return (error);
1552         inp = so->so_pcb;
1553 #ifdef INET6
1554         if (isipv6) {
1555                 inp->inp_vflag |= INP_IPV6;
1556                 inp->in6p_hops = -1;    /* use kernel default */
1557         }
1558         else
1559 #endif
1560         inp->inp_vflag |= INP_IPV4;
1561         tp = tcp_newtcpcb(inp);
1562         if (tp == NULL) {
1563                 /*
1564                  * Make sure the socket is destroyed by the pcbdetach.
1565                  */
1566                 soreference(so);
1567 #ifdef INET6
1568                 if (isipv6)
1569                         in6_pcbdetach(inp);
1570                 else
1571 #endif
1572                 in_pcbdetach(inp);
1573                 sofree(so);     /* from ref above */
1574                 return (ENOBUFS);
1575         }
1576         tp->t_state = TCPS_CLOSED;
1577         /* Keep a reference for asynchronized pru_rcvd */
1578         soreference(so);
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 }