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