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  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
  69  */
  70
  71 #include "opt_ipsec.h"
  72 #include "opt_inet.h"
  73 #include "opt_inet6.h"
  74 #include "opt_tcpdebug.h"
  75
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/kernel.h>
  79 #include <sys/malloc.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/globaldata.h>
  82 #include <sys/thread.h>
  83
  84 #include <sys/mbuf.h>
  85 #ifdef INET6
  86 #include <sys/domain.h>
  87 #endif /* INET6 */
  88 #include <sys/socket.h>
  89 #include <sys/socketvar.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 /*
 163  * TCP attaches to socket via pru_attach(), reserving space,
 164  * and an internet control block.  This is likely occuring on
 165  * cpu0 and may have to move later when we bind/connect.
 166  */
 167 static void
 168 tcp_usr_attach(netmsg_t msg)
 169 {
 170         struct socket *so = msg->base.nm_so;
 171         struct pru_attach_info *ai = msg->attach.nm_ai;
 172         int error;
 173         struct inpcb *inp;
 174         struct tcpcb *tp = 0;
 175         TCPDEBUG0;
 176
 177         soreference(so);
 178         inp = so->so_pcb;
 179         TCPDEBUG1();
 180         if (inp) {
 181                 error = EISCONN;
 182                 goto out;
 183         }
 184
 185         error = tcp_attach(so, ai);
 186         if (error)
 187                 goto out;
 188
 189         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 190                 so->so_linger = TCP_LINGERTIME;
 191         tp = sototcpcb(so);
 192 out:
 193         sofree(so);             /* from ref above */
 194         TCPDEBUG2(PRU_ATTACH);
 195         lwkt_replymsg(&msg->lmsg, error);
 196 }
 197
 198 /*
 199  * pru_detach() detaches the TCP protocol from the socket.
 200  * If the protocol state is non-embryonic, then can't
 201  * do this directly: have to initiate a pru_disconnect(),
 202  * which may finish later; embryonic TCB's can just
 203  * be discarded here.
 204  */
 205 static void
 206 tcp_usr_detach(netmsg_t msg)
 207 {
 208         struct socket *so = msg->base.nm_so;
 209         int error = 0;
 210         struct inpcb *inp;
 211         struct tcpcb *tp;
 212         TCPDEBUG0;
 213
 214         inp = so->so_pcb;
 215
 216         /*
 217          * If the inp is already detached it may have been due to an async
 218          * close.  Just return as if no error occured.
 219          *
 220          * It's possible for the tcpcb (tp) to disconnect from the inp due
 221          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 222          * the detach message has been queued so we may find a NULL tp here.
 223          */
 224         if (inp) {
 225                 if ((tp = intotcpcb(inp)) != NULL) {
 226                         TCPDEBUG1();
 227                         tp = tcp_disconnect(tp);
 228                         TCPDEBUG2(PRU_DETACH);
 229                 }
 230         }
 231         lwkt_replymsg(&msg->lmsg, error);
 232 }
 233
 234 /*
 235  * NOTE: ignore_error is non-zero for certain disconnection races
 236  * which we want to silently allow, otherwise close() may return
 237  * an unexpected error.
 238  *
 239  * NOTE: The variables (msg) and (tp) are assumed.
 240  */
 241 #define COMMON_START(so, inp, ignore_error)                     \
 242         TCPDEBUG0;                                              \
 243                                                                 \
 244         inp = so->so_pcb;                                       \
 245         do {                                                    \
 246                  if (inp == NULL) {                             \
 247                         error = ignore_error ? 0 : EINVAL;      \
 248                         tp = NULL;                              \
 249                         goto out;                               \
 250                  }                                              \
 251                  tp = intotcpcb(inp);                           \
 252                  TCPDEBUG1();                                   \
 253         } while(0)
 254
 255 #define COMMON_END(req)                                         \
 256         out: do {                                               \
 257                 TCPDEBUG2(req);                                 \
 258                 lwkt_replymsg(&msg->lmsg, error);               \
 259                 return;                                         \
 260         } while(0)
 261
 262 /*
 263  * Give the socket an address.
 264  */
 265 static void
 266 tcp_usr_bind(netmsg_t msg)
 267 {
 268         struct socket *so = msg->bind.base.nm_so;
 269         struct sockaddr *nam = msg->bind.nm_nam;
 270         struct thread *td = msg->bind.nm_td;
 271         int error = 0;
 272         struct inpcb *inp;
 273         struct tcpcb *tp;
 274         struct sockaddr_in *sinp;
 275
 276         COMMON_START(so, inp, 0);
 277
 278         /*
 279          * Must check for multicast addresses and disallow binding
 280          * to them.
 281          */
 282         sinp = (struct sockaddr_in *)nam;
 283         if (sinp->sin_family == AF_INET &&
 284             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 285                 error = EAFNOSUPPORT;
 286                 goto out;
 287         }
 288         error = in_pcbbind(inp, nam, td);
 289         if (error)
 290                 goto out;
 291         COMMON_END(PRU_BIND);
 292
 293 }
 294
 295 #ifdef INET6
 296
 297 static void
 298 tcp6_usr_bind(netmsg_t msg)
 299 {
 300         struct socket *so = msg->bind.base.nm_so;
 301         struct sockaddr *nam = msg->bind.nm_nam;
 302         struct thread *td = msg->bind.nm_td;
 303         int error = 0;
 304         struct inpcb *inp;
 305         struct tcpcb *tp;
 306         struct sockaddr_in6 *sin6p;
 307
 308         COMMON_START(so, inp, 0);
 309
 310         /*
 311          * Must check for multicast addresses and disallow binding
 312          * to them.
 313          */
 314         sin6p = (struct sockaddr_in6 *)nam;
 315         if (sin6p->sin6_family == AF_INET6 &&
 316             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 317                 error = EAFNOSUPPORT;
 318                 goto out;
 319         }
 320         inp->inp_vflag &= ~INP_IPV4;
 321         inp->inp_vflag |= INP_IPV6;
 322         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 323                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 324                         inp->inp_vflag |= INP_IPV4;
 325                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 326                         struct sockaddr_in sin;
 327
 328                         in6_sin6_2_sin(&sin, sin6p);
 329                         inp->inp_vflag |= INP_IPV4;
 330                         inp->inp_vflag &= ~INP_IPV6;
 331                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 332                         goto out;
 333                 }
 334         }
 335         error = in6_pcbbind(inp, nam, td);
 336         if (error)
 337                 goto out;
 338         COMMON_END(PRU_BIND);
 339 }
 340 #endif /* INET6 */
 341
 342 #ifdef SMP
 343
 344 struct netmsg_inswildcard {
 345         struct netmsg_base      base;
 346         struct inpcb            *nm_inp;
 347 };
 348
 349 static void
 350 in_pcbinswildcardhash_handler(netmsg_t msg)
 351 {
 352         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 353         int cpu = mycpuid, nextcpu;
 354
 355         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 356
 357         nextcpu = cpu + 1;
 358         if (nextcpu < ncpus2)
 359                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
 360         else
 361                 lwkt_replymsg(&nm->base.lmsg, 0);
 362 }
 363
 364 #endif
 365
 366 /*
 367  * Prepare to accept connections.
 368  */
 369 static void
 370 tcp_usr_listen(netmsg_t msg)
 371 {
 372         struct socket *so = msg->listen.base.nm_so;
 373         struct thread *td = msg->listen.nm_td;
 374         int error = 0;
 375         struct inpcb *inp;
 376         struct tcpcb *tp;
 377 #ifdef SMP
 378         struct netmsg_inswildcard nm;
 379 #endif
 380
 381         COMMON_START(so, inp, 0);
 382         if (inp->inp_lport == 0) {
 383                 error = in_pcbbind(inp, NULL, td);
 384                 if (error)
 385                         goto out;
 386         }
 387
 388         tp->t_state = TCPS_LISTEN;
 389         tp->t_flags |= TF_SYNCACHE;
 390         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 391
 392 #ifdef SMP
 393         /*
 394          * We have to set the flag because we can't have other cpus
 395          * messing with our inp's flags.
 396          */
 397         inp->inp_flags |= INP_WILDCARD_MP;
 398
 399         KKASSERT(so->so_port == cpu_portfn(0));
 400         KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 401         KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 402
 403         netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 404                     MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 405         nm.nm_inp = inp;
 406         lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 407 #endif
 408         in_pcbinswildcardhash(inp);
 409         COMMON_END(PRU_LISTEN);
 410 }
 411
 412 #ifdef INET6
 413
 414 static void
 415 tcp6_usr_listen(netmsg_t msg)
 416 {
 417         struct socket *so = msg->listen.base.nm_so;
 418         struct thread *td = msg->listen.nm_td;
 419         int error = 0;
 420         struct inpcb *inp;
 421         struct tcpcb *tp;
 422 #ifdef SMP
 423         struct netmsg_inswildcard nm;
 424 #endif
 425
 426         COMMON_START(so, inp, 0);
 427         if (inp->inp_lport == 0) {
 428                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 429                         inp->inp_vflag |= INP_IPV4;
 430                 else
 431                         inp->inp_vflag &= ~INP_IPV4;
 432                 error = in6_pcbbind(inp, NULL, td);
 433                 if (error)
 434                         goto out;
 435         }
 436
 437         tp->t_state = TCPS_LISTEN;
 438         tp->t_flags |= TF_SYNCACHE;
 439         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 440
 441 #ifdef SMP
 442         /*
 443          * We have to set the flag because we can't have other cpus
 444          * messing with our inp's flags.
 445          */
 446         inp->inp_flags |= INP_WILDCARD_MP;
 447
 448         KKASSERT(so->so_port == cpu_portfn(0));
 449         KKASSERT(&curthread->td_msgport == cpu_portfn(0));
 450         KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 451
 452         netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 453                     MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 454         nm.nm_inp = inp;
 455         lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
 456 #endif
 457         in_pcbinswildcardhash(inp);
 458         COMMON_END(PRU_LISTEN);
 459 }
 460 #endif /* INET6 */
 461
 462 /*
 463  * Initiate connection to peer.
 464  * Create a template for use in transmissions on this connection.
 465  * Enter SYN_SENT state, and mark socket as connecting.
 466  * Start keep-alive timer, and seed output sequence space.
 467  * Send initial segment on connection.
 468  */
 469 static void
 470 tcp_usr_connect(netmsg_t msg)
 471 {
 472         struct socket *so = msg->connect.base.nm_so;
 473         struct sockaddr *nam = msg->connect.nm_nam;
 474         struct thread *td = msg->connect.nm_td;
 475         int error = 0;
 476         struct inpcb *inp;
 477         struct tcpcb *tp;
 478         struct sockaddr_in *sinp;
 479
 480         COMMON_START(so, inp, 0);
 481
 482         /*
 483          * Must disallow TCP ``connections'' to multicast addresses.
 484          */
 485         sinp = (struct sockaddr_in *)nam;
 486         if (sinp->sin_family == AF_INET
 487             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 488                 error = EAFNOSUPPORT;
 489                 goto out;
 490         }
 491
 492         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 493                 error = EAFNOSUPPORT; /* IPv6 only jail */
 494                 goto out;
 495         }
 496
 497         tcp_connect(msg);
 498         /* msg is invalid now */
 499         return;
 500 out:
 501         if (msg->connect.nm_m) {
 502                 m_freem(msg->connect.nm_m);
 503                 msg->connect.nm_m = NULL;
 504         }
 505         lwkt_replymsg(&msg->lmsg, error);
 506 }
 507
 508 #ifdef INET6
 509
 510 static void
 511 tcp6_usr_connect(netmsg_t msg)
 512 {
 513         struct socket *so = msg->connect.base.nm_so;
 514         struct sockaddr *nam = msg->connect.nm_nam;
 515         struct thread *td = msg->connect.nm_td;
 516         int error = 0;
 517         struct inpcb *inp;
 518         struct tcpcb *tp;
 519         struct sockaddr_in6 *sin6p;
 520
 521         COMMON_START(so, inp, 0);
 522
 523         /*
 524          * Must disallow TCP ``connections'' to multicast addresses.
 525          */
 526         sin6p = (struct sockaddr_in6 *)nam;
 527         if (sin6p->sin6_family == AF_INET6
 528             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 529                 error = EAFNOSUPPORT;
 530                 goto out;
 531         }
 532
 533         if (!prison_remote_ip(td, nam)) {
 534                 error = EAFNOSUPPORT; /* IPv4 only jail */
 535                 goto out;
 536         }
 537
 538         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 539                 struct sockaddr_in *sinp;
 540
 541                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 542                         error = EINVAL;
 543                         goto out;
 544                 }
 545                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 546                 in6_sin6_2_sin(sinp, sin6p);
 547                 inp->inp_vflag |= INP_IPV4;
 548                 inp->inp_vflag &= ~INP_IPV6;
 549                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 550                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
 551                 tcp_connect(msg);
 552                 /* msg is invalid now */
 553                 return;
 554         }
 555         inp->inp_vflag &= ~INP_IPV4;
 556         inp->inp_vflag |= INP_IPV6;
 557         inp->inp_inc.inc_isipv6 = 1;
 558
 559         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
 560         tcp6_connect(msg);
 561         /* msg is invalid now */
 562         return;
 563 out:
 564         if (msg->connect.nm_m) {
 565                 m_freem(msg->connect.nm_m);
 566                 msg->connect.nm_m = NULL;
 567         }
 568         lwkt_replymsg(&msg->lmsg, error);
 569 }
 570
 571 #endif /* INET6 */
 572
 573 /*
 574  * Initiate disconnect from peer.
 575  * If connection never passed embryonic stage, just drop;
 576  * else if don't need to let data drain, then can just drop anyways,
 577  * else have to begin TCP shutdown process: mark socket disconnecting,
 578  * drain unread data, state switch to reflect user close, and
 579  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 580  * when peer sends FIN and acks ours.
 581  *
 582  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 583  */
 584 static void
 585 tcp_usr_disconnect(netmsg_t msg)
 586 {
 587         struct socket *so = msg->disconnect.base.nm_so;
 588         int error = 0;
 589         struct inpcb *inp;
 590         struct tcpcb *tp;
 591
 592         COMMON_START(so, inp, 1);
 593         tp = tcp_disconnect(tp);
 594         COMMON_END(PRU_DISCONNECT);
 595 }
 596
 597 /*
 598  * Accept a connection.  Essentially all the work is
 599  * done at higher levels; just return the address
 600  * of the peer, storing through addr.
 601  */
 602 static void
 603 tcp_usr_accept(netmsg_t msg)
 604 {
 605         struct socket *so = msg->accept.base.nm_so;
 606         struct sockaddr **nam = msg->accept.nm_nam;
 607         int error = 0;
 608         struct inpcb *inp;
 609         struct tcpcb *tp = NULL;
 610         TCPDEBUG0;
 611
 612         inp = so->so_pcb;
 613         if (so->so_state & SS_ISDISCONNECTED) {
 614                 error = ECONNABORTED;
 615                 goto out;
 616         }
 617         if (inp == 0) {
 618                 error = EINVAL;
 619                 goto out;
 620         }
 621
 622         tp = intotcpcb(inp);
 623         TCPDEBUG1();
 624         in_setpeeraddr(so, nam);
 625         COMMON_END(PRU_ACCEPT);
 626 }
 627
 628 #ifdef INET6
 629 static void
 630 tcp6_usr_accept(netmsg_t msg)
 631 {
 632         struct socket *so = msg->accept.base.nm_so;
 633         struct sockaddr **nam = msg->accept.nm_nam;
 634         int error = 0;
 635         struct inpcb *inp;
 636         struct tcpcb *tp = NULL;
 637         TCPDEBUG0;
 638
 639         inp = so->so_pcb;
 640
 641         if (so->so_state & SS_ISDISCONNECTED) {
 642                 error = ECONNABORTED;
 643                 goto out;
 644         }
 645         if (inp == 0) {
 646                 error = EINVAL;
 647                 goto out;
 648         }
 649         tp = intotcpcb(inp);
 650         TCPDEBUG1();
 651         in6_mapped_peeraddr(so, nam);
 652         COMMON_END(PRU_ACCEPT);
 653 }
 654 #endif /* INET6 */
 655 /*
 656  * Mark the connection as being incapable of further output.
 657  */
 658 static void
 659 tcp_usr_shutdown(netmsg_t msg)
 660 {
 661         struct socket *so = msg->shutdown.base.nm_so;
 662         int error = 0;
 663         struct inpcb *inp;
 664         struct tcpcb *tp;
 665
 666         COMMON_START(so, inp, 0);
 667         socantsendmore(so);
 668         tp = tcp_usrclosed(tp);
 669         if (tp)
 670                 error = tcp_output(tp);
 671         COMMON_END(PRU_SHUTDOWN);
 672 }
 673
 674 /*
 675  * After a receive, possibly send window update to peer.
 676  */
 677 static void
 678 tcp_usr_rcvd(netmsg_t msg)
 679 {
 680         struct socket *so = msg->rcvd.base.nm_so;
 681         int error = 0;
 682         struct inpcb *inp;
 683         struct tcpcb *tp;
 684
 685         COMMON_START(so, inp, 0);
 686         tcp_output(tp);
 687         COMMON_END(PRU_RCVD);
 688 }
 689
 690 /*
 691  * Do a send by putting data in output queue and updating urgent
 692  * marker if URG set.  Possibly send more data.  Unlike the other
 693  * pru_*() routines, the mbuf chains are our responsibility.  We
 694  * must either enqueue them or free them.  The other pru_* routines
 695  * generally are caller-frees.
 696  */
 697 static void
 698 tcp_usr_send(netmsg_t msg)
 699 {
 700         struct socket *so = msg->send.base.nm_so;
 701         int flags = msg->send.nm_flags;
 702         struct mbuf *m = msg->send.nm_m;
 703         struct sockaddr *nam = msg->send.nm_addr;
 704         struct mbuf *control = msg->send.nm_control;
 705         struct thread *td = msg->send.nm_td;
 706         int error = 0;
 707         struct inpcb *inp;
 708         struct tcpcb *tp;
 709 #ifdef INET6
 710         int isipv6;
 711 #endif
 712         TCPDEBUG0;
 713
 714         inp = so->so_pcb;
 715
 716         if (inp == NULL) {
 717                 /*
 718                  * OOPS! we lost a race, the TCP session got reset after
 719                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 720                  * network interrupt in the non-critical section of sosend().
 721                  */
 722                 m_freem(m);
 723                 if (control)
 724                         m_freem(control);
 725                 error = ECONNRESET;     /* XXX EPIPE? */
 726                 tp = NULL;
 727                 TCPDEBUG1();
 728                 goto out;
 729         }
 730 #ifdef INET6
 731         isipv6 = nam && nam->sa_family == AF_INET6;
 732 #endif /* INET6 */
 733         tp = intotcpcb(inp);
 734         TCPDEBUG1();
 735         if (control) {
 736                 /* TCP doesn't do control messages (rights, creds, etc) */
 737                 if (control->m_len) {
 738                         m_freem(control);
 739                         m_freem(m);
 740                         error = EINVAL;
 741                         goto out;
 742                 }
 743                 m_freem(control);       /* empty control, just free it */
 744         }
 745
 746         /*
 747          * Don't let too much OOB data build up
 748          */
 749         if (flags & PRUS_OOB) {
 750                 if (ssb_space(&so->so_snd) < -512) {
 751                         m_freem(m);
 752                         error = ENOBUFS;
 753                         goto out;
 754                 }
 755         }
 756
 757         /*
 758          * Do implied connect if not yet connected.  Any data sent
 759          * with the connect is handled by tcp_connect() and friends.
 760          *
 761          * NOTE!  PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
 762          */
 763         if (nam && tp->t_state < TCPS_SYN_SENT) {
 764                 kprintf("implied fallback\n");
 765                 msg->connect.nm_nam = nam;
 766                 msg->connect.nm_td = td;
 767                 msg->connect.nm_m = m;
 768                 msg->connect.nm_flags = flags;
 769                 msg->connect.nm_reconnect = NMSG_RECONNECT_FALLBACK;
 770 #ifdef INET6
 771                 if (isipv6)
 772                         tcp6_connect(msg);
 773                 else
 774 #endif /* INET6 */
 775                         tcp_connect(msg);
 776                 /* msg invalid now */
 777                 return;
 778         }
 779
 780         /*
 781          * Pump the data into the socket.
 782          */
 783         if (m)
 784                 ssb_appendstream(&so->so_snd, m);
 785         if (flags & PRUS_OOB) {
 786                 /*
 787                  * According to RFC961 (Assigned Protocols),
 788                  * the urgent pointer points to the last octet
 789                  * of urgent data.  We continue, however,
 790                  * to consider it to indicate the first octet
 791                  * of data past the urgent section.
 792                  * Otherwise, snd_up should be one lower.
 793                  */
 794                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 795                 tp->t_flags |= TF_FORCE;
 796                 error = tcp_output(tp);
 797                 tp->t_flags &= ~TF_FORCE;
 798         } else {
 799                 if (flags & PRUS_EOF) {
 800                         /*
 801                          * Close the send side of the connection after
 802                          * the data is sent.
 803                          */
 804                         socantsendmore(so);
 805                         tp = tcp_usrclosed(tp);
 806                 }
 807                 if (tp != NULL) {
 808                         if (flags & PRUS_MORETOCOME)
 809                                 tp->t_flags |= TF_MORETOCOME;
 810                         error = tcp_output(tp);
 811                         if (flags & PRUS_MORETOCOME)
 812                                 tp->t_flags &= ~TF_MORETOCOME;
 813                 }
 814         }
 815         COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
 816                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
 817 }
 818
 819 /*
 820  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 821  *       will sofree() it when we return.
 822  */
 823 static void
 824 tcp_usr_abort(netmsg_t msg)
 825 {
 826         struct socket *so = msg->abort.base.nm_so;
 827         int error = 0;
 828         struct inpcb *inp;
 829         struct tcpcb *tp;
 830
 831         COMMON_START(so, inp, 1);
 832         tp = tcp_drop(tp, ECONNABORTED);
 833         COMMON_END(PRU_ABORT);
 834 }
 835
 836 /*
 837  * Receive out-of-band data.
 838  */
 839 static void
 840 tcp_usr_rcvoob(netmsg_t msg)
 841 {
 842         struct socket *so = msg->rcvoob.base.nm_so;
 843         struct mbuf *m = msg->rcvoob.nm_m;
 844         int flags = msg->rcvoob.nm_flags;
 845         int error = 0;
 846         struct inpcb *inp;
 847         struct tcpcb *tp;
 848
 849         COMMON_START(so, inp, 0);
 850         if ((so->so_oobmark == 0 &&
 851              (so->so_state & SS_RCVATMARK) == 0) ||
 852             so->so_options & SO_OOBINLINE ||
 853             tp->t_oobflags & TCPOOB_HADDATA) {
 854                 error = EINVAL;
 855                 goto out;
 856         }
 857         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 858                 error = EWOULDBLOCK;
 859                 goto out;
 860         }
 861         m->m_len = 1;
 862         *mtod(m, caddr_t) = tp->t_iobc;
 863         if ((flags & MSG_PEEK) == 0)
 864                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 865         COMMON_END(PRU_RCVOOB);
 866 }
 867
 868 /* xxx - should be const */
 869 struct pr_usrreqs tcp_usrreqs = {
 870         .pru_abort = tcp_usr_abort,
 871         .pru_accept = tcp_usr_accept,
 872         .pru_attach = tcp_usr_attach,
 873         .pru_bind = tcp_usr_bind,
 874         .pru_connect = tcp_usr_connect,
 875         .pru_connect2 = pr_generic_notsupp,
 876         .pru_control = in_control_dispatch,
 877         .pru_detach = tcp_usr_detach,
 878         .pru_disconnect = tcp_usr_disconnect,
 879         .pru_listen = tcp_usr_listen,
 880         .pru_peeraddr = in_setpeeraddr_dispatch,
 881         .pru_rcvd = tcp_usr_rcvd,
 882         .pru_rcvoob = tcp_usr_rcvoob,
 883         .pru_send = tcp_usr_send,
 884         .pru_sense = pru_sense_null,
 885         .pru_shutdown = tcp_usr_shutdown,
 886         .pru_sockaddr = in_setsockaddr_dispatch,
 887         .pru_sosend = sosend,
 888         .pru_soreceive = soreceive
 889 };
 890
 891 #ifdef INET6
 892 struct pr_usrreqs tcp6_usrreqs = {
 893         .pru_abort = tcp_usr_abort,
 894         .pru_accept = tcp6_usr_accept,
 895         .pru_attach = tcp_usr_attach,
 896         .pru_bind = tcp6_usr_bind,
 897         .pru_connect = tcp6_usr_connect,
 898         .pru_connect2 = pr_generic_notsupp,
 899         .pru_control = in6_control_dispatch,
 900         .pru_detach = tcp_usr_detach,
 901         .pru_disconnect = tcp_usr_disconnect,
 902         .pru_listen = tcp6_usr_listen,
 903         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 904         .pru_rcvd = tcp_usr_rcvd,
 905         .pru_rcvoob = tcp_usr_rcvoob,
 906         .pru_send = tcp_usr_send,
 907         .pru_sense = pru_sense_null,
 908         .pru_shutdown = tcp_usr_shutdown,
 909         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 910         .pru_sosend = sosend,
 911         .pru_soreceive = soreceive
 912 };
 913 #endif /* INET6 */
 914
 915 static int
 916 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 917                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 918 {
 919         struct inpcb *inp = tp->t_inpcb, *oinp;
 920         struct socket *so = inp->inp_socket;
 921         struct route *ro = &inp->inp_route;
 922
 923         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
 924                                  sin->sin_addr, sin->sin_port,
 925                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 926                                   inp->inp_laddr : if_sin->sin_addr),
 927                                 inp->inp_lport, 0, NULL);
 928         if (oinp != NULL) {
 929                 m_freem(m);
 930                 return (EADDRINUSE);
 931         }
 932         if (inp->inp_laddr.s_addr == INADDR_ANY)
 933                 inp->inp_laddr = if_sin->sin_addr;
 934         inp->inp_faddr = sin->sin_addr;
 935         inp->inp_fport = sin->sin_port;
 936         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
 937         in_pcbinsconnhash(inp);
 938
 939         /*
 940          * We are now on the inpcb's owner CPU, if the cached route was
 941          * freed because the rtentry's owner CPU is not the current CPU
 942          * (e.g. in tcp_connect()), then we try to reallocate it here with
 943          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 944          * rtentry.
 945          */
 946         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 947             ro->ro_rt == NULL) {
 948                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 949                 ro->ro_dst.sa_family = AF_INET;
 950                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 951                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 952                         sin->sin_addr;
 953                 rtalloc(ro);
 954         }
 955
 956         /*
 957          * Now that no more errors can occur, change the protocol processing
 958          * port to the current thread (which is the correct thread).
 959          *
 960          * Create TCP timer message now; we are on the tcpcb's owner
 961          * CPU/thread.
 962          */
 963         tcp_create_timermsg(tp, &curthread->td_msgport);
 964
 965         /*
 966          * Compute window scaling to request.  Use a larger scaling then
 967          * needed for the initial receive buffer in case the receive buffer
 968          * gets expanded.
 969          */
 970         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
 971                 tp->request_r_scale = TCP_MIN_WINSHIFT;
 972         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
 973                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
 974         ) {
 975                 tp->request_r_scale++;
 976         }
 977
 978         soisconnecting(so);
 979         tcpstat.tcps_connattempt++;
 980         tp->t_state = TCPS_SYN_SENT;
 981         tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
 982         tp->iss = tcp_new_isn(tp);
 983         tcp_sendseqinit(tp);
 984         if (m) {
 985                 ssb_appendstream(&so->so_snd, m);
 986                 m = NULL;
 987                 if (flags & PRUS_OOB)
 988                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 989         }
 990
 991         /*
 992          * Close the send side of the connection after
 993          * the data is sent if flagged.
 994          */
 995         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
 996                 socantsendmore(so);
 997                 tp = tcp_usrclosed(tp);
 998         }
 999         return (tcp_output(tp));
1000 }
1001
1002 /*
1003  * Common subroutine to open a TCP connection to remote host specified
1004  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1005  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1006  * a local host address (interface).
1007  * Initialize connection parameters and enter SYN-SENT state.
1008  */
1009 static void
1010 tcp_connect(netmsg_t msg)
1011 {
1012         struct socket *so = msg->connect.base.nm_so;
1013         struct sockaddr *nam = msg->connect.nm_nam;
1014         struct thread *td = msg->connect.nm_td;
1015         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1016         struct sockaddr_in *if_sin;
1017         struct inpcb *inp;
1018         struct tcpcb *tp;
1019         int error;
1020 #ifdef SMP
1021         lwkt_port_t port;
1022 #endif
1023
1024         COMMON_START(so, inp, 0);
1025
1026         /*
1027          * Reconnect our pcb if we have to
1028          */
1029         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1030                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1031                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1032         }
1033
1034         /*
1035          * Bind if we have to
1036          */
1037         if (inp->inp_lport == 0) {
1038                 error = in_pcbbind(inp, NULL, td);
1039                 if (error)
1040                         goto out;
1041         }
1042         so = inp->inp_socket;
1043         KKASSERT(so);
1044
1045         /*
1046          * Calculate the correct protocol processing thread.  The connect
1047          * operation must run there.  Set the forwarding port before we
1048          * forward the message or it will get bounced right back to us.
1049          */
1050         error = in_pcbladdr(inp, nam, &if_sin, td);
1051         if (error)
1052                 goto out;
1053
1054 #ifdef SMP
1055         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1056                             (inp->inp_laddr.s_addr ?
1057                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1058                             inp->inp_lport);
1059
1060         if (port != &curthread->td_msgport) {
1061                 struct route *ro = &inp->inp_route;
1062
1063                 /*
1064                  * in_pcbladdr() may have allocated a route entry for us
1065                  * on the current CPU, but we need a route entry on the
1066                  * inpcb's owner CPU, so free it here.
1067                  */
1068                 if (ro->ro_rt != NULL)
1069                         RTFREE(ro->ro_rt);
1070                 bzero(ro, sizeof(*ro));
1071
1072                 /*
1073                  * We are moving the protocol processing port the socket
1074                  * is on, we have to unlink here and re-link on the
1075                  * target cpu.
1076                  */
1077                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1078                 sosetport(so, port);
1079                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1080                 msg->connect.base.nm_dispatch = tcp_connect;
1081
1082                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1083                 /* msg invalid now */
1084                 return;
1085         }
1086 #else
1087         KKASSERT(so->so_port == &curthread->td_msgport);
1088 #endif
1089         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1090                                   msg->connect.nm_m, sin, if_sin);
1091         msg->connect.nm_m = NULL;
1092 out:
1093         if (msg->connect.nm_m) {
1094                 m_freem(msg->connect.nm_m);
1095                 msg->connect.nm_m = NULL;
1096         }
1097         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1098                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1099                 msg->connect.nm_nam = NULL;
1100         }
1101         lwkt_replymsg(&msg->connect.base.lmsg, error);
1102         /* msg invalid now */
1103 }
1104
1105 #ifdef INET6
1106
1107 static void
1108 tcp6_connect(netmsg_t msg)
1109 {
1110         struct tcpcb *tp;
1111         struct socket *so = msg->connect.base.nm_so;
1112         struct sockaddr *nam = msg->connect.nm_nam;
1113         struct thread *td = msg->connect.nm_td;
1114         struct inpcb *inp;
1115         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1116         struct in6_addr *addr6;
1117 #ifdef SMP
1118         lwkt_port_t port;
1119 #endif
1120         int error;
1121
1122         COMMON_START(so, inp, 0);
1123
1124         /*
1125          * Reconnect our pcb if we have to
1126          */
1127         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1128                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1129                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1130         }
1131
1132         /*
1133          * Bind if we have to
1134          */
1135         if (inp->inp_lport == 0) {
1136                 error = in6_pcbbind(inp, NULL, td);
1137                 if (error)
1138                         goto out;
1139         }
1140
1141         /*
1142          * Cannot simply call in_pcbconnect, because there might be an
1143          * earlier incarnation of this same connection still in
1144          * TIME_WAIT state, creating an ADDRINUSE error.
1145          */
1146         error = in6_pcbladdr(inp, nam, &addr6, td);
1147         if (error)
1148                 goto out;
1149
1150 #ifdef SMP
1151         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1152
1153         if (port != &curthread->td_msgport) {
1154                 struct route *ro = &inp->inp_route;
1155
1156                 /*
1157                  * in_pcbladdr() may have allocated a route entry for us
1158                  * on the current CPU, but we need a route entry on the
1159                  * inpcb's owner CPU, so free it here.
1160                  */
1161                 if (ro->ro_rt != NULL)
1162                         RTFREE(ro->ro_rt);
1163                 bzero(ro, sizeof(*ro));
1164
1165                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1166                 sosetport(so, port);
1167                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1168                 msg->connect.base.nm_dispatch = tcp6_connect;
1169
1170                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1171                 /* msg invalid now */
1172                 return;
1173         }
1174 #endif
1175         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1176                                    &msg->connect.nm_m, sin6, addr6);
1177         /* nm_m may still be intact */
1178 out:
1179         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1180                 tcp_connect(msg);
1181                 /* msg invalid now */
1182         } else {
1183                 if (msg->connect.nm_m) {
1184                         m_freem(msg->connect.nm_m);
1185                         msg->connect.nm_m = NULL;
1186                 }
1187                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1188                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1189                         msg->connect.nm_nam = NULL;
1190                 }
1191                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1192                 /* msg invalid now */
1193         }
1194 }
1195
1196 static int
1197 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1198                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1199 {
1200         struct mbuf *m = *mp;
1201         struct inpcb *inp = tp->t_inpcb;
1202         struct socket *so = inp->inp_socket;
1203         struct inpcb *oinp;
1204
1205         /*
1206          * Cannot simply call in_pcbconnect, because there might be an
1207          * earlier incarnation of this same connection still in
1208          * TIME_WAIT state, creating an ADDRINUSE error.
1209          */
1210         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1211                                   &sin6->sin6_addr, sin6->sin6_port,
1212                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1213                                       addr6 : &inp->in6p_laddr),
1214                                   inp->inp_lport,  0, NULL);
1215         if (oinp)
1216                 return (EADDRINUSE);
1217
1218         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1219                 inp->in6p_laddr = *addr6;
1220         inp->in6p_faddr = sin6->sin6_addr;
1221         inp->inp_fport = sin6->sin6_port;
1222         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1223                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1224         in_pcbinsconnhash(inp);
1225
1226         /*
1227          * Now that no more errors can occur, change the protocol processing
1228          * port to the current thread (which is the correct thread).
1229          *
1230          * Create TCP timer message now; we are on the tcpcb's owner
1231          * CPU/thread.
1232          */
1233         tcp_create_timermsg(tp, &curthread->td_msgport);
1234
1235         /* Compute window scaling to request.  */
1236         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1237                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1238         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1239             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1240                 tp->request_r_scale++;
1241         }
1242
1243         soisconnecting(so);
1244         tcpstat.tcps_connattempt++;
1245         tp->t_state = TCPS_SYN_SENT;
1246         tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1247         tp->iss = tcp_new_isn(tp);
1248         tcp_sendseqinit(tp);
1249         if (m) {
1250                 ssb_appendstream(&so->so_snd, m);
1251                 *mp = NULL;
1252                 if (flags & PRUS_OOB)
1253                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1254         }
1255
1256         /*
1257          * Close the send side of the connection after
1258          * the data is sent if flagged.
1259          */
1260         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1261                 socantsendmore(so);
1262                 tp = tcp_usrclosed(tp);
1263         }
1264         return (tcp_output(tp));
1265 }
1266
1267 #endif /* INET6 */
1268
1269 /*
1270  * The new sockopt interface makes it possible for us to block in the
1271  * copyin/out step (if we take a page fault).  Taking a page fault while
1272  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1273  * both now use TSM, there probably isn't any need for this function to
1274  * run in a critical section any more.  This needs more examination.)
1275  */
1276 void
1277 tcp_ctloutput(netmsg_t msg)
1278 {
1279         struct socket *so = msg->base.nm_so;
1280         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1281         int     error, opt, optval;
1282         struct  inpcb *inp;
1283         struct  tcpcb *tp;
1284
1285         error = 0;
1286         inp = so->so_pcb;
1287         if (inp == NULL) {
1288                 error = ECONNRESET;
1289                 goto done;
1290         }
1291
1292         if (sopt->sopt_level != IPPROTO_TCP) {
1293 #ifdef INET6
1294                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1295                         ip6_ctloutput_dispatch(msg);
1296                 else
1297 #endif /* INET6 */
1298                 ip_ctloutput(msg);
1299                 /* msg invalid now */
1300                 return;
1301         }
1302         tp = intotcpcb(inp);
1303
1304         switch (sopt->sopt_dir) {
1305         case SOPT_SET:
1306                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1307                                       sizeof optval);
1308                 if (error)
1309                         break;
1310                 switch (sopt->sopt_name) {
1311                 case TCP_FASTKEEP:
1312                         if (optval > 0) {
1313                                 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1314                                         tp->t_flags |= TF_FASTKEEP;
1315                                         tcp_timer_keep_activity(tp, 0);
1316                                 }
1317                         } else {
1318                                 tp->t_flags &= ~TF_FASTKEEP;
1319                         }
1320                         break;
1321 #ifdef TCP_SIGNATURE
1322                 case TCP_SIGNATURE_ENABLE:
1323                         if (optval > 0)
1324                                 tp->t_flags |= TF_SIGNATURE;
1325                         else
1326                                 tp->t_flags &= ~TF_SIGNATURE;
1327                         break;
1328 #endif /* TCP_SIGNATURE */
1329                 case TCP_NODELAY:
1330                 case TCP_NOOPT:
1331                         switch (sopt->sopt_name) {
1332                         case TCP_NODELAY:
1333                                 opt = TF_NODELAY;
1334                                 break;
1335                         case TCP_NOOPT:
1336                                 opt = TF_NOOPT;
1337                                 break;
1338                         default:
1339                                 opt = 0; /* dead code to fool gcc */
1340                                 break;
1341                         }
1342
1343                         if (optval)
1344                                 tp->t_flags |= opt;
1345                         else
1346                                 tp->t_flags &= ~opt;
1347                         break;
1348
1349                 case TCP_NOPUSH:
1350                         if (optval)
1351                                 tp->t_flags |= TF_NOPUSH;
1352                         else {
1353                                 tp->t_flags &= ~TF_NOPUSH;
1354                                 error = tcp_output(tp);
1355                         }
1356                         break;
1357
1358                 case TCP_MAXSEG:
1359                         /*
1360                          * Must be between 0 and maxseg.  If the requested
1361                          * maxseg is too small to satisfy the desired minmss,
1362                          * pump it up (silently so sysctl modifications of
1363                          * minmss do not create unexpected program failures).
1364                          * Handle degenerate cases.
1365                          */
1366                         if (optval > 0 && optval <= tp->t_maxseg) {
1367                                 if (optval + 40 < tcp_minmss) {
1368                                         optval = tcp_minmss - 40;
1369                                         if (optval < 0)
1370                                                 optval = 1;
1371                                 }
1372                                 tp->t_maxseg = optval;
1373                         } else {
1374                                 error = EINVAL;
1375                         }
1376                         break;
1377
1378                 default:
1379                         error = ENOPROTOOPT;
1380                         break;
1381                 }
1382                 break;
1383
1384         case SOPT_GET:
1385                 switch (sopt->sopt_name) {
1386 #ifdef TCP_SIGNATURE
1387                 case TCP_SIGNATURE_ENABLE:
1388                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1389                         break;
1390 #endif /* TCP_SIGNATURE */
1391                 case TCP_NODELAY:
1392                         optval = tp->t_flags & TF_NODELAY;
1393                         break;
1394                 case TCP_MAXSEG:
1395                         optval = tp->t_maxseg;
1396                         break;
1397                 case TCP_NOOPT:
1398                         optval = tp->t_flags & TF_NOOPT;
1399                         break;
1400                 case TCP_NOPUSH:
1401                         optval = tp->t_flags & TF_NOPUSH;
1402                         break;
1403                 default:
1404                         error = ENOPROTOOPT;
1405                         break;
1406                 }
1407                 if (error == 0)
1408                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1409                 break;
1410         }
1411 done:
1412         lwkt_replymsg(&msg->lmsg, error);
1413 }
1414
1415 /*
1416  * tcp_sendspace and tcp_recvspace are the default send and receive window
1417  * sizes, respectively.  These are obsolescent (this information should
1418  * be set by the route).
1419  *
1420  * Use a default that does not require tcp window scaling to be turned
1421  * on.  Individual programs or the administrator can increase the default.
1422  */
1423 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1424 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1425     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1426 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1427 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1428     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1429
1430 /*
1431  * Attach TCP protocol to socket, allocating internet protocol control
1432  * block, tcp control block, bufer space, and entering LISTEN state
1433  * if to accept connections.
1434  */
1435 static int
1436 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1437 {
1438         struct tcpcb *tp;
1439         struct inpcb *inp;
1440         int error;
1441         int cpu;
1442 #ifdef INET6
1443         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1444 #endif
1445
1446         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1447                 lwkt_gettoken(&so->so_rcv.ssb_token);
1448                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1449                                   ai->sb_rlimit);
1450                 lwkt_reltoken(&so->so_rcv.ssb_token);
1451                 if (error)
1452                         return (error);
1453         }
1454         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1455         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1456         cpu = mycpu->gd_cpuid;
1457
1458         /*
1459          * Set the default port for protocol processing. This will likely
1460          * change when we connect.
1461          */
1462         error = in_pcballoc(so, &tcbinfo[cpu]);
1463         if (error)
1464                 return (error);
1465         inp = so->so_pcb;
1466 #ifdef INET6
1467         if (isipv6) {
1468                 inp->inp_vflag |= INP_IPV6;
1469                 inp->in6p_hops = -1;    /* use kernel default */
1470         }
1471         else
1472 #endif
1473         inp->inp_vflag |= INP_IPV4;
1474         tp = tcp_newtcpcb(inp);
1475         if (tp == NULL) {
1476                 /*
1477                  * Make sure the socket is destroyed by the pcbdetach.
1478                  */
1479                 soreference(so);
1480 #ifdef INET6
1481                 if (isipv6)
1482                         in6_pcbdetach(inp);
1483                 else
1484 #endif
1485                 in_pcbdetach(inp);
1486                 sofree(so);     /* from ref above */
1487                 return (ENOBUFS);
1488         }
1489         tp->t_state = TCPS_CLOSED;
1490         return (0);
1491 }
1492
1493 /*
1494  * Initiate (or continue) disconnect.
1495  * If embryonic state, just send reset (once).
1496  * If in ``let data drain'' option and linger null, just drop.
1497  * Otherwise (hard), mark socket disconnecting and drop
1498  * current input data; switch states based on user close, and
1499  * send segment to peer (with FIN).
1500  */
1501 static struct tcpcb *
1502 tcp_disconnect(struct tcpcb *tp)
1503 {
1504         struct socket *so = tp->t_inpcb->inp_socket;
1505
1506         if (tp->t_state < TCPS_ESTABLISHED) {
1507                 tp = tcp_close(tp);
1508         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1509                 tp = tcp_drop(tp, 0);
1510         } else {
1511                 lwkt_gettoken(&so->so_rcv.ssb_token);
1512                 soisdisconnecting(so);
1513                 sbflush(&so->so_rcv.sb);
1514                 tp = tcp_usrclosed(tp);
1515                 if (tp)
1516                         tcp_output(tp);
1517                 lwkt_reltoken(&so->so_rcv.ssb_token);
1518         }
1519         return (tp);
1520 }
1521
1522 /*
1523  * User issued close, and wish to trail through shutdown states:
1524  * if never received SYN, just forget it.  If got a SYN from peer,
1525  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1526  * If already got a FIN from peer, then almost done; go to LAST_ACK
1527  * state.  In all other cases, have already sent FIN to peer (e.g.
1528  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1529  * for peer to send FIN or not respond to keep-alives, etc.
1530  * We can let the user exit from the close as soon as the FIN is acked.
1531  */
1532 static struct tcpcb *
1533 tcp_usrclosed(struct tcpcb *tp)
1534 {
1535
1536         switch (tp->t_state) {
1537
1538         case TCPS_CLOSED:
1539         case TCPS_LISTEN:
1540                 tp->t_state = TCPS_CLOSED;
1541                 tp = tcp_close(tp);
1542                 break;
1543
1544         case TCPS_SYN_SENT:
1545         case TCPS_SYN_RECEIVED:
1546                 tp->t_flags |= TF_NEEDFIN;
1547                 break;
1548
1549         case TCPS_ESTABLISHED:
1550                 tp->t_state = TCPS_FIN_WAIT_1;
1551                 break;
1552
1553         case TCPS_CLOSE_WAIT:
1554                 tp->t_state = TCPS_LAST_ACK;
1555                 break;
1556         }
1557         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1558                 soisdisconnected(tp->t_inpcb->inp_socket);
1559                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1560                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1561                         tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,
1562                             tcp_timer_2msl);
1563                 }
1564         }
1565         return (tp);
1566 }