sys/netinet/tcp_usrreq.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * Copyright (c) 1982, 1986, 1988, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  *
  38  * Redistribution and use in source and binary forms, with or without
  39  * modification, are permitted provided that the following conditions
  40  * are met:
  41  * 1. Redistributions of source code must retain the above copyright
  42  *    notice, this list of conditions and the following disclaimer.
  43  * 2. Redistributions in binary form must reproduce the above copyright
  44  *    notice, this list of conditions and the following disclaimer in the
  45  *    documentation and/or other materials provided with the distribution.
  46  * 3. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
  63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
  64  */
  65
  66 #include "opt_ipsec.h"
  67 #include "opt_inet.h"
  68 #include "opt_inet6.h"
  69 #include "opt_tcpdebug.h"
  70
  71 #include <sys/param.h>
  72 #include <sys/systm.h>
  73 #include <sys/kernel.h>
  74 #include <sys/malloc.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/globaldata.h>
  77 #include <sys/thread.h>
  78
  79 #include <sys/mbuf.h>
  80 #ifdef INET6
  81 #include <sys/domain.h>
  82 #endif /* INET6 */
  83 #include <sys/socket.h>
  84 #include <sys/socketvar.h>
  85 #include <sys/socketops.h>
  86 #include <sys/protosw.h>
  87
  88 #include <sys/thread2.h>
  89 #include <sys/msgport2.h>
  90 #include <sys/socketvar2.h>
  91
  92 #include <net/if.h>
  93 #include <net/netisr.h>
  94 #include <net/route.h>
  95
  96 #include <net/netmsg2.h>
  97 #include <net/netisr2.h>
  98
  99 #include <netinet/in.h>
 100 #include <netinet/in_systm.h>
 101 #ifdef INET6
 102 #include <netinet/ip6.h>
 103 #endif
 104 #include <netinet/in_pcb.h>
 105 #ifdef INET6
 106 #include <netinet6/in6_pcb.h>
 107 #endif
 108 #include <netinet/in_var.h>
 109 #include <netinet/ip_var.h>
 110 #ifdef INET6
 111 #include <netinet6/ip6_var.h>
 112 #include <netinet6/tcp6_var.h>
 113 #endif
 114 #include <netinet/tcp.h>
 115 #include <netinet/tcp_fsm.h>
 116 #include <netinet/tcp_seq.h>
 117 #include <netinet/tcp_timer.h>
 118 #include <netinet/tcp_timer2.h>
 119 #include <netinet/tcp_var.h>
 120 #include <netinet/tcpip.h>
 121 #ifdef TCPDEBUG
 122 #include <netinet/tcp_debug.h>
 123 #endif
 124
 125 #ifdef IPSEC
 126 #include <netinet6/ipsec.h>
 127 #endif /*IPSEC*/
 128
 129 /*
 130  * TCP protocol interface to socket abstraction.
 131  */
 132 extern  char *tcpstates[];      /* XXX ??? */
 133
 134 static int      tcp_attach (struct socket *, struct pru_attach_info *);
 135 static void     tcp_connect (netmsg_t msg);
 136 #ifdef INET6
 137 static void     tcp6_connect (netmsg_t msg);
 138 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
 139                                 struct mbuf **mp,
 140                                 struct sockaddr_in6 *sin6,
 141                                 struct in6_addr *addr6);
 142 #endif /* INET6 */
 143 static struct tcpcb *
 144                 tcp_disconnect (struct tcpcb *);
 145 static struct tcpcb *
 146                 tcp_usrclosed (struct tcpcb *);
 147
 148 #ifdef TCPDEBUG
 149 #define TCPDEBUG0       int ostate = 0
 150 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
 151 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
 152                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
 153 #else
 154 #define TCPDEBUG0
 155 #define TCPDEBUG1()
 156 #define TCPDEBUG2(req)
 157 #endif
 158
 159 static int      tcp_lport_extension = 1;
 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
 161     &tcp_lport_extension, 0, "");
 162
 163 /*
 164  * For some ill optimized programs, which try to use TCP_NOPUSH
 165  * to improve performance, will have small amount of data sits
 166  * in the sending buffer.  These small amount of data will _not_
 167  * be pushed into the network until more data are written into
 168  * the socket or the socket write side is shutdown.
 169  */
 170 static int      tcp_disable_nopush = 1;
 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
 172     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
 173
 174 /*
 175  * TCP attaches to socket via pru_attach(), reserving space,
 176  * and an internet control block.  This is likely occuring on
 177  * cpu0 and may have to move later when we bind/connect.
 178  */
 179 static void
 180 tcp_usr_attach(netmsg_t msg)
 181 {
 182         struct socket *so = msg->base.nm_so;
 183         struct pru_attach_info *ai = msg->attach.nm_ai;
 184         int error;
 185         struct inpcb *inp;
 186         struct tcpcb *tp = NULL;
 187         TCPDEBUG0;
 188
 189         soreference(so);
 190         inp = so->so_pcb;
 191         TCPDEBUG1();
 192         if (inp) {
 193                 error = EISCONN;
 194                 goto out;
 195         }
 196
 197         error = tcp_attach(so, ai);
 198         if (error)
 199                 goto out;
 200
 201         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 202                 so->so_linger = TCP_LINGERTIME;
 203         tp = sototcpcb(so);
 204 out:
 205         sofree(so);             /* from ref above */
 206         TCPDEBUG2(PRU_ATTACH);
 207         lwkt_replymsg(&msg->lmsg, error);
 208 }
 209
 210 /*
 211  * pru_detach() detaches the TCP protocol from the socket.
 212  * If the protocol state is non-embryonic, then can't
 213  * do this directly: have to initiate a pru_disconnect(),
 214  * which may finish later; embryonic TCB's can just
 215  * be discarded here.
 216  */
 217 static void
 218 tcp_usr_detach(netmsg_t msg)
 219 {
 220         struct socket *so = msg->base.nm_so;
 221         int error = 0;
 222         struct inpcb *inp;
 223         struct tcpcb *tp;
 224         TCPDEBUG0;
 225
 226         inp = so->so_pcb;
 227
 228         /*
 229          * If the inp is already detached it may have been due to an async
 230          * close.  Just return as if no error occured.
 231          *
 232          * It's possible for the tcpcb (tp) to disconnect from the inp due
 233          * to tcp_drop()->tcp_close() being called.  This may occur *after*
 234          * the detach message has been queued so we may find a NULL tp here.
 235          */
 236         if (inp) {
 237                 if ((tp = intotcpcb(inp)) != NULL) {
 238                         TCPDEBUG1();
 239                         tp = tcp_disconnect(tp);
 240                         TCPDEBUG2(PRU_DETACH);
 241                 }
 242         }
 243         lwkt_replymsg(&msg->lmsg, error);
 244 }
 245
 246 /*
 247  * NOTE: ignore_error is non-zero for certain disconnection races
 248  * which we want to silently allow, otherwise close() may return
 249  * an unexpected error.
 250  *
 251  * NOTE: The variables (msg) and (tp) are assumed.
 252  */
 253 #define COMMON_START(so, inp, ignore_error)                     \
 254         TCPDEBUG0;                                              \
 255                                                                 \
 256         inp = so->so_pcb;                                       \
 257         do {                                                    \
 258                 if (inp == NULL) {                              \
 259                         error = ignore_error ? 0 : EINVAL;      \
 260                         tp = NULL;                              \
 261                         goto out;                               \
 262                 }                                               \
 263                 tp = intotcpcb(inp);                            \
 264                 TCPDEBUG1();                                    \
 265         } while(0)
 266
 267 #define COMMON_END1(req, noreply)                               \
 268         out: do {                                               \
 269                 TCPDEBUG2(req);                                 \
 270                 if (!(noreply))                                 \
 271                         lwkt_replymsg(&msg->lmsg, error);       \
 272                 return;                                         \
 273         } while(0)
 274
 275 #define COMMON_END(req)         COMMON_END1((req), 0)
 276
 277 /*
 278  * Give the socket an address.
 279  */
 280 static void
 281 tcp_usr_bind(netmsg_t msg)
 282 {
 283         struct socket *so = msg->bind.base.nm_so;
 284         struct sockaddr *nam = msg->bind.nm_nam;
 285         struct thread *td = msg->bind.nm_td;
 286         int error = 0;
 287         struct inpcb *inp;
 288         struct tcpcb *tp;
 289         struct sockaddr_in *sinp;
 290
 291         COMMON_START(so, inp, 0);
 292
 293         /*
 294          * Must check for multicast addresses and disallow binding
 295          * to them.
 296          */
 297         sinp = (struct sockaddr_in *)nam;
 298         if (sinp->sin_family == AF_INET &&
 299             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 300                 error = EAFNOSUPPORT;
 301                 goto out;
 302         }
 303         error = in_pcbbind(inp, nam, td);
 304         if (error)
 305                 goto out;
 306         COMMON_END(PRU_BIND);
 307
 308 }
 309
 310 #ifdef INET6
 311
 312 static void
 313 tcp6_usr_bind(netmsg_t msg)
 314 {
 315         struct socket *so = msg->bind.base.nm_so;
 316         struct sockaddr *nam = msg->bind.nm_nam;
 317         struct thread *td = msg->bind.nm_td;
 318         int error = 0;
 319         struct inpcb *inp;
 320         struct tcpcb *tp;
 321         struct sockaddr_in6 *sin6p;
 322
 323         COMMON_START(so, inp, 0);
 324
 325         /*
 326          * Must check for multicast addresses and disallow binding
 327          * to them.
 328          */
 329         sin6p = (struct sockaddr_in6 *)nam;
 330         if (sin6p->sin6_family == AF_INET6 &&
 331             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 332                 error = EAFNOSUPPORT;
 333                 goto out;
 334         }
 335         inp->inp_vflag &= ~INP_IPV4;
 336         inp->inp_vflag |= INP_IPV6;
 337         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 338                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
 339                         inp->inp_vflag |= INP_IPV4;
 340                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 341                         struct sockaddr_in sin;
 342
 343                         in6_sin6_2_sin(&sin, sin6p);
 344                         inp->inp_vflag |= INP_IPV4;
 345                         inp->inp_vflag &= ~INP_IPV6;
 346                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
 347                         goto out;
 348                 }
 349         }
 350         error = in6_pcbbind(inp, nam, td);
 351         if (error)
 352                 goto out;
 353         COMMON_END(PRU_BIND);
 354 }
 355 #endif /* INET6 */
 356
 357 struct netmsg_inswildcard {
 358         struct netmsg_base      base;
 359         struct inpcb            *nm_inp;
 360 };
 361
 362 static void
 363 in_pcbinswildcardhash_handler(netmsg_t msg)
 364 {
 365         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 366         int cpu = mycpuid, nextcpu;
 367
 368         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 369
 370         nextcpu = cpu + 1;
 371         if (nextcpu < ncpus2)
 372                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 373         else
 374                 lwkt_replymsg(&nm->base.lmsg, 0);
 375 }
 376
 377 static void
 378 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 379 {
 380         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 381 }
 382
 383 /*
 384  * Prepare to accept connections.
 385  */
 386 static void
 387 tcp_usr_listen(netmsg_t msg)
 388 {
 389         struct socket *so = msg->listen.base.nm_so;
 390         struct thread *td = msg->listen.nm_td;
 391         int error = 0;
 392         struct inpcb *inp;
 393         struct tcpcb *tp;
 394         struct netmsg_inswildcard nm;
 395         lwkt_port_t port0 = netisr_cpuport(0);
 396
 397         COMMON_START(so, inp, 0);
 398
 399         if (&curthread->td_msgport != port0) {
 400                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 401
 402                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 403                     ("already asked to relink"));
 404
 405                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 406                 msg->listen.nm_flags |= PRUL_RELINK;
 407
 408                 /* See the related comment in tcp_connect() */
 409                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 410                 lwkt_forwardmsg(port0, lmsg);
 411                 /* msg invalid now */
 412                 return;
 413         }
 414         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 415
 416         if (msg->listen.nm_flags & PRUL_RELINK) {
 417                 msg->listen.nm_flags &= ~PRUL_RELINK;
 418                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 419         }
 420         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 421
 422         if (tp->t_flags & TF_LISTEN)
 423                 goto out;
 424
 425         if (inp->inp_lport == 0) {
 426                 error = in_pcbbind(inp, NULL, td);
 427                 if (error)
 428                         goto out;
 429         }
 430
 431         tp->t_state = TCPS_LISTEN;
 432         tp->t_flags |= TF_LISTEN;
 433         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 434
 435         if (ncpus2 > 1) {
 436                 /*
 437                  * Put this inpcb into wildcard hash on other cpus.
 438                  */
 439                 ASSERT_INP_NOTINHASH(inp);
 440                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 441                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 442                 nm.nm_inp = inp;
 443                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 444         }
 445         in_pcbinswildcardhash(inp);
 446         COMMON_END(PRU_LISTEN);
 447 }
 448
 449 #ifdef INET6
 450
 451 static void
 452 tcp6_usr_listen(netmsg_t msg)
 453 {
 454         struct socket *so = msg->listen.base.nm_so;
 455         struct thread *td = msg->listen.nm_td;
 456         int error = 0;
 457         struct inpcb *inp;
 458         struct tcpcb *tp;
 459         struct netmsg_inswildcard nm;
 460
 461         COMMON_START(so, inp, 0);
 462
 463         if (tp->t_flags & TF_LISTEN)
 464                 goto out;
 465
 466         if (inp->inp_lport == 0) {
 467                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
 468                         inp->inp_vflag |= INP_IPV4;
 469                 else
 470                         inp->inp_vflag &= ~INP_IPV4;
 471                 error = in6_pcbbind(inp, NULL, td);
 472                 if (error)
 473                         goto out;
 474         }
 475
 476         tp->t_state = TCPS_LISTEN;
 477         tp->t_flags |= TF_LISTEN;
 478         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 479
 480         if (ncpus2 > 1) {
 481                 /*
 482                  * Put this inpcb into wildcard hash on other cpus.
 483                  */
 484                 KKASSERT(so->so_port == netisr_cpuport(0));
 485                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 486                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 487                 ASSERT_INP_NOTINHASH(inp);
 488
 489                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 490                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 491                 nm.nm_inp = inp;
 492                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 493         }
 494         in_pcbinswildcardhash(inp);
 495         COMMON_END(PRU_LISTEN);
 496 }
 497 #endif /* INET6 */
 498
 499 /*
 500  * Initiate connection to peer.
 501  * Create a template for use in transmissions on this connection.
 502  * Enter SYN_SENT state, and mark socket as connecting.
 503  * Start keep-alive timer, and seed output sequence space.
 504  * Send initial segment on connection.
 505  */
 506 static void
 507 tcp_usr_connect(netmsg_t msg)
 508 {
 509         struct socket *so = msg->connect.base.nm_so;
 510         struct sockaddr *nam = msg->connect.nm_nam;
 511         struct thread *td = msg->connect.nm_td;
 512         int error = 0;
 513         struct inpcb *inp;
 514         struct tcpcb *tp;
 515         struct sockaddr_in *sinp;
 516
 517         COMMON_START(so, inp, 0);
 518
 519         /*
 520          * Must disallow TCP ``connections'' to multicast addresses.
 521          */
 522         sinp = (struct sockaddr_in *)nam;
 523         if (sinp->sin_family == AF_INET
 524             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 525                 error = EAFNOSUPPORT;
 526                 goto out;
 527         }
 528
 529         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 530                 error = EAFNOSUPPORT; /* IPv6 only jail */
 531                 goto out;
 532         }
 533
 534         tcp_connect(msg);
 535         /* msg is invalid now */
 536         return;
 537 out:
 538         if (msg->connect.nm_m) {
 539                 m_freem(msg->connect.nm_m);
 540                 msg->connect.nm_m = NULL;
 541         }
 542         if (msg->connect.nm_flags & PRUC_HELDTD)
 543                 lwkt_rele(td);
 544         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 545                 so->so_error = error;
 546                 soisdisconnected(so);
 547         }
 548         lwkt_replymsg(&msg->lmsg, error);
 549 }
 550
 551 #ifdef INET6
 552
 553 static void
 554 tcp6_usr_connect(netmsg_t msg)
 555 {
 556         struct socket *so = msg->connect.base.nm_so;
 557         struct sockaddr *nam = msg->connect.nm_nam;
 558         struct thread *td = msg->connect.nm_td;
 559         int error = 0;
 560         struct inpcb *inp;
 561         struct tcpcb *tp;
 562         struct sockaddr_in6 *sin6p;
 563
 564         COMMON_START(so, inp, 0);
 565
 566         /*
 567          * Must disallow TCP ``connections'' to multicast addresses.
 568          */
 569         sin6p = (struct sockaddr_in6 *)nam;
 570         if (sin6p->sin6_family == AF_INET6
 571             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 572                 error = EAFNOSUPPORT;
 573                 goto out;
 574         }
 575
 576         if (!prison_remote_ip(td, nam)) {
 577                 error = EAFNOSUPPORT; /* IPv4 only jail */
 578                 goto out;
 579         }
 580
 581         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 582                 struct sockaddr_in *sinp;
 583
 584                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
 585                         error = EINVAL;
 586                         goto out;
 587                 }
 588                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
 589                 in6_sin6_2_sin(sinp, sin6p);
 590                 inp->inp_vflag |= INP_IPV4;
 591                 inp->inp_vflag &= ~INP_IPV6;
 592                 msg->connect.nm_nam = (struct sockaddr *)sinp;
 593                 msg->connect.nm_flags |= PRUC_NAMALLOC;
 594                 tcp_connect(msg);
 595                 /* msg is invalid now */
 596                 return;
 597         }
 598         inp->inp_vflag &= ~INP_IPV4;
 599         inp->inp_vflag |= INP_IPV6;
 600         inp->inp_inc.inc_isipv6 = 1;
 601
 602         msg->connect.nm_flags |= PRUC_FALLBACK;
 603         tcp6_connect(msg);
 604         /* msg is invalid now */
 605         return;
 606 out:
 607         if (msg->connect.nm_m) {
 608                 m_freem(msg->connect.nm_m);
 609                 msg->connect.nm_m = NULL;
 610         }
 611         lwkt_replymsg(&msg->lmsg, error);
 612 }
 613
 614 #endif /* INET6 */
 615
 616 /*
 617  * Initiate disconnect from peer.
 618  * If connection never passed embryonic stage, just drop;
 619  * else if don't need to let data drain, then can just drop anyways,
 620  * else have to begin TCP shutdown process: mark socket disconnecting,
 621  * drain unread data, state switch to reflect user close, and
 622  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 623  * when peer sends FIN and acks ours.
 624  *
 625  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 626  */
 627 static void
 628 tcp_usr_disconnect(netmsg_t msg)
 629 {
 630         struct socket *so = msg->disconnect.base.nm_so;
 631         int error = 0;
 632         struct inpcb *inp;
 633         struct tcpcb *tp;
 634
 635         COMMON_START(so, inp, 1);
 636         tp = tcp_disconnect(tp);
 637         COMMON_END(PRU_DISCONNECT);
 638 }
 639
 640 /*
 641  * Accept a connection.  Essentially all the work is
 642  * done at higher levels; just return the address
 643  * of the peer, storing through addr.
 644  */
 645 static void
 646 tcp_usr_accept(netmsg_t msg)
 647 {
 648         struct socket *so = msg->accept.base.nm_so;
 649         struct sockaddr **nam = msg->accept.nm_nam;
 650         int error = 0;
 651         struct inpcb *inp;
 652         struct tcpcb *tp = NULL;
 653         TCPDEBUG0;
 654
 655         inp = so->so_pcb;
 656         if (so->so_state & SS_ISDISCONNECTED) {
 657                 error = ECONNABORTED;
 658                 goto out;
 659         }
 660         if (inp == 0) {
 661                 error = EINVAL;
 662                 goto out;
 663         }
 664
 665         tp = intotcpcb(inp);
 666         TCPDEBUG1();
 667         in_setpeeraddr(so, nam);
 668         COMMON_END(PRU_ACCEPT);
 669 }
 670
 671 #ifdef INET6
 672 static void
 673 tcp6_usr_accept(netmsg_t msg)
 674 {
 675         struct socket *so = msg->accept.base.nm_so;
 676         struct sockaddr **nam = msg->accept.nm_nam;
 677         int error = 0;
 678         struct inpcb *inp;
 679         struct tcpcb *tp = NULL;
 680         TCPDEBUG0;
 681
 682         inp = so->so_pcb;
 683
 684         if (so->so_state & SS_ISDISCONNECTED) {
 685                 error = ECONNABORTED;
 686                 goto out;
 687         }
 688         if (inp == 0) {
 689                 error = EINVAL;
 690                 goto out;
 691         }
 692         tp = intotcpcb(inp);
 693         TCPDEBUG1();
 694         in6_mapped_peeraddr(so, nam);
 695         COMMON_END(PRU_ACCEPT);
 696 }
 697 #endif /* INET6 */
 698 /*
 699  * Mark the connection as being incapable of further output.
 700  */
 701 static void
 702 tcp_usr_shutdown(netmsg_t msg)
 703 {
 704         struct socket *so = msg->shutdown.base.nm_so;
 705         int error = 0;
 706         struct inpcb *inp;
 707         struct tcpcb *tp;
 708
 709         COMMON_START(so, inp, 0);
 710         socantsendmore(so);
 711         tp = tcp_usrclosed(tp);
 712         if (tp)
 713                 error = tcp_output(tp);
 714         COMMON_END(PRU_SHUTDOWN);
 715 }
 716
 717 /*
 718  * After a receive, possibly send window update to peer.
 719  */
 720 static void
 721 tcp_usr_rcvd(netmsg_t msg)
 722 {
 723         struct socket *so = msg->rcvd.base.nm_so;
 724         int error = 0, noreply = 0;
 725         struct inpcb *inp;
 726         struct tcpcb *tp;
 727
 728         COMMON_START(so, inp, 0);
 729
 730         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 731                 noreply = 1;
 732                 so_async_rcvd_reply(so);
 733         }
 734         tcp_output(tp);
 735
 736         COMMON_END1(PRU_RCVD, noreply);
 737 }
 738
 739 /*
 740  * Do a send by putting data in output queue and updating urgent
 741  * marker if URG set.  Possibly send more data.  Unlike the other
 742  * pru_*() routines, the mbuf chains are our responsibility.  We
 743  * must either enqueue them or free them.  The other pru_* routines
 744  * generally are caller-frees.
 745  */
 746 static void
 747 tcp_usr_send(netmsg_t msg)
 748 {
 749         struct socket *so = msg->send.base.nm_so;
 750         int flags = msg->send.nm_flags;
 751         struct mbuf *m = msg->send.nm_m;
 752         int error = 0;
 753         struct inpcb *inp;
 754         struct tcpcb *tp;
 755         TCPDEBUG0;
 756
 757         KKASSERT(msg->send.nm_control == NULL);
 758         KKASSERT(msg->send.nm_addr == NULL);
 759         KKASSERT((flags & PRUS_FREEADDR) == 0);
 760
 761         inp = so->so_pcb;
 762
 763         if (inp == NULL) {
 764                 /*
 765                  * OOPS! we lost a race, the TCP session got reset after
 766                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 767                  * network interrupt in the non-critical section of sosend().
 768                  */
 769                 m_freem(m);
 770                 error = ECONNRESET;     /* XXX EPIPE? */
 771                 tp = NULL;
 772                 TCPDEBUG1();
 773                 goto out;
 774         }
 775         tp = intotcpcb(inp);
 776         TCPDEBUG1();
 777
 778 #ifdef foo
 779         /*
 780          * This is no longer necessary, since:
 781          * - sosendtcp() has already checked it for us
 782          * - It does not work with asynchronized send
 783          */
 784
 785         /*
 786          * Don't let too much OOB data build up
 787          */
 788         if (flags & PRUS_OOB) {
 789                 if (ssb_space(&so->so_snd) < -512) {
 790                         m_freem(m);
 791                         error = ENOBUFS;
 792                         goto out;
 793                 }
 794         }
 795 #endif
 796
 797         /*
 798          * Pump the data into the socket.
 799          */
 800         if (m) {
 801                 ssb_appendstream(&so->so_snd, m);
 802                 sowwakeup(so);
 803         }
 804         if (flags & PRUS_OOB) {
 805                 /*
 806                  * According to RFC961 (Assigned Protocols),
 807                  * the urgent pointer points to the last octet
 808                  * of urgent data.  We continue, however,
 809                  * to consider it to indicate the first octet
 810                  * of data past the urgent section.
 811                  * Otherwise, snd_up should be one lower.
 812                  */
 813                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 814                 tp->t_flags |= TF_FORCE;
 815                 error = tcp_output(tp);
 816                 tp->t_flags &= ~TF_FORCE;
 817         } else {
 818                 if (flags & PRUS_EOF) {
 819                         /*
 820                          * Close the send side of the connection after
 821                          * the data is sent.
 822                          */
 823                         socantsendmore(so);
 824                         tp = tcp_usrclosed(tp);
 825                 }
 826                 if (tp != NULL && !tcp_output_pending(tp)) {
 827                         if (flags & PRUS_MORETOCOME)
 828                                 tp->t_flags |= TF_MORETOCOME;
 829                         error = tcp_output_fair(tp);
 830                         if (flags & PRUS_MORETOCOME)
 831                                 tp->t_flags &= ~TF_MORETOCOME;
 832                 }
 833         }
 834         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 835                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 836                    (flags & PRUS_NOREPLY));
 837 }
 838
 839 /*
 840  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 841  *       will sofree() it when we return.
 842  */
 843 static void
 844 tcp_usr_abort(netmsg_t msg)
 845 {
 846         struct socket *so = msg->abort.base.nm_so;
 847         int error = 0;
 848         struct inpcb *inp;
 849         struct tcpcb *tp;
 850
 851         COMMON_START(so, inp, 1);
 852         tp = tcp_drop(tp, ECONNABORTED);
 853         COMMON_END(PRU_ABORT);
 854 }
 855
 856 /*
 857  * Receive out-of-band data.
 858  */
 859 static void
 860 tcp_usr_rcvoob(netmsg_t msg)
 861 {
 862         struct socket *so = msg->rcvoob.base.nm_so;
 863         struct mbuf *m = msg->rcvoob.nm_m;
 864         int flags = msg->rcvoob.nm_flags;
 865         int error = 0;
 866         struct inpcb *inp;
 867         struct tcpcb *tp;
 868
 869         COMMON_START(so, inp, 0);
 870         if ((so->so_oobmark == 0 &&
 871              (so->so_state & SS_RCVATMARK) == 0) ||
 872             so->so_options & SO_OOBINLINE ||
 873             tp->t_oobflags & TCPOOB_HADDATA) {
 874                 error = EINVAL;
 875                 goto out;
 876         }
 877         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 878                 error = EWOULDBLOCK;
 879                 goto out;
 880         }
 881         m->m_len = 1;
 882         *mtod(m, caddr_t) = tp->t_iobc;
 883         if ((flags & MSG_PEEK) == 0)
 884                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 885         COMMON_END(PRU_RCVOOB);
 886 }
 887
 888 static void
 889 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 890 {
 891         in_savefaddr(so, faddr);
 892 }
 893
 894 #ifdef INET6
 895 static void
 896 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 897 {
 898         in6_mapped_savefaddr(so, faddr);
 899 }
 900 #endif
 901
 902 static int
 903 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 904     struct thread *td __unused)
 905 {
 906         const struct sockaddr_in *sinp;
 907
 908         sinp = (const struct sockaddr_in *)nam;
 909         if (sinp->sin_family == AF_INET &&
 910             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 911                 return EAFNOSUPPORT;
 912
 913         soisconnecting(so);
 914         return 0;
 915 }
 916
 917 /* xxx - should be const */
 918 struct pr_usrreqs tcp_usrreqs = {
 919         .pru_abort = tcp_usr_abort,
 920         .pru_accept = tcp_usr_accept,
 921         .pru_attach = tcp_usr_attach,
 922         .pru_bind = tcp_usr_bind,
 923         .pru_connect = tcp_usr_connect,
 924         .pru_connect2 = pr_generic_notsupp,
 925         .pru_control = in_control_dispatch,
 926         .pru_detach = tcp_usr_detach,
 927         .pru_disconnect = tcp_usr_disconnect,
 928         .pru_listen = tcp_usr_listen,
 929         .pru_peeraddr = in_setpeeraddr_dispatch,
 930         .pru_rcvd = tcp_usr_rcvd,
 931         .pru_rcvoob = tcp_usr_rcvoob,
 932         .pru_send = tcp_usr_send,
 933         .pru_sense = pru_sense_null,
 934         .pru_shutdown = tcp_usr_shutdown,
 935         .pru_sockaddr = in_setsockaddr_dispatch,
 936         .pru_sosend = sosendtcp,
 937         .pru_soreceive = sorecvtcp,
 938         .pru_savefaddr = tcp_usr_savefaddr,
 939         .pru_preconnect = tcp_usr_preconnect
 940 };
 941
 942 #ifdef INET6
 943 struct pr_usrreqs tcp6_usrreqs = {
 944         .pru_abort = tcp_usr_abort,
 945         .pru_accept = tcp6_usr_accept,
 946         .pru_attach = tcp_usr_attach,
 947         .pru_bind = tcp6_usr_bind,
 948         .pru_connect = tcp6_usr_connect,
 949         .pru_connect2 = pr_generic_notsupp,
 950         .pru_control = in6_control_dispatch,
 951         .pru_detach = tcp_usr_detach,
 952         .pru_disconnect = tcp_usr_disconnect,
 953         .pru_listen = tcp6_usr_listen,
 954         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
 955         .pru_rcvd = tcp_usr_rcvd,
 956         .pru_rcvoob = tcp_usr_rcvoob,
 957         .pru_send = tcp_usr_send,
 958         .pru_sense = pru_sense_null,
 959         .pru_shutdown = tcp_usr_shutdown,
 960         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
 961         .pru_sosend = sosendtcp,
 962         .pru_soreceive = sorecvtcp,
 963         .pru_savefaddr = tcp6_usr_savefaddr
 964 };
 965 #endif /* INET6 */
 966
 967 static int
 968 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 969                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 970 {
 971         struct inpcb *inp = tp->t_inpcb, *oinp;
 972         struct socket *so = inp->inp_socket;
 973         struct route *ro = &inp->inp_route;
 974
 975         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
 976             ("pcbinfo mismatch"));
 977
 978         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
 979                                  sin->sin_addr, sin->sin_port,
 980                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 981                                   inp->inp_laddr : if_sin->sin_addr),
 982                                 inp->inp_lport, 0, NULL);
 983         if (oinp != NULL) {
 984                 m_freem(m);
 985                 return (EADDRINUSE);
 986         }
 987         if (inp->inp_laddr.s_addr == INADDR_ANY)
 988                 inp->inp_laddr = if_sin->sin_addr;
 989         inp->inp_faddr = sin->sin_addr;
 990         inp->inp_fport = sin->sin_port;
 991         in_pcbinsconnhash(inp);
 992
 993         /*
 994          * We are now on the inpcb's owner CPU, if the cached route was
 995          * freed because the rtentry's owner CPU is not the current CPU
 996          * (e.g. in tcp_connect()), then we try to reallocate it here with
 997          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 998          * rtentry.
 999          */
1000         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1001             ro->ro_rt == NULL) {
1002                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1003                 ro->ro_dst.sa_family = AF_INET;
1004                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1005                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1006                         sin->sin_addr;
1007                 rtalloc(ro);
1008         }
1009
1010         /*
1011          * Now that no more errors can occur, change the protocol processing
1012          * port to the current thread (which is the correct thread).
1013          *
1014          * Create TCP timer message now; we are on the tcpcb's owner
1015          * CPU/thread.
1016          */
1017         tcp_create_timermsg(tp, &curthread->td_msgport);
1018
1019         /*
1020          * Compute window scaling to request.  Use a larger scaling then
1021          * needed for the initial receive buffer in case the receive buffer
1022          * gets expanded.
1023          */
1024         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1025                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1026         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1027                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1028         ) {
1029                 tp->request_r_scale++;
1030         }
1031
1032         soisconnecting(so);
1033         tcpstat.tcps_connattempt++;
1034         tp->t_state = TCPS_SYN_SENT;
1035         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1036         tp->iss = tcp_new_isn(tp);
1037         tcp_sendseqinit(tp);
1038         if (m) {
1039                 ssb_appendstream(&so->so_snd, m);
1040                 m = NULL;
1041                 if (flags & PRUS_OOB)
1042                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1043         }
1044
1045         /*
1046          * Close the send side of the connection after
1047          * the data is sent if flagged.
1048          */
1049         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1050                 socantsendmore(so);
1051                 tp = tcp_usrclosed(tp);
1052         }
1053         return (tcp_output(tp));
1054 }
1055
1056 /*
1057  * Common subroutine to open a TCP connection to remote host specified
1058  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1059  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1060  * a local host address (interface).
1061  * Initialize connection parameters and enter SYN-SENT state.
1062  */
1063 static void
1064 tcp_connect(netmsg_t msg)
1065 {
1066         struct socket *so = msg->connect.base.nm_so;
1067         struct sockaddr *nam = msg->connect.nm_nam;
1068         struct thread *td = msg->connect.nm_td;
1069         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1070         struct sockaddr_in *if_sin = NULL;
1071         struct inpcb *inp;
1072         struct tcpcb *tp;
1073         int error;
1074         lwkt_port_t port;
1075
1076         COMMON_START(so, inp, 0);
1077
1078         /*
1079          * Reconnect our pcb if we have to
1080          */
1081         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1082                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1083                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1084         }
1085
1086         /*
1087          * Bind if we have to
1088          */
1089         if (inp->inp_lport == 0) {
1090                 if (tcp_lport_extension) {
1091                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1092
1093                         error = in_pcbladdr(inp, nam, &if_sin, td);
1094                         if (error)
1095                                 goto out;
1096                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1097
1098                         error = in_pcbbind_remote(inp, nam, td);
1099                         if (error)
1100                                 goto out;
1101
1102                         msg->connect.nm_flags |= PRUC_HASLADDR;
1103                 } else {
1104                         error = in_pcbbind(inp, NULL, td);
1105                         if (error)
1106                                 goto out;
1107                 }
1108         }
1109
1110         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1111                 /*
1112                  * Calculate the correct protocol processing thread.  The
1113                  * connect operation must run there.  Set the forwarding
1114                  * port before we forward the message or it will get bounced
1115                  * right back to us.
1116                  */
1117                 error = in_pcbladdr(inp, nam, &if_sin, td);
1118                 if (error)
1119                         goto out;
1120         }
1121         KKASSERT(inp->inp_socket == so);
1122
1123         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1124                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1125                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1126                             inp->inp_lport);
1127
1128         if (port != &curthread->td_msgport) {
1129                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1130
1131                 /*
1132                  * in_pcbladdr() may have allocated a route entry for us
1133                  * on the current CPU, but we need a route entry on the
1134                  * inpcb's owner CPU, so free it here.
1135                  */
1136                 in_pcbresetroute(inp);
1137
1138                 /*
1139                  * We are moving the protocol processing port the socket
1140                  * is on, we have to unlink here and re-link on the
1141                  * target cpu.
1142                  */
1143                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1144                 msg->connect.nm_flags |= PRUC_RECONNECT;
1145                 msg->connect.base.nm_dispatch = tcp_connect;
1146
1147                 /*
1148                  * Use message put done receipt to change this socket's
1149                  * so_port, i.e. _after_ this message was put onto the
1150                  * target netisr's msgport but _before_ the message could
1151                  * be pulled from the target netisr's msgport, so that:
1152                  * - The upper half (socket code) will not see the new
1153                  *   msgport before this message reaches the new msgport
1154                  *   and messages for this socket will be ordered.
1155                  * - This message will see the new msgport, when its
1156                  *   handler is called in the target netisr.
1157                  *
1158                  * NOTE:
1159                  * We MUST use messege put done receipt to change this
1160                  * socket's so_port:
1161                  * If we changed the so_port in this netisr after the
1162                  * lwkt_forwardmsg (so messages for this socket will be
1163                  * ordered) and changed the so_port in the target netisr
1164                  * at the very beginning of this message's handler, we
1165                  * would suffer so_port overwritten race, given this
1166                  * message might be forwarded again.
1167                  *
1168                  * NOTE:
1169                  * This mechanism depends on that the netisr's msgport
1170                  * is spin msgport (currently it is :).
1171                  *
1172                  * If the upper half saw the new msgport before this
1173                  * message reached the target netisr's msgport, the
1174                  * messages sent from the upper half could reach the new
1175                  * msgport before this message, thus there would be
1176                  * message reordering.  The worst case could be soclose()
1177                  * saw the new msgport and the detach message could reach
1178                  * the new msgport before this message, i.e. the inpcb
1179                  * could have been destroyed when this message was still
1180                  * pending on or on its way to the new msgport.  Other
1181                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1182                  * since we have unlinked this inpcb from the current
1183                  * pcbinfo first.
1184                  */
1185                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1186                 lwkt_forwardmsg(port, lmsg);
1187                 /* msg invalid now */
1188                 return;
1189         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1190                 /*
1191                  * The original thread is no longer needed; release it.
1192                  */
1193                 lwkt_rele(td);
1194                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1195         }
1196         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1197                                   msg->connect.nm_m, sin, if_sin);
1198         msg->connect.nm_m = NULL;
1199 out:
1200         if (msg->connect.nm_m) {
1201                 m_freem(msg->connect.nm_m);
1202                 msg->connect.nm_m = NULL;
1203         }
1204         if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1205                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1206                 msg->connect.nm_nam = NULL;
1207         }
1208         if (msg->connect.nm_flags & PRUC_HELDTD)
1209                 lwkt_rele(td);
1210         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1211                 so->so_error = error;
1212                 soisdisconnected(so);
1213         }
1214         lwkt_replymsg(&msg->connect.base.lmsg, error);
1215         /* msg invalid now */
1216 }
1217
1218 #ifdef INET6
1219
1220 static void
1221 tcp6_connect(netmsg_t msg)
1222 {
1223         struct tcpcb *tp;
1224         struct socket *so = msg->connect.base.nm_so;
1225         struct sockaddr *nam = msg->connect.nm_nam;
1226         struct thread *td = msg->connect.nm_td;
1227         struct inpcb *inp;
1228         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1229         struct in6_addr *addr6;
1230         lwkt_port_t port;
1231         int error;
1232
1233         COMMON_START(so, inp, 0);
1234
1235         /*
1236          * Reconnect our pcb if we have to
1237          */
1238         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1239                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1240                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1241         }
1242
1243         /*
1244          * Bind if we have to
1245          */
1246         if (inp->inp_lport == 0) {
1247                 error = in6_pcbbind(inp, NULL, td);
1248                 if (error)
1249                         goto out;
1250         }
1251
1252         /*
1253          * Cannot simply call in_pcbconnect, because there might be an
1254          * earlier incarnation of this same connection still in
1255          * TIME_WAIT state, creating an ADDRINUSE error.
1256          */
1257         error = in6_pcbladdr(inp, nam, &addr6, td);
1258         if (error)
1259                 goto out;
1260
1261         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1262
1263         if (port != &curthread->td_msgport) {
1264                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1265
1266                 /*
1267                  * in_pcbladdr() may have allocated a route entry for us
1268                  * on the current CPU, but we need a route entry on the
1269                  * inpcb's owner CPU, so free it here.
1270                  */
1271                 in_pcbresetroute(inp);
1272
1273                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1274                 msg->connect.nm_flags |= PRUC_RECONNECT;
1275                 msg->connect.base.nm_dispatch = tcp6_connect;
1276
1277                 /* See the related comment in tcp_connect() */
1278                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1279                 lwkt_forwardmsg(port, lmsg);
1280                 /* msg invalid now */
1281                 return;
1282         }
1283         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1284                                    &msg->connect.nm_m, sin6, addr6);
1285         /* nm_m may still be intact */
1286 out:
1287         if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1288                 tcp_connect(msg);
1289                 /* msg invalid now */
1290         } else {
1291                 if (msg->connect.nm_m) {
1292                         m_freem(msg->connect.nm_m);
1293                         msg->connect.nm_m = NULL;
1294                 }
1295                 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1296                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1297                         msg->connect.nm_nam = NULL;
1298                 }
1299                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1300                 /* msg invalid now */
1301         }
1302 }
1303
1304 static int
1305 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1306                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1307 {
1308         struct mbuf *m = *mp;
1309         struct inpcb *inp = tp->t_inpcb;
1310         struct socket *so = inp->inp_socket;
1311         struct inpcb *oinp;
1312
1313         /*
1314          * Cannot simply call in_pcbconnect, because there might be an
1315          * earlier incarnation of this same connection still in
1316          * TIME_WAIT state, creating an ADDRINUSE error.
1317          */
1318         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1319                                   &sin6->sin6_addr, sin6->sin6_port,
1320                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1321                                       addr6 : &inp->in6p_laddr),
1322                                   inp->inp_lport,  0, NULL);
1323         if (oinp)
1324                 return (EADDRINUSE);
1325
1326         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1327                 inp->in6p_laddr = *addr6;
1328         inp->in6p_faddr = sin6->sin6_addr;
1329         inp->inp_fport = sin6->sin6_port;
1330         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1331                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1332         in_pcbinsconnhash(inp);
1333
1334         /*
1335          * Now that no more errors can occur, change the protocol processing
1336          * port to the current thread (which is the correct thread).
1337          *
1338          * Create TCP timer message now; we are on the tcpcb's owner
1339          * CPU/thread.
1340          */
1341         tcp_create_timermsg(tp, &curthread->td_msgport);
1342
1343         /* Compute window scaling to request.  */
1344         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1345                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1346         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1347             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1348                 tp->request_r_scale++;
1349         }
1350
1351         soisconnecting(so);
1352         tcpstat.tcps_connattempt++;
1353         tp->t_state = TCPS_SYN_SENT;
1354         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1355         tp->iss = tcp_new_isn(tp);
1356         tcp_sendseqinit(tp);
1357         if (m) {
1358                 ssb_appendstream(&so->so_snd, m);
1359                 *mp = NULL;
1360                 if (flags & PRUS_OOB)
1361                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1362         }
1363
1364         /*
1365          * Close the send side of the connection after
1366          * the data is sent if flagged.
1367          */
1368         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1369                 socantsendmore(so);
1370                 tp = tcp_usrclosed(tp);
1371         }
1372         return (tcp_output(tp));
1373 }
1374
1375 #endif /* INET6 */
1376
1377 /*
1378  * The new sockopt interface makes it possible for us to block in the
1379  * copyin/out step (if we take a page fault).  Taking a page fault while
1380  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1381  * both now use TSM, there probably isn't any need for this function to
1382  * run in a critical section any more.  This needs more examination.)
1383  */
1384 void
1385 tcp_ctloutput(netmsg_t msg)
1386 {
1387         struct socket *so = msg->base.nm_so;
1388         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1389         int     error, opt, optval, opthz;
1390         struct  inpcb *inp;
1391         struct  tcpcb *tp;
1392
1393         error = 0;
1394         inp = so->so_pcb;
1395         if (inp == NULL) {
1396                 error = ECONNRESET;
1397                 goto done;
1398         }
1399         tp = intotcpcb(inp);
1400
1401         /* Get socket's owner cpuid hint */
1402         if (sopt->sopt_level == SOL_SOCKET &&
1403             sopt->sopt_dir == SOPT_GET &&
1404             sopt->sopt_name == SO_CPUHINT) {
1405                 if (tp->t_flags & TF_LISTEN) {
1406                         /*
1407                          * Listen sockets owner cpuid is always 0,
1408                          * which does not make sense if SO_REUSEPORT
1409                          * is not set.
1410                          */
1411                         if (so->so_options & SO_REUSEPORT)
1412                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1413                         else
1414                                 optval = -1; /* no hint */
1415                 } else {
1416                         optval = mycpuid;
1417                 }
1418                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1419                 goto done;
1420         }
1421
1422         if (sopt->sopt_level != IPPROTO_TCP) {
1423                 if (sopt->sopt_level == IPPROTO_IP) {
1424                         switch (sopt->sopt_name) {
1425                         case IP_MULTICAST_IF:
1426                         case IP_MULTICAST_VIF:
1427                         case IP_MULTICAST_TTL:
1428                         case IP_MULTICAST_LOOP:
1429                         case IP_ADD_MEMBERSHIP:
1430                         case IP_DROP_MEMBERSHIP:
1431                                 /*
1432                                  * Multicast does not make sense on
1433                                  * TCP sockets.
1434                                  */
1435                                 error = EOPNOTSUPP;
1436                                 goto done;
1437                         }
1438                 }
1439 #ifdef INET6
1440                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1441                         ip6_ctloutput_dispatch(msg);
1442                 else
1443 #endif /* INET6 */
1444                 ip_ctloutput(msg);
1445                 /* msg invalid now */
1446                 return;
1447         }
1448
1449         switch (sopt->sopt_dir) {
1450         case SOPT_SET:
1451                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1452                                       sizeof optval);
1453                 if (error)
1454                         break;
1455                 switch (sopt->sopt_name) {
1456                 case TCP_FASTKEEP:
1457                         if (optval > 0)
1458                                 tp->t_keepidle = tp->t_keepintvl;
1459                         else
1460                                 tp->t_keepidle = tcp_keepidle;
1461                         tcp_timer_keep_activity(tp, 0);
1462                         break;
1463 #ifdef TCP_SIGNATURE
1464                 case TCP_SIGNATURE_ENABLE:
1465                         if (tp->t_state == TCPS_CLOSED) {
1466                                 /*
1467                                  * This is the only safe state that this
1468                                  * option could be changed.  Some segments
1469                                  * could already have been sent in other
1470                                  * states.
1471                                  */
1472                                 if (optval > 0)
1473                                         tp->t_flags |= TF_SIGNATURE;
1474                                 else
1475                                         tp->t_flags &= ~TF_SIGNATURE;
1476                         } else {
1477                                 error = EOPNOTSUPP;
1478                         }
1479                         break;
1480 #endif /* TCP_SIGNATURE */
1481                 case TCP_NODELAY:
1482                 case TCP_NOOPT:
1483                         switch (sopt->sopt_name) {
1484                         case TCP_NODELAY:
1485                                 opt = TF_NODELAY;
1486                                 break;
1487                         case TCP_NOOPT:
1488                                 opt = TF_NOOPT;
1489                                 break;
1490                         default:
1491                                 opt = 0; /* dead code to fool gcc */
1492                                 break;
1493                         }
1494
1495                         if (optval)
1496                                 tp->t_flags |= opt;
1497                         else
1498                                 tp->t_flags &= ~opt;
1499                         break;
1500
1501                 case TCP_NOPUSH:
1502                         if (tcp_disable_nopush)
1503                                 break;
1504                         if (optval)
1505                                 tp->t_flags |= TF_NOPUSH;
1506                         else {
1507                                 tp->t_flags &= ~TF_NOPUSH;
1508                                 error = tcp_output(tp);
1509                         }
1510                         break;
1511
1512                 case TCP_MAXSEG:
1513                         /*
1514                          * Must be between 0 and maxseg.  If the requested
1515                          * maxseg is too small to satisfy the desired minmss,
1516                          * pump it up (silently so sysctl modifications of
1517                          * minmss do not create unexpected program failures).
1518                          * Handle degenerate cases.
1519                          */
1520                         if (optval > 0 && optval <= tp->t_maxseg) {
1521                                 if (optval + 40 < tcp_minmss) {
1522                                         optval = tcp_minmss - 40;
1523                                         if (optval < 0)
1524                                                 optval = 1;
1525                                 }
1526                                 tp->t_maxseg = optval;
1527                         } else {
1528                                 error = EINVAL;
1529                         }
1530                         break;
1531
1532                 case TCP_KEEPINIT:
1533                         opthz = ((int64_t)optval * hz) / 1000;
1534                         if (opthz >= 1)
1535                                 tp->t_keepinit = opthz;
1536                         else
1537                                 error = EINVAL;
1538                         break;
1539
1540                 case TCP_KEEPIDLE:
1541                         opthz = ((int64_t)optval * hz) / 1000;
1542                         if (opthz >= 1) {
1543                                 tp->t_keepidle = opthz;
1544                                 tcp_timer_keep_activity(tp, 0);
1545                         } else {
1546                                 error = EINVAL;
1547                         }
1548                         break;
1549
1550                 case TCP_KEEPINTVL:
1551                         opthz = ((int64_t)optval * hz) / 1000;
1552                         if (opthz >= 1) {
1553                                 tp->t_keepintvl = opthz;
1554                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1555                         } else {
1556                                 error = EINVAL;
1557                         }
1558                         break;
1559
1560                 case TCP_KEEPCNT:
1561                         if (optval > 0) {
1562                                 tp->t_keepcnt = optval;
1563                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1564                         } else {
1565                                 error = EINVAL;
1566                         }
1567                         break;
1568
1569                 default:
1570                         error = ENOPROTOOPT;
1571                         break;
1572                 }
1573                 break;
1574
1575         case SOPT_GET:
1576                 switch (sopt->sopt_name) {
1577 #ifdef TCP_SIGNATURE
1578                 case TCP_SIGNATURE_ENABLE:
1579                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1580                         break;
1581 #endif /* TCP_SIGNATURE */
1582                 case TCP_NODELAY:
1583                         optval = tp->t_flags & TF_NODELAY;
1584                         break;
1585                 case TCP_MAXSEG:
1586                         optval = tp->t_maxseg;
1587                         break;
1588                 case TCP_NOOPT:
1589                         optval = tp->t_flags & TF_NOOPT;
1590                         break;
1591                 case TCP_NOPUSH:
1592                         optval = tp->t_flags & TF_NOPUSH;
1593                         break;
1594                 case TCP_KEEPINIT:
1595                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1596                         break;
1597                 case TCP_KEEPIDLE:
1598                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1599                         break;
1600                 case TCP_KEEPINTVL:
1601                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1602                         break;
1603                 case TCP_KEEPCNT:
1604                         optval = tp->t_keepcnt;
1605                         break;
1606                 default:
1607                         error = ENOPROTOOPT;
1608                         break;
1609                 }
1610                 if (error == 0)
1611                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1612                 break;
1613         }
1614 done:
1615         lwkt_replymsg(&msg->lmsg, error);
1616 }
1617
1618 /*
1619  * tcp_sendspace and tcp_recvspace are the default send and receive window
1620  * sizes, respectively.  These are obsolescent (this information should
1621  * be set by the route).
1622  *
1623  * Use a default that does not require tcp window scaling to be turned
1624  * on.  Individual programs or the administrator can increase the default.
1625  */
1626 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1627 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1628     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1629 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1630 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1631     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1632
1633 /*
1634  * Attach TCP protocol to socket, allocating internet protocol control
1635  * block, tcp control block, buffer space, and entering CLOSED state.
1636  */
1637 static int
1638 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1639 {
1640         struct tcpcb *tp;
1641         struct inpcb *inp;
1642         int error;
1643         int cpu;
1644 #ifdef INET6
1645         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1646 #endif
1647
1648         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1649                 lwkt_gettoken(&so->so_rcv.ssb_token);
1650                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1651                                   ai->sb_rlimit);
1652                 lwkt_reltoken(&so->so_rcv.ssb_token);
1653                 if (error)
1654                         return (error);
1655         }
1656         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1657         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1658         cpu = mycpu->gd_cpuid;
1659
1660         /*
1661          * Set the default port for protocol processing. This will likely
1662          * change when we connect.
1663          */
1664         error = in_pcballoc(so, &tcbinfo[cpu]);
1665         if (error)
1666                 return (error);
1667         inp = so->so_pcb;
1668 #ifdef INET6
1669         if (isipv6) {
1670                 inp->inp_vflag |= INP_IPV6;
1671                 inp->in6p_hops = -1;    /* use kernel default */
1672         }
1673         else
1674 #endif
1675         inp->inp_vflag |= INP_IPV4;
1676         tp = tcp_newtcpcb(inp);
1677         if (tp == NULL) {
1678                 /*
1679                  * Make sure the socket is destroyed by the pcbdetach.
1680                  */
1681                 soreference(so);
1682 #ifdef INET6
1683                 if (isipv6)
1684                         in6_pcbdetach(inp);
1685                 else
1686 #endif
1687                 in_pcbdetach(inp);
1688                 sofree(so);     /* from ref above */
1689                 return (ENOBUFS);
1690         }
1691         tp->t_state = TCPS_CLOSED;
1692         /* Keep a reference for asynchronized pru_rcvd */
1693         soreference(so);
1694         return (0);
1695 }
1696
1697 /*
1698  * Initiate (or continue) disconnect.
1699  * If embryonic state, just send reset (once).
1700  * If in ``let data drain'' option and linger null, just drop.
1701  * Otherwise (hard), mark socket disconnecting and drop
1702  * current input data; switch states based on user close, and
1703  * send segment to peer (with FIN).
1704  */
1705 static struct tcpcb *
1706 tcp_disconnect(struct tcpcb *tp)
1707 {
1708         struct socket *so = tp->t_inpcb->inp_socket;
1709
1710         if (tp->t_state < TCPS_ESTABLISHED) {
1711                 tp = tcp_close(tp);
1712         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1713                 tp = tcp_drop(tp, 0);
1714         } else {
1715                 lwkt_gettoken(&so->so_rcv.ssb_token);
1716                 soisdisconnecting(so);
1717                 sbflush(&so->so_rcv.sb);
1718                 tp = tcp_usrclosed(tp);
1719                 if (tp)
1720                         tcp_output(tp);
1721                 lwkt_reltoken(&so->so_rcv.ssb_token);
1722         }
1723         return (tp);
1724 }
1725
1726 /*
1727  * User issued close, and wish to trail through shutdown states:
1728  * if never received SYN, just forget it.  If got a SYN from peer,
1729  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1730  * If already got a FIN from peer, then almost done; go to LAST_ACK
1731  * state.  In all other cases, have already sent FIN to peer (e.g.
1732  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1733  * for peer to send FIN or not respond to keep-alives, etc.
1734  * We can let the user exit from the close as soon as the FIN is acked.
1735  */
1736 static struct tcpcb *
1737 tcp_usrclosed(struct tcpcb *tp)
1738 {
1739
1740         switch (tp->t_state) {
1741
1742         case TCPS_CLOSED:
1743         case TCPS_LISTEN:
1744                 tp->t_state = TCPS_CLOSED;
1745                 tp = tcp_close(tp);
1746                 break;
1747
1748         case TCPS_SYN_SENT:
1749         case TCPS_SYN_RECEIVED:
1750                 tp->t_flags |= TF_NEEDFIN;
1751                 break;
1752
1753         case TCPS_ESTABLISHED:
1754                 tp->t_state = TCPS_FIN_WAIT_1;
1755                 break;
1756
1757         case TCPS_CLOSE_WAIT:
1758                 tp->t_state = TCPS_LAST_ACK;
1759                 break;
1760         }
1761         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1762                 soisdisconnected(tp->t_inpcb->inp_socket);
1763                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1764                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1765                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1766                             tcp_timer_2msl);
1767                 }
1768         }
1769         return (tp);
1770 }