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         error = in6_pcbbind(inp, nam, td);
 336         if (error)
 337                 goto out;
 338         COMMON_END(PRU_BIND);
 339 }
 340 #endif /* INET6 */
 341
 342 struct netmsg_inswildcard {
 343         struct netmsg_base      base;
 344         struct inpcb            *nm_inp;
 345 };
 346
 347 static void
 348 in_pcbinswildcardhash_handler(netmsg_t msg)
 349 {
 350         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 351         int cpu = mycpuid, nextcpu;
 352
 353         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 354
 355         nextcpu = cpu + 1;
 356         if (nextcpu < ncpus2)
 357                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 358         else
 359                 lwkt_replymsg(&nm->base.lmsg, 0);
 360 }
 361
 362 static void
 363 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 364 {
 365         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 366 }
 367
 368 /*
 369  * Prepare to accept connections.
 370  */
 371 static void
 372 tcp_usr_listen(netmsg_t msg)
 373 {
 374         struct socket *so = msg->listen.base.nm_so;
 375         struct thread *td = msg->listen.nm_td;
 376         int error = 0;
 377         struct inpcb *inp;
 378         struct tcpcb *tp;
 379         struct netmsg_inswildcard nm;
 380         lwkt_port_t port0 = netisr_cpuport(0);
 381
 382         COMMON_START(so, inp, 0);
 383
 384         if (&curthread->td_msgport != port0) {
 385                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 386
 387                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 388                     ("already asked to relink"));
 389
 390                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 391                 msg->listen.nm_flags |= PRUL_RELINK;
 392
 393                 /* See the related comment in tcp_connect() */
 394                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 395                 lwkt_forwardmsg(port0, lmsg);
 396                 /* msg invalid now */
 397                 return;
 398         }
 399         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 400
 401         if (msg->listen.nm_flags & PRUL_RELINK) {
 402                 msg->listen.nm_flags &= ~PRUL_RELINK;
 403                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 404         }
 405         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 406
 407         if (tp->t_flags & TF_LISTEN)
 408                 goto out;
 409
 410         if (inp->inp_lport == 0) {
 411                 error = in_pcbbind(inp, NULL, td);
 412                 if (error)
 413                         goto out;
 414         }
 415
 416         tp->t_state = TCPS_LISTEN;
 417         tp->t_flags |= TF_LISTEN;
 418         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 419
 420         if (ncpus2 > 1) {
 421                 /*
 422                  * Put this inpcb into wildcard hash on other cpus.
 423                  */
 424                 ASSERT_INP_NOTINHASH(inp);
 425                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 426                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 427                 nm.nm_inp = inp;
 428                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 429         }
 430         in_pcbinswildcardhash(inp);
 431         COMMON_END(PRU_LISTEN);
 432 }
 433
 434 #ifdef INET6
 435
 436 static void
 437 tcp6_usr_listen(netmsg_t msg)
 438 {
 439         struct socket *so = msg->listen.base.nm_so;
 440         struct thread *td = msg->listen.nm_td;
 441         int error = 0;
 442         struct inpcb *inp;
 443         struct tcpcb *tp;
 444         struct netmsg_inswildcard nm;
 445
 446         COMMON_START(so, inp, 0);
 447
 448         if (tp->t_flags & TF_LISTEN)
 449                 goto out;
 450
 451         if (inp->inp_lport == 0) {
 452                 error = in6_pcbbind(inp, NULL, td);
 453                 if (error)
 454                         goto out;
 455         }
 456
 457         tp->t_state = TCPS_LISTEN;
 458         tp->t_flags |= TF_LISTEN;
 459         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 460
 461         if (ncpus2 > 1) {
 462                 /*
 463                  * Put this inpcb into wildcard hash on other cpus.
 464                  */
 465                 KKASSERT(so->so_port == netisr_cpuport(0));
 466                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
 467                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 468                 ASSERT_INP_NOTINHASH(inp);
 469
 470                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 471                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 472                 nm.nm_inp = inp;
 473                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 474         }
 475         in_pcbinswildcardhash(inp);
 476         COMMON_END(PRU_LISTEN);
 477 }
 478 #endif /* INET6 */
 479
 480 /*
 481  * Initiate connection to peer.
 482  * Create a template for use in transmissions on this connection.
 483  * Enter SYN_SENT state, and mark socket as connecting.
 484  * Start keep-alive timer, and seed output sequence space.
 485  * Send initial segment on connection.
 486  */
 487 static void
 488 tcp_usr_connect(netmsg_t msg)
 489 {
 490         struct socket *so = msg->connect.base.nm_so;
 491         struct sockaddr *nam = msg->connect.nm_nam;
 492         struct thread *td = msg->connect.nm_td;
 493         int error = 0;
 494         struct inpcb *inp;
 495         struct tcpcb *tp;
 496         struct sockaddr_in *sinp;
 497
 498         COMMON_START(so, inp, 0);
 499
 500         /*
 501          * Must disallow TCP ``connections'' to multicast addresses.
 502          */
 503         sinp = (struct sockaddr_in *)nam;
 504         if (sinp->sin_family == AF_INET
 505             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 506                 error = EAFNOSUPPORT;
 507                 goto out;
 508         }
 509
 510         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 511                 error = EAFNOSUPPORT; /* IPv6 only jail */
 512                 goto out;
 513         }
 514
 515         tcp_connect(msg);
 516         /* msg is invalid now */
 517         return;
 518 out:
 519         if (msg->connect.nm_m) {
 520                 m_freem(msg->connect.nm_m);
 521                 msg->connect.nm_m = NULL;
 522         }
 523         if (msg->connect.nm_flags & PRUC_HELDTD)
 524                 lwkt_rele(td);
 525         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 526                 so->so_error = error;
 527                 soisdisconnected(so);
 528         }
 529         lwkt_replymsg(&msg->lmsg, error);
 530 }
 531
 532 #ifdef INET6
 533
 534 static void
 535 tcp6_usr_connect(netmsg_t msg)
 536 {
 537         struct socket *so = msg->connect.base.nm_so;
 538         struct sockaddr *nam = msg->connect.nm_nam;
 539         struct thread *td = msg->connect.nm_td;
 540         int error = 0;
 541         struct inpcb *inp;
 542         struct tcpcb *tp;
 543         struct sockaddr_in6 *sin6p;
 544
 545         COMMON_START(so, inp, 0);
 546
 547         /*
 548          * Must disallow TCP ``connections'' to multicast addresses.
 549          */
 550         sin6p = (struct sockaddr_in6 *)nam;
 551         if (sin6p->sin6_family == AF_INET6
 552             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 553                 error = EAFNOSUPPORT;
 554                 goto out;
 555         }
 556
 557         if (!prison_remote_ip(td, nam)) {
 558                 error = EAFNOSUPPORT; /* IPv4 only jail */
 559                 goto out;
 560         }
 561
 562         /* Reject v4-mapped address */
 563         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 564                 error = EADDRNOTAVAIL;
 565                 goto out;
 566         }
 567
 568         inp->inp_inc.inc_isipv6 = 1;
 569         tcp6_connect(msg);
 570         /* msg is invalid now */
 571         return;
 572 out:
 573         if (msg->connect.nm_m) {
 574                 m_freem(msg->connect.nm_m);
 575                 msg->connect.nm_m = NULL;
 576         }
 577         lwkt_replymsg(&msg->lmsg, error);
 578 }
 579
 580 #endif /* INET6 */
 581
 582 /*
 583  * Initiate disconnect from peer.
 584  * If connection never passed embryonic stage, just drop;
 585  * else if don't need to let data drain, then can just drop anyways,
 586  * else have to begin TCP shutdown process: mark socket disconnecting,
 587  * drain unread data, state switch to reflect user close, and
 588  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 589  * when peer sends FIN and acks ours.
 590  *
 591  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 592  */
 593 static void
 594 tcp_usr_disconnect(netmsg_t msg)
 595 {
 596         struct socket *so = msg->disconnect.base.nm_so;
 597         int error = 0;
 598         struct inpcb *inp;
 599         struct tcpcb *tp;
 600
 601         COMMON_START(so, inp, 1);
 602         tp = tcp_disconnect(tp);
 603         COMMON_END(PRU_DISCONNECT);
 604 }
 605
 606 /*
 607  * Accept a connection.  Essentially all the work is
 608  * done at higher levels; just return the address
 609  * of the peer, storing through addr.
 610  */
 611 static void
 612 tcp_usr_accept(netmsg_t msg)
 613 {
 614         struct socket *so = msg->accept.base.nm_so;
 615         struct sockaddr **nam = msg->accept.nm_nam;
 616         int error = 0;
 617         struct inpcb *inp;
 618         struct tcpcb *tp = NULL;
 619         TCPDEBUG0;
 620
 621         inp = so->so_pcb;
 622         if (so->so_state & SS_ISDISCONNECTED) {
 623                 error = ECONNABORTED;
 624                 goto out;
 625         }
 626         if (inp == 0) {
 627                 error = EINVAL;
 628                 goto out;
 629         }
 630
 631         tp = intotcpcb(inp);
 632         TCPDEBUG1();
 633         in_setpeeraddr(so, nam);
 634         COMMON_END(PRU_ACCEPT);
 635 }
 636
 637 #ifdef INET6
 638 static void
 639 tcp6_usr_accept(netmsg_t msg)
 640 {
 641         struct socket *so = msg->accept.base.nm_so;
 642         struct sockaddr **nam = msg->accept.nm_nam;
 643         int error = 0;
 644         struct inpcb *inp;
 645         struct tcpcb *tp = NULL;
 646         TCPDEBUG0;
 647
 648         inp = so->so_pcb;
 649
 650         if (so->so_state & SS_ISDISCONNECTED) {
 651                 error = ECONNABORTED;
 652                 goto out;
 653         }
 654         if (inp == 0) {
 655                 error = EINVAL;
 656                 goto out;
 657         }
 658         tp = intotcpcb(inp);
 659         TCPDEBUG1();
 660         in6_setpeeraddr(so, nam);
 661         COMMON_END(PRU_ACCEPT);
 662 }
 663 #endif /* INET6 */
 664 /*
 665  * Mark the connection as being incapable of further output.
 666  */
 667 static void
 668 tcp_usr_shutdown(netmsg_t msg)
 669 {
 670         struct socket *so = msg->shutdown.base.nm_so;
 671         int error = 0;
 672         struct inpcb *inp;
 673         struct tcpcb *tp;
 674
 675         COMMON_START(so, inp, 0);
 676         socantsendmore(so);
 677         tp = tcp_usrclosed(tp);
 678         if (tp)
 679                 error = tcp_output(tp);
 680         COMMON_END(PRU_SHUTDOWN);
 681 }
 682
 683 /*
 684  * After a receive, possibly send window update to peer.
 685  */
 686 static void
 687 tcp_usr_rcvd(netmsg_t msg)
 688 {
 689         struct socket *so = msg->rcvd.base.nm_so;
 690         int error = 0, noreply = 0;
 691         struct inpcb *inp;
 692         struct tcpcb *tp;
 693
 694         COMMON_START(so, inp, 0);
 695
 696         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 697                 noreply = 1;
 698                 so_async_rcvd_reply(so);
 699         }
 700         tcp_output(tp);
 701
 702         COMMON_END1(PRU_RCVD, noreply);
 703 }
 704
 705 /*
 706  * Do a send by putting data in output queue and updating urgent
 707  * marker if URG set.  Possibly send more data.  Unlike the other
 708  * pru_*() routines, the mbuf chains are our responsibility.  We
 709  * must either enqueue them or free them.  The other pru_* routines
 710  * generally are caller-frees.
 711  */
 712 static void
 713 tcp_usr_send(netmsg_t msg)
 714 {
 715         struct socket *so = msg->send.base.nm_so;
 716         int flags = msg->send.nm_flags;
 717         struct mbuf *m = msg->send.nm_m;
 718         int error = 0;
 719         struct inpcb *inp;
 720         struct tcpcb *tp;
 721         TCPDEBUG0;
 722
 723         KKASSERT(msg->send.nm_control == NULL);
 724         KKASSERT(msg->send.nm_addr == NULL);
 725         KKASSERT((flags & PRUS_FREEADDR) == 0);
 726
 727         inp = so->so_pcb;
 728
 729         if (inp == NULL) {
 730                 /*
 731                  * OOPS! we lost a race, the TCP session got reset after
 732                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 733                  * network interrupt in the non-critical section of sosend().
 734                  */
 735                 m_freem(m);
 736                 error = ECONNRESET;     /* XXX EPIPE? */
 737                 tp = NULL;
 738                 TCPDEBUG1();
 739                 goto out;
 740         }
 741         tp = intotcpcb(inp);
 742         TCPDEBUG1();
 743
 744 #ifdef foo
 745         /*
 746          * This is no longer necessary, since:
 747          * - sosendtcp() has already checked it for us
 748          * - It does not work with asynchronized send
 749          */
 750
 751         /*
 752          * Don't let too much OOB data build up
 753          */
 754         if (flags & PRUS_OOB) {
 755                 if (ssb_space(&so->so_snd) < -512) {
 756                         m_freem(m);
 757                         error = ENOBUFS;
 758                         goto out;
 759                 }
 760         }
 761 #endif
 762
 763         /*
 764          * Pump the data into the socket.
 765          */
 766         if (m) {
 767                 ssb_appendstream(&so->so_snd, m);
 768                 sowwakeup(so);
 769         }
 770         if (flags & PRUS_OOB) {
 771                 /*
 772                  * According to RFC961 (Assigned Protocols),
 773                  * the urgent pointer points to the last octet
 774                  * of urgent data.  We continue, however,
 775                  * to consider it to indicate the first octet
 776                  * of data past the urgent section.
 777                  * Otherwise, snd_up should be one lower.
 778                  */
 779                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 780                 tp->t_flags |= TF_FORCE;
 781                 error = tcp_output(tp);
 782                 tp->t_flags &= ~TF_FORCE;
 783         } else {
 784                 if (flags & PRUS_EOF) {
 785                         /*
 786                          * Close the send side of the connection after
 787                          * the data is sent.
 788                          */
 789                         socantsendmore(so);
 790                         tp = tcp_usrclosed(tp);
 791                 }
 792                 if (tp != NULL && !tcp_output_pending(tp)) {
 793                         if (flags & PRUS_MORETOCOME)
 794                                 tp->t_flags |= TF_MORETOCOME;
 795                         error = tcp_output_fair(tp);
 796                         if (flags & PRUS_MORETOCOME)
 797                                 tp->t_flags &= ~TF_MORETOCOME;
 798                 }
 799         }
 800         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 801                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 802                    (flags & PRUS_NOREPLY));
 803 }
 804
 805 /*
 806  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 807  *       will sofree() it when we return.
 808  */
 809 static void
 810 tcp_usr_abort(netmsg_t msg)
 811 {
 812         struct socket *so = msg->abort.base.nm_so;
 813         int error = 0;
 814         struct inpcb *inp;
 815         struct tcpcb *tp;
 816
 817         COMMON_START(so, inp, 1);
 818         tp = tcp_drop(tp, ECONNABORTED);
 819         COMMON_END(PRU_ABORT);
 820 }
 821
 822 /*
 823  * Receive out-of-band data.
 824  */
 825 static void
 826 tcp_usr_rcvoob(netmsg_t msg)
 827 {
 828         struct socket *so = msg->rcvoob.base.nm_so;
 829         struct mbuf *m = msg->rcvoob.nm_m;
 830         int flags = msg->rcvoob.nm_flags;
 831         int error = 0;
 832         struct inpcb *inp;
 833         struct tcpcb *tp;
 834
 835         COMMON_START(so, inp, 0);
 836         if ((so->so_oobmark == 0 &&
 837              (so->so_state & SS_RCVATMARK) == 0) ||
 838             so->so_options & SO_OOBINLINE ||
 839             tp->t_oobflags & TCPOOB_HADDATA) {
 840                 error = EINVAL;
 841                 goto out;
 842         }
 843         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 844                 error = EWOULDBLOCK;
 845                 goto out;
 846         }
 847         m->m_len = 1;
 848         *mtod(m, caddr_t) = tp->t_iobc;
 849         if ((flags & MSG_PEEK) == 0)
 850                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 851         COMMON_END(PRU_RCVOOB);
 852 }
 853
 854 static void
 855 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 856 {
 857         in_savefaddr(so, faddr);
 858 }
 859
 860 #ifdef INET6
 861 static void
 862 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 863 {
 864         in6_savefaddr(so, faddr);
 865 }
 866 #endif
 867
 868 static int
 869 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 870     struct thread *td __unused)
 871 {
 872         const struct sockaddr_in *sinp;
 873
 874         sinp = (const struct sockaddr_in *)nam;
 875         if (sinp->sin_family == AF_INET &&
 876             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 877                 return EAFNOSUPPORT;
 878
 879         soisconnecting(so);
 880         return 0;
 881 }
 882
 883 /* xxx - should be const */
 884 struct pr_usrreqs tcp_usrreqs = {
 885         .pru_abort = tcp_usr_abort,
 886         .pru_accept = tcp_usr_accept,
 887         .pru_attach = tcp_usr_attach,
 888         .pru_bind = tcp_usr_bind,
 889         .pru_connect = tcp_usr_connect,
 890         .pru_connect2 = pr_generic_notsupp,
 891         .pru_control = in_control_dispatch,
 892         .pru_detach = tcp_usr_detach,
 893         .pru_disconnect = tcp_usr_disconnect,
 894         .pru_listen = tcp_usr_listen,
 895         .pru_peeraddr = in_setpeeraddr_dispatch,
 896         .pru_rcvd = tcp_usr_rcvd,
 897         .pru_rcvoob = tcp_usr_rcvoob,
 898         .pru_send = tcp_usr_send,
 899         .pru_sense = pru_sense_null,
 900         .pru_shutdown = tcp_usr_shutdown,
 901         .pru_sockaddr = in_setsockaddr_dispatch,
 902         .pru_sosend = sosendtcp,
 903         .pru_soreceive = sorecvtcp,
 904         .pru_savefaddr = tcp_usr_savefaddr,
 905         .pru_preconnect = tcp_usr_preconnect
 906 };
 907
 908 #ifdef INET6
 909 struct pr_usrreqs tcp6_usrreqs = {
 910         .pru_abort = tcp_usr_abort,
 911         .pru_accept = tcp6_usr_accept,
 912         .pru_attach = tcp_usr_attach,
 913         .pru_bind = tcp6_usr_bind,
 914         .pru_connect = tcp6_usr_connect,
 915         .pru_connect2 = pr_generic_notsupp,
 916         .pru_control = in6_control_dispatch,
 917         .pru_detach = tcp_usr_detach,
 918         .pru_disconnect = tcp_usr_disconnect,
 919         .pru_listen = tcp6_usr_listen,
 920         .pru_peeraddr = in6_setpeeraddr_dispatch,
 921         .pru_rcvd = tcp_usr_rcvd,
 922         .pru_rcvoob = tcp_usr_rcvoob,
 923         .pru_send = tcp_usr_send,
 924         .pru_sense = pru_sense_null,
 925         .pru_shutdown = tcp_usr_shutdown,
 926         .pru_sockaddr = in6_setsockaddr_dispatch,
 927         .pru_sosend = sosendtcp,
 928         .pru_soreceive = sorecvtcp,
 929         .pru_savefaddr = tcp6_usr_savefaddr
 930 };
 931 #endif /* INET6 */
 932
 933 static int
 934 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
 935                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
 936 {
 937         struct inpcb *inp = tp->t_inpcb, *oinp;
 938         struct socket *so = inp->inp_socket;
 939         struct route *ro = &inp->inp_route;
 940
 941         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
 942             ("pcbinfo mismatch"));
 943
 944         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
 945                                  sin->sin_addr, sin->sin_port,
 946                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
 947                                   inp->inp_laddr : if_sin->sin_addr),
 948                                 inp->inp_lport, 0, NULL);
 949         if (oinp != NULL) {
 950                 m_freem(m);
 951                 return (EADDRINUSE);
 952         }
 953         if (inp->inp_laddr.s_addr == INADDR_ANY)
 954                 inp->inp_laddr = if_sin->sin_addr;
 955         inp->inp_faddr = sin->sin_addr;
 956         inp->inp_fport = sin->sin_port;
 957         in_pcbinsconnhash(inp);
 958
 959         /*
 960          * We are now on the inpcb's owner CPU, if the cached route was
 961          * freed because the rtentry's owner CPU is not the current CPU
 962          * (e.g. in tcp_connect()), then we try to reallocate it here with
 963          * the hope that a rtentry may be cloned from a RTF_PRCLONING
 964          * rtentry.
 965          */
 966         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
 967             ro->ro_rt == NULL) {
 968                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
 969                 ro->ro_dst.sa_family = AF_INET;
 970                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
 971                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
 972                         sin->sin_addr;
 973                 rtalloc(ro);
 974         }
 975
 976         /*
 977          * Now that no more errors can occur, change the protocol processing
 978          * port to the current thread (which is the correct thread).
 979          *
 980          * Create TCP timer message now; we are on the tcpcb's owner
 981          * CPU/thread.
 982          */
 983         tcp_create_timermsg(tp, &curthread->td_msgport);
 984
 985         /*
 986          * Compute window scaling to request.  Use a larger scaling then
 987          * needed for the initial receive buffer in case the receive buffer
 988          * gets expanded.
 989          */
 990         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
 991                 tp->request_r_scale = TCP_MIN_WINSHIFT;
 992         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
 993                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
 994         ) {
 995                 tp->request_r_scale++;
 996         }
 997
 998         soisconnecting(so);
 999         tcpstat.tcps_connattempt++;
1000         tp->t_state = TCPS_SYN_SENT;
1001         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1002         tp->iss = tcp_new_isn(tp);
1003         tcp_sendseqinit(tp);
1004         if (m) {
1005                 ssb_appendstream(&so->so_snd, m);
1006                 m = NULL;
1007                 if (flags & PRUS_OOB)
1008                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1009         }
1010
1011         /*
1012          * Close the send side of the connection after
1013          * the data is sent if flagged.
1014          */
1015         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1016                 socantsendmore(so);
1017                 tp = tcp_usrclosed(tp);
1018         }
1019         return (tcp_output(tp));
1020 }
1021
1022 /*
1023  * Common subroutine to open a TCP connection to remote host specified
1024  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1025  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1026  * a local host address (interface).
1027  * Initialize connection parameters and enter SYN-SENT state.
1028  */
1029 static void
1030 tcp_connect(netmsg_t msg)
1031 {
1032         struct socket *so = msg->connect.base.nm_so;
1033         struct sockaddr *nam = msg->connect.nm_nam;
1034         struct thread *td = msg->connect.nm_td;
1035         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1036         struct sockaddr_in *if_sin = NULL;
1037         struct inpcb *inp;
1038         struct tcpcb *tp;
1039         int error;
1040         lwkt_port_t port;
1041
1042         COMMON_START(so, inp, 0);
1043
1044         /*
1045          * Reconnect our pcb if we have to
1046          */
1047         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1048                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1049                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1050         }
1051
1052         /*
1053          * Bind if we have to
1054          */
1055         if (inp->inp_lport == 0) {
1056                 if (tcp_lport_extension) {
1057                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1058
1059                         error = in_pcbladdr(inp, nam, &if_sin, td);
1060                         if (error)
1061                                 goto out;
1062                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1063
1064                         error = in_pcbbind_remote(inp, nam, td);
1065                         if (error)
1066                                 goto out;
1067
1068                         msg->connect.nm_flags |= PRUC_HASLADDR;
1069                 } else {
1070                         error = in_pcbbind(inp, NULL, td);
1071                         if (error)
1072                                 goto out;
1073                 }
1074         }
1075
1076         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1077                 /*
1078                  * Calculate the correct protocol processing thread.  The
1079                  * connect operation must run there.  Set the forwarding
1080                  * port before we forward the message or it will get bounced
1081                  * right back to us.
1082                  */
1083                 error = in_pcbladdr(inp, nam, &if_sin, td);
1084                 if (error)
1085                         goto out;
1086         }
1087         KKASSERT(inp->inp_socket == so);
1088
1089         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1090                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1091                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1092                             inp->inp_lport);
1093
1094         if (port != &curthread->td_msgport) {
1095                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1096
1097                 /*
1098                  * in_pcbladdr() may have allocated a route entry for us
1099                  * on the current CPU, but we need a route entry on the
1100                  * inpcb's owner CPU, so free it here.
1101                  */
1102                 in_pcbresetroute(inp);
1103
1104                 /*
1105                  * We are moving the protocol processing port the socket
1106                  * is on, we have to unlink here and re-link on the
1107                  * target cpu.
1108                  */
1109                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1110                 msg->connect.nm_flags |= PRUC_RECONNECT;
1111                 msg->connect.base.nm_dispatch = tcp_connect;
1112
1113                 /*
1114                  * Use message put done receipt to change this socket's
1115                  * so_port, i.e. _after_ this message was put onto the
1116                  * target netisr's msgport but _before_ the message could
1117                  * be pulled from the target netisr's msgport, so that:
1118                  * - The upper half (socket code) will not see the new
1119                  *   msgport before this message reaches the new msgport
1120                  *   and messages for this socket will be ordered.
1121                  * - This message will see the new msgport, when its
1122                  *   handler is called in the target netisr.
1123                  *
1124                  * NOTE:
1125                  * We MUST use messege put done receipt to change this
1126                  * socket's so_port:
1127                  * If we changed the so_port in this netisr after the
1128                  * lwkt_forwardmsg (so messages for this socket will be
1129                  * ordered) and changed the so_port in the target netisr
1130                  * at the very beginning of this message's handler, we
1131                  * would suffer so_port overwritten race, given this
1132                  * message might be forwarded again.
1133                  *
1134                  * NOTE:
1135                  * This mechanism depends on that the netisr's msgport
1136                  * is spin msgport (currently it is :).
1137                  *
1138                  * If the upper half saw the new msgport before this
1139                  * message reached the target netisr's msgport, the
1140                  * messages sent from the upper half could reach the new
1141                  * msgport before this message, thus there would be
1142                  * message reordering.  The worst case could be soclose()
1143                  * saw the new msgport and the detach message could reach
1144                  * the new msgport before this message, i.e. the inpcb
1145                  * could have been destroyed when this message was still
1146                  * pending on or on its way to the new msgport.  Other
1147                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1148                  * since we have unlinked this inpcb from the current
1149                  * pcbinfo first.
1150                  */
1151                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1152                 lwkt_forwardmsg(port, lmsg);
1153                 /* msg invalid now */
1154                 return;
1155         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1156                 /*
1157                  * The original thread is no longer needed; release it.
1158                  */
1159                 lwkt_rele(td);
1160                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1161         }
1162         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1163                                   msg->connect.nm_m, sin, if_sin);
1164         msg->connect.nm_m = NULL;
1165 out:
1166         if (msg->connect.nm_m) {
1167                 m_freem(msg->connect.nm_m);
1168                 msg->connect.nm_m = NULL;
1169         }
1170         if (msg->connect.nm_flags & PRUC_HELDTD)
1171                 lwkt_rele(td);
1172         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1173                 so->so_error = error;
1174                 soisdisconnected(so);
1175         }
1176         lwkt_replymsg(&msg->connect.base.lmsg, error);
1177         /* msg invalid now */
1178 }
1179
1180 #ifdef INET6
1181
1182 static void
1183 tcp6_connect(netmsg_t msg)
1184 {
1185         struct tcpcb *tp;
1186         struct socket *so = msg->connect.base.nm_so;
1187         struct sockaddr *nam = msg->connect.nm_nam;
1188         struct thread *td = msg->connect.nm_td;
1189         struct inpcb *inp;
1190         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1191         struct in6_addr *addr6;
1192         lwkt_port_t port;
1193         int error;
1194
1195         COMMON_START(so, inp, 0);
1196
1197         /*
1198          * Reconnect our pcb if we have to
1199          */
1200         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1201                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1202                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1203         }
1204
1205         /*
1206          * Bind if we have to
1207          */
1208         if (inp->inp_lport == 0) {
1209                 error = in6_pcbbind(inp, NULL, td);
1210                 if (error)
1211                         goto out;
1212         }
1213
1214         /*
1215          * Cannot simply call in_pcbconnect, because there might be an
1216          * earlier incarnation of this same connection still in
1217          * TIME_WAIT state, creating an ADDRINUSE error.
1218          */
1219         error = in6_pcbladdr(inp, nam, &addr6, td);
1220         if (error)
1221                 goto out;
1222
1223         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1224
1225         if (port != &curthread->td_msgport) {
1226                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1227
1228                 /*
1229                  * in_pcbladdr() may have allocated a route entry for us
1230                  * on the current CPU, but we need a route entry on the
1231                  * inpcb's owner CPU, so free it here.
1232                  */
1233                 in_pcbresetroute(inp);
1234
1235                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1236                 msg->connect.nm_flags |= PRUC_RECONNECT;
1237                 msg->connect.base.nm_dispatch = tcp6_connect;
1238
1239                 /* See the related comment in tcp_connect() */
1240                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1241                 lwkt_forwardmsg(port, lmsg);
1242                 /* msg invalid now */
1243                 return;
1244         }
1245         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1246                                    &msg->connect.nm_m, sin6, addr6);
1247         /* nm_m may still be intact */
1248 out:
1249         if (msg->connect.nm_m) {
1250                 m_freem(msg->connect.nm_m);
1251                 msg->connect.nm_m = NULL;
1252         }
1253         lwkt_replymsg(&msg->connect.base.lmsg, error);
1254         /* msg invalid now */
1255 }
1256
1257 static int
1258 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1259                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1260 {
1261         struct mbuf *m = *mp;
1262         struct inpcb *inp = tp->t_inpcb;
1263         struct socket *so = inp->inp_socket;
1264         struct inpcb *oinp;
1265
1266         /*
1267          * Cannot simply call in_pcbconnect, because there might be an
1268          * earlier incarnation of this same connection still in
1269          * TIME_WAIT state, creating an ADDRINUSE error.
1270          */
1271         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1272                                   &sin6->sin6_addr, sin6->sin6_port,
1273                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1274                                       addr6 : &inp->in6p_laddr),
1275                                   inp->inp_lport,  0, NULL);
1276         if (oinp)
1277                 return (EADDRINUSE);
1278
1279         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1280                 inp->in6p_laddr = *addr6;
1281         inp->in6p_faddr = sin6->sin6_addr;
1282         inp->inp_fport = sin6->sin6_port;
1283         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1284                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1285         in_pcbinsconnhash(inp);
1286
1287         /*
1288          * Now that no more errors can occur, change the protocol processing
1289          * port to the current thread (which is the correct thread).
1290          *
1291          * Create TCP timer message now; we are on the tcpcb's owner
1292          * CPU/thread.
1293          */
1294         tcp_create_timermsg(tp, &curthread->td_msgport);
1295
1296         /* Compute window scaling to request.  */
1297         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1298                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1299         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1300             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1301                 tp->request_r_scale++;
1302         }
1303
1304         soisconnecting(so);
1305         tcpstat.tcps_connattempt++;
1306         tp->t_state = TCPS_SYN_SENT;
1307         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1308         tp->iss = tcp_new_isn(tp);
1309         tcp_sendseqinit(tp);
1310         if (m) {
1311                 ssb_appendstream(&so->so_snd, m);
1312                 *mp = NULL;
1313                 if (flags & PRUS_OOB)
1314                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1315         }
1316
1317         /*
1318          * Close the send side of the connection after
1319          * the data is sent if flagged.
1320          */
1321         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1322                 socantsendmore(so);
1323                 tp = tcp_usrclosed(tp);
1324         }
1325         return (tcp_output(tp));
1326 }
1327
1328 #endif /* INET6 */
1329
1330 /*
1331  * The new sockopt interface makes it possible for us to block in the
1332  * copyin/out step (if we take a page fault).  Taking a page fault while
1333  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1334  * both now use TSM, there probably isn't any need for this function to
1335  * run in a critical section any more.  This needs more examination.)
1336  */
1337 void
1338 tcp_ctloutput(netmsg_t msg)
1339 {
1340         struct socket *so = msg->base.nm_so;
1341         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1342         int     error, opt, optval, opthz;
1343         struct  inpcb *inp;
1344         struct  tcpcb *tp;
1345
1346         error = 0;
1347         inp = so->so_pcb;
1348         if (inp == NULL) {
1349                 error = ECONNRESET;
1350                 goto done;
1351         }
1352         tp = intotcpcb(inp);
1353
1354         /* Get socket's owner cpuid hint */
1355         if (sopt->sopt_level == SOL_SOCKET &&
1356             sopt->sopt_dir == SOPT_GET &&
1357             sopt->sopt_name == SO_CPUHINT) {
1358                 if (tp->t_flags & TF_LISTEN) {
1359                         /*
1360                          * Listen sockets owner cpuid is always 0,
1361                          * which does not make sense if SO_REUSEPORT
1362                          * is not set.
1363                          */
1364                         if (so->so_options & SO_REUSEPORT)
1365                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1366                         else
1367                                 optval = -1; /* no hint */
1368                 } else {
1369                         optval = mycpuid;
1370                 }
1371                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1372                 goto done;
1373         }
1374
1375         if (sopt->sopt_level != IPPROTO_TCP) {
1376                 if (sopt->sopt_level == IPPROTO_IP) {
1377                         switch (sopt->sopt_name) {
1378                         case IP_MULTICAST_IF:
1379                         case IP_MULTICAST_VIF:
1380                         case IP_MULTICAST_TTL:
1381                         case IP_MULTICAST_LOOP:
1382                         case IP_ADD_MEMBERSHIP:
1383                         case IP_DROP_MEMBERSHIP:
1384                                 /*
1385                                  * Multicast does not make sense on
1386                                  * TCP sockets.
1387                                  */
1388                                 error = EOPNOTSUPP;
1389                                 goto done;
1390                         }
1391                 }
1392 #ifdef INET6
1393                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1394                         ip6_ctloutput_dispatch(msg);
1395                 else
1396 #endif /* INET6 */
1397                 ip_ctloutput(msg);
1398                 /* msg invalid now */
1399                 return;
1400         }
1401
1402         switch (sopt->sopt_dir) {
1403         case SOPT_SET:
1404                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1405                                       sizeof optval);
1406                 if (error)
1407                         break;
1408                 switch (sopt->sopt_name) {
1409                 case TCP_FASTKEEP:
1410                         if (optval > 0)
1411                                 tp->t_keepidle = tp->t_keepintvl;
1412                         else
1413                                 tp->t_keepidle = tcp_keepidle;
1414                         tcp_timer_keep_activity(tp, 0);
1415                         break;
1416 #ifdef TCP_SIGNATURE
1417                 case TCP_SIGNATURE_ENABLE:
1418                         if (tp->t_state == TCPS_CLOSED) {
1419                                 /*
1420                                  * This is the only safe state that this
1421                                  * option could be changed.  Some segments
1422                                  * could already have been sent in other
1423                                  * states.
1424                                  */
1425                                 if (optval > 0)
1426                                         tp->t_flags |= TF_SIGNATURE;
1427                                 else
1428                                         tp->t_flags &= ~TF_SIGNATURE;
1429                         } else {
1430                                 error = EOPNOTSUPP;
1431                         }
1432                         break;
1433 #endif /* TCP_SIGNATURE */
1434                 case TCP_NODELAY:
1435                 case TCP_NOOPT:
1436                         switch (sopt->sopt_name) {
1437                         case TCP_NODELAY:
1438                                 opt = TF_NODELAY;
1439                                 break;
1440                         case TCP_NOOPT:
1441                                 opt = TF_NOOPT;
1442                                 break;
1443                         default:
1444                                 opt = 0; /* dead code to fool gcc */
1445                                 break;
1446                         }
1447
1448                         if (optval)
1449                                 tp->t_flags |= opt;
1450                         else
1451                                 tp->t_flags &= ~opt;
1452                         break;
1453
1454                 case TCP_NOPUSH:
1455                         if (tcp_disable_nopush)
1456                                 break;
1457                         if (optval)
1458                                 tp->t_flags |= TF_NOPUSH;
1459                         else {
1460                                 tp->t_flags &= ~TF_NOPUSH;
1461                                 error = tcp_output(tp);
1462                         }
1463                         break;
1464
1465                 case TCP_MAXSEG:
1466                         /*
1467                          * Must be between 0 and maxseg.  If the requested
1468                          * maxseg is too small to satisfy the desired minmss,
1469                          * pump it up (silently so sysctl modifications of
1470                          * minmss do not create unexpected program failures).
1471                          * Handle degenerate cases.
1472                          */
1473                         if (optval > 0 && optval <= tp->t_maxseg) {
1474                                 if (optval + 40 < tcp_minmss) {
1475                                         optval = tcp_minmss - 40;
1476                                         if (optval < 0)
1477                                                 optval = 1;
1478                                 }
1479                                 tp->t_maxseg = optval;
1480                         } else {
1481                                 error = EINVAL;
1482                         }
1483                         break;
1484
1485                 case TCP_KEEPINIT:
1486                         opthz = ((int64_t)optval * hz) / 1000;
1487                         if (opthz >= 1)
1488                                 tp->t_keepinit = opthz;
1489                         else
1490                                 error = EINVAL;
1491                         break;
1492
1493                 case TCP_KEEPIDLE:
1494                         opthz = ((int64_t)optval * hz) / 1000;
1495                         if (opthz >= 1) {
1496                                 tp->t_keepidle = opthz;
1497                                 tcp_timer_keep_activity(tp, 0);
1498                         } else {
1499                                 error = EINVAL;
1500                         }
1501                         break;
1502
1503                 case TCP_KEEPINTVL:
1504                         opthz = ((int64_t)optval * hz) / 1000;
1505                         if (opthz >= 1) {
1506                                 tp->t_keepintvl = opthz;
1507                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1508                         } else {
1509                                 error = EINVAL;
1510                         }
1511                         break;
1512
1513                 case TCP_KEEPCNT:
1514                         if (optval > 0) {
1515                                 tp->t_keepcnt = optval;
1516                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1517                         } else {
1518                                 error = EINVAL;
1519                         }
1520                         break;
1521
1522                 default:
1523                         error = ENOPROTOOPT;
1524                         break;
1525                 }
1526                 break;
1527
1528         case SOPT_GET:
1529                 switch (sopt->sopt_name) {
1530 #ifdef TCP_SIGNATURE
1531                 case TCP_SIGNATURE_ENABLE:
1532                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1533                         break;
1534 #endif /* TCP_SIGNATURE */
1535                 case TCP_NODELAY:
1536                         optval = tp->t_flags & TF_NODELAY;
1537                         break;
1538                 case TCP_MAXSEG:
1539                         optval = tp->t_maxseg;
1540                         break;
1541                 case TCP_NOOPT:
1542                         optval = tp->t_flags & TF_NOOPT;
1543                         break;
1544                 case TCP_NOPUSH:
1545                         optval = tp->t_flags & TF_NOPUSH;
1546                         break;
1547                 case TCP_KEEPINIT:
1548                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1549                         break;
1550                 case TCP_KEEPIDLE:
1551                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1552                         break;
1553                 case TCP_KEEPINTVL:
1554                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1555                         break;
1556                 case TCP_KEEPCNT:
1557                         optval = tp->t_keepcnt;
1558                         break;
1559                 default:
1560                         error = ENOPROTOOPT;
1561                         break;
1562                 }
1563                 if (error == 0)
1564                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1565                 break;
1566         }
1567 done:
1568         lwkt_replymsg(&msg->lmsg, error);
1569 }
1570
1571 /*
1572  * tcp_sendspace and tcp_recvspace are the default send and receive window
1573  * sizes, respectively.  These are obsolescent (this information should
1574  * be set by the route).
1575  *
1576  * Use a default that does not require tcp window scaling to be turned
1577  * on.  Individual programs or the administrator can increase the default.
1578  */
1579 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1580 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1581     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1582 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1583 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1584     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1585
1586 /*
1587  * Attach TCP protocol to socket, allocating internet protocol control
1588  * block, tcp control block, buffer space, and entering CLOSED state.
1589  */
1590 static int
1591 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1592 {
1593         struct tcpcb *tp;
1594         struct inpcb *inp;
1595         int error;
1596         int cpu;
1597 #ifdef INET6
1598         boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1599 #endif
1600
1601         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1602                 lwkt_gettoken(&so->so_rcv.ssb_token);
1603                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1604                                   ai->sb_rlimit);
1605                 lwkt_reltoken(&so->so_rcv.ssb_token);
1606                 if (error)
1607                         return (error);
1608         }
1609         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1610         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1611         cpu = mycpu->gd_cpuid;
1612
1613         /*
1614          * Set the default port for protocol processing. This will likely
1615          * change when we connect.
1616          */
1617         error = in_pcballoc(so, &tcbinfo[cpu]);
1618         if (error)
1619                 return (error);
1620         inp = so->so_pcb;
1621 #ifdef INET6
1622         if (isipv6)
1623                 inp->in6p_hops = -1;    /* use kernel default */
1624 #endif
1625         tp = tcp_newtcpcb(inp);
1626         if (tp == NULL) {
1627                 /*
1628                  * Make sure the socket is destroyed by the pcbdetach.
1629                  */
1630                 soreference(so);
1631 #ifdef INET6
1632                 if (isipv6)
1633                         in6_pcbdetach(inp);
1634                 else
1635 #endif
1636                 in_pcbdetach(inp);
1637                 sofree(so);     /* from ref above */
1638                 return (ENOBUFS);
1639         }
1640         tp->t_state = TCPS_CLOSED;
1641         /* Keep a reference for asynchronized pru_rcvd */
1642         soreference(so);
1643         return (0);
1644 }
1645
1646 /*
1647  * Initiate (or continue) disconnect.
1648  * If embryonic state, just send reset (once).
1649  * If in ``let data drain'' option and linger null, just drop.
1650  * Otherwise (hard), mark socket disconnecting and drop
1651  * current input data; switch states based on user close, and
1652  * send segment to peer (with FIN).
1653  */
1654 static struct tcpcb *
1655 tcp_disconnect(struct tcpcb *tp)
1656 {
1657         struct socket *so = tp->t_inpcb->inp_socket;
1658
1659         if (tp->t_state < TCPS_ESTABLISHED) {
1660                 tp = tcp_close(tp);
1661         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1662                 tp = tcp_drop(tp, 0);
1663         } else {
1664                 lwkt_gettoken(&so->so_rcv.ssb_token);
1665                 soisdisconnecting(so);
1666                 sbflush(&so->so_rcv.sb);
1667                 tp = tcp_usrclosed(tp);
1668                 if (tp)
1669                         tcp_output(tp);
1670                 lwkt_reltoken(&so->so_rcv.ssb_token);
1671         }
1672         return (tp);
1673 }
1674
1675 /*
1676  * User issued close, and wish to trail through shutdown states:
1677  * if never received SYN, just forget it.  If got a SYN from peer,
1678  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1679  * If already got a FIN from peer, then almost done; go to LAST_ACK
1680  * state.  In all other cases, have already sent FIN to peer (e.g.
1681  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1682  * for peer to send FIN or not respond to keep-alives, etc.
1683  * We can let the user exit from the close as soon as the FIN is acked.
1684  */
1685 static struct tcpcb *
1686 tcp_usrclosed(struct tcpcb *tp)
1687 {
1688
1689         switch (tp->t_state) {
1690
1691         case TCPS_CLOSED:
1692         case TCPS_LISTEN:
1693                 tp->t_state = TCPS_CLOSED;
1694                 tp = tcp_close(tp);
1695                 break;
1696
1697         case TCPS_SYN_SENT:
1698         case TCPS_SYN_RECEIVED:
1699                 tp->t_flags |= TF_NEEDFIN;
1700                 break;
1701
1702         case TCPS_ESTABLISHED:
1703                 tp->t_state = TCPS_FIN_WAIT_1;
1704                 break;
1705
1706         case TCPS_CLOSE_WAIT:
1707                 tp->t_state = TCPS_LAST_ACK;
1708                 break;
1709         }
1710         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1711                 soisdisconnected(tp->t_inpcb->inp_socket);
1712                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1713                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1714                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1715                             tcp_timer_2msl);
1716                 }
1717         }
1718         return (tp);
1719 }