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  * Allocate socket buffer space.
 176  */
 177 static int
 178 tcp_usr_preattach(struct socket *so, int proto __unused,
 179     struct pru_attach_info *ai)
 180 {
 181         int error;
 182
 183         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
 184                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
 185                                   ai->sb_rlimit);
 186                 if (error)
 187                         return (error);
 188         }
 189         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
 190         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
 191
 192         return 0;
 193 }
 194
 195 /*
 196  * TCP attaches to socket via pru_attach(), reserving space,
 197  * and an internet control block.  This socket may move to
 198  * other CPU later when we bind/connect.
 199  */
 200 static void
 201 tcp_usr_attach(netmsg_t msg)
 202 {
 203         struct socket *so = msg->base.nm_so;
 204         struct pru_attach_info *ai = msg->attach.nm_ai;
 205         int error;
 206         struct inpcb *inp;
 207         struct tcpcb *tp = NULL;
 208         TCPDEBUG0;
 209
 210         inp = so->so_pcb;
 211         KASSERT(inp == NULL, ("tcp socket attached"));
 212         TCPDEBUG1();
 213
 214         error = tcp_attach(so, ai);
 215         if (error)
 216                 goto out;
 217
 218         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
 219                 so->so_linger = TCP_LINGERTIME;
 220         tp = sototcpcb(so);
 221 out:
 222         TCPDEBUG2(PRU_ATTACH);
 223         lwkt_replymsg(&msg->lmsg, error);
 224 }
 225
 226 /*
 227  * pru_detach() detaches the TCP protocol from the socket.
 228  * If the protocol state is non-embryonic, then can't
 229  * do this directly: have to initiate a pru_disconnect(),
 230  * which may finish later; embryonic TCB's can just
 231  * be discarded here.
 232  */
 233 static void
 234 tcp_usr_detach(netmsg_t msg)
 235 {
 236         struct socket *so = msg->base.nm_so;
 237         int error = 0;
 238         struct inpcb *inp;
 239         struct tcpcb *tp;
 240         TCPDEBUG0;
 241
 242         inp = so->so_pcb;
 243
 244         /*
 245          * If the inp is already detached or never attached, it may have
 246          * been due to an async close or async attach failure.  Just return
 247          * as if no error occured.
 248          */
 249         if (inp) {
 250                 tp = intotcpcb(inp);
 251                 KASSERT(tp != NULL, ("tcp_usr_detach: tp is NULL"));
 252                 TCPDEBUG1();
 253                 tp = tcp_disconnect(tp);
 254                 TCPDEBUG2(PRU_DETACH);
 255         }
 256         lwkt_replymsg(&msg->lmsg, error);
 257 }
 258
 259 /*
 260  * NOTE: ignore_error is non-zero for certain disconnection races
 261  * which we want to silently allow, otherwise close() may return
 262  * an unexpected error.
 263  *
 264  * NOTE: The variables (msg) and (tp) are assumed.
 265  */
 266 #define COMMON_START(so, inp, ignore_error)                     \
 267         TCPDEBUG0;                                              \
 268                                                                 \
 269         inp = so->so_pcb;                                       \
 270         do {                                                    \
 271                 if (inp == NULL) {                              \
 272                         error = ignore_error ? 0 : EINVAL;      \
 273                         tp = NULL;                              \
 274                         goto out;                               \
 275                 }                                               \
 276                 tp = intotcpcb(inp);                            \
 277                 TCPDEBUG1();                                    \
 278         } while(0)
 279
 280 #define COMMON_END1(req, noreply)                               \
 281         out: do {                                               \
 282                 TCPDEBUG2(req);                                 \
 283                 if (!(noreply))                                 \
 284                         lwkt_replymsg(&msg->lmsg, error);       \
 285                 return;                                         \
 286         } while(0)
 287
 288 #define COMMON_END(req)         COMMON_END1((req), 0)
 289
 290 static void
 291 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
 292 {
 293         sosetport(((struct netmsg_base *)msg)->nm_so, port);
 294 }
 295
 296 /*
 297  * Give the socket an address.
 298  */
 299 static void
 300 tcp_usr_bind(netmsg_t msg)
 301 {
 302         struct socket *so = msg->bind.base.nm_so;
 303         struct sockaddr *nam = msg->bind.nm_nam;
 304         struct thread *td = msg->bind.nm_td;
 305         int error = 0;
 306         struct inpcb *inp;
 307         struct tcpcb *tp;
 308         struct sockaddr_in *sinp;
 309         lwkt_port_t port0 = netisr_cpuport(0);
 310
 311         COMMON_START(so, inp, 0);
 312
 313         /*
 314          * Must check for multicast addresses and disallow binding
 315          * to them.
 316          */
 317         sinp = (struct sockaddr_in *)nam;
 318         if (sinp->sin_family == AF_INET &&
 319             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 320                 error = EAFNOSUPPORT;
 321                 goto out;
 322         }
 323
 324         /*
 325          * Check "already bound" here (in_pcbbind() does the same check
 326          * though), so we don't forward a connected socket to netisr0,
 327          * which would panic in the following in_pcbunlink().
 328          */
 329         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) {
 330                 error = EINVAL; /* already bound */
 331                 goto out;
 332         }
 333
 334         /*
 335          * Use netisr0 to serialize in_pcbbind(), so that pru_detach and
 336          * pru_bind for different sockets on the same local port could be
 337          * properly ordered.  The original race is illustrated here for
 338          * reference.
 339          *
 340          * s1 = socket();
 341          * bind(s1, *.PORT);
 342          * close(s1);  <----- asynchronous
 343          * s2 = socket();
 344          * bind(s2, *.PORT);
 345          *
 346          * All will expect bind(s2, *.PORT) to succeed.  However, it will
 347          * fail, if following sequence happens due to random socket initial
 348          * msgport and asynchronous close(2):
 349          *
 350          *    netisrN                  netisrM
 351          *       :                        :
 352          *       :                    pru_bind(s2) [*.PORT is used by s1]
 353          *  pru_detach(s1)                :
 354          */
 355         if (&curthread->td_msgport != port0) {
 356                 lwkt_msg_t lmsg = &msg->bind.base.lmsg;
 357
 358                 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0,
 359                     ("already asked to relink"));
 360
 361                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 362                 msg->bind.nm_flags |= PRUB_RELINK;
 363
 364                 TCP_STATE_MIGRATE_START(tp);
 365
 366                 /* See the related comment in tcp_connect() */
 367                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 368                 lwkt_forwardmsg(port0, lmsg);
 369                 /* msg invalid now */
 370                 return;
 371         }
 372         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 373
 374         if (msg->bind.nm_flags & PRUB_RELINK) {
 375                 msg->bind.nm_flags &= ~PRUB_RELINK;
 376                 TCP_STATE_MIGRATE_END(tp);
 377                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 378         }
 379         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 380
 381         error = in_pcbbind(inp, nam, td);
 382         if (error)
 383                 goto out;
 384
 385         COMMON_END(PRU_BIND);
 386 }
 387
 388 #ifdef INET6
 389
 390 static void
 391 tcp6_usr_bind(netmsg_t msg)
 392 {
 393         struct socket *so = msg->bind.base.nm_so;
 394         struct sockaddr *nam = msg->bind.nm_nam;
 395         struct thread *td = msg->bind.nm_td;
 396         int error = 0;
 397         struct inpcb *inp;
 398         struct tcpcb *tp;
 399         struct sockaddr_in6 *sin6p;
 400
 401         COMMON_START(so, inp, 0);
 402
 403         /*
 404          * Must check for multicast addresses and disallow binding
 405          * to them.
 406          */
 407         sin6p = (struct sockaddr_in6 *)nam;
 408         if (sin6p->sin6_family == AF_INET6 &&
 409             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 410                 error = EAFNOSUPPORT;
 411                 goto out;
 412         }
 413         error = in6_pcbbind(inp, nam, td);
 414         if (error)
 415                 goto out;
 416         COMMON_END(PRU_BIND);
 417 }
 418 #endif /* INET6 */
 419
 420 struct netmsg_inswildcard {
 421         struct netmsg_base      base;
 422         struct inpcb            *nm_inp;
 423 };
 424
 425 static void
 426 in_pcbinswildcardhash_handler(netmsg_t msg)
 427 {
 428         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
 429         int cpu = mycpuid, nextcpu;
 430
 431         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
 432
 433         nextcpu = cpu + 1;
 434         if (nextcpu < ncpus2)
 435                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
 436         else
 437                 lwkt_replymsg(&nm->base.lmsg, 0);
 438 }
 439
 440 /*
 441  * Prepare to accept connections.
 442  */
 443 static void
 444 tcp_usr_listen(netmsg_t msg)
 445 {
 446         struct socket *so = msg->listen.base.nm_so;
 447         struct thread *td = msg->listen.nm_td;
 448         int error = 0;
 449         struct inpcb *inp;
 450         struct tcpcb *tp;
 451         struct netmsg_inswildcard nm;
 452         lwkt_port_t port0 = netisr_cpuport(0);
 453
 454         COMMON_START(so, inp, 0);
 455
 456         if (&curthread->td_msgport != port0) {
 457                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
 458
 459                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
 460                     ("already asked to relink"));
 461
 462                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
 463                 msg->listen.nm_flags |= PRUL_RELINK;
 464
 465                 TCP_STATE_MIGRATE_START(tp);
 466
 467                 /* See the related comment in tcp_connect() */
 468                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
 469                 lwkt_forwardmsg(port0, lmsg);
 470                 /* msg invalid now */
 471                 return;
 472         }
 473         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
 474
 475         if (msg->listen.nm_flags & PRUL_RELINK) {
 476                 msg->listen.nm_flags &= ~PRUL_RELINK;
 477                 TCP_STATE_MIGRATE_END(tp);
 478                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
 479         }
 480         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
 481
 482         if (tp->t_flags & TF_LISTEN)
 483                 goto out;
 484
 485         if (inp->inp_lport == 0) {
 486                 error = in_pcbbind(inp, NULL, td);
 487                 if (error)
 488                         goto out;
 489         }
 490
 491         TCP_STATE_CHANGE(tp, TCPS_LISTEN);
 492         tp->t_flags |= TF_LISTEN;
 493         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 494
 495         /*
 496          * Create tcpcb per-cpu port cache
 497          *
 498          * NOTE:
 499          * This _must_ be done before installing this inpcb into
 500          * wildcard hash.
 501          */
 502         tcp_pcbport_create(tp);
 503
 504         if (ncpus2 > 1) {
 505                 /*
 506                  * Put this inpcb into wildcard hash on other cpus.
 507                  */
 508                 ASSERT_INP_NOTINHASH(inp);
 509                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 510                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 511                 nm.nm_inp = inp;
 512                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 513         }
 514         in_pcbinswildcardhash(inp);
 515         COMMON_END(PRU_LISTEN);
 516 }
 517
 518 #ifdef INET6
 519
 520 static void
 521 tcp6_usr_listen(netmsg_t msg)
 522 {
 523         struct socket *so = msg->listen.base.nm_so;
 524         struct thread *td = msg->listen.nm_td;
 525         int error = 0;
 526         struct inpcb *inp;
 527         struct tcpcb *tp;
 528         struct netmsg_inswildcard nm;
 529
 530         COMMON_START(so, inp, 0);
 531
 532         if (tp->t_flags & TF_LISTEN)
 533                 goto out;
 534
 535         if (inp->inp_lport == 0) {
 536                 error = in6_pcbbind(inp, NULL, td);
 537                 if (error)
 538                         goto out;
 539         }
 540
 541         TCP_STATE_CHANGE(tp, TCPS_LISTEN);
 542         tp->t_flags |= TF_LISTEN;
 543         tp->tt_msg = NULL; /* Catch any invalid timer usage */
 544
 545         /*
 546          * Create tcpcb per-cpu port cache
 547          *
 548          * NOTE:
 549          * This _must_ be done before installing this inpcb into
 550          * wildcard hash.
 551          */
 552         tcp_pcbport_create(tp);
 553
 554         if (ncpus2 > 1) {
 555                 /*
 556                  * Put this inpcb into wildcard hash on other cpus.
 557                  */
 558                 KKASSERT(so->so_port == netisr_cpuport(0));
 559                 ASSERT_IN_NETISR(0);
 560                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
 561                 ASSERT_INP_NOTINHASH(inp);
 562
 563                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
 564                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
 565                 nm.nm_inp = inp;
 566                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
 567         }
 568         in_pcbinswildcardhash(inp);
 569         COMMON_END(PRU_LISTEN);
 570 }
 571 #endif /* INET6 */
 572
 573 /*
 574  * Initiate connection to peer.
 575  * Create a template for use in transmissions on this connection.
 576  * Enter SYN_SENT state, and mark socket as connecting.
 577  * Start keep-alive timer, and seed output sequence space.
 578  * Send initial segment on connection.
 579  */
 580 static void
 581 tcp_usr_connect(netmsg_t msg)
 582 {
 583         struct socket *so = msg->connect.base.nm_so;
 584         struct sockaddr *nam = msg->connect.nm_nam;
 585         struct thread *td = msg->connect.nm_td;
 586         int error = 0;
 587         struct inpcb *inp;
 588         struct tcpcb *tp;
 589         struct sockaddr_in *sinp;
 590
 591         COMMON_START(so, inp, 0);
 592
 593         /*
 594          * Must disallow TCP ``connections'' to multicast addresses.
 595          */
 596         sinp = (struct sockaddr_in *)nam;
 597         if (sinp->sin_family == AF_INET
 598             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
 599                 error = EAFNOSUPPORT;
 600                 goto out;
 601         }
 602
 603         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
 604                 error = EAFNOSUPPORT; /* IPv6 only jail */
 605                 goto out;
 606         }
 607
 608         tcp_connect(msg);
 609         /* msg is invalid now */
 610         return;
 611 out:
 612         if (msg->connect.nm_m) {
 613                 m_freem(msg->connect.nm_m);
 614                 msg->connect.nm_m = NULL;
 615         }
 616         if (msg->connect.nm_flags & PRUC_HELDTD)
 617                 lwkt_rele(td);
 618         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
 619                 so->so_error = error;
 620                 soisdisconnected(so);
 621         }
 622         lwkt_replymsg(&msg->lmsg, error);
 623 }
 624
 625 #ifdef INET6
 626
 627 static void
 628 tcp6_usr_connect(netmsg_t msg)
 629 {
 630         struct socket *so = msg->connect.base.nm_so;
 631         struct sockaddr *nam = msg->connect.nm_nam;
 632         struct thread *td = msg->connect.nm_td;
 633         int error = 0;
 634         struct inpcb *inp;
 635         struct tcpcb *tp;
 636         struct sockaddr_in6 *sin6p;
 637
 638         COMMON_START(so, inp, 0);
 639
 640         /*
 641          * Must disallow TCP ``connections'' to multicast addresses.
 642          */
 643         sin6p = (struct sockaddr_in6 *)nam;
 644         if (sin6p->sin6_family == AF_INET6
 645             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
 646                 error = EAFNOSUPPORT;
 647                 goto out;
 648         }
 649
 650         if (!prison_remote_ip(td, nam)) {
 651                 error = EAFNOSUPPORT; /* IPv4 only jail */
 652                 goto out;
 653         }
 654
 655         /* Reject v4-mapped address */
 656         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
 657                 error = EADDRNOTAVAIL;
 658                 goto out;
 659         }
 660
 661         inp->inp_inc.inc_isipv6 = 1;
 662         tcp6_connect(msg);
 663         /* msg is invalid now */
 664         return;
 665 out:
 666         if (msg->connect.nm_m) {
 667                 m_freem(msg->connect.nm_m);
 668                 msg->connect.nm_m = NULL;
 669         }
 670         lwkt_replymsg(&msg->lmsg, error);
 671 }
 672
 673 #endif /* INET6 */
 674
 675 /*
 676  * Initiate disconnect from peer.
 677  * If connection never passed embryonic stage, just drop;
 678  * else if don't need to let data drain, then can just drop anyways,
 679  * else have to begin TCP shutdown process: mark socket disconnecting,
 680  * drain unread data, state switch to reflect user close, and
 681  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
 682  * when peer sends FIN and acks ours.
 683  *
 684  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
 685  */
 686 static void
 687 tcp_usr_disconnect(netmsg_t msg)
 688 {
 689         struct socket *so = msg->disconnect.base.nm_so;
 690         int error = 0;
 691         struct inpcb *inp;
 692         struct tcpcb *tp;
 693
 694         COMMON_START(so, inp, 1);
 695         tp = tcp_disconnect(tp);
 696         COMMON_END(PRU_DISCONNECT);
 697 }
 698
 699 /*
 700  * Accept a connection.  Essentially all the work is
 701  * done at higher levels; just return the address
 702  * of the peer, storing through addr.
 703  */
 704 static void
 705 tcp_usr_accept(netmsg_t msg)
 706 {
 707         struct socket *so = msg->accept.base.nm_so;
 708         struct sockaddr **nam = msg->accept.nm_nam;
 709         int error = 0;
 710         struct inpcb *inp;
 711         struct tcpcb *tp = NULL;
 712         TCPDEBUG0;
 713
 714         inp = so->so_pcb;
 715         if (so->so_state & SS_ISDISCONNECTED) {
 716                 error = ECONNABORTED;
 717                 goto out;
 718         }
 719         if (inp == NULL) {
 720                 error = EINVAL;
 721                 goto out;
 722         }
 723
 724         tp = intotcpcb(inp);
 725         TCPDEBUG1();
 726         in_setpeeraddr(so, nam);
 727         COMMON_END(PRU_ACCEPT);
 728 }
 729
 730 #ifdef INET6
 731 static void
 732 tcp6_usr_accept(netmsg_t msg)
 733 {
 734         struct socket *so = msg->accept.base.nm_so;
 735         struct sockaddr **nam = msg->accept.nm_nam;
 736         int error = 0;
 737         struct inpcb *inp;
 738         struct tcpcb *tp = NULL;
 739         TCPDEBUG0;
 740
 741         inp = so->so_pcb;
 742
 743         if (so->so_state & SS_ISDISCONNECTED) {
 744                 error = ECONNABORTED;
 745                 goto out;
 746         }
 747         if (inp == NULL) {
 748                 error = EINVAL;
 749                 goto out;
 750         }
 751         tp = intotcpcb(inp);
 752         TCPDEBUG1();
 753         in6_setpeeraddr(so, nam);
 754         COMMON_END(PRU_ACCEPT);
 755 }
 756 #endif /* INET6 */
 757
 758 /*
 759  * Mark the connection as being incapable of further output.
 760  */
 761 static void
 762 tcp_usr_shutdown(netmsg_t msg)
 763 {
 764         struct socket *so = msg->shutdown.base.nm_so;
 765         int error = 0;
 766         struct inpcb *inp;
 767         struct tcpcb *tp;
 768
 769         COMMON_START(so, inp, 0);
 770         socantsendmore(so);
 771         tp = tcp_usrclosed(tp);
 772         if (tp)
 773                 error = tcp_output(tp);
 774         COMMON_END(PRU_SHUTDOWN);
 775 }
 776
 777 /*
 778  * After a receive, possibly send window update to peer.
 779  */
 780 static void
 781 tcp_usr_rcvd(netmsg_t msg)
 782 {
 783         struct socket *so = msg->rcvd.base.nm_so;
 784         int error = 0, noreply = 0;
 785         struct inpcb *inp;
 786         struct tcpcb *tp;
 787
 788         COMMON_START(so, inp, 0);
 789
 790         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
 791                 noreply = 1;
 792                 so_async_rcvd_reply(so);
 793         }
 794         tcp_output(tp);
 795
 796         COMMON_END1(PRU_RCVD, noreply);
 797 }
 798
 799 /*
 800  * Do a send by putting data in output queue and updating urgent
 801  * marker if URG set.  Possibly send more data.  Unlike the other
 802  * pru_*() routines, the mbuf chains are our responsibility.  We
 803  * must either enqueue them or free them.  The other pru_* routines
 804  * generally are caller-frees.
 805  */
 806 static void
 807 tcp_usr_send(netmsg_t msg)
 808 {
 809         struct socket *so = msg->send.base.nm_so;
 810         int flags = msg->send.nm_flags;
 811         struct mbuf *m = msg->send.nm_m;
 812         int error = 0;
 813         struct inpcb *inp;
 814         struct tcpcb *tp;
 815         TCPDEBUG0;
 816
 817         KKASSERT(msg->send.nm_control == NULL);
 818         KKASSERT(msg->send.nm_addr == NULL);
 819         KKASSERT((flags & PRUS_FREEADDR) == 0);
 820
 821         inp = so->so_pcb;
 822
 823         if (inp == NULL) {
 824                 /*
 825                  * OOPS! we lost a race, the TCP session got reset after
 826                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
 827                  * network interrupt in the non-critical section of sosend().
 828                  */
 829                 m_freem(m);
 830                 error = ECONNRESET;     /* XXX EPIPE? */
 831                 tp = NULL;
 832                 TCPDEBUG1();
 833                 goto out;
 834         }
 835         tp = intotcpcb(inp);
 836         TCPDEBUG1();
 837
 838 #ifdef foo
 839         /*
 840          * This is no longer necessary, since:
 841          * - sosendtcp() has already checked it for us
 842          * - It does not work with asynchronized send
 843          */
 844
 845         /*
 846          * Don't let too much OOB data build up
 847          */
 848         if (flags & PRUS_OOB) {
 849                 if (ssb_space(&so->so_snd) < -512) {
 850                         m_freem(m);
 851                         error = ENOBUFS;
 852                         goto out;
 853                 }
 854         }
 855 #endif
 856
 857         /*
 858          * Pump the data into the socket.
 859          */
 860         if (m) {
 861                 ssb_appendstream(&so->so_snd, m);
 862                 sowwakeup(so);
 863         }
 864         if (flags & PRUS_OOB) {
 865                 /*
 866                  * According to RFC961 (Assigned Protocols),
 867                  * the urgent pointer points to the last octet
 868                  * of urgent data.  We continue, however,
 869                  * to consider it to indicate the first octet
 870                  * of data past the urgent section.
 871                  * Otherwise, snd_up should be one lower.
 872                  */
 873                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
 874                 tp->t_flags |= TF_FORCE;
 875                 error = tcp_output(tp);
 876                 tp->t_flags &= ~TF_FORCE;
 877         } else {
 878                 if (flags & PRUS_EOF) {
 879                         /*
 880                          * Close the send side of the connection after
 881                          * the data is sent.
 882                          */
 883                         socantsendmore(so);
 884                         tp = tcp_usrclosed(tp);
 885                 }
 886                 if (tp != NULL && !tcp_output_pending(tp)) {
 887                         if (flags & PRUS_MORETOCOME)
 888                                 tp->t_flags |= TF_MORETOCOME;
 889                         error = tcp_output_fair(tp);
 890                         if (flags & PRUS_MORETOCOME)
 891                                 tp->t_flags &= ~TF_MORETOCOME;
 892                 }
 893         }
 894         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
 895                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
 896                    (flags & PRUS_NOREPLY));
 897 }
 898
 899 /*
 900  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
 901  *       will sofree() it when we return.
 902  */
 903 static void
 904 tcp_usr_abort(netmsg_t msg)
 905 {
 906         struct socket *so = msg->abort.base.nm_so;
 907         int error = 0;
 908         struct inpcb *inp;
 909         struct tcpcb *tp;
 910
 911         COMMON_START(so, inp, 1);
 912         tp = tcp_drop(tp, ECONNABORTED);
 913         COMMON_END(PRU_ABORT);
 914 }
 915
 916 /*
 917  * Receive out-of-band data.
 918  */
 919 static void
 920 tcp_usr_rcvoob(netmsg_t msg)
 921 {
 922         struct socket *so = msg->rcvoob.base.nm_so;
 923         struct mbuf *m = msg->rcvoob.nm_m;
 924         int flags = msg->rcvoob.nm_flags;
 925         int error = 0;
 926         struct inpcb *inp;
 927         struct tcpcb *tp;
 928
 929         COMMON_START(so, inp, 0);
 930         if ((so->so_oobmark == 0 &&
 931              (so->so_state & SS_RCVATMARK) == 0) ||
 932             so->so_options & SO_OOBINLINE ||
 933             tp->t_oobflags & TCPOOB_HADDATA) {
 934                 error = EINVAL;
 935                 goto out;
 936         }
 937         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
 938                 error = EWOULDBLOCK;
 939                 goto out;
 940         }
 941         m->m_len = 1;
 942         *mtod(m, caddr_t) = tp->t_iobc;
 943         if ((flags & MSG_PEEK) == 0)
 944                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
 945         COMMON_END(PRU_RCVOOB);
 946 }
 947
 948 static void
 949 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 950 {
 951         in_savefaddr(so, faddr);
 952 }
 953
 954 #ifdef INET6
 955 static void
 956 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
 957 {
 958         in6_savefaddr(so, faddr);
 959 }
 960 #endif
 961
 962 static int
 963 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
 964     struct thread *td __unused)
 965 {
 966         const struct sockaddr_in *sinp;
 967
 968         sinp = (const struct sockaddr_in *)nam;
 969         if (sinp->sin_family == AF_INET &&
 970             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
 971                 return EAFNOSUPPORT;
 972
 973         soisconnecting(so);
 974         return 0;
 975 }
 976
 977 /* xxx - should be const */
 978 struct pr_usrreqs tcp_usrreqs = {
 979         .pru_abort = tcp_usr_abort,
 980         .pru_accept = tcp_usr_accept,
 981         .pru_attach = tcp_usr_attach,
 982         .pru_bind = tcp_usr_bind,
 983         .pru_connect = tcp_usr_connect,
 984         .pru_connect2 = pr_generic_notsupp,
 985         .pru_control = in_control_dispatch,
 986         .pru_detach = tcp_usr_detach,
 987         .pru_disconnect = tcp_usr_disconnect,
 988         .pru_listen = tcp_usr_listen,
 989         .pru_peeraddr = in_setpeeraddr_dispatch,
 990         .pru_rcvd = tcp_usr_rcvd,
 991         .pru_rcvoob = tcp_usr_rcvoob,
 992         .pru_send = tcp_usr_send,
 993         .pru_sense = pru_sense_null,
 994         .pru_shutdown = tcp_usr_shutdown,
 995         .pru_sockaddr = in_setsockaddr_dispatch,
 996         .pru_sosend = sosendtcp,
 997         .pru_soreceive = sorecvtcp,
 998         .pru_savefaddr = tcp_usr_savefaddr,
 999         .pru_preconnect = tcp_usr_preconnect,
1000         .pru_preattach = tcp_usr_preattach
1001 };
1002
1003 #ifdef INET6
1004 struct pr_usrreqs tcp6_usrreqs = {
1005         .pru_abort = tcp_usr_abort,
1006         .pru_accept = tcp6_usr_accept,
1007         .pru_attach = tcp_usr_attach,
1008         .pru_bind = tcp6_usr_bind,
1009         .pru_connect = tcp6_usr_connect,
1010         .pru_connect2 = pr_generic_notsupp,
1011         .pru_control = in6_control_dispatch,
1012         .pru_detach = tcp_usr_detach,
1013         .pru_disconnect = tcp_usr_disconnect,
1014         .pru_listen = tcp6_usr_listen,
1015         .pru_peeraddr = in6_setpeeraddr_dispatch,
1016         .pru_rcvd = tcp_usr_rcvd,
1017         .pru_rcvoob = tcp_usr_rcvoob,
1018         .pru_send = tcp_usr_send,
1019         .pru_sense = pru_sense_null,
1020         .pru_shutdown = tcp_usr_shutdown,
1021         .pru_sockaddr = in6_setsockaddr_dispatch,
1022         .pru_sosend = sosendtcp,
1023         .pru_soreceive = sorecvtcp,
1024         .pru_savefaddr = tcp6_usr_savefaddr
1025 };
1026 #endif /* INET6 */
1027
1028 static int
1029 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1030                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
1031 {
1032         struct inpcb *inp = tp->t_inpcb, *oinp;
1033         struct socket *so = inp->inp_socket;
1034         struct route *ro = &inp->inp_route;
1035
1036         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
1037             ("pcbinfo mismatch"));
1038
1039         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1040                                  sin->sin_addr, sin->sin_port,
1041                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
1042                                   inp->inp_laddr : if_sin->sin_addr),
1043                                 inp->inp_lport, 0, NULL);
1044         if (oinp != NULL) {
1045                 m_freem(m);
1046                 return (EADDRINUSE);
1047         }
1048         if (inp->inp_laddr.s_addr == INADDR_ANY)
1049                 inp->inp_laddr = if_sin->sin_addr;
1050         inp->inp_faddr = sin->sin_addr;
1051         inp->inp_fport = sin->sin_port;
1052         in_pcbinsconnhash(inp);
1053
1054         /*
1055          * We are now on the inpcb's owner CPU, if the cached route was
1056          * freed because the rtentry's owner CPU is not the current CPU
1057          * (e.g. in tcp_connect()), then we try to reallocate it here with
1058          * the hope that a rtentry may be cloned from a RTF_PRCLONING
1059          * rtentry.
1060          */
1061         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1062             ro->ro_rt == NULL) {
1063                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1064                 ro->ro_dst.sa_family = AF_INET;
1065                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1066                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1067                         sin->sin_addr;
1068                 rtalloc(ro);
1069         }
1070
1071         /*
1072          * Now that no more errors can occur, change the protocol processing
1073          * port to the current thread (which is the correct thread).
1074          *
1075          * Create TCP timer message now; we are on the tcpcb's owner
1076          * CPU/thread.
1077          */
1078         tcp_create_timermsg(tp, &curthread->td_msgport);
1079
1080         /*
1081          * Compute window scaling to request.  Use a larger scaling then
1082          * needed for the initial receive buffer in case the receive buffer
1083          * gets expanded.
1084          */
1085         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1086                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1087         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1088                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1089         ) {
1090                 tp->request_r_scale++;
1091         }
1092
1093         soisconnecting(so);
1094         tcpstat.tcps_connattempt++;
1095         TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1096         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1097         tp->iss = tcp_new_isn(tp);
1098         tcp_sendseqinit(tp);
1099         if (m) {
1100                 ssb_appendstream(&so->so_snd, m);
1101                 m = NULL;
1102                 if (flags & PRUS_OOB)
1103                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1104         }
1105
1106         /*
1107          * Close the send side of the connection after
1108          * the data is sent if flagged.
1109          */
1110         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1111                 socantsendmore(so);
1112                 tp = tcp_usrclosed(tp);
1113         }
1114         return (tcp_output(tp));
1115 }
1116
1117 /*
1118  * Common subroutine to open a TCP connection to remote host specified
1119  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1120  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1121  * a local host address (interface).
1122  * Initialize connection parameters and enter SYN-SENT state.
1123  */
1124 static void
1125 tcp_connect(netmsg_t msg)
1126 {
1127         struct socket *so = msg->connect.base.nm_so;
1128         struct sockaddr *nam = msg->connect.nm_nam;
1129         struct thread *td = msg->connect.nm_td;
1130         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1131         struct sockaddr_in *if_sin = NULL;
1132         struct inpcb *inp;
1133         struct tcpcb *tp;
1134         int error;
1135         lwkt_port_t port;
1136
1137         COMMON_START(so, inp, 0);
1138
1139         /*
1140          * Reconnect our pcb if we have to
1141          */
1142         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1143                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1144                 TCP_STATE_MIGRATE_END(tp);
1145                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1146         }
1147
1148         /*
1149          * Bind if we have to
1150          */
1151         if (inp->inp_lport == 0) {
1152                 if (tcp_lport_extension) {
1153                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1154
1155                         error = in_pcbladdr(inp, nam, &if_sin, td);
1156                         if (error)
1157                                 goto out;
1158                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1159
1160                         error = in_pcbbind_remote(inp, nam, td);
1161                         if (error)
1162                                 goto out;
1163
1164                         msg->connect.nm_flags |= PRUC_HASLADDR;
1165                 } else {
1166                         error = in_pcbbind(inp, NULL, td);
1167                         if (error)
1168                                 goto out;
1169                 }
1170         }
1171
1172         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1173                 /*
1174                  * Calculate the correct protocol processing thread.  The
1175                  * connect operation must run there.  Set the forwarding
1176                  * port before we forward the message or it will get bounced
1177                  * right back to us.
1178                  */
1179                 error = in_pcbladdr(inp, nam, &if_sin, td);
1180                 if (error)
1181                         goto out;
1182         }
1183         KKASSERT(inp->inp_socket == so);
1184
1185         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1186                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1187                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1188                             inp->inp_lport);
1189
1190         if (port != &curthread->td_msgport) {
1191                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1192
1193                 /*
1194                  * in_pcbladdr() may have allocated a route entry for us
1195                  * on the current CPU, but we need a route entry on the
1196                  * inpcb's owner CPU, so free it here.
1197                  */
1198                 in_pcbresetroute(inp);
1199
1200                 /*
1201                  * We are moving the protocol processing port the socket
1202                  * is on, we have to unlink here and re-link on the
1203                  * target cpu.
1204                  */
1205                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1206                 msg->connect.nm_flags |= PRUC_RECONNECT;
1207                 msg->connect.base.nm_dispatch = tcp_connect;
1208
1209                 TCP_STATE_MIGRATE_START(tp);
1210
1211                 /*
1212                  * Use message put done receipt to change this socket's
1213                  * so_port, i.e. _after_ this message was put onto the
1214                  * target netisr's msgport but _before_ the message could
1215                  * be pulled from the target netisr's msgport, so that:
1216                  * - The upper half (socket code) will not see the new
1217                  *   msgport before this message reaches the new msgport
1218                  *   and messages for this socket will be ordered.
1219                  * - This message will see the new msgport, when its
1220                  *   handler is called in the target netisr.
1221                  *
1222                  * NOTE:
1223                  * We MUST use messege put done receipt to change this
1224                  * socket's so_port:
1225                  * If we changed the so_port in this netisr after the
1226                  * lwkt_forwardmsg (so messages for this socket will be
1227                  * ordered) and changed the so_port in the target netisr
1228                  * at the very beginning of this message's handler, we
1229                  * would suffer so_port overwritten race, given this
1230                  * message might be forwarded again.
1231                  *
1232                  * NOTE:
1233                  * This mechanism depends on that the netisr's msgport
1234                  * is spin msgport (currently it is :).
1235                  *
1236                  * If the upper half saw the new msgport before this
1237                  * message reached the target netisr's msgport, the
1238                  * messages sent from the upper half could reach the new
1239                  * msgport before this message, thus there would be
1240                  * message reordering.  The worst case could be soclose()
1241                  * saw the new msgport and the detach message could reach
1242                  * the new msgport before this message, i.e. the inpcb
1243                  * could have been destroyed when this message was still
1244                  * pending on or on its way to the new msgport.  Other
1245                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1246                  * since we have unlinked this inpcb from the current
1247                  * pcbinfo first.
1248                  */
1249                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1250                 lwkt_forwardmsg(port, lmsg);
1251                 /* msg invalid now */
1252                 return;
1253         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1254                 /*
1255                  * The original thread is no longer needed; release it.
1256                  */
1257                 lwkt_rele(td);
1258                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1259         }
1260         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1261                                   msg->connect.nm_m, sin, if_sin);
1262         msg->connect.nm_m = NULL;
1263 out:
1264         if (msg->connect.nm_m) {
1265                 m_freem(msg->connect.nm_m);
1266                 msg->connect.nm_m = NULL;
1267         }
1268         if (msg->connect.nm_flags & PRUC_HELDTD)
1269                 lwkt_rele(td);
1270         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1271                 so->so_error = error;
1272                 soisdisconnected(so);
1273         }
1274         lwkt_replymsg(&msg->connect.base.lmsg, error);
1275         /* msg invalid now */
1276 }
1277
1278 #ifdef INET6
1279
1280 static void
1281 tcp6_connect(netmsg_t msg)
1282 {
1283         struct tcpcb *tp;
1284         struct socket *so = msg->connect.base.nm_so;
1285         struct sockaddr *nam = msg->connect.nm_nam;
1286         struct thread *td = msg->connect.nm_td;
1287         struct inpcb *inp;
1288         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1289         struct in6_addr *addr6;
1290         lwkt_port_t port;
1291         int error;
1292
1293         COMMON_START(so, inp, 0);
1294
1295         /*
1296          * Reconnect our pcb if we have to
1297          */
1298         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1299                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1300                 TCP_STATE_MIGRATE_END(tp);
1301                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1302         }
1303
1304         /*
1305          * Bind if we have to
1306          */
1307         if (inp->inp_lport == 0) {
1308                 error = in6_pcbbind(inp, NULL, td);
1309                 if (error)
1310                         goto out;
1311         }
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         error = in6_pcbladdr(inp, nam, &addr6, td);
1319         if (error)
1320                 goto out;
1321
1322         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1323
1324         if (port != &curthread->td_msgport) {
1325                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1326
1327                 /*
1328                  * in_pcbladdr() may have allocated a route entry for us
1329                  * on the current CPU, but we need a route entry on the
1330                  * inpcb's owner CPU, so free it here.
1331                  */
1332                 in_pcbresetroute(inp);
1333
1334                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1335                 msg->connect.nm_flags |= PRUC_RECONNECT;
1336                 msg->connect.base.nm_dispatch = tcp6_connect;
1337
1338                 TCP_STATE_MIGRATE_START(tp);
1339
1340                 /* See the related comment in tcp_connect() */
1341                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1342                 lwkt_forwardmsg(port, lmsg);
1343                 /* msg invalid now */
1344                 return;
1345         }
1346         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1347                                    &msg->connect.nm_m, sin6, addr6);
1348         /* nm_m may still be intact */
1349 out:
1350         if (msg->connect.nm_m) {
1351                 m_freem(msg->connect.nm_m);
1352                 msg->connect.nm_m = NULL;
1353         }
1354         lwkt_replymsg(&msg->connect.base.lmsg, error);
1355         /* msg invalid now */
1356 }
1357
1358 static int
1359 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1360                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1361 {
1362         struct mbuf *m = *mp;
1363         struct inpcb *inp = tp->t_inpcb;
1364         struct socket *so = inp->inp_socket;
1365         struct inpcb *oinp;
1366
1367         /*
1368          * Cannot simply call in_pcbconnect, because there might be an
1369          * earlier incarnation of this same connection still in
1370          * TIME_WAIT state, creating an ADDRINUSE error.
1371          */
1372         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1373                                   &sin6->sin6_addr, sin6->sin6_port,
1374                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1375                                       addr6 : &inp->in6p_laddr),
1376                                   inp->inp_lport,  0, NULL);
1377         if (oinp)
1378                 return (EADDRINUSE);
1379
1380         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1381                 inp->in6p_laddr = *addr6;
1382         inp->in6p_faddr = sin6->sin6_addr;
1383         inp->inp_fport = sin6->sin6_port;
1384         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1385                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1386         in_pcbinsconnhash(inp);
1387
1388         /*
1389          * Now that no more errors can occur, change the protocol processing
1390          * port to the current thread (which is the correct thread).
1391          *
1392          * Create TCP timer message now; we are on the tcpcb's owner
1393          * CPU/thread.
1394          */
1395         tcp_create_timermsg(tp, &curthread->td_msgport);
1396
1397         /* Compute window scaling to request.  */
1398         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1399                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1400         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1401             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1402                 tp->request_r_scale++;
1403         }
1404
1405         soisconnecting(so);
1406         tcpstat.tcps_connattempt++;
1407         TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1408         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1409         tp->iss = tcp_new_isn(tp);
1410         tcp_sendseqinit(tp);
1411         if (m) {
1412                 ssb_appendstream(&so->so_snd, m);
1413                 *mp = NULL;
1414                 if (flags & PRUS_OOB)
1415                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1416         }
1417
1418         /*
1419          * Close the send side of the connection after
1420          * the data is sent if flagged.
1421          */
1422         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1423                 socantsendmore(so);
1424                 tp = tcp_usrclosed(tp);
1425         }
1426         return (tcp_output(tp));
1427 }
1428
1429 #endif /* INET6 */
1430
1431 /*
1432  * The new sockopt interface makes it possible for us to block in the
1433  * copyin/out step (if we take a page fault).  Taking a page fault while
1434  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1435  * both now use TSM, there probably isn't any need for this function to
1436  * run in a critical section any more.  This needs more examination.)
1437  */
1438 void
1439 tcp_ctloutput(netmsg_t msg)
1440 {
1441         struct socket *so = msg->base.nm_so;
1442         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1443         int     error, opt, optval, opthz;
1444         struct  inpcb *inp;
1445         struct  tcpcb *tp;
1446
1447         error = 0;
1448         inp = so->so_pcb;
1449         if (inp == NULL) {
1450                 error = ECONNRESET;
1451                 goto done;
1452         }
1453         tp = intotcpcb(inp);
1454
1455         /* Get socket's owner cpuid hint */
1456         if (sopt->sopt_level == SOL_SOCKET &&
1457             sopt->sopt_dir == SOPT_GET &&
1458             sopt->sopt_name == SO_CPUHINT) {
1459                 if (tp->t_flags & TF_LISTEN) {
1460                         /*
1461                          * Listen sockets owner cpuid is always 0,
1462                          * which does not make sense if SO_REUSEPORT
1463                          * is not set.
1464                          */
1465                         if (so->so_options & SO_REUSEPORT)
1466                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1467                         else
1468                                 optval = -1; /* no hint */
1469                 } else {
1470                         optval = mycpuid;
1471                 }
1472                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1473                 goto done;
1474         }
1475
1476         if (sopt->sopt_level != IPPROTO_TCP) {
1477                 if (sopt->sopt_level == IPPROTO_IP) {
1478                         switch (sopt->sopt_name) {
1479                         case IP_MULTICAST_IF:
1480                         case IP_MULTICAST_VIF:
1481                         case IP_MULTICAST_TTL:
1482                         case IP_MULTICAST_LOOP:
1483                         case IP_ADD_MEMBERSHIP:
1484                         case IP_DROP_MEMBERSHIP:
1485                                 /*
1486                                  * Multicast does not make sense on
1487                                  * TCP sockets.
1488                                  */
1489                                 error = EOPNOTSUPP;
1490                                 goto done;
1491                         }
1492                 }
1493 #ifdef INET6
1494                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1495                         ip6_ctloutput_dispatch(msg);
1496                 else
1497 #endif /* INET6 */
1498                 ip_ctloutput(msg);
1499                 /* msg invalid now */
1500                 return;
1501         }
1502
1503         switch (sopt->sopt_dir) {
1504         case SOPT_SET:
1505                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1506                                       sizeof optval);
1507                 if (error)
1508                         break;
1509                 switch (sopt->sopt_name) {
1510                 case TCP_FASTKEEP:
1511                         if (optval > 0)
1512                                 tp->t_keepidle = tp->t_keepintvl;
1513                         else
1514                                 tp->t_keepidle = tcp_keepidle;
1515                         tcp_timer_keep_activity(tp, 0);
1516                         break;
1517 #ifdef TCP_SIGNATURE
1518                 case TCP_SIGNATURE_ENABLE:
1519                         if (tp->t_state == TCPS_CLOSED) {
1520                                 /*
1521                                  * This is the only safe state that this
1522                                  * option could be changed.  Some segments
1523                                  * could already have been sent in other
1524                                  * states.
1525                                  */
1526                                 if (optval > 0)
1527                                         tp->t_flags |= TF_SIGNATURE;
1528                                 else
1529                                         tp->t_flags &= ~TF_SIGNATURE;
1530                         } else {
1531                                 error = EOPNOTSUPP;
1532                         }
1533                         break;
1534 #endif /* TCP_SIGNATURE */
1535                 case TCP_NODELAY:
1536                 case TCP_NOOPT:
1537                         switch (sopt->sopt_name) {
1538                         case TCP_NODELAY:
1539                                 opt = TF_NODELAY;
1540                                 break;
1541                         case TCP_NOOPT:
1542                                 opt = TF_NOOPT;
1543                                 break;
1544                         default:
1545                                 opt = 0; /* dead code to fool gcc */
1546                                 break;
1547                         }
1548
1549                         if (optval)
1550                                 tp->t_flags |= opt;
1551                         else
1552                                 tp->t_flags &= ~opt;
1553                         break;
1554
1555                 case TCP_NOPUSH:
1556                         if (tcp_disable_nopush)
1557                                 break;
1558                         if (optval)
1559                                 tp->t_flags |= TF_NOPUSH;
1560                         else {
1561                                 tp->t_flags &= ~TF_NOPUSH;
1562                                 error = tcp_output(tp);
1563                         }
1564                         break;
1565
1566                 case TCP_MAXSEG:
1567                         /*
1568                          * Must be between 0 and maxseg.  If the requested
1569                          * maxseg is too small to satisfy the desired minmss,
1570                          * pump it up (silently so sysctl modifications of
1571                          * minmss do not create unexpected program failures).
1572                          * Handle degenerate cases.
1573                          */
1574                         if (optval > 0 && optval <= tp->t_maxseg) {
1575                                 if (optval + 40 < tcp_minmss) {
1576                                         optval = tcp_minmss - 40;
1577                                         if (optval < 0)
1578                                                 optval = 1;
1579                                 }
1580                                 tp->t_maxseg = optval;
1581                         } else {
1582                                 error = EINVAL;
1583                         }
1584                         break;
1585
1586                 case TCP_KEEPINIT:
1587                         opthz = ((int64_t)optval * hz) / 1000;
1588                         if (opthz >= 1)
1589                                 tp->t_keepinit = opthz;
1590                         else
1591                                 error = EINVAL;
1592                         break;
1593
1594                 case TCP_KEEPIDLE:
1595                         opthz = ((int64_t)optval * hz) / 1000;
1596                         if (opthz >= 1) {
1597                                 tp->t_keepidle = opthz;
1598                                 tcp_timer_keep_activity(tp, 0);
1599                         } else {
1600                                 error = EINVAL;
1601                         }
1602                         break;
1603
1604                 case TCP_KEEPINTVL:
1605                         opthz = ((int64_t)optval * hz) / 1000;
1606                         if (opthz >= 1) {
1607                                 tp->t_keepintvl = opthz;
1608                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1609                         } else {
1610                                 error = EINVAL;
1611                         }
1612                         break;
1613
1614                 case TCP_KEEPCNT:
1615                         if (optval > 0) {
1616                                 tp->t_keepcnt = optval;
1617                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1618                         } else {
1619                                 error = EINVAL;
1620                         }
1621                         break;
1622
1623                 default:
1624                         error = ENOPROTOOPT;
1625                         break;
1626                 }
1627                 break;
1628
1629         case SOPT_GET:
1630                 switch (sopt->sopt_name) {
1631 #ifdef TCP_SIGNATURE
1632                 case TCP_SIGNATURE_ENABLE:
1633                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1634                         break;
1635 #endif /* TCP_SIGNATURE */
1636                 case TCP_NODELAY:
1637                         optval = tp->t_flags & TF_NODELAY;
1638                         break;
1639                 case TCP_MAXSEG:
1640                         optval = tp->t_maxseg;
1641                         break;
1642                 case TCP_NOOPT:
1643                         optval = tp->t_flags & TF_NOOPT;
1644                         break;
1645                 case TCP_NOPUSH:
1646                         optval = tp->t_flags & TF_NOPUSH;
1647                         break;
1648                 case TCP_KEEPINIT:
1649                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1650                         break;
1651                 case TCP_KEEPIDLE:
1652                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1653                         break;
1654                 case TCP_KEEPINTVL:
1655                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1656                         break;
1657                 case TCP_KEEPCNT:
1658                         optval = tp->t_keepcnt;
1659                         break;
1660                 default:
1661                         error = ENOPROTOOPT;
1662                         break;
1663                 }
1664                 if (error == 0)
1665                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1666                 break;
1667         }
1668 done:
1669         lwkt_replymsg(&msg->lmsg, error);
1670 }
1671
1672 /*
1673  * tcp_sendspace and tcp_recvspace are the default send and receive window
1674  * sizes, respectively.  These are obsolescent (this information should
1675  * be set by the route).
1676  *
1677  * Use a default that does not require tcp window scaling to be turned
1678  * on.  Individual programs or the administrator can increase the default.
1679  */
1680 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1681 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1682     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1683 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1684 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1685     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1686
1687 /*
1688  * Attach TCP protocol to socket, allocating internet protocol control
1689  * block, tcp control block, buffer space, and entering CLOSED state.
1690  */
1691 static int
1692 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1693 {
1694         struct inpcb *inp;
1695         int error;
1696         int cpu;
1697 #ifdef INET6
1698         boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1699 #endif
1700
1701         if (ai != NULL) {
1702                 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1703                 if (error)
1704                         return (error);
1705         } else {
1706                 /* Post attach; do nothing */
1707         }
1708
1709         cpu = mycpu->gd_cpuid;
1710
1711         /*
1712          * Set the default pcbinfo.  This will likely change when we
1713          * bind/connect.
1714          */
1715         error = in_pcballoc(so, &tcbinfo[cpu]);
1716         if (error)
1717                 return (error);
1718         inp = so->so_pcb;
1719 #ifdef INET6
1720         if (isipv6)
1721                 inp->in6p_hops = -1;    /* use kernel default */
1722 #endif
1723         tcp_newtcpcb(inp);
1724         /* Keep a reference for asynchronized pru_rcvd */
1725         soreference(so);
1726         return (0);
1727 }
1728
1729 /*
1730  * Initiate (or continue) disconnect.
1731  * If embryonic state, just send reset (once).
1732  * If in ``let data drain'' option and linger null, just drop.
1733  * Otherwise (hard), mark socket disconnecting and drop
1734  * current input data; switch states based on user close, and
1735  * send segment to peer (with FIN).
1736  */
1737 static struct tcpcb *
1738 tcp_disconnect(struct tcpcb *tp)
1739 {
1740         struct socket *so = tp->t_inpcb->inp_socket;
1741
1742         if (tp->t_state < TCPS_ESTABLISHED) {
1743                 tp = tcp_close(tp);
1744         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1745                 tp = tcp_drop(tp, 0);
1746         } else {
1747                 lwkt_gettoken(&so->so_rcv.ssb_token);
1748                 soisdisconnecting(so);
1749                 sbflush(&so->so_rcv.sb);
1750                 tp = tcp_usrclosed(tp);
1751                 if (tp)
1752                         tcp_output(tp);
1753                 lwkt_reltoken(&so->so_rcv.ssb_token);
1754         }
1755         return (tp);
1756 }
1757
1758 /*
1759  * User issued close, and wish to trail through shutdown states:
1760  * if never received SYN, just forget it.  If got a SYN from peer,
1761  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1762  * If already got a FIN from peer, then almost done; go to LAST_ACK
1763  * state.  In all other cases, have already sent FIN to peer (e.g.
1764  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1765  * for peer to send FIN or not respond to keep-alives, etc.
1766  * We can let the user exit from the close as soon as the FIN is acked.
1767  */
1768 static struct tcpcb *
1769 tcp_usrclosed(struct tcpcb *tp)
1770 {
1771
1772         switch (tp->t_state) {
1773
1774         case TCPS_CLOSED:
1775         case TCPS_LISTEN:
1776                 TCP_STATE_CHANGE(tp, TCPS_CLOSED);
1777                 tp = tcp_close(tp);
1778                 break;
1779
1780         case TCPS_SYN_SENT:
1781         case TCPS_SYN_RECEIVED:
1782                 tp->t_flags |= TF_NEEDFIN;
1783                 break;
1784
1785         case TCPS_ESTABLISHED:
1786                 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1787                 break;
1788
1789         case TCPS_CLOSE_WAIT:
1790                 TCP_STATE_CHANGE(tp, TCPS_LAST_ACK);
1791                 break;
1792         }
1793         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1794                 soisdisconnected(tp->t_inpcb->inp_socket);
1795                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1796                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1797                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1798                             tcp_timer_2msl);
1799                 }
1800         }
1801         return (tp);
1802 }