1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 67 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $ 68 * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $ 69 */ 70 71 #include "opt_ipsec.h" 72 #include "opt_inet6.h" 73 #include "opt_tcpdebug.h" 74 75 #include <sys/param.h> 76 #include <sys/systm.h> 77 #include <sys/kernel.h> 78 #include <sys/malloc.h> 79 #include <sys/sysctl.h> 80 #include <sys/globaldata.h> 81 #include <sys/thread.h> 82 83 #include <sys/mbuf.h> 84 #ifdef INET6 85 #include <sys/domain.h> 86 #endif /* INET6 */ 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/protosw.h> 90 91 #include <sys/thread2.h> 92 #include <sys/msgport2.h> 93 94 #include <net/if.h> 95 #include <net/netisr.h> 96 #include <net/route.h> 97 98 #include <net/netmsg2.h> 99 100 #include <netinet/in.h> 101 #include <netinet/in_systm.h> 102 #ifdef INET6 103 #include <netinet/ip6.h> 104 #endif 105 #include <netinet/in_pcb.h> 106 #ifdef INET6 107 #include <netinet6/in6_pcb.h> 108 #endif 109 #include <netinet/in_var.h> 110 #include <netinet/ip_var.h> 111 #ifdef INET6 112 #include <netinet6/ip6_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 int tcp_connect (struct tcpcb *, struct sockaddr *, 136 struct thread *); 137 #ifdef INET6 138 static int tcp6_connect (struct tcpcb *, struct sockaddr *, 139 struct thread *); 140 #endif /* INET6 */ 141 static struct tcpcb * 142 tcp_disconnect (struct tcpcb *); 143 static struct tcpcb * 144 tcp_usrclosed (struct tcpcb *); 145 146 #ifdef TCPDEBUG 147 #define TCPDEBUG0 int ostate = 0 148 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 149 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 150 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 151 #else 152 #define TCPDEBUG0 153 #define TCPDEBUG1() 154 #define TCPDEBUG2(req) 155 #endif 156 157 /* 158 * TCP attaches to socket via pru_attach(), reserving space, 159 * and an internet control block. 160 */ 161 static int 162 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai) 163 { 164 int error; 165 struct inpcb *inp; 166 struct tcpcb *tp = 0; 167 TCPDEBUG0; 168 169 crit_enter(); 170 inp = so->so_pcb; 171 TCPDEBUG1(); 172 if (inp) { 173 error = EISCONN; 174 goto out; 175 } 176 177 error = tcp_attach(so, ai); 178 if (error) 179 goto out; 180 181 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 182 so->so_linger = TCP_LINGERTIME; 183 tp = sototcpcb(so); 184 out: 185 TCPDEBUG2(PRU_ATTACH); 186 crit_exit(); 187 return error; 188 } 189 190 /* 191 * pru_detach() detaches the TCP protocol from the socket. 192 * If the protocol state is non-embryonic, then can't 193 * do this directly: have to initiate a pru_disconnect(), 194 * which may finish later; embryonic TCB's can just 195 * be discarded here. 196 */ 197 static int 198 tcp_usr_detach(struct socket *so) 199 { 200 int error = 0; 201 struct inpcb *inp; 202 struct tcpcb *tp; 203 TCPDEBUG0; 204 205 crit_enter(); 206 inp = so->so_pcb; 207 208 /* 209 * If the inp is already detached it may have been due to an async 210 * close. Just return as if no error occured. 211 */ 212 if (inp == NULL) { 213 crit_exit(); 214 return 0; 215 } 216 217 /* 218 * It's possible for the tcpcb (tp) to disconnect from the inp due 219 * to tcp_drop()->tcp_close() being called. This may occur *after* 220 * the detach message has been queued so we may find a NULL tp here. 221 */ 222 if ((tp = intotcpcb(inp)) != NULL) { 223 TCPDEBUG1(); 224 tp = tcp_disconnect(tp); 225 TCPDEBUG2(PRU_DETACH); 226 } 227 crit_exit(); 228 return error; 229 } 230 231 /* 232 * Note: ignore_error is non-zero for certain disconnection races 233 * which we want to silently allow, otherwise close() may return 234 * an unexpected error. 235 */ 236 #define COMMON_START(so, inp, ignore_error) \ 237 TCPDEBUG0; \ 238 \ 239 crit_enter(); \ 240 inp = so->so_pcb; \ 241 do { \ 242 if (inp == NULL) { \ 243 crit_exit(); \ 244 return (ignore_error ? 0 : EINVAL); \ 245 } \ 246 tp = intotcpcb(inp); \ 247 TCPDEBUG1(); \ 248 } while(0) 249 250 #define COMMON_END(req) out: TCPDEBUG2(req); crit_exit(); return error; goto out 251 252 253 /* 254 * Give the socket an address. 255 */ 256 static int 257 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 258 { 259 int error = 0; 260 struct inpcb *inp; 261 struct tcpcb *tp; 262 struct sockaddr_in *sinp; 263 264 COMMON_START(so, inp, 0); 265 266 /* 267 * Must check for multicast addresses and disallow binding 268 * to them. 269 */ 270 sinp = (struct sockaddr_in *)nam; 271 if (sinp->sin_family == AF_INET && 272 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 273 error = EAFNOSUPPORT; 274 goto out; 275 } 276 error = in_pcbbind(inp, nam, td); 277 if (error) 278 goto out; 279 COMMON_END(PRU_BIND); 280 281 } 282 283 #ifdef INET6 284 static int 285 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 286 { 287 int error = 0; 288 struct inpcb *inp; 289 struct tcpcb *tp; 290 struct sockaddr_in6 *sin6p; 291 292 COMMON_START(so, inp, 0); 293 294 /* 295 * Must check for multicast addresses and disallow binding 296 * to them. 297 */ 298 sin6p = (struct sockaddr_in6 *)nam; 299 if (sin6p->sin6_family == AF_INET6 && 300 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 301 error = EAFNOSUPPORT; 302 goto out; 303 } 304 inp->inp_vflag &= ~INP_IPV4; 305 inp->inp_vflag |= INP_IPV6; 306 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 307 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 308 inp->inp_vflag |= INP_IPV4; 309 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 310 struct sockaddr_in sin; 311 312 in6_sin6_2_sin(&sin, sin6p); 313 inp->inp_vflag |= INP_IPV4; 314 inp->inp_vflag &= ~INP_IPV6; 315 error = in_pcbbind(inp, (struct sockaddr *)&sin, td); 316 goto out; 317 } 318 } 319 error = in6_pcbbind(inp, nam, td); 320 if (error) 321 goto out; 322 COMMON_END(PRU_BIND); 323 } 324 #endif /* INET6 */ 325 326 #ifdef SMP 327 struct netmsg_inswildcard { 328 struct netmsg nm_netmsg; 329 struct inpcb *nm_inp; 330 struct inpcbinfo *nm_pcbinfo; 331 }; 332 333 static void 334 in_pcbinswildcardhash_handler(struct netmsg *msg0) 335 { 336 struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0; 337 338 in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo); 339 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0); 340 } 341 #endif 342 343 /* 344 * Prepare to accept connections. 345 */ 346 static int 347 tcp_usr_listen(struct socket *so, struct thread *td) 348 { 349 int error = 0; 350 struct inpcb *inp; 351 struct tcpcb *tp; 352 #ifdef SMP 353 int cpu; 354 #endif 355 356 COMMON_START(so, inp, 0); 357 if (inp->inp_lport == 0) { 358 error = in_pcbbind(inp, NULL, td); 359 if (error != 0) 360 goto out; 361 } 362 363 tp->t_state = TCPS_LISTEN; 364 tp->tt_msg = NULL; /* Catch any invalid timer usage */ 365 #ifdef SMP 366 /* 367 * We have to set the flag because we can't have other cpus 368 * messing with our inp's flags. 369 */ 370 inp->inp_flags |= INP_WILDCARD_MP; 371 for (cpu = 0; cpu < ncpus2; cpu++) { 372 struct netmsg_inswildcard *msg; 373 374 if (cpu == mycpu->gd_cpuid) { 375 in_pcbinswildcardhash(inp); 376 continue; 377 } 378 379 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG, 380 M_INTWAIT); 381 netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0, 382 in_pcbinswildcardhash_handler); 383 msg->nm_inp = inp; 384 msg->nm_pcbinfo = &tcbinfo[cpu]; 385 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg); 386 } 387 #else 388 in_pcbinswildcardhash(inp); 389 #endif 390 COMMON_END(PRU_LISTEN); 391 } 392 393 #ifdef INET6 394 static int 395 tcp6_usr_listen(struct socket *so, struct thread *td) 396 { 397 int error = 0; 398 struct inpcb *inp; 399 struct tcpcb *tp; 400 #ifdef SMP 401 int cpu; 402 #endif 403 404 COMMON_START(so, inp, 0); 405 if (inp->inp_lport == 0) { 406 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY)) 407 inp->inp_vflag |= INP_IPV4; 408 else 409 inp->inp_vflag &= ~INP_IPV4; 410 error = in6_pcbbind(inp, (struct sockaddr *)0, td); 411 } 412 if (error == 0) 413 tp->t_state = TCPS_LISTEN; 414 #ifdef SMP 415 /* 416 * We have to set the flag because we can't have other cpus 417 * messing with our inp's flags. 418 */ 419 inp->inp_flags |= INP_WILDCARD_MP; 420 for (cpu = 0; cpu < ncpus2; cpu++) { 421 struct netmsg_inswildcard *msg; 422 423 if (cpu == mycpu->gd_cpuid) { 424 in_pcbinswildcardhash(inp); 425 continue; 426 } 427 428 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG, 429 M_INTWAIT); 430 netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0, 431 in_pcbinswildcardhash_handler); 432 msg->nm_inp = inp; 433 msg->nm_pcbinfo = &tcbinfo[cpu]; 434 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg); 435 } 436 #else 437 in_pcbinswildcardhash(inp); 438 #endif 439 COMMON_END(PRU_LISTEN); 440 } 441 #endif /* INET6 */ 442 443 #ifdef SMP 444 static void 445 tcp_output_dispatch(struct netmsg *nmsg) 446 { 447 struct lwkt_msg *msg = &nmsg->nm_lmsg; 448 struct tcpcb *tp = msg->u.ms_resultp; 449 int error; 450 451 error = tcp_output(tp); 452 lwkt_replymsg(msg, error); 453 } 454 #endif 455 456 /* 457 * Initiate connection to peer. 458 * Create a template for use in transmissions on this connection. 459 * Enter SYN_SENT state, and mark socket as connecting. 460 * Start keep-alive timer, and seed output sequence space. 461 * Send initial segment on connection. 462 */ 463 static int 464 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 465 { 466 int error = 0; 467 struct inpcb *inp; 468 struct tcpcb *tp; 469 struct sockaddr_in *sinp; 470 #ifdef SMP 471 lwkt_port_t port; 472 #endif 473 474 COMMON_START(so, inp, 0); 475 476 /* 477 * Must disallow TCP ``connections'' to multicast addresses. 478 */ 479 sinp = (struct sockaddr_in *)nam; 480 if (sinp->sin_family == AF_INET 481 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 482 error = EAFNOSUPPORT; 483 goto out; 484 } 485 486 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) { 487 error = EAFNOSUPPORT; /* IPv6 only jail */ 488 goto out; 489 } 490 491 if ((error = tcp_connect(tp, nam, td)) != 0) 492 goto out; 493 494 #ifdef SMP 495 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 496 inp->inp_laddr.s_addr, inp->inp_lport); 497 if (port != &curthread->td_msgport) { 498 struct netmsg nmsg; 499 struct lwkt_msg *msg; 500 501 netmsg_init(&nmsg, &curthread->td_msgport, 0, 502 tcp_output_dispatch); 503 msg = &nmsg.nm_lmsg; 504 msg->u.ms_resultp = tp; 505 506 error = lwkt_domsg(port, msg, 0); 507 } else 508 #endif 509 error = tcp_output(tp); 510 COMMON_END(PRU_CONNECT); 511 } 512 513 #ifdef INET6 514 static int 515 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 516 { 517 int error = 0; 518 struct inpcb *inp; 519 struct tcpcb *tp; 520 struct sockaddr_in6 *sin6p; 521 522 COMMON_START(so, inp, 0); 523 524 /* 525 * Must disallow TCP ``connections'' to multicast addresses. 526 */ 527 sin6p = (struct sockaddr_in6 *)nam; 528 if (sin6p->sin6_family == AF_INET6 529 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 530 error = EAFNOSUPPORT; 531 goto out; 532 } 533 534 if (!prison_remote_ip(td, nam)) { 535 error = EAFNOSUPPORT; /* IPv4 only jail */ 536 goto out; 537 } 538 539 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 540 struct sockaddr_in sin; 541 542 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 543 error = EINVAL; 544 goto out; 545 } 546 547 in6_sin6_2_sin(&sin, sin6p); 548 inp->inp_vflag |= INP_IPV4; 549 inp->inp_vflag &= ~INP_IPV6; 550 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 551 goto out; 552 error = tcp_output(tp); 553 goto out; 554 } 555 inp->inp_vflag &= ~INP_IPV4; 556 inp->inp_vflag |= INP_IPV6; 557 inp->inp_inc.inc_isipv6 = 1; 558 if ((error = tcp6_connect(tp, nam, td)) != 0) 559 goto out; 560 error = tcp_output(tp); 561 COMMON_END(PRU_CONNECT); 562 } 563 #endif /* INET6 */ 564 565 /* 566 * Initiate disconnect from peer. 567 * If connection never passed embryonic stage, just drop; 568 * else if don't need to let data drain, then can just drop anyways, 569 * else have to begin TCP shutdown process: mark socket disconnecting, 570 * drain unread data, state switch to reflect user close, and 571 * send segment (e.g. FIN) to peer. Socket will be really disconnected 572 * when peer sends FIN and acks ours. 573 * 574 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 575 */ 576 static int 577 tcp_usr_disconnect(struct socket *so) 578 { 579 int error = 0; 580 struct inpcb *inp; 581 struct tcpcb *tp; 582 583 COMMON_START(so, inp, 1); 584 tp = tcp_disconnect(tp); 585 COMMON_END(PRU_DISCONNECT); 586 } 587 588 /* 589 * Accept a connection. Essentially all the work is 590 * done at higher levels; just return the address 591 * of the peer, storing through addr. 592 */ 593 static int 594 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 595 { 596 int error = 0; 597 struct inpcb *inp; 598 struct tcpcb *tp = NULL; 599 TCPDEBUG0; 600 601 crit_enter(); 602 inp = so->so_pcb; 603 if (so->so_state & SS_ISDISCONNECTED) { 604 error = ECONNABORTED; 605 goto out; 606 } 607 if (inp == 0) { 608 crit_exit(); 609 return (EINVAL); 610 } 611 tp = intotcpcb(inp); 612 TCPDEBUG1(); 613 in_setpeeraddr(so, nam); 614 COMMON_END(PRU_ACCEPT); 615 } 616 617 #ifdef INET6 618 static int 619 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 620 { 621 int error = 0; 622 struct inpcb *inp; 623 struct tcpcb *tp = NULL; 624 TCPDEBUG0; 625 626 crit_enter(); 627 inp = so->so_pcb; 628 629 if (so->so_state & SS_ISDISCONNECTED) { 630 error = ECONNABORTED; 631 goto out; 632 } 633 if (inp == 0) { 634 crit_exit(); 635 return (EINVAL); 636 } 637 tp = intotcpcb(inp); 638 TCPDEBUG1(); 639 in6_mapped_peeraddr(so, nam); 640 COMMON_END(PRU_ACCEPT); 641 } 642 #endif /* INET6 */ 643 /* 644 * Mark the connection as being incapable of further output. 645 */ 646 static int 647 tcp_usr_shutdown(struct socket *so) 648 { 649 int error = 0; 650 struct inpcb *inp; 651 struct tcpcb *tp; 652 653 COMMON_START(so, inp, 0); 654 socantsendmore(so); 655 tp = tcp_usrclosed(tp); 656 if (tp) 657 error = tcp_output(tp); 658 COMMON_END(PRU_SHUTDOWN); 659 } 660 661 /* 662 * After a receive, possibly send window update to peer. 663 */ 664 static int 665 tcp_usr_rcvd(struct socket *so, int flags) 666 { 667 int error = 0; 668 struct inpcb *inp; 669 struct tcpcb *tp; 670 671 COMMON_START(so, inp, 0); 672 tcp_output(tp); 673 COMMON_END(PRU_RCVD); 674 } 675 676 /* 677 * Do a send by putting data in output queue and updating urgent 678 * marker if URG set. Possibly send more data. Unlike the other 679 * pru_*() routines, the mbuf chains are our responsibility. We 680 * must either enqueue them or free them. The other pru_* routines 681 * generally are caller-frees. 682 */ 683 static int 684 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 685 struct sockaddr *nam, struct mbuf *control, struct thread *td) 686 { 687 int error = 0; 688 struct inpcb *inp; 689 struct tcpcb *tp; 690 #ifdef INET6 691 int isipv6; 692 #endif 693 TCPDEBUG0; 694 695 crit_enter(); 696 inp = so->so_pcb; 697 698 if (inp == NULL) { 699 /* 700 * OOPS! we lost a race, the TCP session got reset after 701 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a 702 * network interrupt in the non-critical section of sosend(). 703 */ 704 if (m) 705 m_freem(m); 706 if (control) 707 m_freem(control); 708 error = ECONNRESET; /* XXX EPIPE? */ 709 tp = NULL; 710 TCPDEBUG1(); 711 goto out; 712 } 713 #ifdef INET6 714 isipv6 = nam && nam->sa_family == AF_INET6; 715 #endif /* INET6 */ 716 tp = intotcpcb(inp); 717 TCPDEBUG1(); 718 if (control) { 719 /* TCP doesn't do control messages (rights, creds, etc) */ 720 if (control->m_len) { 721 m_freem(control); 722 if (m) 723 m_freem(m); 724 error = EINVAL; 725 goto out; 726 } 727 m_freem(control); /* empty control, just free it */ 728 } 729 if(!(flags & PRUS_OOB)) { 730 ssb_appendstream(&so->so_snd, m); 731 if (nam && tp->t_state < TCPS_SYN_SENT) { 732 /* 733 * Do implied connect if not yet connected, 734 * initialize window to default value, and 735 * initialize maxseg/maxopd using peer's cached 736 * MSS. 737 */ 738 #ifdef INET6 739 if (isipv6) 740 error = tcp6_connect(tp, nam, td); 741 else 742 #endif /* INET6 */ 743 error = tcp_connect(tp, nam, td); 744 if (error) 745 goto out; 746 tp->snd_wnd = TTCP_CLIENT_SND_WND; 747 tcp_mss(tp, -1); 748 } 749 750 if (flags & PRUS_EOF) { 751 /* 752 * Close the send side of the connection after 753 * the data is sent. 754 */ 755 socantsendmore(so); 756 tp = tcp_usrclosed(tp); 757 } 758 if (tp != NULL) { 759 if (flags & PRUS_MORETOCOME) 760 tp->t_flags |= TF_MORETOCOME; 761 error = tcp_output(tp); 762 if (flags & PRUS_MORETOCOME) 763 tp->t_flags &= ~TF_MORETOCOME; 764 } 765 } else { 766 if (ssb_space(&so->so_snd) < -512) { 767 m_freem(m); 768 error = ENOBUFS; 769 goto out; 770 } 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 ssb_appendstream(&so->so_snd, m); 780 if (nam && tp->t_state < TCPS_SYN_SENT) { 781 /* 782 * Do implied connect if not yet connected, 783 * initialize window to default value, and 784 * initialize maxseg/maxopd using peer's cached 785 * MSS. 786 */ 787 #ifdef INET6 788 if (isipv6) 789 error = tcp6_connect(tp, nam, td); 790 else 791 #endif /* INET6 */ 792 error = tcp_connect(tp, nam, td); 793 if (error) 794 goto out; 795 tp->snd_wnd = TTCP_CLIENT_SND_WND; 796 tcp_mss(tp, -1); 797 } 798 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 799 tp->t_flags |= TF_FORCE; 800 error = tcp_output(tp); 801 tp->t_flags &= ~TF_FORCE; 802 } 803 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB : 804 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 805 } 806 807 /* 808 * Abort the TCP. 809 */ 810 static int 811 tcp_usr_abort(struct socket *so) 812 { 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 int 826 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 827 { 828 int error = 0; 829 struct inpcb *inp; 830 struct tcpcb *tp; 831 832 COMMON_START(so, inp, 0); 833 if ((so->so_oobmark == 0 && 834 (so->so_state & SS_RCVATMARK) == 0) || 835 so->so_options & SO_OOBINLINE || 836 tp->t_oobflags & TCPOOB_HADDATA) { 837 error = EINVAL; 838 goto out; 839 } 840 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 841 error = EWOULDBLOCK; 842 goto out; 843 } 844 m->m_len = 1; 845 *mtod(m, caddr_t) = tp->t_iobc; 846 if ((flags & MSG_PEEK) == 0) 847 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 848 COMMON_END(PRU_RCVOOB); 849 } 850 851 /* xxx - should be const */ 852 struct pr_usrreqs tcp_usrreqs = { 853 .pru_abort = tcp_usr_abort, 854 .pru_accept = tcp_usr_accept, 855 .pru_attach = tcp_usr_attach, 856 .pru_bind = tcp_usr_bind, 857 .pru_connect = tcp_usr_connect, 858 .pru_connect2 = pru_connect2_notsupp, 859 .pru_control = in_control, 860 .pru_detach = tcp_usr_detach, 861 .pru_disconnect = tcp_usr_disconnect, 862 .pru_listen = tcp_usr_listen, 863 .pru_peeraddr = in_setpeeraddr, 864 .pru_rcvd = tcp_usr_rcvd, 865 .pru_rcvoob = tcp_usr_rcvoob, 866 .pru_send = tcp_usr_send, 867 .pru_sense = pru_sense_null, 868 .pru_shutdown = tcp_usr_shutdown, 869 .pru_sockaddr = in_setsockaddr, 870 .pru_sosend = sosend, 871 .pru_soreceive = soreceive, 872 .pru_sopoll = sopoll 873 }; 874 875 #ifdef INET6 876 struct pr_usrreqs tcp6_usrreqs = { 877 .pru_abort = tcp_usr_abort, 878 .pru_accept = tcp6_usr_accept, 879 .pru_attach = tcp_usr_attach, 880 .pru_bind = tcp6_usr_bind, 881 .pru_connect = tcp6_usr_connect, 882 .pru_connect2 = pru_connect2_notsupp, 883 .pru_control = in6_control, 884 .pru_detach = tcp_usr_detach, 885 .pru_disconnect = tcp_usr_disconnect, 886 .pru_listen = tcp6_usr_listen, 887 .pru_peeraddr = in6_mapped_peeraddr, 888 .pru_rcvd = tcp_usr_rcvd, 889 .pru_rcvoob = tcp_usr_rcvoob, 890 .pru_send = tcp_usr_send, 891 .pru_sense = pru_sense_null, 892 .pru_shutdown = tcp_usr_shutdown, 893 .pru_sockaddr = in6_mapped_sockaddr, 894 .pru_sosend = sosend, 895 .pru_soreceive = soreceive, 896 .pru_sopoll = sopoll 897 }; 898 #endif /* INET6 */ 899 900 static int 901 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin, 902 struct sockaddr_in *if_sin) 903 { 904 struct inpcb *inp = tp->t_inpcb, *oinp; 905 struct socket *so = inp->inp_socket; 906 struct route *ro = &inp->inp_route; 907 struct tcpcb *otp; 908 struct rmxp_tao *taop; 909 struct rmxp_tao tao_noncached; 910 911 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid], 912 sin->sin_addr, sin->sin_port, 913 inp->inp_laddr.s_addr != INADDR_ANY ? 914 inp->inp_laddr : if_sin->sin_addr, 915 inp->inp_lport, 0, NULL); 916 if (oinp != NULL) { 917 if (oinp != inp && (otp = intotcpcb(oinp)) != NULL && 918 otp->t_state == TCPS_TIME_WAIT && 919 (ticks - otp->t_starttime) < tcp_msl && 920 (otp->t_flags & TF_RCVD_CC)) 921 tcp_close(otp); 922 else 923 return (EADDRINUSE); 924 } 925 if (inp->inp_laddr.s_addr == INADDR_ANY) 926 inp->inp_laddr = if_sin->sin_addr; 927 inp->inp_faddr = sin->sin_addr; 928 inp->inp_fport = sin->sin_port; 929 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid]; 930 in_pcbinsconnhash(inp); 931 932 /* 933 * We are now on the inpcb's owner CPU, if the cached route was 934 * freed because the rtentry's owner CPU is not the current CPU 935 * (e.g. in tcp_connect()), then we try to reallocate it here with 936 * the hope that a rtentry may be cloned from a RTF_PRCLONING 937 * rtentry. 938 */ 939 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 940 ro->ro_rt == NULL) { 941 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 942 ro->ro_dst.sa_family = AF_INET; 943 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 944 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = 945 sin->sin_addr; 946 rtalloc(ro); 947 } 948 949 tcp_create_timermsg(tp); 950 951 /* Compute window scaling to request. */ 952 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 953 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) 954 tp->request_r_scale++; 955 956 soisconnecting(so); 957 tcpstat.tcps_connattempt++; 958 tp->t_state = TCPS_SYN_SENT; 959 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 960 tp->iss = tcp_new_isn(tp); 961 tcp_sendseqinit(tp); 962 963 /* 964 * Generate a CC value for this connection and 965 * check whether CC or CCnew should be used. 966 */ 967 if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) { 968 taop = &tao_noncached; 969 bzero(taop, sizeof *taop); 970 } 971 972 tp->cc_send = CC_INC(tcp_ccgen); 973 if (taop->tao_ccsent != 0 && 974 CC_GEQ(tp->cc_send, taop->tao_ccsent)) { 975 taop->tao_ccsent = tp->cc_send; 976 } else { 977 taop->tao_ccsent = 0; 978 tp->t_flags |= TF_SENDCCNEW; 979 } 980 981 return (0); 982 } 983 984 #ifdef SMP 985 986 struct netmsg_tcp_connect { 987 struct netmsg nm_netmsg; 988 struct tcpcb *nm_tp; 989 struct sockaddr_in *nm_sin; 990 struct sockaddr_in *nm_ifsin; 991 }; 992 993 static void 994 tcp_connect_handler(netmsg_t netmsg) 995 { 996 struct netmsg_tcp_connect *msg = (void *)netmsg; 997 int error; 998 999 error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin); 1000 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error); 1001 } 1002 1003 #endif 1004 1005 /* 1006 * Common subroutine to open a TCP connection to remote host specified 1007 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1008 * port number if needed. Call in_pcbladdr to do the routing and to choose 1009 * a local host address (interface). If there is an existing incarnation 1010 * of the same connection in TIME-WAIT state and if the remote host was 1011 * sending CC options and if the connection duration was < MSL, then 1012 * truncate the previous TIME-WAIT state and proceed. 1013 * Initialize connection parameters and enter SYN-SENT state. 1014 */ 1015 static int 1016 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1017 { 1018 struct inpcb *inp = tp->t_inpcb; 1019 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1020 struct sockaddr_in *if_sin; 1021 int error; 1022 #ifdef SMP 1023 lwkt_port_t port; 1024 #endif 1025 1026 if (inp->inp_lport == 0) { 1027 error = in_pcbbind(inp, (struct sockaddr *)NULL, td); 1028 if (error) 1029 return (error); 1030 } 1031 1032 /* 1033 * Cannot simply call in_pcbconnect, because there might be an 1034 * earlier incarnation of this same connection still in 1035 * TIME_WAIT state, creating an ADDRINUSE error. 1036 */ 1037 error = in_pcbladdr(inp, nam, &if_sin, td); 1038 if (error) 1039 return (error); 1040 1041 #ifdef SMP 1042 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port, 1043 inp->inp_laddr.s_addr ? 1044 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, 1045 inp->inp_lport); 1046 1047 if (port != &curthread->td_msgport) { 1048 struct netmsg_tcp_connect msg; 1049 struct route *ro = &inp->inp_route; 1050 1051 /* 1052 * in_pcbladdr() may have allocated a route entry for us 1053 * on the current CPU, but we need a route entry on the 1054 * inpcb's owner CPU, so free it here. 1055 */ 1056 if (ro->ro_rt != NULL) 1057 RTFREE(ro->ro_rt); 1058 bzero(ro, sizeof(*ro)); 1059 1060 netmsg_init(&msg.nm_netmsg, &curthread->td_msgport, 0, 1061 tcp_connect_handler); 1062 msg.nm_tp = tp; 1063 msg.nm_sin = sin; 1064 msg.nm_ifsin = if_sin; 1065 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0); 1066 } else 1067 #endif 1068 error = tcp_connect_oncpu(tp, sin, if_sin); 1069 1070 return (error); 1071 } 1072 1073 #ifdef INET6 1074 static int 1075 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1076 { 1077 struct inpcb *inp = tp->t_inpcb, *oinp; 1078 struct socket *so = inp->inp_socket; 1079 struct tcpcb *otp; 1080 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1081 struct in6_addr *addr6; 1082 struct rmxp_tao *taop; 1083 struct rmxp_tao tao_noncached; 1084 int error; 1085 1086 if (inp->inp_lport == 0) { 1087 error = in6_pcbbind(inp, (struct sockaddr *)0, td); 1088 if (error) 1089 return error; 1090 } 1091 1092 /* 1093 * Cannot simply call in_pcbconnect, because there might be an 1094 * earlier incarnation of this same connection still in 1095 * TIME_WAIT state, creating an ADDRINUSE error. 1096 */ 1097 error = in6_pcbladdr(inp, nam, &addr6, td); 1098 if (error) 1099 return error; 1100 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo, 1101 &sin6->sin6_addr, sin6->sin6_port, 1102 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ? 1103 addr6 : &inp->in6p_laddr, 1104 inp->inp_lport, 0, NULL); 1105 if (oinp) { 1106 if (oinp != inp && (otp = intotcpcb(oinp)) != NULL && 1107 otp->t_state == TCPS_TIME_WAIT && 1108 (ticks - otp->t_starttime) < tcp_msl && 1109 (otp->t_flags & TF_RCVD_CC)) 1110 otp = tcp_close(otp); 1111 else 1112 return (EADDRINUSE); 1113 } 1114 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1115 inp->in6p_laddr = *addr6; 1116 inp->in6p_faddr = sin6->sin6_addr; 1117 inp->inp_fport = sin6->sin6_port; 1118 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0) 1119 inp->in6p_flowinfo = sin6->sin6_flowinfo; 1120 in_pcbinsconnhash(inp); 1121 1122 /* Compute window scaling to request. */ 1123 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1124 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) 1125 tp->request_r_scale++; 1126 1127 soisconnecting(so); 1128 tcpstat.tcps_connattempt++; 1129 tp->t_state = TCPS_SYN_SENT; 1130 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 1131 tp->iss = tcp_new_isn(tp); 1132 tcp_sendseqinit(tp); 1133 1134 /* 1135 * Generate a CC value for this connection and 1136 * check whether CC or CCnew should be used. 1137 */ 1138 if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) { 1139 taop = &tao_noncached; 1140 bzero(taop, sizeof *taop); 1141 } 1142 1143 tp->cc_send = CC_INC(tcp_ccgen); 1144 if (taop->tao_ccsent != 0 && 1145 CC_GEQ(tp->cc_send, taop->tao_ccsent)) { 1146 taop->tao_ccsent = tp->cc_send; 1147 } else { 1148 taop->tao_ccsent = 0; 1149 tp->t_flags |= TF_SENDCCNEW; 1150 } 1151 1152 return (0); 1153 } 1154 #endif /* INET6 */ 1155 1156 /* 1157 * The new sockopt interface makes it possible for us to block in the 1158 * copyin/out step (if we take a page fault). Taking a page fault while 1159 * in a critical section is probably a Bad Thing. (Since sockets and pcbs 1160 * both now use TSM, there probably isn't any need for this function to 1161 * run in a critical section any more. This needs more examination.) 1162 */ 1163 int 1164 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1165 { 1166 int error, opt, optval; 1167 struct inpcb *inp; 1168 struct tcpcb *tp; 1169 1170 error = 0; 1171 crit_enter(); /* XXX */ 1172 inp = so->so_pcb; 1173 if (inp == NULL) { 1174 crit_exit(); 1175 return (ECONNRESET); 1176 } 1177 if (sopt->sopt_level != IPPROTO_TCP) { 1178 #ifdef INET6 1179 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1180 error = ip6_ctloutput(so, sopt); 1181 else 1182 #endif /* INET6 */ 1183 error = ip_ctloutput(so, sopt); 1184 crit_exit(); 1185 return (error); 1186 } 1187 tp = intotcpcb(inp); 1188 1189 switch (sopt->sopt_dir) { 1190 case SOPT_SET: 1191 error = soopt_to_kbuf(sopt, &optval, sizeof optval, 1192 sizeof optval); 1193 if (error) 1194 break; 1195 switch (sopt->sopt_name) { 1196 case TCP_NODELAY: 1197 case TCP_NOOPT: 1198 switch (sopt->sopt_name) { 1199 case TCP_NODELAY: 1200 opt = TF_NODELAY; 1201 break; 1202 case TCP_NOOPT: 1203 opt = TF_NOOPT; 1204 break; 1205 default: 1206 opt = 0; /* dead code to fool gcc */ 1207 break; 1208 } 1209 1210 if (optval) 1211 tp->t_flags |= opt; 1212 else 1213 tp->t_flags &= ~opt; 1214 break; 1215 1216 case TCP_NOPUSH: 1217 if (optval) 1218 tp->t_flags |= TF_NOPUSH; 1219 else { 1220 tp->t_flags &= ~TF_NOPUSH; 1221 error = tcp_output(tp); 1222 } 1223 break; 1224 1225 case TCP_MAXSEG: 1226 if (optval > 0 && optval <= tp->t_maxseg) 1227 tp->t_maxseg = optval; 1228 else 1229 error = EINVAL; 1230 break; 1231 1232 default: 1233 error = ENOPROTOOPT; 1234 break; 1235 } 1236 break; 1237 1238 case SOPT_GET: 1239 switch (sopt->sopt_name) { 1240 case TCP_NODELAY: 1241 optval = tp->t_flags & TF_NODELAY; 1242 break; 1243 case TCP_MAXSEG: 1244 optval = tp->t_maxseg; 1245 break; 1246 case TCP_NOOPT: 1247 optval = tp->t_flags & TF_NOOPT; 1248 break; 1249 case TCP_NOPUSH: 1250 optval = tp->t_flags & TF_NOPUSH; 1251 break; 1252 default: 1253 error = ENOPROTOOPT; 1254 break; 1255 } 1256 if (error == 0) 1257 soopt_from_kbuf(sopt, &optval, sizeof optval); 1258 break; 1259 } 1260 crit_exit(); 1261 return (error); 1262 } 1263 1264 /* 1265 * tcp_sendspace and tcp_recvspace are the default send and receive window 1266 * sizes, respectively. These are obsolescent (this information should 1267 * be set by the route). 1268 * 1269 * Use a default that does not require tcp window scaling to be turned 1270 * on. Individual programs or the administrator can increase the default. 1271 */ 1272 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1273 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1274 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1275 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1276 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1277 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1278 1279 /* 1280 * Attach TCP protocol to socket, allocating 1281 * internet protocol control block, tcp control block, 1282 * bufer space, and entering LISTEN state if to accept connections. 1283 */ 1284 static int 1285 tcp_attach(struct socket *so, struct pru_attach_info *ai) 1286 { 1287 struct tcpcb *tp; 1288 struct inpcb *inp; 1289 int error; 1290 int cpu; 1291 #ifdef INET6 1292 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1293 #endif 1294 1295 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 1296 error = soreserve(so, tcp_sendspace, tcp_recvspace, 1297 ai->sb_rlimit); 1298 if (error) 1299 return (error); 1300 } 1301 cpu = mycpu->gd_cpuid; 1302 error = in_pcballoc(so, &tcbinfo[cpu]); 1303 if (error) 1304 return (error); 1305 inp = so->so_pcb; 1306 #ifdef INET6 1307 if (isipv6) { 1308 inp->inp_vflag |= INP_IPV6; 1309 inp->in6p_hops = -1; /* use kernel default */ 1310 } 1311 else 1312 #endif 1313 inp->inp_vflag |= INP_IPV4; 1314 tp = tcp_newtcpcb(inp); 1315 if (tp == 0) { 1316 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 1317 1318 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 1319 #ifdef INET6 1320 if (isipv6) 1321 in6_pcbdetach(inp); 1322 else 1323 #endif 1324 in_pcbdetach(inp); 1325 so->so_state |= nofd; 1326 return (ENOBUFS); 1327 } 1328 tp->t_state = TCPS_CLOSED; 1329 return (0); 1330 } 1331 1332 /* 1333 * Initiate (or continue) disconnect. 1334 * If embryonic state, just send reset (once). 1335 * If in ``let data drain'' option and linger null, just drop. 1336 * Otherwise (hard), mark socket disconnecting and drop 1337 * current input data; switch states based on user close, and 1338 * send segment to peer (with FIN). 1339 */ 1340 static struct tcpcb * 1341 tcp_disconnect(struct tcpcb *tp) 1342 { 1343 struct socket *so = tp->t_inpcb->inp_socket; 1344 1345 if (tp->t_state < TCPS_ESTABLISHED) 1346 tp = tcp_close(tp); 1347 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 1348 tp = tcp_drop(tp, 0); 1349 else { 1350 soisdisconnecting(so); 1351 sbflush(&so->so_rcv.sb); 1352 tp = tcp_usrclosed(tp); 1353 if (tp) 1354 tcp_output(tp); 1355 } 1356 return (tp); 1357 } 1358 1359 /* 1360 * User issued close, and wish to trail through shutdown states: 1361 * if never received SYN, just forget it. If got a SYN from peer, 1362 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1363 * If already got a FIN from peer, then almost done; go to LAST_ACK 1364 * state. In all other cases, have already sent FIN to peer (e.g. 1365 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1366 * for peer to send FIN or not respond to keep-alives, etc. 1367 * We can let the user exit from the close as soon as the FIN is acked. 1368 */ 1369 static struct tcpcb * 1370 tcp_usrclosed(struct tcpcb *tp) 1371 { 1372 1373 switch (tp->t_state) { 1374 1375 case TCPS_CLOSED: 1376 case TCPS_LISTEN: 1377 tp->t_state = TCPS_CLOSED; 1378 tp = tcp_close(tp); 1379 break; 1380 1381 case TCPS_SYN_SENT: 1382 case TCPS_SYN_RECEIVED: 1383 tp->t_flags |= TF_NEEDFIN; 1384 break; 1385 1386 case TCPS_ESTABLISHED: 1387 tp->t_state = TCPS_FIN_WAIT_1; 1388 break; 1389 1390 case TCPS_CLOSE_WAIT: 1391 tp->t_state = TCPS_LAST_ACK; 1392 break; 1393 } 1394 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1395 soisdisconnected(tp->t_inpcb->inp_socket); 1396 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1397 if (tp->t_state == TCPS_FIN_WAIT_2) { 1398 tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle, 1399 tcp_timer_2msl); 1400 } 1401 } 1402 return (tp); 1403 } 1404