tcp_usrreq.c revision 127526
1/* 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 34 * $FreeBSD: head/sys/netinet/tcp_usrreq.c 127526 2004-03-28 15:48:00Z pjd $ 35 */ 36 37#include "opt_ipsec.h" 38#include "opt_inet.h" 39#include "opt_inet6.h" 40#include "opt_tcpdebug.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/malloc.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/mbuf.h> 48#ifdef INET6 49#include <sys/domain.h> 50#endif /* INET6 */ 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53#include <sys/protosw.h> 54#include <sys/proc.h> 55#include <sys/jail.h> 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#ifdef INET6 63#include <netinet/ip6.h> 64#endif 65#include <netinet/in_pcb.h> 66#ifdef INET6 67#include <netinet6/in6_pcb.h> 68#endif 69#include <netinet/in_var.h> 70#include <netinet/ip_var.h> 71#ifdef INET6 72#include <netinet6/ip6_var.h> 73#endif 74#include <netinet/tcp.h> 75#include <netinet/tcp_fsm.h> 76#include <netinet/tcp_seq.h> 77#include <netinet/tcp_timer.h> 78#include <netinet/tcp_var.h> 79#include <netinet/tcpip.h> 80#ifdef TCPDEBUG 81#include <netinet/tcp_debug.h> 82#endif 83 84#ifdef IPSEC 85#include <netinet6/ipsec.h> 86#endif /*IPSEC*/ 87 88/* 89 * TCP protocol interface to socket abstraction. 90 */ 91extern char *tcpstates[]; /* XXX ??? */ 92 93static int tcp_attach(struct socket *); 94static int tcp_connect(struct tcpcb *, struct sockaddr *, 95 struct thread *td); 96#ifdef INET6 97static int tcp6_connect(struct tcpcb *, struct sockaddr *, 98 struct thread *td); 99#endif /* INET6 */ 100static struct tcpcb * 101 tcp_disconnect(struct tcpcb *); 102static struct tcpcb * 103 tcp_usrclosed(struct tcpcb *); 104 105#ifdef TCPDEBUG 106#define TCPDEBUG0 int ostate = 0 107#define TCPDEBUG1() ostate = tp ? tp->t_state : 0 108#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 109 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 110#else 111#define TCPDEBUG0 112#define TCPDEBUG1() 113#define TCPDEBUG2(req) 114#endif 115 116/* 117 * TCP attaches to socket via pru_attach(), reserving space, 118 * and an internet control block. 119 */ 120static int 121tcp_usr_attach(struct socket *so, int proto, struct thread *td) 122{ 123 int s = splnet(); 124 int error; 125 struct inpcb *inp; 126 struct tcpcb *tp = 0; 127 TCPDEBUG0; 128 129 INP_INFO_WLOCK(&tcbinfo); 130 TCPDEBUG1(); 131 inp = sotoinpcb(so); 132 if (inp) { 133 error = EISCONN; 134 goto out; 135 } 136 137 error = tcp_attach(so); 138 if (error) 139 goto out; 140 141 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 142 so->so_linger = TCP_LINGERTIME; 143 144 inp = sotoinpcb(so); 145 tp = intotcpcb(inp); 146out: 147 TCPDEBUG2(PRU_ATTACH); 148 INP_INFO_WUNLOCK(&tcbinfo); 149 splx(s); 150 return error; 151} 152 153/* 154 * pru_detach() detaches the TCP protocol from the socket. 155 * If the protocol state is non-embryonic, then can't 156 * do this directly: have to initiate a pru_disconnect(), 157 * which may finish later; embryonic TCB's can just 158 * be discarded here. 159 */ 160static int 161tcp_usr_detach(struct socket *so) 162{ 163 int s = splnet(); 164 int error = 0; 165 struct inpcb *inp; 166 struct tcpcb *tp; 167 TCPDEBUG0; 168 169 INP_INFO_WLOCK(&tcbinfo); 170 inp = sotoinpcb(so); 171 if (inp == 0) { 172 INP_INFO_WUNLOCK(&tcbinfo); 173 splx(s); 174 return EINVAL; /* XXX */ 175 } 176 INP_LOCK(inp); 177 tp = intotcpcb(inp); 178 TCPDEBUG1(); 179 tp = tcp_disconnect(tp); 180 181 TCPDEBUG2(PRU_DETACH); 182 if (tp) 183 INP_UNLOCK(inp); 184 INP_INFO_WUNLOCK(&tcbinfo); 185 splx(s); 186 return error; 187} 188 189#define INI_NOLOCK 0 190#define INI_READ 1 191#define INI_WRITE 2 192 193#define COMMON_START() \ 194 TCPDEBUG0; \ 195 do { \ 196 if (inirw == INI_READ) \ 197 INP_INFO_RLOCK(&tcbinfo); \ 198 else if (inirw == INI_WRITE) \ 199 INP_INFO_WLOCK(&tcbinfo); \ 200 inp = sotoinpcb(so); \ 201 if (inp == 0) { \ 202 if (inirw == INI_READ) \ 203 INP_INFO_RUNLOCK(&tcbinfo); \ 204 else if (inirw == INI_WRITE) \ 205 INP_INFO_WUNLOCK(&tcbinfo); \ 206 splx(s); \ 207 return EINVAL; \ 208 } \ 209 INP_LOCK(inp); \ 210 if (inirw == INI_READ) \ 211 INP_INFO_RUNLOCK(&tcbinfo); \ 212 tp = intotcpcb(inp); \ 213 TCPDEBUG1(); \ 214} while(0) 215 216#define COMMON_END(req) \ 217out: TCPDEBUG2(req); \ 218 do { \ 219 if (tp) \ 220 INP_UNLOCK(inp); \ 221 if (inirw == INI_WRITE) \ 222 INP_INFO_WUNLOCK(&tcbinfo); \ 223 splx(s); \ 224 return error; \ 225 goto out; \ 226} while(0) 227 228/* 229 * Give the socket an address. 230 */ 231static int 232tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 233{ 234 int s = splnet(); 235 int error = 0; 236 struct inpcb *inp; 237 struct tcpcb *tp; 238 struct sockaddr_in *sinp; 239 const int inirw = INI_WRITE; 240 241 COMMON_START(); 242 243 /* 244 * Must check for multicast addresses and disallow binding 245 * to them. 246 */ 247 sinp = (struct sockaddr_in *)nam; 248 if (nam->sa_len != sizeof (*sinp)) 249 return (EINVAL); 250 if (sinp->sin_family == AF_INET && 251 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 252 error = EAFNOSUPPORT; 253 goto out; 254 } 255 error = in_pcbbind(inp, nam, td->td_ucred); 256 if (error) 257 goto out; 258 COMMON_END(PRU_BIND); 259} 260 261#ifdef INET6 262static int 263tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 264{ 265 int s = splnet(); 266 int error = 0; 267 struct inpcb *inp; 268 struct tcpcb *tp; 269 struct sockaddr_in6 *sin6p; 270 const int inirw = INI_WRITE; 271 272 COMMON_START(); 273 274 /* 275 * Must check for multicast addresses and disallow binding 276 * to them. 277 */ 278 sin6p = (struct sockaddr_in6 *)nam; 279 if (nam->sa_len != sizeof (*sin6p)) 280 return (EINVAL); 281 if (sin6p->sin6_family == AF_INET6 && 282 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 283 error = EAFNOSUPPORT; 284 goto out; 285 } 286 inp->inp_vflag &= ~INP_IPV4; 287 inp->inp_vflag |= INP_IPV6; 288 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 289 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 290 inp->inp_vflag |= INP_IPV4; 291 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 292 struct sockaddr_in sin; 293 294 in6_sin6_2_sin(&sin, sin6p); 295 inp->inp_vflag |= INP_IPV4; 296 inp->inp_vflag &= ~INP_IPV6; 297 error = in_pcbbind(inp, (struct sockaddr *)&sin, 298 td->td_ucred); 299 goto out; 300 } 301 } 302 error = in6_pcbbind(inp, nam, td->td_ucred); 303 if (error) 304 goto out; 305 COMMON_END(PRU_BIND); 306} 307#endif /* INET6 */ 308 309/* 310 * Prepare to accept connections. 311 */ 312static int 313tcp_usr_listen(struct socket *so, struct thread *td) 314{ 315 int s = splnet(); 316 int error = 0; 317 struct inpcb *inp; 318 struct tcpcb *tp; 319 const int inirw = INI_WRITE; 320 321 COMMON_START(); 322 if (inp->inp_lport == 0) 323 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 324 if (error == 0) 325 tp->t_state = TCPS_LISTEN; 326 COMMON_END(PRU_LISTEN); 327} 328 329#ifdef INET6 330static int 331tcp6_usr_listen(struct socket *so, struct thread *td) 332{ 333 int s = splnet(); 334 int error = 0; 335 struct inpcb *inp; 336 struct tcpcb *tp; 337 const int inirw = INI_WRITE; 338 339 COMMON_START(); 340 if (inp->inp_lport == 0) { 341 inp->inp_vflag &= ~INP_IPV4; 342 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 343 inp->inp_vflag |= INP_IPV4; 344 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 345 } 346 if (error == 0) 347 tp->t_state = TCPS_LISTEN; 348 COMMON_END(PRU_LISTEN); 349} 350#endif /* INET6 */ 351 352/* 353 * Initiate connection to peer. 354 * Create a template for use in transmissions on this connection. 355 * Enter SYN_SENT state, and mark socket as connecting. 356 * Start keep-alive timer, and seed output sequence space. 357 * Send initial segment on connection. 358 */ 359static int 360tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 361{ 362 int s = splnet(); 363 int error = 0; 364 struct inpcb *inp; 365 struct tcpcb *tp; 366 struct sockaddr_in *sinp; 367 const int inirw = INI_WRITE; 368 369 COMMON_START(); 370 371 /* 372 * Must disallow TCP ``connections'' to multicast addresses. 373 */ 374 sinp = (struct sockaddr_in *)nam; 375 if (nam->sa_len != sizeof (*sinp)) 376 return (EINVAL); 377 if (sinp->sin_family == AF_INET 378 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 379 error = EAFNOSUPPORT; 380 goto out; 381 } 382 383 if (td && jailed(td->td_ucred)) 384 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 385 386 if ((error = tcp_connect(tp, nam, td)) != 0) 387 goto out; 388 error = tcp_output(tp); 389 COMMON_END(PRU_CONNECT); 390} 391 392#ifdef INET6 393static int 394tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 395{ 396 int s = splnet(); 397 int error = 0; 398 struct inpcb *inp; 399 struct tcpcb *tp; 400 struct sockaddr_in6 *sin6p; 401 const int inirw = INI_WRITE; 402 403 COMMON_START(); 404 405 /* 406 * Must disallow TCP ``connections'' to multicast addresses. 407 */ 408 sin6p = (struct sockaddr_in6 *)nam; 409 if (nam->sa_len != sizeof (*sin6p)) 410 return (EINVAL); 411 if (sin6p->sin6_family == AF_INET6 412 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 413 error = EAFNOSUPPORT; 414 goto out; 415 } 416 417 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 418 struct sockaddr_in sin; 419 420 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 421 error = EINVAL; 422 goto out; 423 } 424 425 in6_sin6_2_sin(&sin, sin6p); 426 inp->inp_vflag |= INP_IPV4; 427 inp->inp_vflag &= ~INP_IPV6; 428 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 429 goto out; 430 error = tcp_output(tp); 431 goto out; 432 } 433 inp->inp_vflag &= ~INP_IPV4; 434 inp->inp_vflag |= INP_IPV6; 435 inp->inp_inc.inc_isipv6 = 1; 436 if ((error = tcp6_connect(tp, nam, td)) != 0) 437 goto out; 438 error = tcp_output(tp); 439 COMMON_END(PRU_CONNECT); 440} 441#endif /* INET6 */ 442 443/* 444 * Initiate disconnect from peer. 445 * If connection never passed embryonic stage, just drop; 446 * else if don't need to let data drain, then can just drop anyways, 447 * else have to begin TCP shutdown process: mark socket disconnecting, 448 * drain unread data, state switch to reflect user close, and 449 * send segment (e.g. FIN) to peer. Socket will be really disconnected 450 * when peer sends FIN and acks ours. 451 * 452 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 453 */ 454static int 455tcp_usr_disconnect(struct socket *so) 456{ 457 int s = splnet(); 458 int error = 0; 459 struct inpcb *inp; 460 struct tcpcb *tp; 461 const int inirw = INI_WRITE; 462 463 COMMON_START(); 464 tp = tcp_disconnect(tp); 465 COMMON_END(PRU_DISCONNECT); 466} 467 468/* 469 * Accept a connection. Essentially all the work is 470 * done at higher levels; just return the address 471 * of the peer, storing through addr. 472 */ 473static int 474tcp_usr_accept(struct socket *so, struct sockaddr **nam) 475{ 476 int s; 477 int error = 0; 478 struct inpcb *inp = NULL; 479 struct tcpcb *tp = NULL; 480 struct in_addr addr; 481 in_port_t port = 0; 482 TCPDEBUG0; 483 484 if (so->so_state & SS_ISDISCONNECTED) { 485 error = ECONNABORTED; 486 goto out; 487 } 488 489 s = splnet(); 490 INP_INFO_RLOCK(&tcbinfo); 491 inp = sotoinpcb(so); 492 if (!inp) { 493 INP_INFO_RUNLOCK(&tcbinfo); 494 splx(s); 495 return (EINVAL); 496 } 497 INP_LOCK(inp); 498 INP_INFO_RUNLOCK(&tcbinfo); 499 tp = intotcpcb(inp); 500 TCPDEBUG1(); 501 502 /* 503 * We inline in_setpeeraddr and COMMON_END here, so that we can 504 * copy the data of interest and defer the malloc until after we 505 * release the lock. 506 */ 507 port = inp->inp_fport; 508 addr = inp->inp_faddr; 509 510out: TCPDEBUG2(PRU_ACCEPT); 511 if (tp) 512 INP_UNLOCK(inp); 513 splx(s); 514 if (error == 0) 515 *nam = in_sockaddr(port, &addr); 516 return error; 517} 518 519#ifdef INET6 520static int 521tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 522{ 523 int s; 524 struct inpcb *inp = NULL; 525 int error = 0; 526 struct tcpcb *tp = NULL; 527 struct in_addr addr; 528 struct in6_addr addr6; 529 in_port_t port = 0; 530 int v4 = 0; 531 TCPDEBUG0; 532 533 if (so->so_state & SS_ISDISCONNECTED) { 534 error = ECONNABORTED; 535 goto out; 536 } 537 538 s = splnet(); 539 INP_INFO_RLOCK(&tcbinfo); 540 inp = sotoinpcb(so); 541 if (inp == 0) { 542 INP_INFO_RUNLOCK(&tcbinfo); 543 splx(s); 544 return (EINVAL); 545 } 546 INP_LOCK(inp); 547 INP_INFO_RUNLOCK(&tcbinfo); 548 tp = intotcpcb(inp); 549 TCPDEBUG1(); 550 /* 551 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 552 * copy the data of interest and defer the malloc until after we 553 * release the lock. 554 */ 555 if (inp->inp_vflag & INP_IPV4) { 556 v4 = 1; 557 port = inp->inp_fport; 558 addr = inp->inp_faddr; 559 } else { 560 port = inp->inp_fport; 561 addr6 = inp->in6p_faddr; 562 } 563 564out: TCPDEBUG2(PRU_ACCEPT); 565 if (tp) 566 INP_UNLOCK(inp); 567 splx(s); 568 if (error == 0) { 569 if (v4) 570 *nam = in6_v4mapsin6_sockaddr(port, &addr); 571 else 572 *nam = in6_sockaddr(port, &addr6); 573 } 574 return error; 575} 576#endif /* INET6 */ 577 578/* 579 * This is the wrapper function for in_setsockaddr. We just pass down 580 * the pcbinfo for in_setsockaddr to lock. We don't want to do the locking 581 * here because in_setsockaddr will call malloc and can block. 582 */ 583static int 584tcp_sockaddr(struct socket *so, struct sockaddr **nam) 585{ 586 return (in_setsockaddr(so, nam, &tcbinfo)); 587} 588 589/* 590 * This is the wrapper function for in_setpeeraddr. We just pass down 591 * the pcbinfo for in_setpeeraddr to lock. 592 */ 593static int 594tcp_peeraddr(struct socket *so, struct sockaddr **nam) 595{ 596 return (in_setpeeraddr(so, nam, &tcbinfo)); 597} 598 599/* 600 * Mark the connection as being incapable of further output. 601 */ 602static int 603tcp_usr_shutdown(struct socket *so) 604{ 605 int s = splnet(); 606 int error = 0; 607 struct inpcb *inp; 608 struct tcpcb *tp; 609 const int inirw = INI_WRITE; 610 611 COMMON_START(); 612 socantsendmore(so); 613 tp = tcp_usrclosed(tp); 614 if (tp) 615 error = tcp_output(tp); 616 COMMON_END(PRU_SHUTDOWN); 617} 618 619/* 620 * After a receive, possibly send window update to peer. 621 */ 622static int 623tcp_usr_rcvd(struct socket *so, int flags) 624{ 625 int s = splnet(); 626 int error = 0; 627 struct inpcb *inp; 628 struct tcpcb *tp; 629 const int inirw = INI_READ; 630 631 COMMON_START(); 632 tcp_output(tp); 633 COMMON_END(PRU_RCVD); 634} 635 636/* 637 * Do a send by putting data in output queue and updating urgent 638 * marker if URG set. Possibly send more data. Unlike the other 639 * pru_*() routines, the mbuf chains are our responsibility. We 640 * must either enqueue them or free them. The other pru_* routines 641 * generally are caller-frees. 642 */ 643static int 644tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 645 struct sockaddr *nam, struct mbuf *control, struct thread *td) 646{ 647 int s = splnet(); 648 int error = 0; 649 struct inpcb *inp; 650 struct tcpcb *tp; 651 const int inirw = INI_WRITE; 652#ifdef INET6 653 int isipv6; 654#endif 655 TCPDEBUG0; 656 657 /* 658 * Need write lock here because this function might call 659 * tcp_connect or tcp_usrclosed. 660 * We really want to have to this function upgrade from read lock 661 * to write lock. XXX 662 */ 663 INP_INFO_WLOCK(&tcbinfo); 664 inp = sotoinpcb(so); 665 if (inp == NULL) { 666 /* 667 * OOPS! we lost a race, the TCP session got reset after 668 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a 669 * network interrupt in the non-splnet() section of sosend(). 670 */ 671 if (m) 672 m_freem(m); 673 if (control) 674 m_freem(control); 675 error = ECONNRESET; /* XXX EPIPE? */ 676 tp = NULL; 677 TCPDEBUG1(); 678 goto out; 679 } 680 INP_LOCK(inp); 681#ifdef INET6 682 isipv6 = nam && nam->sa_family == AF_INET6; 683#endif /* INET6 */ 684 tp = intotcpcb(inp); 685 TCPDEBUG1(); 686 if (control) { 687 /* TCP doesn't do control messages (rights, creds, etc) */ 688 if (control->m_len) { 689 m_freem(control); 690 if (m) 691 m_freem(m); 692 error = EINVAL; 693 goto out; 694 } 695 m_freem(control); /* empty control, just free it */ 696 } 697 if (!(flags & PRUS_OOB)) { 698 sbappendstream(&so->so_snd, m); 699 if (nam && tp->t_state < TCPS_SYN_SENT) { 700 /* 701 * Do implied connect if not yet connected, 702 * initialize window to default value, and 703 * initialize maxseg/maxopd using peer's cached 704 * MSS. 705 */ 706#ifdef INET6 707 if (isipv6) 708 error = tcp6_connect(tp, nam, td); 709 else 710#endif /* INET6 */ 711 error = tcp_connect(tp, nam, td); 712 if (error) 713 goto out; 714 tp->snd_wnd = TTCP_CLIENT_SND_WND; 715 tcp_mss(tp, -1); 716 } 717 718 if (flags & PRUS_EOF) { 719 /* 720 * Close the send side of the connection after 721 * the data is sent. 722 */ 723 socantsendmore(so); 724 tp = tcp_usrclosed(tp); 725 } 726 if (tp != NULL) { 727 if (flags & PRUS_MORETOCOME) 728 tp->t_flags |= TF_MORETOCOME; 729 error = tcp_output(tp); 730 if (flags & PRUS_MORETOCOME) 731 tp->t_flags &= ~TF_MORETOCOME; 732 } 733 } else { 734 if (sbspace(&so->so_snd) < -512) { 735 m_freem(m); 736 error = ENOBUFS; 737 goto out; 738 } 739 /* 740 * According to RFC961 (Assigned Protocols), 741 * the urgent pointer points to the last octet 742 * of urgent data. We continue, however, 743 * to consider it to indicate the first octet 744 * of data past the urgent section. 745 * Otherwise, snd_up should be one lower. 746 */ 747 sbappendstream(&so->so_snd, m); 748 if (nam && tp->t_state < TCPS_SYN_SENT) { 749 /* 750 * Do implied connect if not yet connected, 751 * initialize window to default value, and 752 * initialize maxseg/maxopd using peer's cached 753 * MSS. 754 */ 755#ifdef INET6 756 if (isipv6) 757 error = tcp6_connect(tp, nam, td); 758 else 759#endif /* INET6 */ 760 error = tcp_connect(tp, nam, td); 761 if (error) 762 goto out; 763 tp->snd_wnd = TTCP_CLIENT_SND_WND; 764 tcp_mss(tp, -1); 765 } 766 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 767 tp->t_force = 1; 768 error = tcp_output(tp); 769 tp->t_force = 0; 770 } 771 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB : 772 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 773} 774 775/* 776 * Abort the TCP. 777 */ 778static int 779tcp_usr_abort(struct socket *so) 780{ 781 int s = splnet(); 782 int error = 0; 783 struct inpcb *inp; 784 struct tcpcb *tp; 785 const int inirw = INI_WRITE; 786 787 COMMON_START(); 788 tp = tcp_drop(tp, ECONNABORTED); 789 COMMON_END(PRU_ABORT); 790} 791 792/* 793 * Receive out-of-band data. 794 */ 795static int 796tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 797{ 798 int s = splnet(); 799 int error = 0; 800 struct inpcb *inp; 801 struct tcpcb *tp; 802 const int inirw = INI_READ; 803 804 COMMON_START(); 805 if ((so->so_oobmark == 0 && 806 (so->so_state & SS_RCVATMARK) == 0) || 807 so->so_options & SO_OOBINLINE || 808 tp->t_oobflags & TCPOOB_HADDATA) { 809 error = EINVAL; 810 goto out; 811 } 812 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 813 error = EWOULDBLOCK; 814 goto out; 815 } 816 m->m_len = 1; 817 *mtod(m, caddr_t) = tp->t_iobc; 818 if ((flags & MSG_PEEK) == 0) 819 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 820 COMMON_END(PRU_RCVOOB); 821} 822 823/* xxx - should be const */ 824struct pr_usrreqs tcp_usrreqs = { 825 tcp_usr_abort, tcp_usr_accept, tcp_usr_attach, tcp_usr_bind, 826 tcp_usr_connect, pru_connect2_notsupp, in_control, tcp_usr_detach, 827 tcp_usr_disconnect, tcp_usr_listen, tcp_peeraddr, tcp_usr_rcvd, 828 tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown, 829 tcp_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel 830}; 831 832#ifdef INET6 833struct pr_usrreqs tcp6_usrreqs = { 834 tcp_usr_abort, tcp6_usr_accept, tcp_usr_attach, tcp6_usr_bind, 835 tcp6_usr_connect, pru_connect2_notsupp, in6_control, tcp_usr_detach, 836 tcp_usr_disconnect, tcp6_usr_listen, in6_mapped_peeraddr, tcp_usr_rcvd, 837 tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown, 838 in6_mapped_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel 839}; 840#endif /* INET6 */ 841 842/* 843 * Common subroutine to open a TCP connection to remote host specified 844 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 845 * port number if needed. Call in_pcbconnect_setup to do the routing and 846 * to choose a local host address (interface). If there is an existing 847 * incarnation of the same connection in TIME-WAIT state and if the remote 848 * host was sending CC options and if the connection duration was < MSL, then 849 * truncate the previous TIME-WAIT state and proceed. 850 * Initialize connection parameters and enter SYN-SENT state. 851 */ 852static int 853tcp_connect(tp, nam, td) 854 register struct tcpcb *tp; 855 struct sockaddr *nam; 856 struct thread *td; 857{ 858 struct inpcb *inp = tp->t_inpcb, *oinp; 859 struct socket *so = inp->inp_socket; 860 struct tcptw *otw; 861 struct rmxp_tao tao; 862 struct in_addr laddr; 863 u_short lport; 864 int error; 865 866 bzero(&tao, sizeof(tao)); 867 868 if (inp->inp_lport == 0) { 869 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 870 if (error) 871 return error; 872 } 873 874 /* 875 * Cannot simply call in_pcbconnect, because there might be an 876 * earlier incarnation of this same connection still in 877 * TIME_WAIT state, creating an ADDRINUSE error. 878 */ 879 laddr = inp->inp_laddr; 880 lport = inp->inp_lport; 881 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 882 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 883 if (error && oinp == NULL) 884 return error; 885 if (oinp) { 886 if (oinp != inp && 887 (oinp->inp_vflag & INP_TIMEWAIT) && 888 (ticks - (otw = intotw(oinp))->t_starttime) < tcp_msl && 889 otw->cc_recv != 0) { 890 inp->inp_faddr = oinp->inp_faddr; 891 inp->inp_fport = oinp->inp_fport; 892 (void) tcp_twclose(otw, 0); 893 } else 894 return EADDRINUSE; 895 } 896 inp->inp_laddr = laddr; 897 in_pcbrehash(inp); 898 899 /* Compute window scaling to request. */ 900 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 901 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 902 tp->request_r_scale++; 903 904 soisconnecting(so); 905 tcpstat.tcps_connattempt++; 906 tp->t_state = TCPS_SYN_SENT; 907 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 908 tp->iss = tcp_new_isn(tp); 909 tp->t_bw_rtseq = tp->iss; 910 tcp_sendseqinit(tp); 911 912 /* 913 * Generate a CC value for this connection and 914 * check whether CC or CCnew should be used. 915 */ 916 if (tcp_do_rfc1644) 917 tcp_hc_gettao(&inp->inp_inc, &tao); 918 919 tp->cc_send = CC_INC(tcp_ccgen); 920 if (tao.tao_ccsent != 0 && 921 CC_GEQ(tp->cc_send, tao.tao_ccsent)) { 922 tao.tao_ccsent = tp->cc_send; 923 } else { 924 tao.tao_ccsent = 0; 925 tp->t_flags |= TF_SENDCCNEW; 926 } 927 928 if (tcp_do_rfc1644) 929 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, 930 tao.tao_ccsent, 0); 931 932 return 0; 933} 934 935#ifdef INET6 936static int 937tcp6_connect(tp, nam, td) 938 register struct tcpcb *tp; 939 struct sockaddr *nam; 940 struct thread *td; 941{ 942 struct inpcb *inp = tp->t_inpcb, *oinp; 943 struct socket *so = inp->inp_socket; 944 struct tcptw *otw; 945 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 946 struct in6_addr *addr6; 947 struct rmxp_tao tao; 948 int error; 949 950 bzero(&tao, sizeof(tao)); 951 952 if (inp->inp_lport == 0) { 953 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 954 if (error) 955 return error; 956 } 957 958 /* 959 * Cannot simply call in_pcbconnect, because there might be an 960 * earlier incarnation of this same connection still in 961 * TIME_WAIT state, creating an ADDRINUSE error. 962 */ 963 error = in6_pcbladdr(inp, nam, &addr6); 964 if (error) 965 return error; 966 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 967 &sin6->sin6_addr, sin6->sin6_port, 968 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 969 ? addr6 970 : &inp->in6p_laddr, 971 inp->inp_lport, 0, NULL); 972 if (oinp) { 973 if (oinp != inp && 974 (oinp->inp_vflag & INP_TIMEWAIT) && 975 (ticks - (otw = intotw(oinp))->t_starttime) < tcp_msl && 976 otw->cc_recv != 0) { 977 inp->inp_faddr = oinp->inp_faddr; 978 inp->inp_fport = oinp->inp_fport; 979 (void) tcp_twclose(otw, 0); 980 } else 981 return EADDRINUSE; 982 } 983 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 984 inp->in6p_laddr = *addr6; 985 inp->in6p_faddr = sin6->sin6_addr; 986 inp->inp_fport = sin6->sin6_port; 987 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0) 988 inp->in6p_flowinfo = sin6->sin6_flowinfo; 989 in_pcbrehash(inp); 990 991 /* Compute window scaling to request. */ 992 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 993 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 994 tp->request_r_scale++; 995 996 soisconnecting(so); 997 tcpstat.tcps_connattempt++; 998 tp->t_state = TCPS_SYN_SENT; 999 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 1000 tp->iss = tcp_new_isn(tp); 1001 tp->t_bw_rtseq = tp->iss; 1002 tcp_sendseqinit(tp); 1003 1004 /* 1005 * Generate a CC value for this connection and 1006 * check whether CC or CCnew should be used. 1007 */ 1008 if (tcp_do_rfc1644) 1009 tcp_hc_gettao(&inp->inp_inc, &tao); 1010 1011 tp->cc_send = CC_INC(tcp_ccgen); 1012 if (tao.tao_ccsent != 0 && 1013 CC_GEQ(tp->cc_send, tao.tao_ccsent)) { 1014 tao.tao_ccsent = tp->cc_send; 1015 } else { 1016 tao.tao_ccsent = 0; 1017 tp->t_flags |= TF_SENDCCNEW; 1018 } 1019 if (tcp_do_rfc1644) 1020 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, 1021 tao.tao_ccsent, 0); 1022 1023 return 0; 1024} 1025#endif /* INET6 */ 1026 1027/* 1028 * The new sockopt interface makes it possible for us to block in the 1029 * copyin/out step (if we take a page fault). Taking a page fault at 1030 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1031 * use TSM, there probably isn't any need for this function to run at 1032 * splnet() any more. This needs more examination.) 1033 */ 1034int 1035tcp_ctloutput(so, sopt) 1036 struct socket *so; 1037 struct sockopt *sopt; 1038{ 1039 int error, opt, optval, s; 1040 struct inpcb *inp; 1041 struct tcpcb *tp; 1042 1043 error = 0; 1044 s = splnet(); /* XXX */ 1045 INP_INFO_RLOCK(&tcbinfo); 1046 inp = sotoinpcb(so); 1047 if (inp == NULL) { 1048 INP_INFO_RUNLOCK(&tcbinfo); 1049 splx(s); 1050 return (ECONNRESET); 1051 } 1052 INP_LOCK(inp); 1053 INP_INFO_RUNLOCK(&tcbinfo); 1054 if (sopt->sopt_level != IPPROTO_TCP) { 1055#ifdef INET6 1056 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1057 error = ip6_ctloutput(so, sopt); 1058 else 1059#endif /* INET6 */ 1060 error = ip_ctloutput(so, sopt); 1061 INP_UNLOCK(inp); 1062 splx(s); 1063 return (error); 1064 } 1065 tp = intotcpcb(inp); 1066 1067 switch (sopt->sopt_dir) { 1068 case SOPT_SET: 1069 switch (sopt->sopt_name) { 1070#ifdef TCP_SIGNATURE 1071 case TCP_MD5SIG: 1072 error = sooptcopyin(sopt, &optval, sizeof optval, 1073 sizeof optval); 1074 if (error) 1075 break; 1076 1077 if (optval > 0) 1078 tp->t_flags |= TF_SIGNATURE; 1079 else 1080 tp->t_flags &= ~TF_SIGNATURE; 1081 break; 1082#endif /* TCP_SIGNATURE */ 1083 case TCP_NODELAY: 1084 case TCP_NOOPT: 1085 error = sooptcopyin(sopt, &optval, sizeof optval, 1086 sizeof optval); 1087 if (error) 1088 break; 1089 1090 switch (sopt->sopt_name) { 1091 case TCP_NODELAY: 1092 opt = TF_NODELAY; 1093 break; 1094 case TCP_NOOPT: 1095 opt = TF_NOOPT; 1096 break; 1097 default: 1098 opt = 0; /* dead code to fool gcc */ 1099 break; 1100 } 1101 1102 if (optval) 1103 tp->t_flags |= opt; 1104 else 1105 tp->t_flags &= ~opt; 1106 break; 1107 1108 case TCP_NOPUSH: 1109 error = sooptcopyin(sopt, &optval, sizeof optval, 1110 sizeof optval); 1111 if (error) 1112 break; 1113 1114 if (optval) 1115 tp->t_flags |= TF_NOPUSH; 1116 else { 1117 tp->t_flags &= ~TF_NOPUSH; 1118 error = tcp_output(tp); 1119 } 1120 break; 1121 1122 case TCP_MAXSEG: 1123 error = sooptcopyin(sopt, &optval, sizeof optval, 1124 sizeof optval); 1125 if (error) 1126 break; 1127 1128 if (optval > 0 && optval <= tp->t_maxseg && 1129 optval + 40 >= tcp_minmss) 1130 tp->t_maxseg = optval; 1131 else 1132 error = EINVAL; 1133 break; 1134 1135 default: 1136 error = ENOPROTOOPT; 1137 break; 1138 } 1139 break; 1140 1141 case SOPT_GET: 1142 switch (sopt->sopt_name) { 1143#ifdef TCP_SIGNATURE 1144 case TCP_MD5SIG: 1145 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1146 break; 1147#endif 1148 case TCP_NODELAY: 1149 optval = tp->t_flags & TF_NODELAY; 1150 break; 1151 case TCP_MAXSEG: 1152 optval = tp->t_maxseg; 1153 break; 1154 case TCP_NOOPT: 1155 optval = tp->t_flags & TF_NOOPT; 1156 break; 1157 case TCP_NOPUSH: 1158 optval = tp->t_flags & TF_NOPUSH; 1159 break; 1160 default: 1161 error = ENOPROTOOPT; 1162 break; 1163 } 1164 if (error == 0) 1165 error = sooptcopyout(sopt, &optval, sizeof optval); 1166 break; 1167 } 1168 INP_UNLOCK(inp); 1169 splx(s); 1170 return (error); 1171} 1172 1173/* 1174 * tcp_sendspace and tcp_recvspace are the default send and receive window 1175 * sizes, respectively. These are obsolescent (this information should 1176 * be set by the route). 1177 */ 1178u_long tcp_sendspace = 1024*32; 1179SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1180 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1181u_long tcp_recvspace = 1024*64; 1182SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1183 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1184 1185/* 1186 * Attach TCP protocol to socket, allocating 1187 * internet protocol control block, tcp control block, 1188 * bufer space, and entering LISTEN state if to accept connections. 1189 */ 1190static int 1191tcp_attach(so) 1192 struct socket *so; 1193{ 1194 register struct tcpcb *tp; 1195 struct inpcb *inp; 1196 int error; 1197#ifdef INET6 1198 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1199#endif 1200 1201 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1202 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1203 if (error) 1204 return (error); 1205 } 1206 error = in_pcballoc(so, &tcbinfo, "tcpinp"); 1207 if (error) 1208 return (error); 1209 inp = sotoinpcb(so); 1210#ifdef INET6 1211 if (isipv6) { 1212 inp->inp_vflag |= INP_IPV6; 1213 inp->in6p_hops = -1; /* use kernel default */ 1214 } 1215 else 1216#endif 1217 inp->inp_vflag |= INP_IPV4; 1218 tp = tcp_newtcpcb(inp); 1219 if (tp == 0) { 1220 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 1221 1222 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 1223#ifdef INET6 1224 if (isipv6) 1225 in6_pcbdetach(inp); 1226 else 1227#endif 1228 in_pcbdetach(inp); 1229 so->so_state |= nofd; 1230 return (ENOBUFS); 1231 } 1232 tp->t_state = TCPS_CLOSED; 1233 return (0); 1234} 1235 1236/* 1237 * Initiate (or continue) disconnect. 1238 * If embryonic state, just send reset (once). 1239 * If in ``let data drain'' option and linger null, just drop. 1240 * Otherwise (hard), mark socket disconnecting and drop 1241 * current input data; switch states based on user close, and 1242 * send segment to peer (with FIN). 1243 */ 1244static struct tcpcb * 1245tcp_disconnect(tp) 1246 register struct tcpcb *tp; 1247{ 1248 struct socket *so = tp->t_inpcb->inp_socket; 1249 1250 if (tp->t_state < TCPS_ESTABLISHED) 1251 tp = tcp_close(tp); 1252 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 1253 tp = tcp_drop(tp, 0); 1254 else { 1255 soisdisconnecting(so); 1256 sbflush(&so->so_rcv); 1257 tp = tcp_usrclosed(tp); 1258 if (tp) 1259 (void) tcp_output(tp); 1260 } 1261 return (tp); 1262} 1263 1264/* 1265 * User issued close, and wish to trail through shutdown states: 1266 * if never received SYN, just forget it. If got a SYN from peer, 1267 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1268 * If already got a FIN from peer, then almost done; go to LAST_ACK 1269 * state. In all other cases, have already sent FIN to peer (e.g. 1270 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1271 * for peer to send FIN or not respond to keep-alives, etc. 1272 * We can let the user exit from the close as soon as the FIN is acked. 1273 */ 1274static struct tcpcb * 1275tcp_usrclosed(tp) 1276 register struct tcpcb *tp; 1277{ 1278 1279 switch (tp->t_state) { 1280 1281 case TCPS_CLOSED: 1282 case TCPS_LISTEN: 1283 tp->t_state = TCPS_CLOSED; 1284 tp = tcp_close(tp); 1285 break; 1286 1287 case TCPS_SYN_SENT: 1288 case TCPS_SYN_RECEIVED: 1289 tp->t_flags |= TF_NEEDFIN; 1290 break; 1291 1292 case TCPS_ESTABLISHED: 1293 tp->t_state = TCPS_FIN_WAIT_1; 1294 break; 1295 1296 case TCPS_CLOSE_WAIT: 1297 tp->t_state = TCPS_LAST_ACK; 1298 break; 1299 } 1300 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1301 soisdisconnected(tp->t_inpcb->inp_socket); 1302 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1303 if (tp->t_state == TCPS_FIN_WAIT_2) 1304 callout_reset(tp->tt_2msl, tcp_maxidle, 1305 tcp_timer_2msl, tp); 1306 } 1307 return (tp); 1308} 1309 1310