tcp_usrreq.c revision 157366
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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 30 * $FreeBSD: head/sys/netinet/tcp_usrreq.c 157366 2006-04-01 15:15:05Z rwatson $ 31 */ 32 33#include "opt_inet.h" 34#include "opt_inet6.h" 35#include "opt_tcpdebug.h" 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/malloc.h> 40#include <sys/kernel.h> 41#include <sys/sysctl.h> 42#include <sys/mbuf.h> 43#ifdef INET6 44#include <sys/domain.h> 45#endif /* INET6 */ 46#include <sys/socket.h> 47#include <sys/socketvar.h> 48#include <sys/protosw.h> 49#include <sys/proc.h> 50#include <sys/jail.h> 51 52#include <net/if.h> 53#include <net/route.h> 54 55#include <netinet/in.h> 56#include <netinet/in_systm.h> 57#ifdef INET6 58#include <netinet/ip6.h> 59#endif 60#include <netinet/in_pcb.h> 61#ifdef INET6 62#include <netinet6/in6_pcb.h> 63#endif 64#include <netinet/in_var.h> 65#include <netinet/ip_var.h> 66#ifdef INET6 67#include <netinet6/ip6_var.h> 68#include <netinet6/scope6_var.h> 69#endif 70#include <netinet/tcp.h> 71#include <netinet/tcp_fsm.h> 72#include <netinet/tcp_seq.h> 73#include <netinet/tcp_timer.h> 74#include <netinet/tcp_var.h> 75#include <netinet/tcpip.h> 76#ifdef TCPDEBUG 77#include <netinet/tcp_debug.h> 78#endif 79 80/* 81 * TCP protocol interface to socket abstraction. 82 */ 83extern char *tcpstates[]; /* XXX ??? */ 84 85static int tcp_attach(struct socket *); 86static int tcp_connect(struct tcpcb *, struct sockaddr *, 87 struct thread *td); 88#ifdef INET6 89static int tcp6_connect(struct tcpcb *, struct sockaddr *, 90 struct thread *td); 91#endif /* INET6 */ 92static struct tcpcb * 93 tcp_disconnect(struct tcpcb *); 94static struct tcpcb * 95 tcp_usrclosed(struct tcpcb *); 96static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 97 98#ifdef TCPDEBUG 99#define TCPDEBUG0 int ostate = 0 100#define TCPDEBUG1() ostate = tp ? tp->t_state : 0 101#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 102 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 103#else 104#define TCPDEBUG0 105#define TCPDEBUG1() 106#define TCPDEBUG2(req) 107#endif 108 109/* 110 * TCP attaches to socket via pru_attach(), reserving space, 111 * and an internet control block. 112 */ 113static int 114tcp_usr_attach(struct socket *so, int proto, struct thread *td) 115{ 116 int error; 117 struct inpcb *inp; 118 struct tcpcb *tp = 0; 119 TCPDEBUG0; 120 121 INP_INFO_WLOCK(&tcbinfo); 122 TCPDEBUG1(); 123 inp = sotoinpcb(so); 124 if (inp) { 125 error = EISCONN; 126 goto out; 127 } 128 129 error = tcp_attach(so); 130 if (error) 131 goto out; 132 133 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 134 so->so_linger = TCP_LINGERTIME; 135 136 inp = sotoinpcb(so); 137 tp = intotcpcb(inp); 138out: 139 TCPDEBUG2(PRU_ATTACH); 140 INP_INFO_WUNLOCK(&tcbinfo); 141 return error; 142} 143 144/* 145 * pru_detach() detaches the TCP protocol from the socket. 146 * If the protocol state is non-embryonic, then can't 147 * do this directly: have to initiate a pru_disconnect(), 148 * which may finish later; embryonic TCB's can just 149 * be discarded here. 150 */ 151static int 152tcp_usr_detach(struct socket *so) 153{ 154 int error = 0; 155 struct inpcb *inp; 156 struct tcpcb *tp; 157 TCPDEBUG0; 158 159 INP_INFO_WLOCK(&tcbinfo); 160 inp = sotoinpcb(so); 161 if (inp == NULL) { 162 INP_INFO_WUNLOCK(&tcbinfo); 163 return error; 164 } 165 INP_LOCK(inp); 166 tp = intotcpcb(inp); 167 TCPDEBUG1(); 168 tp = tcp_disconnect(tp); 169 170 TCPDEBUG2(PRU_DETACH); 171 if (tp) 172 INP_UNLOCK(inp); 173 INP_INFO_WUNLOCK(&tcbinfo); 174 return error; 175} 176 177#define INI_NOLOCK 0 178#define INI_READ 1 179#define INI_WRITE 2 180 181#define COMMON_START() \ 182 TCPDEBUG0; \ 183 do { \ 184 if (inirw == INI_READ) \ 185 INP_INFO_RLOCK(&tcbinfo); \ 186 else if (inirw == INI_WRITE) \ 187 INP_INFO_WLOCK(&tcbinfo); \ 188 inp = sotoinpcb(so); \ 189 if (inp == 0) { \ 190 if (inirw == INI_READ) \ 191 INP_INFO_RUNLOCK(&tcbinfo); \ 192 else if (inirw == INI_WRITE) \ 193 INP_INFO_WUNLOCK(&tcbinfo); \ 194 return EINVAL; \ 195 } \ 196 INP_LOCK(inp); \ 197 if (inirw == INI_READ) \ 198 INP_INFO_RUNLOCK(&tcbinfo); \ 199 tp = intotcpcb(inp); \ 200 TCPDEBUG1(); \ 201} while(0) 202 203#define COMMON_END(req) \ 204out: TCPDEBUG2(req); \ 205 do { \ 206 if (tp) \ 207 INP_UNLOCK(inp); \ 208 if (inirw == INI_WRITE) \ 209 INP_INFO_WUNLOCK(&tcbinfo); \ 210 return error; \ 211 goto out; \ 212} while(0) 213 214/* 215 * Give the socket an address. 216 */ 217static int 218tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 219{ 220 int error = 0; 221 struct inpcb *inp; 222 struct tcpcb *tp; 223 struct sockaddr_in *sinp; 224 const int inirw = INI_WRITE; 225 226 sinp = (struct sockaddr_in *)nam; 227 if (nam->sa_len != sizeof (*sinp)) 228 return (EINVAL); 229 /* 230 * Must check for multicast addresses and disallow binding 231 * to them. 232 */ 233 if (sinp->sin_family == AF_INET && 234 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 235 return (EAFNOSUPPORT); 236 237 COMMON_START(); 238 error = in_pcbbind(inp, nam, td->td_ucred); 239 if (error) 240 goto out; 241 COMMON_END(PRU_BIND); 242} 243 244#ifdef INET6 245static int 246tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 247{ 248 int error = 0; 249 struct inpcb *inp; 250 struct tcpcb *tp; 251 struct sockaddr_in6 *sin6p; 252 const int inirw = INI_WRITE; 253 254 sin6p = (struct sockaddr_in6 *)nam; 255 if (nam->sa_len != sizeof (*sin6p)) 256 return (EINVAL); 257 /* 258 * Must check for multicast addresses and disallow binding 259 * to them. 260 */ 261 if (sin6p->sin6_family == AF_INET6 && 262 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 263 return (EAFNOSUPPORT); 264 265 COMMON_START(); 266 inp->inp_vflag &= ~INP_IPV4; 267 inp->inp_vflag |= INP_IPV6; 268 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 269 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 270 inp->inp_vflag |= INP_IPV4; 271 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 272 struct sockaddr_in sin; 273 274 in6_sin6_2_sin(&sin, sin6p); 275 inp->inp_vflag |= INP_IPV4; 276 inp->inp_vflag &= ~INP_IPV6; 277 error = in_pcbbind(inp, (struct sockaddr *)&sin, 278 td->td_ucred); 279 goto out; 280 } 281 } 282 error = in6_pcbbind(inp, nam, td->td_ucred); 283 if (error) 284 goto out; 285 COMMON_END(PRU_BIND); 286} 287#endif /* INET6 */ 288 289/* 290 * Prepare to accept connections. 291 */ 292static int 293tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 294{ 295 int error = 0; 296 struct inpcb *inp; 297 struct tcpcb *tp; 298 const int inirw = INI_WRITE; 299 300 COMMON_START(); 301 SOCK_LOCK(so); 302 error = solisten_proto_check(so); 303 if (error == 0 && inp->inp_lport == 0) 304 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 305 if (error == 0) { 306 tp->t_state = TCPS_LISTEN; 307 solisten_proto(so, backlog); 308 } 309 SOCK_UNLOCK(so); 310 COMMON_END(PRU_LISTEN); 311} 312 313#ifdef INET6 314static int 315tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 316{ 317 int error = 0; 318 struct inpcb *inp; 319 struct tcpcb *tp; 320 const int inirw = INI_WRITE; 321 322 COMMON_START(); 323 SOCK_LOCK(so); 324 error = solisten_proto_check(so); 325 if (error == 0 && inp->inp_lport == 0) { 326 inp->inp_vflag &= ~INP_IPV4; 327 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 328 inp->inp_vflag |= INP_IPV4; 329 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 330 } 331 if (error == 0) { 332 tp->t_state = TCPS_LISTEN; 333 solisten_proto(so, backlog); 334 } 335 SOCK_UNLOCK(so); 336 COMMON_END(PRU_LISTEN); 337} 338#endif /* INET6 */ 339 340/* 341 * Initiate connection to peer. 342 * Create a template for use in transmissions on this connection. 343 * Enter SYN_SENT state, and mark socket as connecting. 344 * Start keep-alive timer, and seed output sequence space. 345 * Send initial segment on connection. 346 */ 347static int 348tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 349{ 350 int error = 0; 351 struct inpcb *inp; 352 struct tcpcb *tp; 353 struct sockaddr_in *sinp; 354 const int inirw = INI_WRITE; 355 356 sinp = (struct sockaddr_in *)nam; 357 if (nam->sa_len != sizeof (*sinp)) 358 return (EINVAL); 359 /* 360 * Must disallow TCP ``connections'' to multicast addresses. 361 */ 362 if (sinp->sin_family == AF_INET 363 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 364 return (EAFNOSUPPORT); 365 if (jailed(td->td_ucred)) 366 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 367 368 COMMON_START(); 369 if ((error = tcp_connect(tp, nam, td)) != 0) 370 goto out; 371 error = tcp_output(tp); 372 COMMON_END(PRU_CONNECT); 373} 374 375#ifdef INET6 376static int 377tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 378{ 379 int error = 0; 380 struct inpcb *inp; 381 struct tcpcb *tp; 382 struct sockaddr_in6 *sin6p; 383 const int inirw = INI_WRITE; 384 385 sin6p = (struct sockaddr_in6 *)nam; 386 if (nam->sa_len != sizeof (*sin6p)) 387 return (EINVAL); 388 /* 389 * Must disallow TCP ``connections'' to multicast addresses. 390 */ 391 if (sin6p->sin6_family == AF_INET6 392 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 393 return (EAFNOSUPPORT); 394 395 COMMON_START(); 396 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 397 struct sockaddr_in sin; 398 399 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 400 error = EINVAL; 401 goto out; 402 } 403 404 in6_sin6_2_sin(&sin, sin6p); 405 inp->inp_vflag |= INP_IPV4; 406 inp->inp_vflag &= ~INP_IPV6; 407 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 408 goto out; 409 error = tcp_output(tp); 410 goto out; 411 } 412 inp->inp_vflag &= ~INP_IPV4; 413 inp->inp_vflag |= INP_IPV6; 414 inp->inp_inc.inc_isipv6 = 1; 415 if ((error = tcp6_connect(tp, nam, td)) != 0) 416 goto out; 417 error = tcp_output(tp); 418 COMMON_END(PRU_CONNECT); 419} 420#endif /* INET6 */ 421 422/* 423 * Initiate disconnect from peer. 424 * If connection never passed embryonic stage, just drop; 425 * else if don't need to let data drain, then can just drop anyways, 426 * else have to begin TCP shutdown process: mark socket disconnecting, 427 * drain unread data, state switch to reflect user close, and 428 * send segment (e.g. FIN) to peer. Socket will be really disconnected 429 * when peer sends FIN and acks ours. 430 * 431 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 432 */ 433static int 434tcp_usr_disconnect(struct socket *so) 435{ 436 int error = 0; 437 struct inpcb *inp; 438 struct tcpcb *tp; 439 const int inirw = INI_WRITE; 440 441 COMMON_START(); 442 tp = tcp_disconnect(tp); 443 COMMON_END(PRU_DISCONNECT); 444} 445 446/* 447 * Accept a connection. Essentially all the work is 448 * done at higher levels; just return the address 449 * of the peer, storing through addr. 450 */ 451static int 452tcp_usr_accept(struct socket *so, struct sockaddr **nam) 453{ 454 int error = 0; 455 struct inpcb *inp = NULL; 456 struct tcpcb *tp = NULL; 457 struct in_addr addr; 458 in_port_t port = 0; 459 TCPDEBUG0; 460 461 if (so->so_state & SS_ISDISCONNECTED) { 462 error = ECONNABORTED; 463 goto out; 464 } 465 466 INP_INFO_RLOCK(&tcbinfo); 467 inp = sotoinpcb(so); 468 if (!inp) { 469 INP_INFO_RUNLOCK(&tcbinfo); 470 return (EINVAL); 471 } 472 INP_LOCK(inp); 473 INP_INFO_RUNLOCK(&tcbinfo); 474 tp = intotcpcb(inp); 475 TCPDEBUG1(); 476 477 /* 478 * We inline in_setpeeraddr and COMMON_END here, so that we can 479 * copy the data of interest and defer the malloc until after we 480 * release the lock. 481 */ 482 port = inp->inp_fport; 483 addr = inp->inp_faddr; 484 485out: TCPDEBUG2(PRU_ACCEPT); 486 if (tp) 487 INP_UNLOCK(inp); 488 if (error == 0) 489 *nam = in_sockaddr(port, &addr); 490 return error; 491} 492 493#ifdef INET6 494static int 495tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 496{ 497 struct inpcb *inp = NULL; 498 int error = 0; 499 struct tcpcb *tp = NULL; 500 struct in_addr addr; 501 struct in6_addr addr6; 502 in_port_t port = 0; 503 int v4 = 0; 504 TCPDEBUG0; 505 506 if (so->so_state & SS_ISDISCONNECTED) { 507 error = ECONNABORTED; 508 goto out; 509 } 510 511 INP_INFO_RLOCK(&tcbinfo); 512 inp = sotoinpcb(so); 513 if (inp == 0) { 514 INP_INFO_RUNLOCK(&tcbinfo); 515 return (EINVAL); 516 } 517 INP_LOCK(inp); 518 INP_INFO_RUNLOCK(&tcbinfo); 519 tp = intotcpcb(inp); 520 TCPDEBUG1(); 521 /* 522 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 523 * copy the data of interest and defer the malloc until after we 524 * release the lock. 525 */ 526 if (inp->inp_vflag & INP_IPV4) { 527 v4 = 1; 528 port = inp->inp_fport; 529 addr = inp->inp_faddr; 530 } else { 531 port = inp->inp_fport; 532 addr6 = inp->in6p_faddr; 533 } 534 535out: TCPDEBUG2(PRU_ACCEPT); 536 if (tp) 537 INP_UNLOCK(inp); 538 if (error == 0) { 539 if (v4) 540 *nam = in6_v4mapsin6_sockaddr(port, &addr); 541 else 542 *nam = in6_sockaddr(port, &addr6); 543 } 544 return error; 545} 546#endif /* INET6 */ 547 548/* 549 * This is the wrapper function for in_setsockaddr. We just pass down 550 * the pcbinfo for in_setsockaddr to lock. We don't want to do the locking 551 * here because in_setsockaddr will call malloc and can block. 552 */ 553static int 554tcp_sockaddr(struct socket *so, struct sockaddr **nam) 555{ 556 return (in_setsockaddr(so, nam, &tcbinfo)); 557} 558 559/* 560 * This is the wrapper function for in_setpeeraddr. We just pass down 561 * the pcbinfo for in_setpeeraddr to lock. 562 */ 563static int 564tcp_peeraddr(struct socket *so, struct sockaddr **nam) 565{ 566 return (in_setpeeraddr(so, nam, &tcbinfo)); 567} 568 569/* 570 * Mark the connection as being incapable of further output. 571 */ 572static int 573tcp_usr_shutdown(struct socket *so) 574{ 575 int error = 0; 576 struct inpcb *inp; 577 struct tcpcb *tp; 578 const int inirw = INI_WRITE; 579 580 COMMON_START(); 581 socantsendmore(so); 582 tp = tcp_usrclosed(tp); 583 if (tp) 584 error = tcp_output(tp); 585 COMMON_END(PRU_SHUTDOWN); 586} 587 588/* 589 * After a receive, possibly send window update to peer. 590 */ 591static int 592tcp_usr_rcvd(struct socket *so, int flags) 593{ 594 int error = 0; 595 struct inpcb *inp; 596 struct tcpcb *tp; 597 const int inirw = INI_READ; 598 599 COMMON_START(); 600 tcp_output(tp); 601 COMMON_END(PRU_RCVD); 602} 603 604/* 605 * Do a send by putting data in output queue and updating urgent 606 * marker if URG set. Possibly send more data. Unlike the other 607 * pru_*() routines, the mbuf chains are our responsibility. We 608 * must either enqueue them or free them. The other pru_* routines 609 * generally are caller-frees. 610 */ 611static int 612tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 613 struct sockaddr *nam, struct mbuf *control, struct thread *td) 614{ 615 int error = 0; 616 struct inpcb *inp; 617 struct tcpcb *tp; 618 int unlocked = 0; 619#ifdef INET6 620 int isipv6; 621#endif 622 TCPDEBUG0; 623 624 /* 625 * Need write lock here because this function might call 626 * tcp_connect or tcp_usrclosed. 627 * We really want to have to this function upgrade from read lock 628 * to write lock. XXX 629 */ 630 INP_INFO_WLOCK(&tcbinfo); 631 inp = sotoinpcb(so); 632 if (inp == NULL) { 633 /* 634 * OOPS! we lost a race, the TCP session got reset after 635 * we checked SBS_CANTSENDMORE, eg: while doing uiomove or a 636 * network interrupt in the non-splnet() section of sosend(). 637 */ 638 if (m) 639 m_freem(m); 640 if (control) 641 m_freem(control); 642 error = ECONNRESET; /* XXX EPIPE? */ 643 tp = NULL; 644 TCPDEBUG1(); 645 goto out; 646 } 647 INP_LOCK(inp); 648#ifdef INET6 649 isipv6 = nam && nam->sa_family == AF_INET6; 650#endif /* INET6 */ 651 tp = intotcpcb(inp); 652 TCPDEBUG1(); 653 if (control) { 654 /* TCP doesn't do control messages (rights, creds, etc) */ 655 if (control->m_len) { 656 m_freem(control); 657 if (m) 658 m_freem(m); 659 error = EINVAL; 660 goto out; 661 } 662 m_freem(control); /* empty control, just free it */ 663 } 664 if (!(flags & PRUS_OOB)) { 665 sbappendstream(&so->so_snd, m); 666 if (nam && tp->t_state < TCPS_SYN_SENT) { 667 /* 668 * Do implied connect if not yet connected, 669 * initialize window to default value, and 670 * initialize maxseg/maxopd using peer's cached 671 * MSS. 672 */ 673#ifdef INET6 674 if (isipv6) 675 error = tcp6_connect(tp, nam, td); 676 else 677#endif /* INET6 */ 678 error = tcp_connect(tp, nam, td); 679 if (error) 680 goto out; 681 tp->snd_wnd = TTCP_CLIENT_SND_WND; 682 tcp_mss(tp, -1); 683 } 684 685 if (flags & PRUS_EOF) { 686 /* 687 * Close the send side of the connection after 688 * the data is sent. 689 */ 690 socantsendmore(so); 691 tp = tcp_usrclosed(tp); 692 } 693 INP_INFO_WUNLOCK(&tcbinfo); 694 unlocked = 1; 695 if (tp != NULL) { 696 if (flags & PRUS_MORETOCOME) 697 tp->t_flags |= TF_MORETOCOME; 698 error = tcp_output(tp); 699 if (flags & PRUS_MORETOCOME) 700 tp->t_flags &= ~TF_MORETOCOME; 701 } 702 } else { 703 SOCKBUF_LOCK(&so->so_snd); 704 if (sbspace(&so->so_snd) < -512) { 705 SOCKBUF_UNLOCK(&so->so_snd); 706 m_freem(m); 707 error = ENOBUFS; 708 goto out; 709 } 710 /* 711 * According to RFC961 (Assigned Protocols), 712 * the urgent pointer points to the last octet 713 * of urgent data. We continue, however, 714 * to consider it to indicate the first octet 715 * of data past the urgent section. 716 * Otherwise, snd_up should be one lower. 717 */ 718 sbappendstream_locked(&so->so_snd, m); 719 SOCKBUF_UNLOCK(&so->so_snd); 720 if (nam && tp->t_state < TCPS_SYN_SENT) { 721 /* 722 * Do implied connect if not yet connected, 723 * initialize window to default value, and 724 * initialize maxseg/maxopd using peer's cached 725 * MSS. 726 */ 727#ifdef INET6 728 if (isipv6) 729 error = tcp6_connect(tp, nam, td); 730 else 731#endif /* INET6 */ 732 error = tcp_connect(tp, nam, td); 733 if (error) 734 goto out; 735 tp->snd_wnd = TTCP_CLIENT_SND_WND; 736 tcp_mss(tp, -1); 737 } 738 INP_INFO_WUNLOCK(&tcbinfo); 739 unlocked = 1; 740 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 741 tp->t_flags |= TF_FORCEDATA; 742 error = tcp_output(tp); 743 tp->t_flags &= ~TF_FORCEDATA; 744 } 745out: 746 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 747 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 748 if (tp) 749 INP_UNLOCK(inp); 750 if (!unlocked) 751 INP_INFO_WUNLOCK(&tcbinfo); 752 return (error); 753} 754 755/* 756 * Abort the TCP. 757 */ 758static void 759tcp_usr_abort(struct socket *so) 760{ 761 struct inpcb *inp; 762 struct tcpcb *tp; 763 TCPDEBUG0; 764 765 INP_INFO_WLOCK(&tcbinfo); 766 inp = sotoinpcb(so); 767 if (inp == NULL) 768 return; 769 INP_LOCK(inp); 770 tp = intotcpcb(inp); 771 TCPDEBUG1(); 772 tp = tcp_drop(tp, ECONNABORTED); 773 TCPDEBUG2(PRU_ABORT); 774 if (tp) 775 INP_UNLOCK(inp); 776 INP_INFO_WUNLOCK(&tcbinfo); 777} 778 779/* 780 * Receive out-of-band data. 781 */ 782static int 783tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 784{ 785 int error = 0; 786 struct inpcb *inp; 787 struct tcpcb *tp; 788 const int inirw = INI_READ; 789 790 COMMON_START(); 791 if ((so->so_oobmark == 0 && 792 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 793 so->so_options & SO_OOBINLINE || 794 tp->t_oobflags & TCPOOB_HADDATA) { 795 error = EINVAL; 796 goto out; 797 } 798 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 799 error = EWOULDBLOCK; 800 goto out; 801 } 802 m->m_len = 1; 803 *mtod(m, caddr_t) = tp->t_iobc; 804 if ((flags & MSG_PEEK) == 0) 805 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 806 COMMON_END(PRU_RCVOOB); 807} 808 809struct pr_usrreqs tcp_usrreqs = { 810 .pru_abort = tcp_usr_abort, 811 .pru_accept = tcp_usr_accept, 812 .pru_attach = tcp_usr_attach, 813 .pru_bind = tcp_usr_bind, 814 .pru_connect = tcp_usr_connect, 815 .pru_control = in_control, 816 .pru_detach = tcp_usr_detach, 817 .pru_disconnect = tcp_usr_disconnect, 818 .pru_listen = tcp_usr_listen, 819 .pru_peeraddr = tcp_peeraddr, 820 .pru_rcvd = tcp_usr_rcvd, 821 .pru_rcvoob = tcp_usr_rcvoob, 822 .pru_send = tcp_usr_send, 823 .pru_shutdown = tcp_usr_shutdown, 824 .pru_sockaddr = tcp_sockaddr, 825 .pru_sosetlabel = in_pcbsosetlabel 826}; 827 828#ifdef INET6 829struct pr_usrreqs tcp6_usrreqs = { 830 .pru_abort = tcp_usr_abort, 831 .pru_accept = tcp6_usr_accept, 832 .pru_attach = tcp_usr_attach, 833 .pru_bind = tcp6_usr_bind, 834 .pru_connect = tcp6_usr_connect, 835 .pru_control = in6_control, 836 .pru_detach = tcp_usr_detach, 837 .pru_disconnect = tcp_usr_disconnect, 838 .pru_listen = tcp6_usr_listen, 839 .pru_peeraddr = in6_mapped_peeraddr, 840 .pru_rcvd = tcp_usr_rcvd, 841 .pru_rcvoob = tcp_usr_rcvoob, 842 .pru_send = tcp_usr_send, 843 .pru_shutdown = tcp_usr_shutdown, 844 .pru_sockaddr = in6_mapped_sockaddr, 845 .pru_sosetlabel = in_pcbsosetlabel 846}; 847#endif /* INET6 */ 848 849/* 850 * Common subroutine to open a TCP connection to remote host specified 851 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 852 * port number if needed. Call in_pcbconnect_setup to do the routing and 853 * to choose a local host address (interface). If there is an existing 854 * incarnation of the same connection in TIME-WAIT state and if the remote 855 * host was sending CC options and if the connection duration was < MSL, then 856 * truncate the previous TIME-WAIT state and proceed. 857 * Initialize connection parameters and enter SYN-SENT state. 858 */ 859static int 860tcp_connect(tp, nam, td) 861 register struct tcpcb *tp; 862 struct sockaddr *nam; 863 struct thread *td; 864{ 865 struct inpcb *inp = tp->t_inpcb, *oinp; 866 struct socket *so = inp->inp_socket; 867 struct in_addr laddr; 868 u_short lport; 869 int error; 870 871 if (inp->inp_lport == 0) { 872 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 873 if (error) 874 return error; 875 } 876 877 /* 878 * Cannot simply call in_pcbconnect, because there might be an 879 * earlier incarnation of this same connection still in 880 * TIME_WAIT state, creating an ADDRINUSE error. 881 */ 882 laddr = inp->inp_laddr; 883 lport = inp->inp_lport; 884 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 885 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 886 if (error && oinp == NULL) 887 return error; 888 if (oinp) 889 return EADDRINUSE; 890 inp->inp_laddr = laddr; 891 in_pcbrehash(inp); 892 893 /* Compute window scaling to request. */ 894 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 895 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 896 tp->request_r_scale++; 897 898 soisconnecting(so); 899 tcpstat.tcps_connattempt++; 900 tp->t_state = TCPS_SYN_SENT; 901 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 902 tp->iss = tcp_new_isn(tp); 903 tp->t_bw_rtseq = tp->iss; 904 tcp_sendseqinit(tp); 905 906 return 0; 907} 908 909#ifdef INET6 910static int 911tcp6_connect(tp, nam, td) 912 register struct tcpcb *tp; 913 struct sockaddr *nam; 914 struct thread *td; 915{ 916 struct inpcb *inp = tp->t_inpcb, *oinp; 917 struct socket *so = inp->inp_socket; 918 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 919 struct in6_addr *addr6; 920 int error; 921 922 if (inp->inp_lport == 0) { 923 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 924 if (error) 925 return error; 926 } 927 928 /* 929 * Cannot simply call in_pcbconnect, because there might be an 930 * earlier incarnation of this same connection still in 931 * TIME_WAIT state, creating an ADDRINUSE error. 932 * in6_pcbladdr() also handles scope zone IDs. 933 */ 934 error = in6_pcbladdr(inp, nam, &addr6); 935 if (error) 936 return error; 937 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 938 &sin6->sin6_addr, sin6->sin6_port, 939 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 940 ? addr6 941 : &inp->in6p_laddr, 942 inp->inp_lport, 0, NULL); 943 if (oinp) 944 return EADDRINUSE; 945 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 946 inp->in6p_laddr = *addr6; 947 inp->in6p_faddr = sin6->sin6_addr; 948 inp->inp_fport = sin6->sin6_port; 949 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 950 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 951 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 952 inp->in6p_flowinfo |= 953 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 954 in_pcbrehash(inp); 955 956 /* Compute window scaling to request. */ 957 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 958 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 959 tp->request_r_scale++; 960 961 soisconnecting(so); 962 tcpstat.tcps_connattempt++; 963 tp->t_state = TCPS_SYN_SENT; 964 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 965 tp->iss = tcp_new_isn(tp); 966 tp->t_bw_rtseq = tp->iss; 967 tcp_sendseqinit(tp); 968 969 return 0; 970} 971#endif /* INET6 */ 972 973/* 974 * Export TCP internal state information via a struct tcp_info, based on the 975 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 976 * (TCP state machine, etc). We export all information using FreeBSD-native 977 * constants -- for example, the numeric values for tcpi_state will differ 978 * from Linux. 979 */ 980static void 981tcp_fill_info(tp, ti) 982 struct tcpcb *tp; 983 struct tcp_info *ti; 984{ 985 986 INP_LOCK_ASSERT(tp->t_inpcb); 987 bzero(ti, sizeof(*ti)); 988 989 ti->tcpi_state = tp->t_state; 990 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 991 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 992 if (tp->sack_enable) 993 ti->tcpi_options |= TCPI_OPT_SACK; 994 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 995 ti->tcpi_options |= TCPI_OPT_WSCALE; 996 ti->tcpi_snd_wscale = tp->snd_scale; 997 ti->tcpi_rcv_wscale = tp->rcv_scale; 998 } 999 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1000 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1001 1002 /* 1003 * FreeBSD-specific extension fields for tcp_info. 1004 */ 1005 ti->tcpi_rcv_space = tp->rcv_wnd; 1006 ti->tcpi_snd_wnd = tp->snd_wnd; 1007 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1008} 1009 1010/* 1011 * The new sockopt interface makes it possible for us to block in the 1012 * copyin/out step (if we take a page fault). Taking a page fault at 1013 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1014 * use TSM, there probably isn't any need for this function to run at 1015 * splnet() any more. This needs more examination.) 1016 * 1017 * XXXRW: The locking here is wrong; we may take a page fault while holding 1018 * the inpcb lock. 1019 */ 1020int 1021tcp_ctloutput(so, sopt) 1022 struct socket *so; 1023 struct sockopt *sopt; 1024{ 1025 int error, opt, optval; 1026 struct inpcb *inp; 1027 struct tcpcb *tp; 1028 struct tcp_info ti; 1029 1030 error = 0; 1031 INP_INFO_RLOCK(&tcbinfo); 1032 inp = sotoinpcb(so); 1033 if (inp == NULL) { 1034 INP_INFO_RUNLOCK(&tcbinfo); 1035 return (ECONNRESET); 1036 } 1037 INP_LOCK(inp); 1038 INP_INFO_RUNLOCK(&tcbinfo); 1039 if (sopt->sopt_level != IPPROTO_TCP) { 1040 INP_UNLOCK(inp); 1041#ifdef INET6 1042 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1043 error = ip6_ctloutput(so, sopt); 1044 else 1045#endif /* INET6 */ 1046 error = ip_ctloutput(so, sopt); 1047 return (error); 1048 } 1049 tp = intotcpcb(inp); 1050 1051 switch (sopt->sopt_dir) { 1052 case SOPT_SET: 1053 switch (sopt->sopt_name) { 1054#ifdef TCP_SIGNATURE 1055 case TCP_MD5SIG: 1056 error = sooptcopyin(sopt, &optval, sizeof optval, 1057 sizeof optval); 1058 if (error) 1059 break; 1060 1061 if (optval > 0) 1062 tp->t_flags |= TF_SIGNATURE; 1063 else 1064 tp->t_flags &= ~TF_SIGNATURE; 1065 break; 1066#endif /* TCP_SIGNATURE */ 1067 case TCP_NODELAY: 1068 case TCP_NOOPT: 1069 error = sooptcopyin(sopt, &optval, sizeof optval, 1070 sizeof optval); 1071 if (error) 1072 break; 1073 1074 switch (sopt->sopt_name) { 1075 case TCP_NODELAY: 1076 opt = TF_NODELAY; 1077 break; 1078 case TCP_NOOPT: 1079 opt = TF_NOOPT; 1080 break; 1081 default: 1082 opt = 0; /* dead code to fool gcc */ 1083 break; 1084 } 1085 1086 if (optval) 1087 tp->t_flags |= opt; 1088 else 1089 tp->t_flags &= ~opt; 1090 break; 1091 1092 case TCP_NOPUSH: 1093 error = sooptcopyin(sopt, &optval, sizeof optval, 1094 sizeof optval); 1095 if (error) 1096 break; 1097 1098 if (optval) 1099 tp->t_flags |= TF_NOPUSH; 1100 else { 1101 tp->t_flags &= ~TF_NOPUSH; 1102 error = tcp_output(tp); 1103 } 1104 break; 1105 1106 case TCP_MAXSEG: 1107 error = sooptcopyin(sopt, &optval, sizeof optval, 1108 sizeof optval); 1109 if (error) 1110 break; 1111 1112 if (optval > 0 && optval <= tp->t_maxseg && 1113 optval + 40 >= tcp_minmss) 1114 tp->t_maxseg = optval; 1115 else 1116 error = EINVAL; 1117 break; 1118 1119 case TCP_INFO: 1120 error = EINVAL; 1121 break; 1122 1123 default: 1124 error = ENOPROTOOPT; 1125 break; 1126 } 1127 break; 1128 1129 case SOPT_GET: 1130 switch (sopt->sopt_name) { 1131#ifdef TCP_SIGNATURE 1132 case TCP_MD5SIG: 1133 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1134 error = sooptcopyout(sopt, &optval, sizeof optval); 1135 break; 1136#endif 1137 case TCP_NODELAY: 1138 optval = tp->t_flags & TF_NODELAY; 1139 error = sooptcopyout(sopt, &optval, sizeof optval); 1140 break; 1141 case TCP_MAXSEG: 1142 optval = tp->t_maxseg; 1143 error = sooptcopyout(sopt, &optval, sizeof optval); 1144 break; 1145 case TCP_NOOPT: 1146 optval = tp->t_flags & TF_NOOPT; 1147 error = sooptcopyout(sopt, &optval, sizeof optval); 1148 break; 1149 case TCP_NOPUSH: 1150 optval = tp->t_flags & TF_NOPUSH; 1151 error = sooptcopyout(sopt, &optval, sizeof optval); 1152 break; 1153 case TCP_INFO: 1154 tcp_fill_info(tp, &ti); 1155 error = sooptcopyout(sopt, &ti, sizeof ti); 1156 break; 1157 default: 1158 error = ENOPROTOOPT; 1159 break; 1160 } 1161 break; 1162 } 1163 INP_UNLOCK(inp); 1164 return (error); 1165} 1166 1167/* 1168 * tcp_sendspace and tcp_recvspace are the default send and receive window 1169 * sizes, respectively. These are obsolescent (this information should 1170 * be set by the route). 1171 */ 1172u_long tcp_sendspace = 1024*32; 1173SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1174 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1175u_long tcp_recvspace = 1024*64; 1176SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1177 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1178 1179/* 1180 * Attach TCP protocol to socket, allocating 1181 * internet protocol control block, tcp control block, 1182 * bufer space, and entering LISTEN state if to accept connections. 1183 */ 1184static int 1185tcp_attach(so) 1186 struct socket *so; 1187{ 1188 register struct tcpcb *tp; 1189 struct inpcb *inp; 1190 int error; 1191#ifdef INET6 1192 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1193#endif 1194 1195 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1196 1197 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1198 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1199 if (error) 1200 return (error); 1201 } 1202 error = in_pcballoc(so, &tcbinfo, "tcpinp"); 1203 if (error) 1204 return (error); 1205 inp = sotoinpcb(so); 1206#ifdef INET6 1207 if (isipv6) { 1208 inp->inp_vflag |= INP_IPV6; 1209 inp->in6p_hops = -1; /* use kernel default */ 1210 } 1211 else 1212#endif 1213 inp->inp_vflag |= INP_IPV4; 1214 tp = tcp_newtcpcb(inp); 1215 if (tp == 0) { 1216 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 1217 1218 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 1219 1220 INP_LOCK(inp); 1221#ifdef INET6 1222 if (isipv6) 1223 in6_pcbdetach(inp); 1224 else 1225#endif 1226 in_pcbdetach(inp); 1227 so->so_state |= nofd; 1228 return (ENOBUFS); 1229 } 1230 tp->t_state = TCPS_CLOSED; 1231 return (0); 1232} 1233 1234/* 1235 * Initiate (or continue) disconnect. 1236 * If embryonic state, just send reset (once). 1237 * If in ``let data drain'' option and linger null, just drop. 1238 * Otherwise (hard), mark socket disconnecting and drop 1239 * current input data; switch states based on user close, and 1240 * send segment to peer (with FIN). 1241 */ 1242static struct tcpcb * 1243tcp_disconnect(tp) 1244 register struct tcpcb *tp; 1245{ 1246 struct inpcb *inp = tp->t_inpcb; 1247 struct socket *so = inp->inp_socket; 1248 1249 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1250 INP_LOCK_ASSERT(inp); 1251 1252 if (tp->t_state < TCPS_ESTABLISHED) 1253 tp = tcp_close(tp); 1254 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 1255 tp = tcp_drop(tp, 0); 1256 else { 1257 soisdisconnecting(so); 1258 sbflush(&so->so_rcv); 1259 tp = tcp_usrclosed(tp); 1260 if (tp) 1261 (void) tcp_output(tp); 1262 } 1263 return (tp); 1264} 1265 1266/* 1267 * User issued close, and wish to trail through shutdown states: 1268 * if never received SYN, just forget it. If got a SYN from peer, 1269 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1270 * If already got a FIN from peer, then almost done; go to LAST_ACK 1271 * state. In all other cases, have already sent FIN to peer (e.g. 1272 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1273 * for peer to send FIN or not respond to keep-alives, etc. 1274 * We can let the user exit from the close as soon as the FIN is acked. 1275 */ 1276static struct tcpcb * 1277tcp_usrclosed(tp) 1278 register struct tcpcb *tp; 1279{ 1280 1281 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1282 INP_LOCK_ASSERT(tp->t_inpcb); 1283 1284 switch (tp->t_state) { 1285 1286 case TCPS_CLOSED: 1287 case TCPS_LISTEN: 1288 tp->t_state = TCPS_CLOSED; 1289 tp = tcp_close(tp); 1290 break; 1291 1292 case TCPS_SYN_SENT: 1293 case TCPS_SYN_RECEIVED: 1294 tp->t_flags |= TF_NEEDFIN; 1295 break; 1296 1297 case TCPS_ESTABLISHED: 1298 tp->t_state = TCPS_FIN_WAIT_1; 1299 break; 1300 1301 case TCPS_CLOSE_WAIT: 1302 tp->t_state = TCPS_LAST_ACK; 1303 break; 1304 } 1305 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1306 soisdisconnected(tp->t_inpcb->inp_socket); 1307 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1308 if (tp->t_state == TCPS_FIN_WAIT_2) 1309 callout_reset(tp->tt_2msl, tcp_maxidle, 1310 tcp_timer_2msl, tp); 1311 } 1312 return (tp); 1313} 1314