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 * $FreeBSD$ 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 */ 83 extern char *tcpstates[]; /* XXX ??? */ 84 85 static int tcp_attach(struct socket *); 86 static int tcp_connect(struct tcpcb *, struct sockaddr *, 87 struct thread *td); 88 #ifdef INET6 89 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 90 struct thread *td); 91 #endif /* INET6 */ 92 static struct tcpcb * 93 tcp_disconnect(struct tcpcb *); 94 static struct tcpcb * 95 tcp_usrclosed(struct tcpcb *); 96 static 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 */ 113 static int 114 tcp_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); 138 out: 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 */ 151 static int 152 tcp_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) \ 204 out: 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 */ 217 static int 218 tcp_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 245 static int 246 tcp6_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 */ 292 static int 293 tcp_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 314 static int 315 tcp6_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 */ 347 static int 348 tcp_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 376 static int 377 tcp6_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 */ 433 static int 434 tcp_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 */ 451 static int 452 tcp_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; 484 485 out: 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 494 static int 495 tcp6_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 } 534 535 out: 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 */ 553 static int 554 tcp_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 */ 563 static int 564 tcp_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 */ 572 static int 573 tcp_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 */ 591 static int 592 tcp_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 */ 611 static int 612 tcp_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 } 745 out: 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 */ 758 static int 759 tcp_usr_abort(struct socket *so) 760 { 761 int error = 0; 762 struct inpcb *inp; 763 struct tcpcb *tp; 764 const int inirw = INI_WRITE; 765 766 COMMON_START(); 767 tp = tcp_drop(tp, ECONNABORTED); 768 COMMON_END(PRU_ABORT); 769 } 770 771 /* 772 * Receive out-of-band data. 773 */ 774 static int 775 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 776 { 777 int error = 0; 778 struct inpcb *inp; 779 struct tcpcb *tp; 780 const int inirw = INI_READ; 781 782 COMMON_START(); 783 if ((so->so_oobmark == 0 && 784 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 785 so->so_options & SO_OOBINLINE || 786 tp->t_oobflags & TCPOOB_HADDATA) { 787 error = EINVAL; 788 goto out; 789 } 790 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 791 error = EWOULDBLOCK; 792 goto out; 793 } 794 m->m_len = 1; 795 *mtod(m, caddr_t) = tp->t_iobc; 796 if ((flags & MSG_PEEK) == 0) 797 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 798 COMMON_END(PRU_RCVOOB); 799 } 800 801 struct pr_usrreqs tcp_usrreqs = { 802 .pru_abort = tcp_usr_abort, 803 .pru_accept = tcp_usr_accept, 804 .pru_attach = tcp_usr_attach, 805 .pru_bind = tcp_usr_bind, 806 .pru_connect = tcp_usr_connect, 807 .pru_control = in_control, 808 .pru_detach = tcp_usr_detach, 809 .pru_disconnect = tcp_usr_disconnect, 810 .pru_listen = tcp_usr_listen, 811 .pru_peeraddr = tcp_peeraddr, 812 .pru_rcvd = tcp_usr_rcvd, 813 .pru_rcvoob = tcp_usr_rcvoob, 814 .pru_send = tcp_usr_send, 815 .pru_shutdown = tcp_usr_shutdown, 816 .pru_sockaddr = tcp_sockaddr, 817 .pru_sosetlabel = in_pcbsosetlabel 818 }; 819 820 #ifdef INET6 821 struct pr_usrreqs tcp6_usrreqs = { 822 .pru_abort = tcp_usr_abort, 823 .pru_accept = tcp6_usr_accept, 824 .pru_attach = tcp_usr_attach, 825 .pru_bind = tcp6_usr_bind, 826 .pru_connect = tcp6_usr_connect, 827 .pru_control = in6_control, 828 .pru_detach = tcp_usr_detach, 829 .pru_disconnect = tcp_usr_disconnect, 830 .pru_listen = tcp6_usr_listen, 831 .pru_peeraddr = in6_mapped_peeraddr, 832 .pru_rcvd = tcp_usr_rcvd, 833 .pru_rcvoob = tcp_usr_rcvoob, 834 .pru_send = tcp_usr_send, 835 .pru_shutdown = tcp_usr_shutdown, 836 .pru_sockaddr = in6_mapped_sockaddr, 837 .pru_sosetlabel = in_pcbsosetlabel 838 }; 839 #endif /* INET6 */ 840 841 /* 842 * Common subroutine to open a TCP connection to remote host specified 843 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 844 * port number if needed. Call in_pcbconnect_setup to do the routing and 845 * to choose a local host address (interface). If there is an existing 846 * incarnation of the same connection in TIME-WAIT state and if the remote 847 * host was sending CC options and if the connection duration was < MSL, then 848 * truncate the previous TIME-WAIT state and proceed. 849 * Initialize connection parameters and enter SYN-SENT state. 850 */ 851 static int 852 tcp_connect(tp, nam, td) 853 register struct tcpcb *tp; 854 struct sockaddr *nam; 855 struct thread *td; 856 { 857 struct inpcb *inp = tp->t_inpcb, *oinp; 858 struct socket *so = inp->inp_socket; 859 struct in_addr laddr; 860 u_short lport; 861 int error; 862 863 if (inp->inp_lport == 0) { 864 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 865 if (error) 866 return error; 867 } 868 869 /* 870 * Cannot simply call in_pcbconnect, because there might be an 871 * earlier incarnation of this same connection still in 872 * TIME_WAIT state, creating an ADDRINUSE error. 873 */ 874 laddr = inp->inp_laddr; 875 lport = inp->inp_lport; 876 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 877 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 878 if (error && oinp == NULL) 879 return error; 880 if (oinp) 881 return EADDRINUSE; 882 inp->inp_laddr = laddr; 883 in_pcbrehash(inp); 884 885 /* Compute window scaling to request. */ 886 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 887 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 888 tp->request_r_scale++; 889 890 soisconnecting(so); 891 tcpstat.tcps_connattempt++; 892 tp->t_state = TCPS_SYN_SENT; 893 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 894 tp->iss = tcp_new_isn(tp); 895 tp->t_bw_rtseq = tp->iss; 896 tcp_sendseqinit(tp); 897 898 return 0; 899 } 900 901 #ifdef INET6 902 static int 903 tcp6_connect(tp, nam, td) 904 register struct tcpcb *tp; 905 struct sockaddr *nam; 906 struct thread *td; 907 { 908 struct inpcb *inp = tp->t_inpcb, *oinp; 909 struct socket *so = inp->inp_socket; 910 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 911 struct in6_addr *addr6; 912 int error; 913 914 if (inp->inp_lport == 0) { 915 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 916 if (error) 917 return error; 918 } 919 920 /* 921 * Cannot simply call in_pcbconnect, because there might be an 922 * earlier incarnation of this same connection still in 923 * TIME_WAIT state, creating an ADDRINUSE error. 924 * in6_pcbladdr() also handles scope zone IDs. 925 */ 926 error = in6_pcbladdr(inp, nam, &addr6); 927 if (error) 928 return error; 929 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 930 &sin6->sin6_addr, sin6->sin6_port, 931 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 932 ? addr6 933 : &inp->in6p_laddr, 934 inp->inp_lport, 0, NULL); 935 if (oinp) 936 return EADDRINUSE; 937 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 938 inp->in6p_laddr = *addr6; 939 inp->in6p_faddr = sin6->sin6_addr; 940 inp->inp_fport = sin6->sin6_port; 941 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 942 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 943 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 944 inp->in6p_flowinfo |= 945 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 946 in_pcbrehash(inp); 947 948 /* Compute window scaling to request. */ 949 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 950 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 951 tp->request_r_scale++; 952 953 soisconnecting(so); 954 tcpstat.tcps_connattempt++; 955 tp->t_state = TCPS_SYN_SENT; 956 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 957 tp->iss = tcp_new_isn(tp); 958 tp->t_bw_rtseq = tp->iss; 959 tcp_sendseqinit(tp); 960 961 return 0; 962 } 963 #endif /* INET6 */ 964 965 /* 966 * Export TCP internal state information via a struct tcp_info, based on the 967 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 968 * (TCP state machine, etc). We export all information using FreeBSD-native 969 * constants -- for example, the numeric values for tcpi_state will differ 970 * from Linux. 971 */ 972 static void 973 tcp_fill_info(tp, ti) 974 struct tcpcb *tp; 975 struct tcp_info *ti; 976 { 977 978 INP_LOCK_ASSERT(tp->t_inpcb); 979 bzero(ti, sizeof(*ti)); 980 981 ti->tcpi_state = tp->t_state; 982 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 983 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 984 if (tp->sack_enable) 985 ti->tcpi_options |= TCPI_OPT_SACK; 986 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 987 ti->tcpi_options |= TCPI_OPT_WSCALE; 988 ti->tcpi_snd_wscale = tp->snd_scale; 989 ti->tcpi_rcv_wscale = tp->rcv_scale; 990 } 991 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 992 ti->tcpi_snd_cwnd = tp->snd_cwnd; 993 994 /* 995 * FreeBSD-specific extension fields for tcp_info. 996 */ 997 ti->tcpi_rcv_space = tp->rcv_wnd; 998 ti->tcpi_snd_wnd = tp->snd_wnd; 999 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1000 } 1001 1002 /* 1003 * The new sockopt interface makes it possible for us to block in the 1004 * copyin/out step (if we take a page fault). Taking a page fault at 1005 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1006 * use TSM, there probably isn't any need for this function to run at 1007 * splnet() any more. This needs more examination.) 1008 * 1009 * XXXRW: The locking here is wrong; we may take a page fault while holding 1010 * the inpcb lock. 1011 */ 1012 int 1013 tcp_ctloutput(so, sopt) 1014 struct socket *so; 1015 struct sockopt *sopt; 1016 { 1017 int error, opt, optval; 1018 struct inpcb *inp; 1019 struct tcpcb *tp; 1020 struct tcp_info ti; 1021 1022 error = 0; 1023 INP_INFO_RLOCK(&tcbinfo); 1024 inp = sotoinpcb(so); 1025 if (inp == NULL) { 1026 INP_INFO_RUNLOCK(&tcbinfo); 1027 return (ECONNRESET); 1028 } 1029 INP_LOCK(inp); 1030 INP_INFO_RUNLOCK(&tcbinfo); 1031 if (sopt->sopt_level != IPPROTO_TCP) { 1032 INP_UNLOCK(inp); 1033 #ifdef INET6 1034 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1035 error = ip6_ctloutput(so, sopt); 1036 else 1037 #endif /* INET6 */ 1038 error = ip_ctloutput(so, sopt); 1039 return (error); 1040 } 1041 tp = intotcpcb(inp); 1042 1043 switch (sopt->sopt_dir) { 1044 case SOPT_SET: 1045 switch (sopt->sopt_name) { 1046 #ifdef TCP_SIGNATURE 1047 case TCP_MD5SIG: 1048 error = sooptcopyin(sopt, &optval, sizeof optval, 1049 sizeof optval); 1050 if (error) 1051 break; 1052 1053 if (optval > 0) 1054 tp->t_flags |= TF_SIGNATURE; 1055 else 1056 tp->t_flags &= ~TF_SIGNATURE; 1057 break; 1058 #endif /* TCP_SIGNATURE */ 1059 case TCP_NODELAY: 1060 case TCP_NOOPT: 1061 error = sooptcopyin(sopt, &optval, sizeof optval, 1062 sizeof optval); 1063 if (error) 1064 break; 1065 1066 switch (sopt->sopt_name) { 1067 case TCP_NODELAY: 1068 opt = TF_NODELAY; 1069 break; 1070 case TCP_NOOPT: 1071 opt = TF_NOOPT; 1072 break; 1073 default: 1074 opt = 0; /* dead code to fool gcc */ 1075 break; 1076 } 1077 1078 if (optval) 1079 tp->t_flags |= opt; 1080 else 1081 tp->t_flags &= ~opt; 1082 break; 1083 1084 case TCP_NOPUSH: 1085 error = sooptcopyin(sopt, &optval, sizeof optval, 1086 sizeof optval); 1087 if (error) 1088 break; 1089 1090 if (optval) 1091 tp->t_flags |= TF_NOPUSH; 1092 else { 1093 tp->t_flags &= ~TF_NOPUSH; 1094 error = tcp_output(tp); 1095 } 1096 break; 1097 1098 case TCP_MAXSEG: 1099 error = sooptcopyin(sopt, &optval, sizeof optval, 1100 sizeof optval); 1101 if (error) 1102 break; 1103 1104 if (optval > 0 && optval <= tp->t_maxseg && 1105 optval + 40 >= tcp_minmss) 1106 tp->t_maxseg = optval; 1107 else 1108 error = EINVAL; 1109 break; 1110 1111 case TCP_INFO: 1112 error = EINVAL; 1113 break; 1114 1115 default: 1116 error = ENOPROTOOPT; 1117 break; 1118 } 1119 break; 1120 1121 case SOPT_GET: 1122 switch (sopt->sopt_name) { 1123 #ifdef TCP_SIGNATURE 1124 case TCP_MD5SIG: 1125 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1126 error = sooptcopyout(sopt, &optval, sizeof optval); 1127 break; 1128 #endif 1129 case TCP_NODELAY: 1130 optval = tp->t_flags & TF_NODELAY; 1131 error = sooptcopyout(sopt, &optval, sizeof optval); 1132 break; 1133 case TCP_MAXSEG: 1134 optval = tp->t_maxseg; 1135 error = sooptcopyout(sopt, &optval, sizeof optval); 1136 break; 1137 case TCP_NOOPT: 1138 optval = tp->t_flags & TF_NOOPT; 1139 error = sooptcopyout(sopt, &optval, sizeof optval); 1140 break; 1141 case TCP_NOPUSH: 1142 optval = tp->t_flags & TF_NOPUSH; 1143 error = sooptcopyout(sopt, &optval, sizeof optval); 1144 break; 1145 case TCP_INFO: 1146 tcp_fill_info(tp, &ti); 1147 error = sooptcopyout(sopt, &ti, sizeof ti); 1148 break; 1149 default: 1150 error = ENOPROTOOPT; 1151 break; 1152 } 1153 break; 1154 } 1155 INP_UNLOCK(inp); 1156 return (error); 1157 } 1158 1159 /* 1160 * tcp_sendspace and tcp_recvspace are the default send and receive window 1161 * sizes, respectively. These are obsolescent (this information should 1162 * be set by the route). 1163 */ 1164 u_long tcp_sendspace = 1024*32; 1165 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1166 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1167 u_long tcp_recvspace = 1024*64; 1168 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1169 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1170 1171 /* 1172 * Attach TCP protocol to socket, allocating 1173 * internet protocol control block, tcp control block, 1174 * bufer space, and entering LISTEN state if to accept connections. 1175 */ 1176 static int 1177 tcp_attach(so) 1178 struct socket *so; 1179 { 1180 register struct tcpcb *tp; 1181 struct inpcb *inp; 1182 int error; 1183 #ifdef INET6 1184 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1185 #endif 1186 1187 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1188 1189 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1190 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1191 if (error) 1192 return (error); 1193 } 1194 error = in_pcballoc(so, &tcbinfo, "tcpinp"); 1195 if (error) 1196 return (error); 1197 inp = sotoinpcb(so); 1198 #ifdef INET6 1199 if (isipv6) { 1200 inp->inp_vflag |= INP_IPV6; 1201 inp->in6p_hops = -1; /* use kernel default */ 1202 } 1203 else 1204 #endif 1205 inp->inp_vflag |= INP_IPV4; 1206 tp = tcp_newtcpcb(inp); 1207 if (tp == 0) { 1208 int nofd = so->so_state & SS_NOFDREF; /* XXX */ 1209 1210 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */ 1211 1212 INP_LOCK(inp); 1213 #ifdef INET6 1214 if (isipv6) 1215 in6_pcbdetach(inp); 1216 else 1217 #endif 1218 in_pcbdetach(inp); 1219 so->so_state |= nofd; 1220 return (ENOBUFS); 1221 } 1222 tp->t_state = TCPS_CLOSED; 1223 return (0); 1224 } 1225 1226 /* 1227 * Initiate (or continue) disconnect. 1228 * If embryonic state, just send reset (once). 1229 * If in ``let data drain'' option and linger null, just drop. 1230 * Otherwise (hard), mark socket disconnecting and drop 1231 * current input data; switch states based on user close, and 1232 * send segment to peer (with FIN). 1233 */ 1234 static struct tcpcb * 1235 tcp_disconnect(tp) 1236 register struct tcpcb *tp; 1237 { 1238 struct inpcb *inp = tp->t_inpcb; 1239 struct socket *so = inp->inp_socket; 1240 1241 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1242 INP_LOCK_ASSERT(inp); 1243 1244 if (tp->t_state < TCPS_ESTABLISHED) 1245 tp = tcp_close(tp); 1246 else if ((so->so_options & SO_LINGER) && so->so_linger == 0) 1247 tp = tcp_drop(tp, 0); 1248 else { 1249 soisdisconnecting(so); 1250 sbflush(&so->so_rcv); 1251 tp = tcp_usrclosed(tp); 1252 if (tp) 1253 (void) tcp_output(tp); 1254 } 1255 return (tp); 1256 } 1257 1258 /* 1259 * User issued close, and wish to trail through shutdown states: 1260 * if never received SYN, just forget it. If got a SYN from peer, 1261 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1262 * If already got a FIN from peer, then almost done; go to LAST_ACK 1263 * state. In all other cases, have already sent FIN to peer (e.g. 1264 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1265 * for peer to send FIN or not respond to keep-alives, etc. 1266 * We can let the user exit from the close as soon as the FIN is acked. 1267 */ 1268 static struct tcpcb * 1269 tcp_usrclosed(tp) 1270 register struct tcpcb *tp; 1271 { 1272 1273 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1274 INP_LOCK_ASSERT(tp->t_inpcb); 1275 1276 switch (tp->t_state) { 1277 1278 case TCPS_CLOSED: 1279 case TCPS_LISTEN: 1280 tp->t_state = TCPS_CLOSED; 1281 tp = tcp_close(tp); 1282 break; 1283 1284 case TCPS_SYN_SENT: 1285 case TCPS_SYN_RECEIVED: 1286 tp->t_flags |= TF_NEEDFIN; 1287 break; 1288 1289 case TCPS_ESTABLISHED: 1290 tp->t_state = TCPS_FIN_WAIT_1; 1291 break; 1292 1293 case TCPS_CLOSE_WAIT: 1294 tp->t_state = TCPS_LAST_ACK; 1295 break; 1296 } 1297 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1298 soisdisconnected(tp->t_inpcb->inp_socket); 1299 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1300 if (tp->t_state == TCPS_FIN_WAIT_2) 1301 callout_reset(tp->tt_2msl, tcp_maxidle, 1302 tcp_timer_2msl, tp); 1303 } 1304 return (tp); 1305 } 1306