tcp_usrreq.c revision 175438
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2006-2007 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 32 */ 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: head/sys/netinet/tcp_usrreq.c 175438 2008-01-18 12:19:50Z rwatson $"); 36 37#include "opt_ddb.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#ifdef DDB 58#include <ddb/ddb.h> 59#endif 60 61#include <net/if.h> 62#include <net/route.h> 63 64#include <netinet/in.h> 65#include <netinet/in_systm.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#endif 69#include <netinet/in_pcb.h> 70#ifdef INET6 71#include <netinet6/in6_pcb.h> 72#endif 73#include <netinet/in_var.h> 74#include <netinet/ip_var.h> 75#ifdef INET6 76#include <netinet6/ip6_var.h> 77#include <netinet6/scope6_var.h> 78#endif 79#include <netinet/tcp.h> 80#include <netinet/tcp_fsm.h> 81#include <netinet/tcp_seq.h> 82#include <netinet/tcp_timer.h> 83#include <netinet/tcp_var.h> 84#include <netinet/tcpip.h> 85#ifdef TCPDEBUG 86#include <netinet/tcp_debug.h> 87#endif 88#include <netinet/tcp_offload.h> 89 90/* 91 * TCP protocol interface to socket abstraction. 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 void tcp_disconnect(struct tcpcb *); 101static void tcp_usrclosed(struct tcpcb *); 102static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 103 104#ifdef TCPDEBUG 105#define TCPDEBUG0 int ostate = 0 106#define TCPDEBUG1() ostate = tp ? tp->t_state : 0 107#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 108 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 109#else 110#define TCPDEBUG0 111#define TCPDEBUG1() 112#define TCPDEBUG2(req) 113#endif 114 115/* 116 * TCP attaches to socket via pru_attach(), reserving space, 117 * and an internet control block. 118 */ 119static int 120tcp_usr_attach(struct socket *so, int proto, struct thread *td) 121{ 122 struct inpcb *inp; 123 struct tcpcb *tp = NULL; 124 int error; 125 TCPDEBUG0; 126 127 inp = sotoinpcb(so); 128 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 129 TCPDEBUG1(); 130 131 error = tcp_attach(so); 132 if (error) 133 goto out; 134 135 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 136 so->so_linger = TCP_LINGERTIME; 137 138 inp = sotoinpcb(so); 139 tp = intotcpcb(inp); 140out: 141 TCPDEBUG2(PRU_ATTACH); 142 return error; 143} 144 145/* 146 * tcp_detach is called when the socket layer loses its final reference 147 * to the socket, be it a file descriptor reference, a reference from TCP, 148 * etc. At this point, there is only one case in which we will keep around 149 * inpcb state: time wait. 150 * 151 * This function can probably be re-absorbed back into tcp_usr_detach() now 152 * that there is a single detach path. 153 */ 154static void 155tcp_detach(struct socket *so, struct inpcb *inp) 156{ 157 struct tcpcb *tp; 158#ifdef INET6 159 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 160#endif 161 162 INP_INFO_WLOCK_ASSERT(&tcbinfo); 163 INP_LOCK_ASSERT(inp); 164 165 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 166 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 167 168 tp = intotcpcb(inp); 169 170 if (inp->inp_vflag & INP_TIMEWAIT) { 171 /* 172 * There are two cases to handle: one in which the time wait 173 * state is being discarded (INP_DROPPED), and one in which 174 * this connection will remain in timewait. In the former, 175 * it is time to discard all state (except tcptw, which has 176 * already been discarded by the timewait close code, which 177 * should be further up the call stack somewhere). In the 178 * latter case, we detach from the socket, but leave the pcb 179 * present until timewait ends. 180 * 181 * XXXRW: Would it be cleaner to free the tcptw here? 182 */ 183 if (inp->inp_vflag & INP_DROPPED) { 184 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 185 "INP_DROPPED && tp != NULL")); 186#ifdef INET6 187 if (isipv6) { 188 in6_pcbdetach(inp); 189 in6_pcbfree(inp); 190 } else { 191#endif 192 in_pcbdetach(inp); 193 in_pcbfree(inp); 194#ifdef INET6 195 } 196#endif 197 } else { 198#ifdef INET6 199 if (isipv6) 200 in6_pcbdetach(inp); 201 else 202#endif 203 in_pcbdetach(inp); 204 INP_UNLOCK(inp); 205 } 206 } else { 207 /* 208 * If the connection is not in timewait, we consider two 209 * two conditions: one in which no further processing is 210 * necessary (dropped || embryonic), and one in which TCP is 211 * not yet done, but no longer requires the socket, so the 212 * pcb will persist for the time being. 213 * 214 * XXXRW: Does the second case still occur? 215 */ 216 if (inp->inp_vflag & INP_DROPPED || 217 tp->t_state < TCPS_SYN_SENT) { 218 tcp_discardcb(tp); 219#ifdef INET6 220 if (isipv6) { 221 in6_pcbdetach(inp); 222 in6_pcbfree(inp); 223 } else { 224#endif 225 in_pcbdetach(inp); 226 in_pcbfree(inp); 227#ifdef INET6 228 } 229#endif 230 } else { 231#ifdef INET6 232 if (isipv6) 233 in6_pcbdetach(inp); 234 else 235#endif 236 in_pcbdetach(inp); 237 } 238 } 239} 240 241/* 242 * pru_detach() detaches the TCP protocol from the socket. 243 * If the protocol state is non-embryonic, then can't 244 * do this directly: have to initiate a pru_disconnect(), 245 * which may finish later; embryonic TCB's can just 246 * be discarded here. 247 */ 248static void 249tcp_usr_detach(struct socket *so) 250{ 251 struct inpcb *inp; 252 253 inp = sotoinpcb(so); 254 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 255 INP_INFO_WLOCK(&tcbinfo); 256 INP_LOCK(inp); 257 KASSERT(inp->inp_socket != NULL, 258 ("tcp_usr_detach: inp_socket == NULL")); 259 tcp_detach(so, inp); 260 INP_INFO_WUNLOCK(&tcbinfo); 261} 262 263/* 264 * Give the socket an address. 265 */ 266static int 267tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 268{ 269 int error = 0; 270 struct inpcb *inp; 271 struct tcpcb *tp = NULL; 272 struct sockaddr_in *sinp; 273 274 sinp = (struct sockaddr_in *)nam; 275 if (nam->sa_len != sizeof (*sinp)) 276 return (EINVAL); 277 /* 278 * Must check for multicast addresses and disallow binding 279 * to them. 280 */ 281 if (sinp->sin_family == AF_INET && 282 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 283 return (EAFNOSUPPORT); 284 285 TCPDEBUG0; 286 INP_INFO_WLOCK(&tcbinfo); 287 inp = sotoinpcb(so); 288 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 289 INP_LOCK(inp); 290 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 291 error = EINVAL; 292 goto out; 293 } 294 tp = intotcpcb(inp); 295 TCPDEBUG1(); 296 error = in_pcbbind(inp, nam, td->td_ucred); 297out: 298 TCPDEBUG2(PRU_BIND); 299 INP_UNLOCK(inp); 300 INP_INFO_WUNLOCK(&tcbinfo); 301 302 return (error); 303} 304 305#ifdef INET6 306static int 307tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 308{ 309 int error = 0; 310 struct inpcb *inp; 311 struct tcpcb *tp = NULL; 312 struct sockaddr_in6 *sin6p; 313 314 sin6p = (struct sockaddr_in6 *)nam; 315 if (nam->sa_len != sizeof (*sin6p)) 316 return (EINVAL); 317 /* 318 * Must check for multicast addresses and disallow binding 319 * to them. 320 */ 321 if (sin6p->sin6_family == AF_INET6 && 322 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 323 return (EAFNOSUPPORT); 324 325 TCPDEBUG0; 326 INP_INFO_WLOCK(&tcbinfo); 327 inp = sotoinpcb(so); 328 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 329 INP_LOCK(inp); 330 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 331 error = EINVAL; 332 goto out; 333 } 334 tp = intotcpcb(inp); 335 TCPDEBUG1(); 336 inp->inp_vflag &= ~INP_IPV4; 337 inp->inp_vflag |= INP_IPV6; 338 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 339 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 340 inp->inp_vflag |= INP_IPV4; 341 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 342 struct sockaddr_in sin; 343 344 in6_sin6_2_sin(&sin, sin6p); 345 inp->inp_vflag |= INP_IPV4; 346 inp->inp_vflag &= ~INP_IPV6; 347 error = in_pcbbind(inp, (struct sockaddr *)&sin, 348 td->td_ucred); 349 goto out; 350 } 351 } 352 error = in6_pcbbind(inp, nam, td->td_ucred); 353out: 354 TCPDEBUG2(PRU_BIND); 355 INP_UNLOCK(inp); 356 INP_INFO_WUNLOCK(&tcbinfo); 357 return (error); 358} 359#endif /* INET6 */ 360 361/* 362 * Prepare to accept connections. 363 */ 364static int 365tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 366{ 367 int error = 0; 368 struct inpcb *inp; 369 struct tcpcb *tp = NULL; 370 371 TCPDEBUG0; 372 INP_INFO_WLOCK(&tcbinfo); 373 inp = sotoinpcb(so); 374 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 375 INP_LOCK(inp); 376 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 377 error = EINVAL; 378 goto out; 379 } 380 tp = intotcpcb(inp); 381 TCPDEBUG1(); 382 SOCK_LOCK(so); 383 error = solisten_proto_check(so); 384 if (error == 0 && inp->inp_lport == 0) 385 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 386 if (error == 0) { 387 tp->t_state = TCPS_LISTEN; 388 solisten_proto(so, backlog); 389 tcp_offload_listen_open(tp); 390 } 391 SOCK_UNLOCK(so); 392 393out: 394 TCPDEBUG2(PRU_LISTEN); 395 INP_UNLOCK(inp); 396 INP_INFO_WUNLOCK(&tcbinfo); 397 return (error); 398} 399 400#ifdef INET6 401static int 402tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 403{ 404 int error = 0; 405 struct inpcb *inp; 406 struct tcpcb *tp = NULL; 407 408 TCPDEBUG0; 409 INP_INFO_WLOCK(&tcbinfo); 410 inp = sotoinpcb(so); 411 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 412 INP_LOCK(inp); 413 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 414 error = EINVAL; 415 goto out; 416 } 417 tp = intotcpcb(inp); 418 TCPDEBUG1(); 419 SOCK_LOCK(so); 420 error = solisten_proto_check(so); 421 if (error == 0 && inp->inp_lport == 0) { 422 inp->inp_vflag &= ~INP_IPV4; 423 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 424 inp->inp_vflag |= INP_IPV4; 425 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 426 } 427 if (error == 0) { 428 tp->t_state = TCPS_LISTEN; 429 solisten_proto(so, backlog); 430 } 431 SOCK_UNLOCK(so); 432 433out: 434 TCPDEBUG2(PRU_LISTEN); 435 INP_UNLOCK(inp); 436 INP_INFO_WUNLOCK(&tcbinfo); 437 return (error); 438} 439#endif /* INET6 */ 440 441/* 442 * Initiate connection to peer. 443 * Create a template for use in transmissions on this connection. 444 * Enter SYN_SENT state, and mark socket as connecting. 445 * Start keep-alive timer, and seed output sequence space. 446 * Send initial segment on connection. 447 */ 448static int 449tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 450{ 451 int error = 0; 452 struct inpcb *inp; 453 struct tcpcb *tp = NULL; 454 struct sockaddr_in *sinp; 455 456 sinp = (struct sockaddr_in *)nam; 457 if (nam->sa_len != sizeof (*sinp)) 458 return (EINVAL); 459 /* 460 * Must disallow TCP ``connections'' to multicast addresses. 461 */ 462 if (sinp->sin_family == AF_INET 463 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 464 return (EAFNOSUPPORT); 465 if (jailed(td->td_ucred)) 466 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 467 468 TCPDEBUG0; 469 INP_INFO_WLOCK(&tcbinfo); 470 inp = sotoinpcb(so); 471 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 472 INP_LOCK(inp); 473 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 474 error = EINVAL; 475 goto out; 476 } 477 tp = intotcpcb(inp); 478 TCPDEBUG1(); 479 if ((error = tcp_connect(tp, nam, td)) != 0) 480 goto out; 481 error = tcp_output_connect(so, nam); 482out: 483 TCPDEBUG2(PRU_CONNECT); 484 INP_UNLOCK(inp); 485 INP_INFO_WUNLOCK(&tcbinfo); 486 return (error); 487} 488 489#ifdef INET6 490static int 491tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 492{ 493 int error = 0; 494 struct inpcb *inp; 495 struct tcpcb *tp = NULL; 496 struct sockaddr_in6 *sin6p; 497 498 TCPDEBUG0; 499 500 sin6p = (struct sockaddr_in6 *)nam; 501 if (nam->sa_len != sizeof (*sin6p)) 502 return (EINVAL); 503 /* 504 * Must disallow TCP ``connections'' to multicast addresses. 505 */ 506 if (sin6p->sin6_family == AF_INET6 507 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 508 return (EAFNOSUPPORT); 509 510 INP_INFO_WLOCK(&tcbinfo); 511 inp = sotoinpcb(so); 512 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 513 INP_LOCK(inp); 514 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 515 error = EINVAL; 516 goto out; 517 } 518 tp = intotcpcb(inp); 519 TCPDEBUG1(); 520 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 521 struct sockaddr_in sin; 522 523 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 524 error = EINVAL; 525 goto out; 526 } 527 528 in6_sin6_2_sin(&sin, sin6p); 529 inp->inp_vflag |= INP_IPV4; 530 inp->inp_vflag &= ~INP_IPV6; 531 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 532 goto out; 533 error = tcp_output_connect(so, nam); 534 goto out; 535 } 536 inp->inp_vflag &= ~INP_IPV4; 537 inp->inp_vflag |= INP_IPV6; 538 inp->inp_inc.inc_isipv6 = 1; 539 if ((error = tcp6_connect(tp, nam, td)) != 0) 540 goto out; 541 error = tcp_output_connect(so, nam); 542 543out: 544 TCPDEBUG2(PRU_CONNECT); 545 INP_UNLOCK(inp); 546 INP_INFO_WUNLOCK(&tcbinfo); 547 return (error); 548} 549#endif /* INET6 */ 550 551/* 552 * Initiate disconnect from peer. 553 * If connection never passed embryonic stage, just drop; 554 * else if don't need to let data drain, then can just drop anyways, 555 * else have to begin TCP shutdown process: mark socket disconnecting, 556 * drain unread data, state switch to reflect user close, and 557 * send segment (e.g. FIN) to peer. Socket will be really disconnected 558 * when peer sends FIN and acks ours. 559 * 560 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 561 */ 562static int 563tcp_usr_disconnect(struct socket *so) 564{ 565 struct inpcb *inp; 566 struct tcpcb *tp = NULL; 567 int error = 0; 568 569 TCPDEBUG0; 570 INP_INFO_WLOCK(&tcbinfo); 571 inp = sotoinpcb(so); 572 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 573 INP_LOCK(inp); 574 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 575 error = ECONNRESET; 576 goto out; 577 } 578 tp = intotcpcb(inp); 579 TCPDEBUG1(); 580 tcp_disconnect(tp); 581out: 582 TCPDEBUG2(PRU_DISCONNECT); 583 INP_UNLOCK(inp); 584 INP_INFO_WUNLOCK(&tcbinfo); 585 return (error); 586} 587 588/* 589 * Accept a connection. Essentially all the work is 590 * done at higher levels; just return the address 591 * of the peer, storing through addr. 592 */ 593static int 594tcp_usr_accept(struct socket *so, struct sockaddr **nam) 595{ 596 int error = 0; 597 struct inpcb *inp = NULL; 598 struct tcpcb *tp = NULL; 599 struct in_addr addr; 600 in_port_t port = 0; 601 TCPDEBUG0; 602 603 if (so->so_state & SS_ISDISCONNECTED) 604 return (ECONNABORTED); 605 606 inp = sotoinpcb(so); 607 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 608 INP_LOCK(inp); 609 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 610 error = ECONNABORTED; 611 goto out; 612 } 613 tp = intotcpcb(inp); 614 TCPDEBUG1(); 615 616 /* 617 * We inline in_getpeeraddr and COMMON_END here, so that we can 618 * copy the data of interest and defer the malloc until after we 619 * release the lock. 620 */ 621 port = inp->inp_fport; 622 addr = inp->inp_faddr; 623 624out: 625 TCPDEBUG2(PRU_ACCEPT); 626 INP_UNLOCK(inp); 627 if (error == 0) 628 *nam = in_sockaddr(port, &addr); 629 return error; 630} 631 632#ifdef INET6 633static int 634tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 635{ 636 struct inpcb *inp = NULL; 637 int error = 0; 638 struct tcpcb *tp = NULL; 639 struct in_addr addr; 640 struct in6_addr addr6; 641 in_port_t port = 0; 642 int v4 = 0; 643 TCPDEBUG0; 644 645 if (so->so_state & SS_ISDISCONNECTED) 646 return (ECONNABORTED); 647 648 inp = sotoinpcb(so); 649 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 650 INP_LOCK(inp); 651 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 652 error = ECONNABORTED; 653 goto out; 654 } 655 tp = intotcpcb(inp); 656 TCPDEBUG1(); 657 658 /* 659 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 660 * copy the data of interest and defer the malloc until after we 661 * release the lock. 662 */ 663 if (inp->inp_vflag & INP_IPV4) { 664 v4 = 1; 665 port = inp->inp_fport; 666 addr = inp->inp_faddr; 667 } else { 668 port = inp->inp_fport; 669 addr6 = inp->in6p_faddr; 670 } 671 672out: 673 TCPDEBUG2(PRU_ACCEPT); 674 INP_UNLOCK(inp); 675 if (error == 0) { 676 if (v4) 677 *nam = in6_v4mapsin6_sockaddr(port, &addr); 678 else 679 *nam = in6_sockaddr(port, &addr6); 680 } 681 return error; 682} 683#endif /* INET6 */ 684 685/* 686 * Mark the connection as being incapable of further output. 687 */ 688static int 689tcp_usr_shutdown(struct socket *so) 690{ 691 int error = 0; 692 struct inpcb *inp; 693 struct tcpcb *tp = NULL; 694 695 TCPDEBUG0; 696 INP_INFO_WLOCK(&tcbinfo); 697 inp = sotoinpcb(so); 698 KASSERT(inp != NULL, ("inp == NULL")); 699 INP_LOCK(inp); 700 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 701 error = ECONNRESET; 702 goto out; 703 } 704 tp = intotcpcb(inp); 705 TCPDEBUG1(); 706 socantsendmore(so); 707 tcp_usrclosed(tp); 708 error = tcp_output_disconnect(tp); 709 710out: 711 TCPDEBUG2(PRU_SHUTDOWN); 712 INP_UNLOCK(inp); 713 INP_INFO_WUNLOCK(&tcbinfo); 714 715 return (error); 716} 717 718/* 719 * After a receive, possibly send window update to peer. 720 */ 721static int 722tcp_usr_rcvd(struct socket *so, int flags) 723{ 724 struct inpcb *inp; 725 struct tcpcb *tp = NULL; 726 int error = 0; 727 728 TCPDEBUG0; 729 inp = sotoinpcb(so); 730 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 731 INP_LOCK(inp); 732 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 733 error = ECONNRESET; 734 goto out; 735 } 736 tp = intotcpcb(inp); 737 TCPDEBUG1(); 738 tcp_output_rcvd(tp); 739 740out: 741 TCPDEBUG2(PRU_RCVD); 742 INP_UNLOCK(inp); 743 return (error); 744} 745 746/* 747 * Do a send by putting data in output queue and updating urgent 748 * marker if URG set. Possibly send more data. Unlike the other 749 * pru_*() routines, the mbuf chains are our responsibility. We 750 * must either enqueue them or free them. The other pru_* routines 751 * generally are caller-frees. 752 */ 753static int 754tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 755 struct sockaddr *nam, struct mbuf *control, struct thread *td) 756{ 757 int error = 0; 758 struct inpcb *inp; 759 struct tcpcb *tp = NULL; 760 int headlocked = 0; 761#ifdef INET6 762 int isipv6; 763#endif 764 TCPDEBUG0; 765 766 /* 767 * We require the pcbinfo lock in two cases: 768 * 769 * (1) An implied connect is taking place, which can result in 770 * binding IPs and ports and hence modification of the pcb hash 771 * chains. 772 * 773 * (2) PRUS_EOF is set, resulting in explicit close on the send. 774 */ 775 if ((nam != NULL) || (flags & PRUS_EOF)) { 776 INP_INFO_WLOCK(&tcbinfo); 777 headlocked = 1; 778 } 779 inp = sotoinpcb(so); 780 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 781 INP_LOCK(inp); 782 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 783 if (control) 784 m_freem(control); 785 if (m) 786 m_freem(m); 787 error = ECONNRESET; 788 goto out; 789 } 790#ifdef INET6 791 isipv6 = nam && nam->sa_family == AF_INET6; 792#endif /* INET6 */ 793 tp = intotcpcb(inp); 794 TCPDEBUG1(); 795 if (control) { 796 /* TCP doesn't do control messages (rights, creds, etc) */ 797 if (control->m_len) { 798 m_freem(control); 799 if (m) 800 m_freem(m); 801 error = EINVAL; 802 goto out; 803 } 804 m_freem(control); /* empty control, just free it */ 805 } 806 if (!(flags & PRUS_OOB)) { 807 sbappendstream(&so->so_snd, m); 808 if (nam && tp->t_state < TCPS_SYN_SENT) { 809 /* 810 * Do implied connect if not yet connected, 811 * initialize window to default value, and 812 * initialize maxseg/maxopd using peer's cached 813 * MSS. 814 */ 815 INP_INFO_WLOCK_ASSERT(&tcbinfo); 816#ifdef INET6 817 if (isipv6) 818 error = tcp6_connect(tp, nam, td); 819 else 820#endif /* INET6 */ 821 error = tcp_connect(tp, nam, td); 822 if (error) 823 goto out; 824 tp->snd_wnd = TTCP_CLIENT_SND_WND; 825 tcp_mss(tp, -1); 826 } 827 if (flags & PRUS_EOF) { 828 /* 829 * Close the send side of the connection after 830 * the data is sent. 831 */ 832 INP_INFO_WLOCK_ASSERT(&tcbinfo); 833 socantsendmore(so); 834 tcp_usrclosed(tp); 835 } 836 if (headlocked) { 837 INP_INFO_WUNLOCK(&tcbinfo); 838 headlocked = 0; 839 } 840 if (tp != NULL) { 841 if (flags & PRUS_MORETOCOME) 842 tp->t_flags |= TF_MORETOCOME; 843 error = tcp_output_send(tp); 844 if (flags & PRUS_MORETOCOME) 845 tp->t_flags &= ~TF_MORETOCOME; 846 } 847 } else { 848 /* 849 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 850 */ 851 SOCKBUF_LOCK(&so->so_snd); 852 if (sbspace(&so->so_snd) < -512) { 853 SOCKBUF_UNLOCK(&so->so_snd); 854 m_freem(m); 855 error = ENOBUFS; 856 goto out; 857 } 858 /* 859 * According to RFC961 (Assigned Protocols), 860 * the urgent pointer points to the last octet 861 * of urgent data. We continue, however, 862 * to consider it to indicate the first octet 863 * of data past the urgent section. 864 * Otherwise, snd_up should be one lower. 865 */ 866 sbappendstream_locked(&so->so_snd, m); 867 SOCKBUF_UNLOCK(&so->so_snd); 868 if (nam && tp->t_state < TCPS_SYN_SENT) { 869 /* 870 * Do implied connect if not yet connected, 871 * initialize window to default value, and 872 * initialize maxseg/maxopd using peer's cached 873 * MSS. 874 */ 875 INP_INFO_WLOCK_ASSERT(&tcbinfo); 876#ifdef INET6 877 if (isipv6) 878 error = tcp6_connect(tp, nam, td); 879 else 880#endif /* INET6 */ 881 error = tcp_connect(tp, nam, td); 882 if (error) 883 goto out; 884 tp->snd_wnd = TTCP_CLIENT_SND_WND; 885 tcp_mss(tp, -1); 886 INP_INFO_WUNLOCK(&tcbinfo); 887 headlocked = 0; 888 } else if (nam) { 889 INP_INFO_WUNLOCK(&tcbinfo); 890 headlocked = 0; 891 } 892 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 893 tp->t_flags |= TF_FORCEDATA; 894 error = tcp_output_send(tp); 895 tp->t_flags &= ~TF_FORCEDATA; 896 } 897out: 898 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 899 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 900 INP_UNLOCK(inp); 901 if (headlocked) 902 INP_INFO_WUNLOCK(&tcbinfo); 903 return (error); 904} 905 906/* 907 * Abort the TCP. Drop the connection abruptly. 908 */ 909static void 910tcp_usr_abort(struct socket *so) 911{ 912 struct inpcb *inp; 913 struct tcpcb *tp = NULL; 914 TCPDEBUG0; 915 916 inp = sotoinpcb(so); 917 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 918 919 INP_INFO_WLOCK(&tcbinfo); 920 INP_LOCK(inp); 921 KASSERT(inp->inp_socket != NULL, 922 ("tcp_usr_abort: inp_socket == NULL")); 923 924 /* 925 * If we still have full TCP state, and we're not dropped, drop. 926 */ 927 if (!(inp->inp_vflag & INP_TIMEWAIT) && 928 !(inp->inp_vflag & INP_DROPPED)) { 929 tp = intotcpcb(inp); 930 TCPDEBUG1(); 931 tcp_drop(tp, ECONNABORTED); 932 TCPDEBUG2(PRU_ABORT); 933 } 934 if (!(inp->inp_vflag & INP_DROPPED)) { 935 SOCK_LOCK(so); 936 so->so_state |= SS_PROTOREF; 937 SOCK_UNLOCK(so); 938 inp->inp_vflag |= INP_SOCKREF; 939 } 940 INP_UNLOCK(inp); 941 INP_INFO_WUNLOCK(&tcbinfo); 942} 943 944/* 945 * TCP socket is closed. Start friendly disconnect. 946 */ 947static void 948tcp_usr_close(struct socket *so) 949{ 950 struct inpcb *inp; 951 struct tcpcb *tp = NULL; 952 TCPDEBUG0; 953 954 inp = sotoinpcb(so); 955 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 956 957 INP_INFO_WLOCK(&tcbinfo); 958 INP_LOCK(inp); 959 KASSERT(inp->inp_socket != NULL, 960 ("tcp_usr_close: inp_socket == NULL")); 961 962 /* 963 * If we still have full TCP state, and we're not dropped, initiate 964 * a disconnect. 965 */ 966 if (!(inp->inp_vflag & INP_TIMEWAIT) && 967 !(inp->inp_vflag & INP_DROPPED)) { 968 tp = intotcpcb(inp); 969 TCPDEBUG1(); 970 tcp_disconnect(tp); 971 TCPDEBUG2(PRU_CLOSE); 972 } 973 if (!(inp->inp_vflag & INP_DROPPED)) { 974 SOCK_LOCK(so); 975 so->so_state |= SS_PROTOREF; 976 SOCK_UNLOCK(so); 977 inp->inp_vflag |= INP_SOCKREF; 978 } 979 INP_UNLOCK(inp); 980 INP_INFO_WUNLOCK(&tcbinfo); 981} 982 983/* 984 * Receive out-of-band data. 985 */ 986static int 987tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 988{ 989 int error = 0; 990 struct inpcb *inp; 991 struct tcpcb *tp = NULL; 992 993 TCPDEBUG0; 994 inp = sotoinpcb(so); 995 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 996 INP_LOCK(inp); 997 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 998 error = ECONNRESET; 999 goto out; 1000 } 1001 tp = intotcpcb(inp); 1002 TCPDEBUG1(); 1003 if ((so->so_oobmark == 0 && 1004 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1005 so->so_options & SO_OOBINLINE || 1006 tp->t_oobflags & TCPOOB_HADDATA) { 1007 error = EINVAL; 1008 goto out; 1009 } 1010 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1011 error = EWOULDBLOCK; 1012 goto out; 1013 } 1014 m->m_len = 1; 1015 *mtod(m, caddr_t) = tp->t_iobc; 1016 if ((flags & MSG_PEEK) == 0) 1017 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1018 1019out: 1020 TCPDEBUG2(PRU_RCVOOB); 1021 INP_UNLOCK(inp); 1022 return (error); 1023} 1024 1025struct pr_usrreqs tcp_usrreqs = { 1026 .pru_abort = tcp_usr_abort, 1027 .pru_accept = tcp_usr_accept, 1028 .pru_attach = tcp_usr_attach, 1029 .pru_bind = tcp_usr_bind, 1030 .pru_connect = tcp_usr_connect, 1031 .pru_control = in_control, 1032 .pru_detach = tcp_usr_detach, 1033 .pru_disconnect = tcp_usr_disconnect, 1034 .pru_listen = tcp_usr_listen, 1035 .pru_peeraddr = in_getpeeraddr, 1036 .pru_rcvd = tcp_usr_rcvd, 1037 .pru_rcvoob = tcp_usr_rcvoob, 1038 .pru_send = tcp_usr_send, 1039 .pru_shutdown = tcp_usr_shutdown, 1040 .pru_sockaddr = in_getsockaddr, 1041 .pru_sosetlabel = in_pcbsosetlabel, 1042 .pru_close = tcp_usr_close, 1043}; 1044 1045#ifdef INET6 1046struct pr_usrreqs tcp6_usrreqs = { 1047 .pru_abort = tcp_usr_abort, 1048 .pru_accept = tcp6_usr_accept, 1049 .pru_attach = tcp_usr_attach, 1050 .pru_bind = tcp6_usr_bind, 1051 .pru_connect = tcp6_usr_connect, 1052 .pru_control = in6_control, 1053 .pru_detach = tcp_usr_detach, 1054 .pru_disconnect = tcp_usr_disconnect, 1055 .pru_listen = tcp6_usr_listen, 1056 .pru_peeraddr = in6_mapped_peeraddr, 1057 .pru_rcvd = tcp_usr_rcvd, 1058 .pru_rcvoob = tcp_usr_rcvoob, 1059 .pru_send = tcp_usr_send, 1060 .pru_shutdown = tcp_usr_shutdown, 1061 .pru_sockaddr = in6_mapped_sockaddr, 1062 .pru_sosetlabel = in_pcbsosetlabel, 1063 .pru_close = tcp_usr_close, 1064}; 1065#endif /* INET6 */ 1066 1067/* 1068 * Common subroutine to open a TCP connection to remote host specified 1069 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1070 * port number if needed. Call in_pcbconnect_setup to do the routing and 1071 * to choose a local host address (interface). If there is an existing 1072 * incarnation of the same connection in TIME-WAIT state and if the remote 1073 * host was sending CC options and if the connection duration was < MSL, then 1074 * truncate the previous TIME-WAIT state and proceed. 1075 * Initialize connection parameters and enter SYN-SENT state. 1076 */ 1077static int 1078tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1079{ 1080 struct inpcb *inp = tp->t_inpcb, *oinp; 1081 struct socket *so = inp->inp_socket; 1082 struct in_addr laddr; 1083 u_short lport; 1084 int error; 1085 1086 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1087 INP_LOCK_ASSERT(inp); 1088 1089 if (inp->inp_lport == 0) { 1090 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1091 if (error) 1092 return error; 1093 } 1094 1095 /* 1096 * Cannot simply call in_pcbconnect, because there might be an 1097 * earlier incarnation of this same connection still in 1098 * TIME_WAIT state, creating an ADDRINUSE error. 1099 */ 1100 laddr = inp->inp_laddr; 1101 lport = inp->inp_lport; 1102 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1103 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1104 if (error && oinp == NULL) 1105 return error; 1106 if (oinp) 1107 return EADDRINUSE; 1108 inp->inp_laddr = laddr; 1109 in_pcbrehash(inp); 1110 1111 /* 1112 * Compute window scaling to request: 1113 * Scale to fit into sweet spot. See tcp_syncache.c. 1114 * XXX: This should move to tcp_output(). 1115 */ 1116 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1117 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1118 tp->request_r_scale++; 1119 1120 soisconnecting(so); 1121 tcpstat.tcps_connattempt++; 1122 tp->t_state = TCPS_SYN_SENT; 1123 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1124 tp->iss = tcp_new_isn(tp); 1125 tp->t_bw_rtseq = tp->iss; 1126 tcp_sendseqinit(tp); 1127 1128 return 0; 1129} 1130 1131#ifdef INET6 1132static int 1133tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1134{ 1135 struct inpcb *inp = tp->t_inpcb, *oinp; 1136 struct socket *so = inp->inp_socket; 1137 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1138 struct in6_addr *addr6; 1139 int error; 1140 1141 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1142 INP_LOCK_ASSERT(inp); 1143 1144 if (inp->inp_lport == 0) { 1145 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1146 if (error) 1147 return error; 1148 } 1149 1150 /* 1151 * Cannot simply call in_pcbconnect, because there might be an 1152 * earlier incarnation of this same connection still in 1153 * TIME_WAIT state, creating an ADDRINUSE error. 1154 * in6_pcbladdr() also handles scope zone IDs. 1155 */ 1156 error = in6_pcbladdr(inp, nam, &addr6); 1157 if (error) 1158 return error; 1159 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1160 &sin6->sin6_addr, sin6->sin6_port, 1161 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1162 ? addr6 1163 : &inp->in6p_laddr, 1164 inp->inp_lport, 0, NULL); 1165 if (oinp) 1166 return EADDRINUSE; 1167 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1168 inp->in6p_laddr = *addr6; 1169 inp->in6p_faddr = sin6->sin6_addr; 1170 inp->inp_fport = sin6->sin6_port; 1171 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1172 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 1173 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 1174 inp->in6p_flowinfo |= 1175 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1176 in_pcbrehash(inp); 1177 1178 /* Compute window scaling to request. */ 1179 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1180 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 1181 tp->request_r_scale++; 1182 1183 soisconnecting(so); 1184 tcpstat.tcps_connattempt++; 1185 tp->t_state = TCPS_SYN_SENT; 1186 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1187 tp->iss = tcp_new_isn(tp); 1188 tp->t_bw_rtseq = tp->iss; 1189 tcp_sendseqinit(tp); 1190 1191 return 0; 1192} 1193#endif /* INET6 */ 1194 1195/* 1196 * Export TCP internal state information via a struct tcp_info, based on the 1197 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1198 * (TCP state machine, etc). We export all information using FreeBSD-native 1199 * constants -- for example, the numeric values for tcpi_state will differ 1200 * from Linux. 1201 */ 1202static void 1203tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1204{ 1205 1206 INP_LOCK_ASSERT(tp->t_inpcb); 1207 bzero(ti, sizeof(*ti)); 1208 1209 ti->tcpi_state = tp->t_state; 1210 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1211 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1212 if (tp->t_flags & TF_SACK_PERMIT) 1213 ti->tcpi_options |= TCPI_OPT_SACK; 1214 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1215 ti->tcpi_options |= TCPI_OPT_WSCALE; 1216 ti->tcpi_snd_wscale = tp->snd_scale; 1217 ti->tcpi_rcv_wscale = tp->rcv_scale; 1218 } 1219 1220 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1221 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1222 1223 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1224 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1225 1226 /* 1227 * FreeBSD-specific extension fields for tcp_info. 1228 */ 1229 ti->tcpi_rcv_space = tp->rcv_wnd; 1230 ti->tcpi_snd_wnd = tp->snd_wnd; 1231 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1232} 1233 1234/* 1235 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1236 * socket option arguments. When it re-acquires the lock after the copy, it 1237 * has to revalidate that the connection is still valid for the socket 1238 * option. 1239 */ 1240#define INP_LOCK_RECHECK(inp) do { \ 1241 INP_LOCK(inp); \ 1242 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { \ 1243 INP_UNLOCK(inp); \ 1244 return (ECONNRESET); \ 1245 } \ 1246 tp = intotcpcb(inp); \ 1247} while(0) 1248 1249int 1250tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1251{ 1252 int error, opt, optval; 1253 struct inpcb *inp; 1254 struct tcpcb *tp; 1255 struct tcp_info ti; 1256 1257 error = 0; 1258 inp = sotoinpcb(so); 1259 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1260 INP_LOCK(inp); 1261 if (sopt->sopt_level != IPPROTO_TCP) { 1262#ifdef INET6 1263 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 1264 INP_UNLOCK(inp); 1265 error = ip6_ctloutput(so, sopt); 1266 } else { 1267#endif /* INET6 */ 1268 INP_UNLOCK(inp); 1269 error = ip_ctloutput(so, sopt); 1270#ifdef INET6 1271 } 1272#endif 1273 return (error); 1274 } 1275 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1276 INP_UNLOCK(inp); 1277 return (ECONNRESET); 1278 } 1279 1280 switch (sopt->sopt_dir) { 1281 case SOPT_SET: 1282 switch (sopt->sopt_name) { 1283#ifdef TCP_SIGNATURE 1284 case TCP_MD5SIG: 1285 INP_UNLOCK(inp); 1286 error = sooptcopyin(sopt, &optval, sizeof optval, 1287 sizeof optval); 1288 if (error) 1289 return (error); 1290 1291 INP_LOCK_RECHECK(inp); 1292 if (optval > 0) 1293 tp->t_flags |= TF_SIGNATURE; 1294 else 1295 tp->t_flags &= ~TF_SIGNATURE; 1296 INP_UNLOCK(inp); 1297 break; 1298#endif /* TCP_SIGNATURE */ 1299 case TCP_NODELAY: 1300 case TCP_NOOPT: 1301 INP_UNLOCK(inp); 1302 error = sooptcopyin(sopt, &optval, sizeof optval, 1303 sizeof optval); 1304 if (error) 1305 return (error); 1306 1307 INP_LOCK_RECHECK(inp); 1308 switch (sopt->sopt_name) { 1309 case TCP_NODELAY: 1310 opt = TF_NODELAY; 1311 break; 1312 case TCP_NOOPT: 1313 opt = TF_NOOPT; 1314 break; 1315 default: 1316 opt = 0; /* dead code to fool gcc */ 1317 break; 1318 } 1319 1320 if (optval) 1321 tp->t_flags |= opt; 1322 else 1323 tp->t_flags &= ~opt; 1324 INP_UNLOCK(inp); 1325 break; 1326 1327 case TCP_NOPUSH: 1328 INP_UNLOCK(inp); 1329 error = sooptcopyin(sopt, &optval, sizeof optval, 1330 sizeof optval); 1331 if (error) 1332 return (error); 1333 1334 INP_LOCK_RECHECK(inp); 1335 if (optval) 1336 tp->t_flags |= TF_NOPUSH; 1337 else { 1338 tp->t_flags &= ~TF_NOPUSH; 1339 error = tcp_output(tp); 1340 } 1341 INP_UNLOCK(inp); 1342 break; 1343 1344 case TCP_MAXSEG: 1345 INP_UNLOCK(inp); 1346 error = sooptcopyin(sopt, &optval, sizeof optval, 1347 sizeof optval); 1348 if (error) 1349 return (error); 1350 1351 INP_LOCK_RECHECK(inp); 1352 if (optval > 0 && optval <= tp->t_maxseg && 1353 optval + 40 >= tcp_minmss) 1354 tp->t_maxseg = optval; 1355 else 1356 error = EINVAL; 1357 INP_UNLOCK(inp); 1358 break; 1359 1360 case TCP_INFO: 1361 INP_UNLOCK(inp); 1362 error = EINVAL; 1363 break; 1364 1365 default: 1366 INP_UNLOCK(inp); 1367 error = ENOPROTOOPT; 1368 break; 1369 } 1370 break; 1371 1372 case SOPT_GET: 1373 tp = intotcpcb(inp); 1374 switch (sopt->sopt_name) { 1375#ifdef TCP_SIGNATURE 1376 case TCP_MD5SIG: 1377 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1378 INP_UNLOCK(inp); 1379 error = sooptcopyout(sopt, &optval, sizeof optval); 1380 break; 1381#endif 1382 1383 case TCP_NODELAY: 1384 optval = tp->t_flags & TF_NODELAY; 1385 INP_UNLOCK(inp); 1386 error = sooptcopyout(sopt, &optval, sizeof optval); 1387 break; 1388 case TCP_MAXSEG: 1389 optval = tp->t_maxseg; 1390 INP_UNLOCK(inp); 1391 error = sooptcopyout(sopt, &optval, sizeof optval); 1392 break; 1393 case TCP_NOOPT: 1394 optval = tp->t_flags & TF_NOOPT; 1395 INP_UNLOCK(inp); 1396 error = sooptcopyout(sopt, &optval, sizeof optval); 1397 break; 1398 case TCP_NOPUSH: 1399 optval = tp->t_flags & TF_NOPUSH; 1400 INP_UNLOCK(inp); 1401 error = sooptcopyout(sopt, &optval, sizeof optval); 1402 break; 1403 case TCP_INFO: 1404 tcp_fill_info(tp, &ti); 1405 INP_UNLOCK(inp); 1406 error = sooptcopyout(sopt, &ti, sizeof ti); 1407 break; 1408 default: 1409 INP_UNLOCK(inp); 1410 error = ENOPROTOOPT; 1411 break; 1412 } 1413 break; 1414 } 1415 return (error); 1416} 1417#undef INP_LOCK_RECHECK 1418 1419/* 1420 * tcp_sendspace and tcp_recvspace are the default send and receive window 1421 * sizes, respectively. These are obsolescent (this information should 1422 * be set by the route). 1423 */ 1424u_long tcp_sendspace = 1024*32; 1425SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1426 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1427u_long tcp_recvspace = 1024*64; 1428SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1429 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1430 1431/* 1432 * Attach TCP protocol to socket, allocating 1433 * internet protocol control block, tcp control block, 1434 * bufer space, and entering LISTEN state if to accept connections. 1435 */ 1436static int 1437tcp_attach(struct socket *so) 1438{ 1439 struct tcpcb *tp; 1440 struct inpcb *inp; 1441 int error; 1442#ifdef INET6 1443 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1444#endif 1445 1446 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1447 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1448 if (error) 1449 return (error); 1450 } 1451 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1452 so->so_snd.sb_flags |= SB_AUTOSIZE; 1453 INP_INFO_WLOCK(&tcbinfo); 1454 error = in_pcballoc(so, &tcbinfo); 1455 if (error) { 1456 INP_INFO_WUNLOCK(&tcbinfo); 1457 return (error); 1458 } 1459 inp = sotoinpcb(so); 1460#ifdef INET6 1461 if (isipv6) { 1462 inp->inp_vflag |= INP_IPV6; 1463 inp->in6p_hops = -1; /* use kernel default */ 1464 } 1465 else 1466#endif 1467 inp->inp_vflag |= INP_IPV4; 1468 tp = tcp_newtcpcb(inp); 1469 if (tp == NULL) { 1470#ifdef INET6 1471 if (isipv6) { 1472 in6_pcbdetach(inp); 1473 in6_pcbfree(inp); 1474 } else { 1475#endif 1476 in_pcbdetach(inp); 1477 in_pcbfree(inp); 1478#ifdef INET6 1479 } 1480#endif 1481 INP_INFO_WUNLOCK(&tcbinfo); 1482 return (ENOBUFS); 1483 } 1484 tp->t_state = TCPS_CLOSED; 1485 INP_UNLOCK(inp); 1486 INP_INFO_WUNLOCK(&tcbinfo); 1487 return (0); 1488} 1489 1490/* 1491 * Initiate (or continue) disconnect. 1492 * If embryonic state, just send reset (once). 1493 * If in ``let data drain'' option and linger null, just drop. 1494 * Otherwise (hard), mark socket disconnecting and drop 1495 * current input data; switch states based on user close, and 1496 * send segment to peer (with FIN). 1497 */ 1498static void 1499tcp_disconnect(struct tcpcb *tp) 1500{ 1501 struct inpcb *inp = tp->t_inpcb; 1502 struct socket *so = inp->inp_socket; 1503 1504 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1505 INP_LOCK_ASSERT(inp); 1506 1507 /* 1508 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1509 * socket is still open. 1510 */ 1511 if (tp->t_state < TCPS_ESTABLISHED) { 1512 tp = tcp_close(tp); 1513 KASSERT(tp != NULL, 1514 ("tcp_disconnect: tcp_close() returned NULL")); 1515 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1516 tp = tcp_drop(tp, 0); 1517 KASSERT(tp != NULL, 1518 ("tcp_disconnect: tcp_drop() returned NULL")); 1519 } else { 1520 soisdisconnecting(so); 1521 sbflush(&so->so_rcv); 1522 tcp_usrclosed(tp); 1523 if (!(inp->inp_vflag & INP_DROPPED)) 1524 tcp_output_disconnect(tp); 1525 } 1526} 1527 1528/* 1529 * User issued close, and wish to trail through shutdown states: 1530 * if never received SYN, just forget it. If got a SYN from peer, 1531 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1532 * If already got a FIN from peer, then almost done; go to LAST_ACK 1533 * state. In all other cases, have already sent FIN to peer (e.g. 1534 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1535 * for peer to send FIN or not respond to keep-alives, etc. 1536 * We can let the user exit from the close as soon as the FIN is acked. 1537 */ 1538static void 1539tcp_usrclosed(struct tcpcb *tp) 1540{ 1541 1542 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1543 INP_LOCK_ASSERT(tp->t_inpcb); 1544 1545 switch (tp->t_state) { 1546 case TCPS_LISTEN: 1547 tcp_offload_listen_close(tp); 1548 /* FALLTHROUGH */ 1549 case TCPS_CLOSED: 1550 tp->t_state = TCPS_CLOSED; 1551 tp = tcp_close(tp); 1552 /* 1553 * tcp_close() should never return NULL here as the socket is 1554 * still open. 1555 */ 1556 KASSERT(tp != NULL, 1557 ("tcp_usrclosed: tcp_close() returned NULL")); 1558 break; 1559 1560 case TCPS_SYN_SENT: 1561 case TCPS_SYN_RECEIVED: 1562 tp->t_flags |= TF_NEEDFIN; 1563 break; 1564 1565 case TCPS_ESTABLISHED: 1566 tp->t_state = TCPS_FIN_WAIT_1; 1567 break; 1568 1569 case TCPS_CLOSE_WAIT: 1570 tp->t_state = TCPS_LAST_ACK; 1571 break; 1572 } 1573 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1574 soisdisconnected(tp->t_inpcb->inp_socket); 1575 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1576 if (tp->t_state == TCPS_FIN_WAIT_2) { 1577 int timeout; 1578 1579 timeout = (tcp_fast_finwait2_recycle) ? 1580 tcp_finwait2_timeout : tcp_maxidle; 1581 tcp_timer_activate(tp, TT_2MSL, timeout); 1582 } 1583 } 1584} 1585 1586#ifdef DDB 1587static void 1588db_print_indent(int indent) 1589{ 1590 int i; 1591 1592 for (i = 0; i < indent; i++) 1593 db_printf(" "); 1594} 1595 1596static void 1597db_print_tstate(int t_state) 1598{ 1599 1600 switch (t_state) { 1601 case TCPS_CLOSED: 1602 db_printf("TCPS_CLOSED"); 1603 return; 1604 1605 case TCPS_LISTEN: 1606 db_printf("TCPS_LISTEN"); 1607 return; 1608 1609 case TCPS_SYN_SENT: 1610 db_printf("TCPS_SYN_SENT"); 1611 return; 1612 1613 case TCPS_SYN_RECEIVED: 1614 db_printf("TCPS_SYN_RECEIVED"); 1615 return; 1616 1617 case TCPS_ESTABLISHED: 1618 db_printf("TCPS_ESTABLISHED"); 1619 return; 1620 1621 case TCPS_CLOSE_WAIT: 1622 db_printf("TCPS_CLOSE_WAIT"); 1623 return; 1624 1625 case TCPS_FIN_WAIT_1: 1626 db_printf("TCPS_FIN_WAIT_1"); 1627 return; 1628 1629 case TCPS_CLOSING: 1630 db_printf("TCPS_CLOSING"); 1631 return; 1632 1633 case TCPS_LAST_ACK: 1634 db_printf("TCPS_LAST_ACK"); 1635 return; 1636 1637 case TCPS_FIN_WAIT_2: 1638 db_printf("TCPS_FIN_WAIT_2"); 1639 return; 1640 1641 case TCPS_TIME_WAIT: 1642 db_printf("TCPS_TIME_WAIT"); 1643 return; 1644 1645 default: 1646 db_printf("unknown"); 1647 return; 1648 } 1649} 1650 1651static void 1652db_print_tflags(u_int t_flags) 1653{ 1654 int comma; 1655 1656 comma = 0; 1657 if (t_flags & TF_ACKNOW) { 1658 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1659 comma = 1; 1660 } 1661 if (t_flags & TF_DELACK) { 1662 db_printf("%sTF_DELACK", comma ? ", " : ""); 1663 comma = 1; 1664 } 1665 if (t_flags & TF_NODELAY) { 1666 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1667 comma = 1; 1668 } 1669 if (t_flags & TF_NOOPT) { 1670 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1671 comma = 1; 1672 } 1673 if (t_flags & TF_SENTFIN) { 1674 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1675 comma = 1; 1676 } 1677 if (t_flags & TF_REQ_SCALE) { 1678 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1679 comma = 1; 1680 } 1681 if (t_flags & TF_RCVD_SCALE) { 1682 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1683 comma = 1; 1684 } 1685 if (t_flags & TF_REQ_TSTMP) { 1686 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1687 comma = 1; 1688 } 1689 if (t_flags & TF_RCVD_TSTMP) { 1690 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1691 comma = 1; 1692 } 1693 if (t_flags & TF_SACK_PERMIT) { 1694 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1695 comma = 1; 1696 } 1697 if (t_flags & TF_NEEDSYN) { 1698 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1699 comma = 1; 1700 } 1701 if (t_flags & TF_NEEDFIN) { 1702 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1703 comma = 1; 1704 } 1705 if (t_flags & TF_NOPUSH) { 1706 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1707 comma = 1; 1708 } 1709 if (t_flags & TF_NOPUSH) { 1710 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1711 comma = 1; 1712 } 1713 if (t_flags & TF_MORETOCOME) { 1714 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1715 comma = 1; 1716 } 1717 if (t_flags & TF_LQ_OVERFLOW) { 1718 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1719 comma = 1; 1720 } 1721 if (t_flags & TF_LASTIDLE) { 1722 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1723 comma = 1; 1724 } 1725 if (t_flags & TF_RXWIN0SENT) { 1726 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1727 comma = 1; 1728 } 1729 if (t_flags & TF_FASTRECOVERY) { 1730 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1731 comma = 1; 1732 } 1733 if (t_flags & TF_WASFRECOVERY) { 1734 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1735 comma = 1; 1736 } 1737 if (t_flags & TF_SIGNATURE) { 1738 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1739 comma = 1; 1740 } 1741 if (t_flags & TF_FORCEDATA) { 1742 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1743 comma = 1; 1744 } 1745 if (t_flags & TF_TSO) { 1746 db_printf("%sTF_TSO", comma ? ", " : ""); 1747 comma = 1; 1748 } 1749} 1750 1751static void 1752db_print_toobflags(char t_oobflags) 1753{ 1754 int comma; 1755 1756 comma = 0; 1757 if (t_oobflags & TCPOOB_HAVEDATA) { 1758 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1759 comma = 1; 1760 } 1761 if (t_oobflags & TCPOOB_HADDATA) { 1762 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1763 comma = 1; 1764 } 1765} 1766 1767static void 1768db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1769{ 1770 1771 db_print_indent(indent); 1772 db_printf("%s at %p\n", name, tp); 1773 1774 indent += 2; 1775 1776 db_print_indent(indent); 1777 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1778 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1779 1780 db_print_indent(indent); 1781 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1782 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1783 1784 db_print_indent(indent); 1785 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1786 &tp->t_timers->tt_delack, tp->t_inpcb); 1787 1788 db_print_indent(indent); 1789 db_printf("t_state: %d (", tp->t_state); 1790 db_print_tstate(tp->t_state); 1791 db_printf(")\n"); 1792 1793 db_print_indent(indent); 1794 db_printf("t_flags: 0x%x (", tp->t_flags); 1795 db_print_tflags(tp->t_flags); 1796 db_printf(")\n"); 1797 1798 db_print_indent(indent); 1799 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1800 tp->snd_una, tp->snd_max, tp->snd_nxt); 1801 1802 db_print_indent(indent); 1803 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1804 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1805 1806 db_print_indent(indent); 1807 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1808 tp->iss, tp->irs, tp->rcv_nxt); 1809 1810 db_print_indent(indent); 1811 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1812 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1813 1814 db_print_indent(indent); 1815 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1816 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1817 1818 db_print_indent(indent); 1819 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1820 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1821 tp->snd_recover); 1822 1823 db_print_indent(indent); 1824 db_printf("t_maxopd: %u t_rcvtime: %lu t_startime: %lu\n", 1825 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1826 1827 db_print_indent(indent); 1828 db_printf("t_rttime: %d t_rtsq: 0x%08x t_bw_rtttime: %d\n", 1829 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1830 1831 db_print_indent(indent); 1832 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1833 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1834 tp->t_srtt); 1835 1836 db_print_indent(indent); 1837 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1838 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1839 tp->t_rttbest); 1840 1841 db_print_indent(indent); 1842 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1843 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1844 1845 db_print_indent(indent); 1846 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1847 db_print_toobflags(tp->t_oobflags); 1848 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1849 1850 db_print_indent(indent); 1851 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1852 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1853 1854 db_print_indent(indent); 1855 db_printf("ts_recent: %u ts_recent_age: %lu\n", 1856 tp->ts_recent, tp->ts_recent_age); 1857 1858 db_print_indent(indent); 1859 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1860 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1861 1862 db_print_indent(indent); 1863 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1864 "t_badrxtwin: %lu\n", tp->snd_ssthresh_prev, 1865 tp->snd_recover_prev, tp->t_badrxtwin); 1866 1867 db_print_indent(indent); 1868 db_printf("snd_numholes: %d snd_holes first: %p\n", 1869 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1870 1871 db_print_indent(indent); 1872 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1873 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1874 1875 /* Skip sackblks, sackhint. */ 1876 1877 db_print_indent(indent); 1878 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1879 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1880} 1881 1882DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1883{ 1884 struct tcpcb *tp; 1885 1886 if (!have_addr) { 1887 db_printf("usage: show tcpcb <addr>\n"); 1888 return; 1889 } 1890 tp = (struct tcpcb *)addr; 1891 1892 db_print_tcpcb(tp, "tcpcb", 0); 1893} 1894#endif 1895