tcp_usrreq.c revision 196019
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 196019 2009-08-01 19:26:27Z 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#include <net/vnet.h> 64 65#include <netinet/in.h> 66#include <netinet/in_systm.h> 67#ifdef INET6 68#include <netinet/ip6.h> 69#endif 70#include <netinet/in_pcb.h> 71#ifdef INET6 72#include <netinet6/in6_pcb.h> 73#endif 74#include <netinet/in_var.h> 75#include <netinet/ip_var.h> 76#ifdef INET6 77#include <netinet6/ip6_var.h> 78#include <netinet6/scope6_var.h> 79#endif 80#include <netinet/tcp.h> 81#include <netinet/tcp_fsm.h> 82#include <netinet/tcp_seq.h> 83#include <netinet/tcp_timer.h> 84#include <netinet/tcp_var.h> 85#include <netinet/tcpip.h> 86#ifdef TCPDEBUG 87#include <netinet/tcp_debug.h> 88#endif 89#include <netinet/tcp_offload.h> 90 91/* 92 * TCP protocol interface to socket abstraction. 93 */ 94static int tcp_attach(struct socket *); 95static int tcp_connect(struct tcpcb *, struct sockaddr *, 96 struct thread *td); 97#ifdef INET6 98static int tcp6_connect(struct tcpcb *, struct sockaddr *, 99 struct thread *td); 100#endif /* INET6 */ 101static void tcp_disconnect(struct tcpcb *); 102static void tcp_usrclosed(struct tcpcb *); 103static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 104 105#ifdef TCPDEBUG 106#define TCPDEBUG0 int ostate = 0 107#define TCPDEBUG1() ostate = tp ? tp->t_state : 0 108#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 109 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 110#else 111#define TCPDEBUG0 112#define TCPDEBUG1() 113#define TCPDEBUG2(req) 114#endif 115 116/* 117 * TCP attaches to socket via pru_attach(), reserving space, 118 * and an internet control block. 119 */ 120static int 121tcp_usr_attach(struct socket *so, int proto, struct thread *td) 122{ 123 struct inpcb *inp; 124 struct tcpcb *tp = NULL; 125 int error; 126 TCPDEBUG0; 127 128 inp = sotoinpcb(so); 129 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 130 TCPDEBUG1(); 131 132 error = tcp_attach(so); 133 if (error) 134 goto out; 135 136 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 137 so->so_linger = TCP_LINGERTIME; 138 139 inp = sotoinpcb(so); 140 tp = intotcpcb(inp); 141out: 142 TCPDEBUG2(PRU_ATTACH); 143 return error; 144} 145 146/* 147 * tcp_detach is called when the socket layer loses its final reference 148 * to the socket, be it a file descriptor reference, a reference from TCP, 149 * etc. At this point, there is only one case in which we will keep around 150 * inpcb state: time wait. 151 * 152 * This function can probably be re-absorbed back into tcp_usr_detach() now 153 * that there is a single detach path. 154 */ 155static void 156tcp_detach(struct socket *so, struct inpcb *inp) 157{ 158 struct tcpcb *tp; 159 160 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 161 INP_WLOCK_ASSERT(inp); 162 163 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 164 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 165 166 tp = intotcpcb(inp); 167 168 if (inp->inp_flags & INP_TIMEWAIT) { 169 /* 170 * There are two cases to handle: one in which the time wait 171 * state is being discarded (INP_DROPPED), and one in which 172 * this connection will remain in timewait. In the former, 173 * it is time to discard all state (except tcptw, which has 174 * already been discarded by the timewait close code, which 175 * should be further up the call stack somewhere). In the 176 * latter case, we detach from the socket, but leave the pcb 177 * present until timewait ends. 178 * 179 * XXXRW: Would it be cleaner to free the tcptw here? 180 */ 181 if (inp->inp_flags & INP_DROPPED) { 182 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 183 "INP_DROPPED && tp != NULL")); 184 in_pcbdetach(inp); 185 in_pcbfree(inp); 186 } else { 187 in_pcbdetach(inp); 188 INP_WUNLOCK(inp); 189 } 190 } else { 191 /* 192 * If the connection is not in timewait, we consider two 193 * two conditions: one in which no further processing is 194 * necessary (dropped || embryonic), and one in which TCP is 195 * not yet done, but no longer requires the socket, so the 196 * pcb will persist for the time being. 197 * 198 * XXXRW: Does the second case still occur? 199 */ 200 if (inp->inp_flags & INP_DROPPED || 201 tp->t_state < TCPS_SYN_SENT) { 202 tcp_discardcb(tp); 203 in_pcbdetach(inp); 204 in_pcbfree(inp); 205 } else 206 in_pcbdetach(inp); 207 } 208} 209 210/* 211 * pru_detach() detaches the TCP protocol from the socket. 212 * If the protocol state is non-embryonic, then can't 213 * do this directly: have to initiate a pru_disconnect(), 214 * which may finish later; embryonic TCB's can just 215 * be discarded here. 216 */ 217static void 218tcp_usr_detach(struct socket *so) 219{ 220 struct inpcb *inp; 221 222 inp = sotoinpcb(so); 223 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 224 INP_INFO_WLOCK(&V_tcbinfo); 225 INP_WLOCK(inp); 226 KASSERT(inp->inp_socket != NULL, 227 ("tcp_usr_detach: inp_socket == NULL")); 228 tcp_detach(so, inp); 229 INP_INFO_WUNLOCK(&V_tcbinfo); 230} 231 232/* 233 * Give the socket an address. 234 */ 235static int 236tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 237{ 238 int error = 0; 239 struct inpcb *inp; 240 struct tcpcb *tp = NULL; 241 struct sockaddr_in *sinp; 242 243 sinp = (struct sockaddr_in *)nam; 244 if (nam->sa_len != sizeof (*sinp)) 245 return (EINVAL); 246 /* 247 * Must check for multicast addresses and disallow binding 248 * to them. 249 */ 250 if (sinp->sin_family == AF_INET && 251 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 252 return (EAFNOSUPPORT); 253 254 TCPDEBUG0; 255 INP_INFO_WLOCK(&V_tcbinfo); 256 inp = sotoinpcb(so); 257 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 258 INP_WLOCK(inp); 259 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 260 error = EINVAL; 261 goto out; 262 } 263 tp = intotcpcb(inp); 264 TCPDEBUG1(); 265 error = in_pcbbind(inp, nam, td->td_ucred); 266out: 267 TCPDEBUG2(PRU_BIND); 268 INP_WUNLOCK(inp); 269 INP_INFO_WUNLOCK(&V_tcbinfo); 270 271 return (error); 272} 273 274#ifdef INET6 275static int 276tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 277{ 278 int error = 0; 279 struct inpcb *inp; 280 struct tcpcb *tp = NULL; 281 struct sockaddr_in6 *sin6p; 282 283 sin6p = (struct sockaddr_in6 *)nam; 284 if (nam->sa_len != sizeof (*sin6p)) 285 return (EINVAL); 286 /* 287 * Must check for multicast addresses and disallow binding 288 * to them. 289 */ 290 if (sin6p->sin6_family == AF_INET6 && 291 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 292 return (EAFNOSUPPORT); 293 294 TCPDEBUG0; 295 INP_INFO_WLOCK(&V_tcbinfo); 296 inp = sotoinpcb(so); 297 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 298 INP_WLOCK(inp); 299 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 300 error = EINVAL; 301 goto out; 302 } 303 tp = intotcpcb(inp); 304 TCPDEBUG1(); 305 inp->inp_vflag &= ~INP_IPV4; 306 inp->inp_vflag |= INP_IPV6; 307 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 308 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 309 inp->inp_vflag |= INP_IPV4; 310 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 311 struct sockaddr_in sin; 312 313 in6_sin6_2_sin(&sin, sin6p); 314 inp->inp_vflag |= INP_IPV4; 315 inp->inp_vflag &= ~INP_IPV6; 316 error = in_pcbbind(inp, (struct sockaddr *)&sin, 317 td->td_ucred); 318 goto out; 319 } 320 } 321 error = in6_pcbbind(inp, nam, td->td_ucred); 322out: 323 TCPDEBUG2(PRU_BIND); 324 INP_WUNLOCK(inp); 325 INP_INFO_WUNLOCK(&V_tcbinfo); 326 return (error); 327} 328#endif /* INET6 */ 329 330/* 331 * Prepare to accept connections. 332 */ 333static int 334tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 335{ 336 int error = 0; 337 struct inpcb *inp; 338 struct tcpcb *tp = NULL; 339 340 TCPDEBUG0; 341 INP_INFO_WLOCK(&V_tcbinfo); 342 inp = sotoinpcb(so); 343 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 344 INP_WLOCK(inp); 345 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 346 error = EINVAL; 347 goto out; 348 } 349 tp = intotcpcb(inp); 350 TCPDEBUG1(); 351 SOCK_LOCK(so); 352 error = solisten_proto_check(so); 353 if (error == 0 && inp->inp_lport == 0) 354 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 355 if (error == 0) { 356 tp->t_state = TCPS_LISTEN; 357 solisten_proto(so, backlog); 358 tcp_offload_listen_open(tp); 359 } 360 SOCK_UNLOCK(so); 361 362out: 363 TCPDEBUG2(PRU_LISTEN); 364 INP_WUNLOCK(inp); 365 INP_INFO_WUNLOCK(&V_tcbinfo); 366 return (error); 367} 368 369#ifdef INET6 370static int 371tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 372{ 373 int error = 0; 374 struct inpcb *inp; 375 struct tcpcb *tp = NULL; 376 377 TCPDEBUG0; 378 INP_INFO_WLOCK(&V_tcbinfo); 379 inp = sotoinpcb(so); 380 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 381 INP_WLOCK(inp); 382 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 383 error = EINVAL; 384 goto out; 385 } 386 tp = intotcpcb(inp); 387 TCPDEBUG1(); 388 SOCK_LOCK(so); 389 error = solisten_proto_check(so); 390 if (error == 0 && inp->inp_lport == 0) { 391 inp->inp_vflag &= ~INP_IPV4; 392 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 393 inp->inp_vflag |= INP_IPV4; 394 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 395 } 396 if (error == 0) { 397 tp->t_state = TCPS_LISTEN; 398 solisten_proto(so, backlog); 399 } 400 SOCK_UNLOCK(so); 401 402out: 403 TCPDEBUG2(PRU_LISTEN); 404 INP_WUNLOCK(inp); 405 INP_INFO_WUNLOCK(&V_tcbinfo); 406 return (error); 407} 408#endif /* INET6 */ 409 410/* 411 * Initiate connection to peer. 412 * Create a template for use in transmissions on this connection. 413 * Enter SYN_SENT state, and mark socket as connecting. 414 * Start keep-alive timer, and seed output sequence space. 415 * Send initial segment on connection. 416 */ 417static int 418tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 419{ 420 int error = 0; 421 struct inpcb *inp; 422 struct tcpcb *tp = NULL; 423 struct sockaddr_in *sinp; 424 425 sinp = (struct sockaddr_in *)nam; 426 if (nam->sa_len != sizeof (*sinp)) 427 return (EINVAL); 428 /* 429 * Must disallow TCP ``connections'' to multicast addresses. 430 */ 431 if (sinp->sin_family == AF_INET 432 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 433 return (EAFNOSUPPORT); 434 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 435 return (error); 436 437 TCPDEBUG0; 438 INP_INFO_WLOCK(&V_tcbinfo); 439 inp = sotoinpcb(so); 440 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 441 INP_WLOCK(inp); 442 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 443 error = EINVAL; 444 goto out; 445 } 446 tp = intotcpcb(inp); 447 TCPDEBUG1(); 448 if ((error = tcp_connect(tp, nam, td)) != 0) 449 goto out; 450 error = tcp_output_connect(so, nam); 451out: 452 TCPDEBUG2(PRU_CONNECT); 453 INP_WUNLOCK(inp); 454 INP_INFO_WUNLOCK(&V_tcbinfo); 455 return (error); 456} 457 458#ifdef INET6 459static int 460tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 461{ 462 int error = 0; 463 struct inpcb *inp; 464 struct tcpcb *tp = NULL; 465 struct sockaddr_in6 *sin6p; 466 467 TCPDEBUG0; 468 469 sin6p = (struct sockaddr_in6 *)nam; 470 if (nam->sa_len != sizeof (*sin6p)) 471 return (EINVAL); 472 /* 473 * Must disallow TCP ``connections'' to multicast addresses. 474 */ 475 if (sin6p->sin6_family == AF_INET6 476 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 477 return (EAFNOSUPPORT); 478 479 INP_INFO_WLOCK(&V_tcbinfo); 480 inp = sotoinpcb(so); 481 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 482 INP_WLOCK(inp); 483 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 484 error = EINVAL; 485 goto out; 486 } 487 tp = intotcpcb(inp); 488 TCPDEBUG1(); 489 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 490 struct sockaddr_in sin; 491 492 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 493 error = EINVAL; 494 goto out; 495 } 496 497 in6_sin6_2_sin(&sin, sin6p); 498 inp->inp_vflag |= INP_IPV4; 499 inp->inp_vflag &= ~INP_IPV6; 500 if ((error = prison_remote_ip4(td->td_ucred, 501 &sin.sin_addr)) != 0) 502 goto out; 503 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 504 goto out; 505 error = tcp_output_connect(so, nam); 506 goto out; 507 } 508 inp->inp_vflag &= ~INP_IPV4; 509 inp->inp_vflag |= INP_IPV6; 510 inp->inp_inc.inc_flags |= INC_ISIPV6; 511 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0) 512 goto out; 513 if ((error = tcp6_connect(tp, nam, td)) != 0) 514 goto out; 515 error = tcp_output_connect(so, nam); 516 517out: 518 TCPDEBUG2(PRU_CONNECT); 519 INP_WUNLOCK(inp); 520 INP_INFO_WUNLOCK(&V_tcbinfo); 521 return (error); 522} 523#endif /* INET6 */ 524 525/* 526 * Initiate disconnect from peer. 527 * If connection never passed embryonic stage, just drop; 528 * else if don't need to let data drain, then can just drop anyways, 529 * else have to begin TCP shutdown process: mark socket disconnecting, 530 * drain unread data, state switch to reflect user close, and 531 * send segment (e.g. FIN) to peer. Socket will be really disconnected 532 * when peer sends FIN and acks ours. 533 * 534 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 535 */ 536static int 537tcp_usr_disconnect(struct socket *so) 538{ 539 struct inpcb *inp; 540 struct tcpcb *tp = NULL; 541 int error = 0; 542 543 TCPDEBUG0; 544 INP_INFO_WLOCK(&V_tcbinfo); 545 inp = sotoinpcb(so); 546 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 547 INP_WLOCK(inp); 548 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 549 error = ECONNRESET; 550 goto out; 551 } 552 tp = intotcpcb(inp); 553 TCPDEBUG1(); 554 tcp_disconnect(tp); 555out: 556 TCPDEBUG2(PRU_DISCONNECT); 557 INP_WUNLOCK(inp); 558 INP_INFO_WUNLOCK(&V_tcbinfo); 559 return (error); 560} 561 562/* 563 * Accept a connection. Essentially all the work is 564 * done at higher levels; just return the address 565 * of the peer, storing through addr. 566 */ 567static int 568tcp_usr_accept(struct socket *so, struct sockaddr **nam) 569{ 570 int error = 0; 571 struct inpcb *inp = NULL; 572 struct tcpcb *tp = NULL; 573 struct in_addr addr; 574 in_port_t port = 0; 575 TCPDEBUG0; 576 577 if (so->so_state & SS_ISDISCONNECTED) 578 return (ECONNABORTED); 579 580 inp = sotoinpcb(so); 581 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 582 INP_INFO_RLOCK(&V_tcbinfo); 583 INP_WLOCK(inp); 584 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 585 error = ECONNABORTED; 586 goto out; 587 } 588 tp = intotcpcb(inp); 589 TCPDEBUG1(); 590 591 /* 592 * We inline in_getpeeraddr and COMMON_END here, so that we can 593 * copy the data of interest and defer the malloc until after we 594 * release the lock. 595 */ 596 port = inp->inp_fport; 597 addr = inp->inp_faddr; 598 599out: 600 TCPDEBUG2(PRU_ACCEPT); 601 INP_WUNLOCK(inp); 602 INP_INFO_RUNLOCK(&V_tcbinfo); 603 if (error == 0) 604 *nam = in_sockaddr(port, &addr); 605 return error; 606} 607 608#ifdef INET6 609static int 610tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 611{ 612 struct inpcb *inp = NULL; 613 int error = 0; 614 struct tcpcb *tp = NULL; 615 struct in_addr addr; 616 struct in6_addr addr6; 617 in_port_t port = 0; 618 int v4 = 0; 619 TCPDEBUG0; 620 621 if (so->so_state & SS_ISDISCONNECTED) 622 return (ECONNABORTED); 623 624 inp = sotoinpcb(so); 625 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 626 INP_WLOCK(inp); 627 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 628 error = ECONNABORTED; 629 goto out; 630 } 631 tp = intotcpcb(inp); 632 TCPDEBUG1(); 633 634 /* 635 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 636 * copy the data of interest and defer the malloc until after we 637 * release the lock. 638 */ 639 if (inp->inp_vflag & INP_IPV4) { 640 v4 = 1; 641 port = inp->inp_fport; 642 addr = inp->inp_faddr; 643 } else { 644 port = inp->inp_fport; 645 addr6 = inp->in6p_faddr; 646 } 647 648out: 649 TCPDEBUG2(PRU_ACCEPT); 650 INP_WUNLOCK(inp); 651 if (error == 0) { 652 if (v4) 653 *nam = in6_v4mapsin6_sockaddr(port, &addr); 654 else 655 *nam = in6_sockaddr(port, &addr6); 656 } 657 return error; 658} 659#endif /* INET6 */ 660 661/* 662 * Mark the connection as being incapable of further output. 663 */ 664static int 665tcp_usr_shutdown(struct socket *so) 666{ 667 int error = 0; 668 struct inpcb *inp; 669 struct tcpcb *tp = NULL; 670 671 TCPDEBUG0; 672 INP_INFO_WLOCK(&V_tcbinfo); 673 inp = sotoinpcb(so); 674 KASSERT(inp != NULL, ("inp == NULL")); 675 INP_WLOCK(inp); 676 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 677 error = ECONNRESET; 678 goto out; 679 } 680 tp = intotcpcb(inp); 681 TCPDEBUG1(); 682 socantsendmore(so); 683 tcp_usrclosed(tp); 684 if (!(inp->inp_flags & INP_DROPPED)) 685 error = tcp_output_disconnect(tp); 686 687out: 688 TCPDEBUG2(PRU_SHUTDOWN); 689 INP_WUNLOCK(inp); 690 INP_INFO_WUNLOCK(&V_tcbinfo); 691 692 return (error); 693} 694 695/* 696 * After a receive, possibly send window update to peer. 697 */ 698static int 699tcp_usr_rcvd(struct socket *so, int flags) 700{ 701 struct inpcb *inp; 702 struct tcpcb *tp = NULL; 703 int error = 0; 704 705 TCPDEBUG0; 706 inp = sotoinpcb(so); 707 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 708 INP_WLOCK(inp); 709 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 710 error = ECONNRESET; 711 goto out; 712 } 713 tp = intotcpcb(inp); 714 TCPDEBUG1(); 715 tcp_output_rcvd(tp); 716 717out: 718 TCPDEBUG2(PRU_RCVD); 719 INP_WUNLOCK(inp); 720 return (error); 721} 722 723/* 724 * Do a send by putting data in output queue and updating urgent 725 * marker if URG set. Possibly send more data. Unlike the other 726 * pru_*() routines, the mbuf chains are our responsibility. We 727 * must either enqueue them or free them. The other pru_* routines 728 * generally are caller-frees. 729 */ 730static int 731tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 732 struct sockaddr *nam, struct mbuf *control, struct thread *td) 733{ 734 int error = 0; 735 struct inpcb *inp; 736 struct tcpcb *tp = NULL; 737 int headlocked = 0; 738#ifdef INET6 739 int isipv6; 740#endif 741 TCPDEBUG0; 742 743 /* 744 * We require the pcbinfo lock in two cases: 745 * 746 * (1) An implied connect is taking place, which can result in 747 * binding IPs and ports and hence modification of the pcb hash 748 * chains. 749 * 750 * (2) PRUS_EOF is set, resulting in explicit close on the send. 751 */ 752 if ((nam != NULL) || (flags & PRUS_EOF)) { 753 INP_INFO_WLOCK(&V_tcbinfo); 754 headlocked = 1; 755 } 756 inp = sotoinpcb(so); 757 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 758 INP_WLOCK(inp); 759 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 760 if (control) 761 m_freem(control); 762 if (m) 763 m_freem(m); 764 error = ECONNRESET; 765 goto out; 766 } 767#ifdef INET6 768 isipv6 = nam && nam->sa_family == AF_INET6; 769#endif /* INET6 */ 770 tp = intotcpcb(inp); 771 TCPDEBUG1(); 772 if (control) { 773 /* TCP doesn't do control messages (rights, creds, etc) */ 774 if (control->m_len) { 775 m_freem(control); 776 if (m) 777 m_freem(m); 778 error = EINVAL; 779 goto out; 780 } 781 m_freem(control); /* empty control, just free it */ 782 } 783 if (!(flags & PRUS_OOB)) { 784 sbappendstream(&so->so_snd, m); 785 if (nam && tp->t_state < TCPS_SYN_SENT) { 786 /* 787 * Do implied connect if not yet connected, 788 * initialize window to default value, and 789 * initialize maxseg/maxopd using peer's cached 790 * MSS. 791 */ 792 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 793#ifdef INET6 794 if (isipv6) 795 error = tcp6_connect(tp, nam, td); 796 else 797#endif /* INET6 */ 798 error = tcp_connect(tp, nam, td); 799 if (error) 800 goto out; 801 tp->snd_wnd = TTCP_CLIENT_SND_WND; 802 tcp_mss(tp, -1); 803 } 804 if (flags & PRUS_EOF) { 805 /* 806 * Close the send side of the connection after 807 * the data is sent. 808 */ 809 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 810 socantsendmore(so); 811 tcp_usrclosed(tp); 812 } 813 if (headlocked) { 814 INP_INFO_WUNLOCK(&V_tcbinfo); 815 headlocked = 0; 816 } 817 if (!(inp->inp_flags & INP_DROPPED)) { 818 if (flags & PRUS_MORETOCOME) 819 tp->t_flags |= TF_MORETOCOME; 820 error = tcp_output_send(tp); 821 if (flags & PRUS_MORETOCOME) 822 tp->t_flags &= ~TF_MORETOCOME; 823 } 824 } else { 825 /* 826 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 827 */ 828 SOCKBUF_LOCK(&so->so_snd); 829 if (sbspace(&so->so_snd) < -512) { 830 SOCKBUF_UNLOCK(&so->so_snd); 831 m_freem(m); 832 error = ENOBUFS; 833 goto out; 834 } 835 /* 836 * According to RFC961 (Assigned Protocols), 837 * the urgent pointer points to the last octet 838 * of urgent data. We continue, however, 839 * to consider it to indicate the first octet 840 * of data past the urgent section. 841 * Otherwise, snd_up should be one lower. 842 */ 843 sbappendstream_locked(&so->so_snd, m); 844 SOCKBUF_UNLOCK(&so->so_snd); 845 if (nam && tp->t_state < TCPS_SYN_SENT) { 846 /* 847 * Do implied connect if not yet connected, 848 * initialize window to default value, and 849 * initialize maxseg/maxopd using peer's cached 850 * MSS. 851 */ 852 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 853#ifdef INET6 854 if (isipv6) 855 error = tcp6_connect(tp, nam, td); 856 else 857#endif /* INET6 */ 858 error = tcp_connect(tp, nam, td); 859 if (error) 860 goto out; 861 tp->snd_wnd = TTCP_CLIENT_SND_WND; 862 tcp_mss(tp, -1); 863 INP_INFO_WUNLOCK(&V_tcbinfo); 864 headlocked = 0; 865 } else if (nam) { 866 INP_INFO_WUNLOCK(&V_tcbinfo); 867 headlocked = 0; 868 } 869 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 870 tp->t_flags |= TF_FORCEDATA; 871 error = tcp_output_send(tp); 872 tp->t_flags &= ~TF_FORCEDATA; 873 } 874out: 875 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 876 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 877 INP_WUNLOCK(inp); 878 if (headlocked) 879 INP_INFO_WUNLOCK(&V_tcbinfo); 880 return (error); 881} 882 883/* 884 * Abort the TCP. Drop the connection abruptly. 885 */ 886static void 887tcp_usr_abort(struct socket *so) 888{ 889 struct inpcb *inp; 890 struct tcpcb *tp = NULL; 891 TCPDEBUG0; 892 893 inp = sotoinpcb(so); 894 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 895 896 INP_INFO_WLOCK(&V_tcbinfo); 897 INP_WLOCK(inp); 898 KASSERT(inp->inp_socket != NULL, 899 ("tcp_usr_abort: inp_socket == NULL")); 900 901 /* 902 * If we still have full TCP state, and we're not dropped, drop. 903 */ 904 if (!(inp->inp_flags & INP_TIMEWAIT) && 905 !(inp->inp_flags & INP_DROPPED)) { 906 tp = intotcpcb(inp); 907 TCPDEBUG1(); 908 tcp_drop(tp, ECONNABORTED); 909 TCPDEBUG2(PRU_ABORT); 910 } 911 if (!(inp->inp_flags & INP_DROPPED)) { 912 SOCK_LOCK(so); 913 so->so_state |= SS_PROTOREF; 914 SOCK_UNLOCK(so); 915 inp->inp_flags |= INP_SOCKREF; 916 } 917 INP_WUNLOCK(inp); 918 INP_INFO_WUNLOCK(&V_tcbinfo); 919} 920 921/* 922 * TCP socket is closed. Start friendly disconnect. 923 */ 924static void 925tcp_usr_close(struct socket *so) 926{ 927 struct inpcb *inp; 928 struct tcpcb *tp = NULL; 929 TCPDEBUG0; 930 931 inp = sotoinpcb(so); 932 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 933 934 INP_INFO_WLOCK(&V_tcbinfo); 935 INP_WLOCK(inp); 936 KASSERT(inp->inp_socket != NULL, 937 ("tcp_usr_close: inp_socket == NULL")); 938 939 /* 940 * If we still have full TCP state, and we're not dropped, initiate 941 * a disconnect. 942 */ 943 if (!(inp->inp_flags & INP_TIMEWAIT) && 944 !(inp->inp_flags & INP_DROPPED)) { 945 tp = intotcpcb(inp); 946 TCPDEBUG1(); 947 tcp_disconnect(tp); 948 TCPDEBUG2(PRU_CLOSE); 949 } 950 if (!(inp->inp_flags & INP_DROPPED)) { 951 SOCK_LOCK(so); 952 so->so_state |= SS_PROTOREF; 953 SOCK_UNLOCK(so); 954 inp->inp_flags |= INP_SOCKREF; 955 } 956 INP_WUNLOCK(inp); 957 INP_INFO_WUNLOCK(&V_tcbinfo); 958} 959 960/* 961 * Receive out-of-band data. 962 */ 963static int 964tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 965{ 966 int error = 0; 967 struct inpcb *inp; 968 struct tcpcb *tp = NULL; 969 970 TCPDEBUG0; 971 inp = sotoinpcb(so); 972 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 973 INP_WLOCK(inp); 974 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 975 error = ECONNRESET; 976 goto out; 977 } 978 tp = intotcpcb(inp); 979 TCPDEBUG1(); 980 if ((so->so_oobmark == 0 && 981 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 982 so->so_options & SO_OOBINLINE || 983 tp->t_oobflags & TCPOOB_HADDATA) { 984 error = EINVAL; 985 goto out; 986 } 987 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 988 error = EWOULDBLOCK; 989 goto out; 990 } 991 m->m_len = 1; 992 *mtod(m, caddr_t) = tp->t_iobc; 993 if ((flags & MSG_PEEK) == 0) 994 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 995 996out: 997 TCPDEBUG2(PRU_RCVOOB); 998 INP_WUNLOCK(inp); 999 return (error); 1000} 1001 1002struct pr_usrreqs tcp_usrreqs = { 1003 .pru_abort = tcp_usr_abort, 1004 .pru_accept = tcp_usr_accept, 1005 .pru_attach = tcp_usr_attach, 1006 .pru_bind = tcp_usr_bind, 1007 .pru_connect = tcp_usr_connect, 1008 .pru_control = in_control, 1009 .pru_detach = tcp_usr_detach, 1010 .pru_disconnect = tcp_usr_disconnect, 1011 .pru_listen = tcp_usr_listen, 1012 .pru_peeraddr = in_getpeeraddr, 1013 .pru_rcvd = tcp_usr_rcvd, 1014 .pru_rcvoob = tcp_usr_rcvoob, 1015 .pru_send = tcp_usr_send, 1016 .pru_shutdown = tcp_usr_shutdown, 1017 .pru_sockaddr = in_getsockaddr, 1018#if 0 1019 .pru_soreceive = soreceive_stream, 1020#endif 1021 .pru_sosetlabel = in_pcbsosetlabel, 1022 .pru_close = tcp_usr_close, 1023}; 1024 1025#ifdef INET6 1026struct pr_usrreqs tcp6_usrreqs = { 1027 .pru_abort = tcp_usr_abort, 1028 .pru_accept = tcp6_usr_accept, 1029 .pru_attach = tcp_usr_attach, 1030 .pru_bind = tcp6_usr_bind, 1031 .pru_connect = tcp6_usr_connect, 1032 .pru_control = in6_control, 1033 .pru_detach = tcp_usr_detach, 1034 .pru_disconnect = tcp_usr_disconnect, 1035 .pru_listen = tcp6_usr_listen, 1036 .pru_peeraddr = in6_mapped_peeraddr, 1037 .pru_rcvd = tcp_usr_rcvd, 1038 .pru_rcvoob = tcp_usr_rcvoob, 1039 .pru_send = tcp_usr_send, 1040 .pru_shutdown = tcp_usr_shutdown, 1041 .pru_sockaddr = in6_mapped_sockaddr, 1042#if 0 1043 .pru_soreceive = soreceive_stream, 1044#endif 1045 .pru_sosetlabel = in_pcbsosetlabel, 1046 .pru_close = tcp_usr_close, 1047}; 1048#endif /* INET6 */ 1049 1050/* 1051 * Common subroutine to open a TCP connection to remote host specified 1052 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1053 * port number if needed. Call in_pcbconnect_setup to do the routing and 1054 * to choose a local host address (interface). If there is an existing 1055 * incarnation of the same connection in TIME-WAIT state and if the remote 1056 * host was sending CC options and if the connection duration was < MSL, then 1057 * truncate the previous TIME-WAIT state and proceed. 1058 * Initialize connection parameters and enter SYN-SENT state. 1059 */ 1060static int 1061tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1062{ 1063 struct inpcb *inp = tp->t_inpcb, *oinp; 1064 struct socket *so = inp->inp_socket; 1065 struct in_addr laddr; 1066 u_short lport; 1067 int error; 1068 1069 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1070 INP_WLOCK_ASSERT(inp); 1071 1072 if (inp->inp_lport == 0) { 1073 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1074 if (error) 1075 return error; 1076 } 1077 1078 /* 1079 * Cannot simply call in_pcbconnect, because there might be an 1080 * earlier incarnation of this same connection still in 1081 * TIME_WAIT state, creating an ADDRINUSE error. 1082 */ 1083 laddr = inp->inp_laddr; 1084 lport = inp->inp_lport; 1085 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1086 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1087 if (error && oinp == NULL) 1088 return error; 1089 if (oinp) 1090 return EADDRINUSE; 1091 inp->inp_laddr = laddr; 1092 in_pcbrehash(inp); 1093 1094 /* 1095 * Compute window scaling to request: 1096 * Scale to fit into sweet spot. See tcp_syncache.c. 1097 * XXX: This should move to tcp_output(). 1098 */ 1099 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1100 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1101 tp->request_r_scale++; 1102 1103 soisconnecting(so); 1104 TCPSTAT_INC(tcps_connattempt); 1105 tp->t_state = TCPS_SYN_SENT; 1106 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1107 tp->iss = tcp_new_isn(tp); 1108 tp->t_bw_rtseq = tp->iss; 1109 tcp_sendseqinit(tp); 1110 1111 return 0; 1112} 1113 1114#ifdef INET6 1115static int 1116tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1117{ 1118 struct inpcb *inp = tp->t_inpcb, *oinp; 1119 struct socket *so = inp->inp_socket; 1120 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1121 struct in6_addr addr6; 1122 int error; 1123 1124 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1125 INP_WLOCK_ASSERT(inp); 1126 1127 if (inp->inp_lport == 0) { 1128 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1129 if (error) 1130 return error; 1131 } 1132 1133 /* 1134 * Cannot simply call in_pcbconnect, because there might be an 1135 * earlier incarnation of this same connection still in 1136 * TIME_WAIT state, creating an ADDRINUSE error. 1137 * in6_pcbladdr() also handles scope zone IDs. 1138 */ 1139 error = in6_pcbladdr(inp, nam, &addr6); 1140 if (error) 1141 return error; 1142 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1143 &sin6->sin6_addr, sin6->sin6_port, 1144 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1145 ? &addr6 1146 : &inp->in6p_laddr, 1147 inp->inp_lport, 0, NULL); 1148 if (oinp) 1149 return EADDRINUSE; 1150 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1151 inp->in6p_laddr = addr6; 1152 inp->in6p_faddr = sin6->sin6_addr; 1153 inp->inp_fport = sin6->sin6_port; 1154 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1155 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; 1156 if (inp->inp_flags & IN6P_AUTOFLOWLABEL) 1157 inp->inp_flow |= 1158 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1159 in_pcbrehash(inp); 1160 1161 /* Compute window scaling to request. */ 1162 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1163 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1164 tp->request_r_scale++; 1165 1166 soisconnecting(so); 1167 TCPSTAT_INC(tcps_connattempt); 1168 tp->t_state = TCPS_SYN_SENT; 1169 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1170 tp->iss = tcp_new_isn(tp); 1171 tp->t_bw_rtseq = tp->iss; 1172 tcp_sendseqinit(tp); 1173 1174 return 0; 1175} 1176#endif /* INET6 */ 1177 1178/* 1179 * Export TCP internal state information via a struct tcp_info, based on the 1180 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1181 * (TCP state machine, etc). We export all information using FreeBSD-native 1182 * constants -- for example, the numeric values for tcpi_state will differ 1183 * from Linux. 1184 */ 1185static void 1186tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1187{ 1188 1189 INP_WLOCK_ASSERT(tp->t_inpcb); 1190 bzero(ti, sizeof(*ti)); 1191 1192 ti->tcpi_state = tp->t_state; 1193 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1194 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1195 if (tp->t_flags & TF_SACK_PERMIT) 1196 ti->tcpi_options |= TCPI_OPT_SACK; 1197 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1198 ti->tcpi_options |= TCPI_OPT_WSCALE; 1199 ti->tcpi_snd_wscale = tp->snd_scale; 1200 ti->tcpi_rcv_wscale = tp->rcv_scale; 1201 } 1202 1203 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1204 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1205 1206 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1207 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1208 1209 /* 1210 * FreeBSD-specific extension fields for tcp_info. 1211 */ 1212 ti->tcpi_rcv_space = tp->rcv_wnd; 1213 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1214 ti->tcpi_snd_wnd = tp->snd_wnd; 1215 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1216 ti->tcpi_snd_nxt = tp->snd_nxt; 1217 ti->__tcpi_snd_mss = tp->t_maxseg; 1218 ti->__tcpi_rcv_mss = tp->t_maxseg; 1219 if (tp->t_flags & TF_TOE) 1220 ti->tcpi_options |= TCPI_OPT_TOE; 1221} 1222 1223/* 1224 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1225 * socket option arguments. When it re-acquires the lock after the copy, it 1226 * has to revalidate that the connection is still valid for the socket 1227 * option. 1228 */ 1229#define INP_WLOCK_RECHECK(inp) do { \ 1230 INP_WLOCK(inp); \ 1231 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1232 INP_WUNLOCK(inp); \ 1233 return (ECONNRESET); \ 1234 } \ 1235 tp = intotcpcb(inp); \ 1236} while(0) 1237 1238int 1239tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1240{ 1241 int error, opt, optval; 1242 struct inpcb *inp; 1243 struct tcpcb *tp; 1244 struct tcp_info ti; 1245 1246 error = 0; 1247 inp = sotoinpcb(so); 1248 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1249 INP_WLOCK(inp); 1250 if (sopt->sopt_level != IPPROTO_TCP) { 1251#ifdef INET6 1252 if (inp->inp_vflag & INP_IPV6PROTO) { 1253 INP_WUNLOCK(inp); 1254 error = ip6_ctloutput(so, sopt); 1255 } else { 1256#endif /* INET6 */ 1257 INP_WUNLOCK(inp); 1258 error = ip_ctloutput(so, sopt); 1259#ifdef INET6 1260 } 1261#endif 1262 return (error); 1263 } 1264 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1265 INP_WUNLOCK(inp); 1266 return (ECONNRESET); 1267 } 1268 1269 switch (sopt->sopt_dir) { 1270 case SOPT_SET: 1271 switch (sopt->sopt_name) { 1272#ifdef TCP_SIGNATURE 1273 case TCP_MD5SIG: 1274 INP_WUNLOCK(inp); 1275 error = sooptcopyin(sopt, &optval, sizeof optval, 1276 sizeof optval); 1277 if (error) 1278 return (error); 1279 1280 INP_WLOCK_RECHECK(inp); 1281 if (optval > 0) 1282 tp->t_flags |= TF_SIGNATURE; 1283 else 1284 tp->t_flags &= ~TF_SIGNATURE; 1285 INP_WUNLOCK(inp); 1286 break; 1287#endif /* TCP_SIGNATURE */ 1288 case TCP_NODELAY: 1289 case TCP_NOOPT: 1290 INP_WUNLOCK(inp); 1291 error = sooptcopyin(sopt, &optval, sizeof optval, 1292 sizeof optval); 1293 if (error) 1294 return (error); 1295 1296 INP_WLOCK_RECHECK(inp); 1297 switch (sopt->sopt_name) { 1298 case TCP_NODELAY: 1299 opt = TF_NODELAY; 1300 break; 1301 case TCP_NOOPT: 1302 opt = TF_NOOPT; 1303 break; 1304 default: 1305 opt = 0; /* dead code to fool gcc */ 1306 break; 1307 } 1308 1309 if (optval) 1310 tp->t_flags |= opt; 1311 else 1312 tp->t_flags &= ~opt; 1313 INP_WUNLOCK(inp); 1314 break; 1315 1316 case TCP_NOPUSH: 1317 INP_WUNLOCK(inp); 1318 error = sooptcopyin(sopt, &optval, sizeof optval, 1319 sizeof optval); 1320 if (error) 1321 return (error); 1322 1323 INP_WLOCK_RECHECK(inp); 1324 if (optval) 1325 tp->t_flags |= TF_NOPUSH; 1326 else { 1327 tp->t_flags &= ~TF_NOPUSH; 1328 error = tcp_output(tp); 1329 } 1330 INP_WUNLOCK(inp); 1331 break; 1332 1333 case TCP_MAXSEG: 1334 INP_WUNLOCK(inp); 1335 error = sooptcopyin(sopt, &optval, sizeof optval, 1336 sizeof optval); 1337 if (error) 1338 return (error); 1339 1340 INP_WLOCK_RECHECK(inp); 1341 if (optval > 0 && optval <= tp->t_maxseg && 1342 optval + 40 >= V_tcp_minmss) 1343 tp->t_maxseg = optval; 1344 else 1345 error = EINVAL; 1346 INP_WUNLOCK(inp); 1347 break; 1348 1349 case TCP_INFO: 1350 INP_WUNLOCK(inp); 1351 error = EINVAL; 1352 break; 1353 1354 default: 1355 INP_WUNLOCK(inp); 1356 error = ENOPROTOOPT; 1357 break; 1358 } 1359 break; 1360 1361 case SOPT_GET: 1362 tp = intotcpcb(inp); 1363 switch (sopt->sopt_name) { 1364#ifdef TCP_SIGNATURE 1365 case TCP_MD5SIG: 1366 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1367 INP_WUNLOCK(inp); 1368 error = sooptcopyout(sopt, &optval, sizeof optval); 1369 break; 1370#endif 1371 1372 case TCP_NODELAY: 1373 optval = tp->t_flags & TF_NODELAY; 1374 INP_WUNLOCK(inp); 1375 error = sooptcopyout(sopt, &optval, sizeof optval); 1376 break; 1377 case TCP_MAXSEG: 1378 optval = tp->t_maxseg; 1379 INP_WUNLOCK(inp); 1380 error = sooptcopyout(sopt, &optval, sizeof optval); 1381 break; 1382 case TCP_NOOPT: 1383 optval = tp->t_flags & TF_NOOPT; 1384 INP_WUNLOCK(inp); 1385 error = sooptcopyout(sopt, &optval, sizeof optval); 1386 break; 1387 case TCP_NOPUSH: 1388 optval = tp->t_flags & TF_NOPUSH; 1389 INP_WUNLOCK(inp); 1390 error = sooptcopyout(sopt, &optval, sizeof optval); 1391 break; 1392 case TCP_INFO: 1393 tcp_fill_info(tp, &ti); 1394 INP_WUNLOCK(inp); 1395 error = sooptcopyout(sopt, &ti, sizeof ti); 1396 break; 1397 default: 1398 INP_WUNLOCK(inp); 1399 error = ENOPROTOOPT; 1400 break; 1401 } 1402 break; 1403 } 1404 return (error); 1405} 1406#undef INP_WLOCK_RECHECK 1407 1408/* 1409 * tcp_sendspace and tcp_recvspace are the default send and receive window 1410 * sizes, respectively. These are obsolescent (this information should 1411 * be set by the route). 1412 */ 1413u_long tcp_sendspace = 1024*32; 1414SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1415 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1416u_long tcp_recvspace = 1024*64; 1417SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1418 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1419 1420/* 1421 * Attach TCP protocol to socket, allocating 1422 * internet protocol control block, tcp control block, 1423 * bufer space, and entering LISTEN state if to accept connections. 1424 */ 1425static int 1426tcp_attach(struct socket *so) 1427{ 1428 struct tcpcb *tp; 1429 struct inpcb *inp; 1430 int error; 1431 1432 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1433 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1434 if (error) 1435 return (error); 1436 } 1437 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1438 so->so_snd.sb_flags |= SB_AUTOSIZE; 1439 INP_INFO_WLOCK(&V_tcbinfo); 1440 error = in_pcballoc(so, &V_tcbinfo); 1441 if (error) { 1442 INP_INFO_WUNLOCK(&V_tcbinfo); 1443 return (error); 1444 } 1445 inp = sotoinpcb(so); 1446#ifdef INET6 1447 if (inp->inp_vflag & INP_IPV6PROTO) { 1448 inp->inp_vflag |= INP_IPV6; 1449 inp->in6p_hops = -1; /* use kernel default */ 1450 } 1451 else 1452#endif 1453 inp->inp_vflag |= INP_IPV4; 1454 tp = tcp_newtcpcb(inp); 1455 if (tp == NULL) { 1456 in_pcbdetach(inp); 1457 in_pcbfree(inp); 1458 INP_INFO_WUNLOCK(&V_tcbinfo); 1459 return (ENOBUFS); 1460 } 1461 tp->t_state = TCPS_CLOSED; 1462 INP_WUNLOCK(inp); 1463 INP_INFO_WUNLOCK(&V_tcbinfo); 1464 return (0); 1465} 1466 1467/* 1468 * Initiate (or continue) disconnect. 1469 * If embryonic state, just send reset (once). 1470 * If in ``let data drain'' option and linger null, just drop. 1471 * Otherwise (hard), mark socket disconnecting and drop 1472 * current input data; switch states based on user close, and 1473 * send segment to peer (with FIN). 1474 */ 1475static void 1476tcp_disconnect(struct tcpcb *tp) 1477{ 1478 struct inpcb *inp = tp->t_inpcb; 1479 struct socket *so = inp->inp_socket; 1480 1481 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1482 INP_WLOCK_ASSERT(inp); 1483 1484 /* 1485 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1486 * socket is still open. 1487 */ 1488 if (tp->t_state < TCPS_ESTABLISHED) { 1489 tp = tcp_close(tp); 1490 KASSERT(tp != NULL, 1491 ("tcp_disconnect: tcp_close() returned NULL")); 1492 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1493 tp = tcp_drop(tp, 0); 1494 KASSERT(tp != NULL, 1495 ("tcp_disconnect: tcp_drop() returned NULL")); 1496 } else { 1497 soisdisconnecting(so); 1498 sbflush(&so->so_rcv); 1499 tcp_usrclosed(tp); 1500 if (!(inp->inp_flags & INP_DROPPED)) 1501 tcp_output_disconnect(tp); 1502 } 1503} 1504 1505/* 1506 * User issued close, and wish to trail through shutdown states: 1507 * if never received SYN, just forget it. If got a SYN from peer, 1508 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1509 * If already got a FIN from peer, then almost done; go to LAST_ACK 1510 * state. In all other cases, have already sent FIN to peer (e.g. 1511 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1512 * for peer to send FIN or not respond to keep-alives, etc. 1513 * We can let the user exit from the close as soon as the FIN is acked. 1514 */ 1515static void 1516tcp_usrclosed(struct tcpcb *tp) 1517{ 1518 1519 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1520 INP_WLOCK_ASSERT(tp->t_inpcb); 1521 1522 switch (tp->t_state) { 1523 case TCPS_LISTEN: 1524 tcp_offload_listen_close(tp); 1525 /* FALLTHROUGH */ 1526 case TCPS_CLOSED: 1527 tp->t_state = TCPS_CLOSED; 1528 tp = tcp_close(tp); 1529 /* 1530 * tcp_close() should never return NULL here as the socket is 1531 * still open. 1532 */ 1533 KASSERT(tp != NULL, 1534 ("tcp_usrclosed: tcp_close() returned NULL")); 1535 break; 1536 1537 case TCPS_SYN_SENT: 1538 case TCPS_SYN_RECEIVED: 1539 tp->t_flags |= TF_NEEDFIN; 1540 break; 1541 1542 case TCPS_ESTABLISHED: 1543 tp->t_state = TCPS_FIN_WAIT_1; 1544 break; 1545 1546 case TCPS_CLOSE_WAIT: 1547 tp->t_state = TCPS_LAST_ACK; 1548 break; 1549 } 1550 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1551 soisdisconnected(tp->t_inpcb->inp_socket); 1552 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1553 if (tp->t_state == TCPS_FIN_WAIT_2) { 1554 int timeout; 1555 1556 timeout = (tcp_fast_finwait2_recycle) ? 1557 tcp_finwait2_timeout : tcp_maxidle; 1558 tcp_timer_activate(tp, TT_2MSL, timeout); 1559 } 1560 } 1561} 1562 1563#ifdef DDB 1564static void 1565db_print_indent(int indent) 1566{ 1567 int i; 1568 1569 for (i = 0; i < indent; i++) 1570 db_printf(" "); 1571} 1572 1573static void 1574db_print_tstate(int t_state) 1575{ 1576 1577 switch (t_state) { 1578 case TCPS_CLOSED: 1579 db_printf("TCPS_CLOSED"); 1580 return; 1581 1582 case TCPS_LISTEN: 1583 db_printf("TCPS_LISTEN"); 1584 return; 1585 1586 case TCPS_SYN_SENT: 1587 db_printf("TCPS_SYN_SENT"); 1588 return; 1589 1590 case TCPS_SYN_RECEIVED: 1591 db_printf("TCPS_SYN_RECEIVED"); 1592 return; 1593 1594 case TCPS_ESTABLISHED: 1595 db_printf("TCPS_ESTABLISHED"); 1596 return; 1597 1598 case TCPS_CLOSE_WAIT: 1599 db_printf("TCPS_CLOSE_WAIT"); 1600 return; 1601 1602 case TCPS_FIN_WAIT_1: 1603 db_printf("TCPS_FIN_WAIT_1"); 1604 return; 1605 1606 case TCPS_CLOSING: 1607 db_printf("TCPS_CLOSING"); 1608 return; 1609 1610 case TCPS_LAST_ACK: 1611 db_printf("TCPS_LAST_ACK"); 1612 return; 1613 1614 case TCPS_FIN_WAIT_2: 1615 db_printf("TCPS_FIN_WAIT_2"); 1616 return; 1617 1618 case TCPS_TIME_WAIT: 1619 db_printf("TCPS_TIME_WAIT"); 1620 return; 1621 1622 default: 1623 db_printf("unknown"); 1624 return; 1625 } 1626} 1627 1628static void 1629db_print_tflags(u_int t_flags) 1630{ 1631 int comma; 1632 1633 comma = 0; 1634 if (t_flags & TF_ACKNOW) { 1635 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1636 comma = 1; 1637 } 1638 if (t_flags & TF_DELACK) { 1639 db_printf("%sTF_DELACK", comma ? ", " : ""); 1640 comma = 1; 1641 } 1642 if (t_flags & TF_NODELAY) { 1643 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1644 comma = 1; 1645 } 1646 if (t_flags & TF_NOOPT) { 1647 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1648 comma = 1; 1649 } 1650 if (t_flags & TF_SENTFIN) { 1651 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1652 comma = 1; 1653 } 1654 if (t_flags & TF_REQ_SCALE) { 1655 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1656 comma = 1; 1657 } 1658 if (t_flags & TF_RCVD_SCALE) { 1659 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1660 comma = 1; 1661 } 1662 if (t_flags & TF_REQ_TSTMP) { 1663 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1664 comma = 1; 1665 } 1666 if (t_flags & TF_RCVD_TSTMP) { 1667 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1668 comma = 1; 1669 } 1670 if (t_flags & TF_SACK_PERMIT) { 1671 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1672 comma = 1; 1673 } 1674 if (t_flags & TF_NEEDSYN) { 1675 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1676 comma = 1; 1677 } 1678 if (t_flags & TF_NEEDFIN) { 1679 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1680 comma = 1; 1681 } 1682 if (t_flags & TF_NOPUSH) { 1683 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1684 comma = 1; 1685 } 1686 if (t_flags & TF_NOPUSH) { 1687 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1688 comma = 1; 1689 } 1690 if (t_flags & TF_MORETOCOME) { 1691 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1692 comma = 1; 1693 } 1694 if (t_flags & TF_LQ_OVERFLOW) { 1695 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1696 comma = 1; 1697 } 1698 if (t_flags & TF_LASTIDLE) { 1699 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1700 comma = 1; 1701 } 1702 if (t_flags & TF_RXWIN0SENT) { 1703 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1704 comma = 1; 1705 } 1706 if (t_flags & TF_FASTRECOVERY) { 1707 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1708 comma = 1; 1709 } 1710 if (t_flags & TF_WASFRECOVERY) { 1711 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1712 comma = 1; 1713 } 1714 if (t_flags & TF_SIGNATURE) { 1715 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1716 comma = 1; 1717 } 1718 if (t_flags & TF_FORCEDATA) { 1719 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1720 comma = 1; 1721 } 1722 if (t_flags & TF_TSO) { 1723 db_printf("%sTF_TSO", comma ? ", " : ""); 1724 comma = 1; 1725 } 1726 if (t_flags & TF_ECN_PERMIT) { 1727 db_printf("%sTF_ECN_PERMIT", comma ? ", " : ""); 1728 comma = 1; 1729 } 1730} 1731 1732static void 1733db_print_toobflags(char t_oobflags) 1734{ 1735 int comma; 1736 1737 comma = 0; 1738 if (t_oobflags & TCPOOB_HAVEDATA) { 1739 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1740 comma = 1; 1741 } 1742 if (t_oobflags & TCPOOB_HADDATA) { 1743 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1744 comma = 1; 1745 } 1746} 1747 1748static void 1749db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1750{ 1751 1752 db_print_indent(indent); 1753 db_printf("%s at %p\n", name, tp); 1754 1755 indent += 2; 1756 1757 db_print_indent(indent); 1758 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1759 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1760 1761 db_print_indent(indent); 1762 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1763 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1764 1765 db_print_indent(indent); 1766 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1767 &tp->t_timers->tt_delack, tp->t_inpcb); 1768 1769 db_print_indent(indent); 1770 db_printf("t_state: %d (", tp->t_state); 1771 db_print_tstate(tp->t_state); 1772 db_printf(")\n"); 1773 1774 db_print_indent(indent); 1775 db_printf("t_flags: 0x%x (", tp->t_flags); 1776 db_print_tflags(tp->t_flags); 1777 db_printf(")\n"); 1778 1779 db_print_indent(indent); 1780 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1781 tp->snd_una, tp->snd_max, tp->snd_nxt); 1782 1783 db_print_indent(indent); 1784 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1785 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1786 1787 db_print_indent(indent); 1788 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1789 tp->iss, tp->irs, tp->rcv_nxt); 1790 1791 db_print_indent(indent); 1792 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1793 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1794 1795 db_print_indent(indent); 1796 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1797 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1798 1799 db_print_indent(indent); 1800 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1801 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1802 tp->snd_recover); 1803 1804 db_print_indent(indent); 1805 db_printf("t_maxopd: %u t_rcvtime: %u t_startime: %u\n", 1806 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1807 1808 db_print_indent(indent); 1809 db_printf("t_rttime: %u t_rtsq: 0x%08x t_bw_rtttime: %u\n", 1810 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1811 1812 db_print_indent(indent); 1813 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1814 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1815 tp->t_srtt); 1816 1817 db_print_indent(indent); 1818 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1819 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1820 tp->t_rttbest); 1821 1822 db_print_indent(indent); 1823 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1824 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1825 1826 db_print_indent(indent); 1827 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1828 db_print_toobflags(tp->t_oobflags); 1829 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1830 1831 db_print_indent(indent); 1832 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1833 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1834 1835 db_print_indent(indent); 1836 db_printf("ts_recent: %u ts_recent_age: %u\n", 1837 tp->ts_recent, tp->ts_recent_age); 1838 1839 db_print_indent(indent); 1840 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1841 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1842 1843 db_print_indent(indent); 1844 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1845 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 1846 tp->snd_recover_prev, tp->t_badrxtwin); 1847 1848 db_print_indent(indent); 1849 db_printf("snd_numholes: %d snd_holes first: %p\n", 1850 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1851 1852 db_print_indent(indent); 1853 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1854 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1855 1856 /* Skip sackblks, sackhint. */ 1857 1858 db_print_indent(indent); 1859 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1860 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1861} 1862 1863DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1864{ 1865 struct tcpcb *tp; 1866 1867 if (!have_addr) { 1868 db_printf("usage: show tcpcb <addr>\n"); 1869 return; 1870 } 1871 tp = (struct tcpcb *)addr; 1872 1873 db_print_tcpcb(tp, "tcpcb", 0); 1874} 1875#endif 1876