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("$FreeBSD$"); 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 */ 93 static int tcp_attach(struct socket *); 94 static int tcp_connect(struct tcpcb *, struct sockaddr *, 95 struct thread *td); 96 #ifdef INET6 97 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 98 struct thread *td); 99 #endif /* INET6 */ 100 static void tcp_disconnect(struct tcpcb *); 101 static void tcp_usrclosed(struct tcpcb *); 102 static 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 */ 119 static int 120 tcp_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); 140 out: 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 */ 154 static void 155 tcp_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 */ 248 static void 249 tcp_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 */ 266 static int 267 tcp_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); 297 out: 298 TCPDEBUG2(PRU_BIND); 299 INP_UNLOCK(inp); 300 INP_INFO_WUNLOCK(&tcbinfo); 301 302 return (error); 303 } 304 305 #ifdef INET6 306 static int 307 tcp6_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); 353 out: 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 */ 364 static int 365 tcp_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); 392 393 out: 394 TCPDEBUG2(PRU_LISTEN); 395 INP_UNLOCK(inp); 396 INP_INFO_WUNLOCK(&tcbinfo); 397 return (error); 398 } 399 400 #ifdef INET6 401 static int 402 tcp6_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); 432 433 out: 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 */ 448 static int 449 tcp_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); 482 out: 483 TCPDEBUG2(PRU_CONNECT); 484 INP_UNLOCK(inp); 485 INP_INFO_WUNLOCK(&tcbinfo); 486 return (error); 487 } 488 489 #ifdef INET6 490 static int 491 tcp6_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); 542 543 out: 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 */ 562 static int 563 tcp_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); 581 out: 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 */ 593 static int 594 tcp_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; 623 624 out: 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 633 static int 634 tcp6_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 } 671 672 out: 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 */ 688 static int 689 tcp_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); 709 710 out: 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 */ 721 static int 722 tcp_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); 739 740 out: 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 */ 753 static int 754 tcp_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 } 897 out: 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 */ 909 static void 910 tcp_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 */ 947 static void 948 tcp_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 */ 986 static int 987 tcp_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 1019 out: 1020 TCPDEBUG2(PRU_RCVOOB); 1021 INP_UNLOCK(inp); 1022 return (error); 1023 } 1024 1025 struct 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 1046 struct 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 */ 1077 static int 1078 tcp_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 1132 static int 1133 tcp6_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 */ 1202 static void 1203 tcp_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 * The new sockopt interface makes it possible for us to block in the 1236 * copyin/out step (if we take a page fault). Taking a page fault at 1237 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1238 * use TSM, there probably isn't any need for this function to run at 1239 * splnet() any more. This needs more examination.) 1240 * 1241 * XXXRW: The locking here is wrong; we may take a page fault while holding 1242 * the inpcb lock. 1243 */ 1244 int 1245 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1246 { 1247 int error, opt, optval; 1248 struct inpcb *inp; 1249 struct tcpcb *tp; 1250 struct tcp_info ti; 1251 1252 error = 0; 1253 inp = sotoinpcb(so); 1254 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1255 INP_LOCK(inp); 1256 if (sopt->sopt_level != IPPROTO_TCP) { 1257 INP_UNLOCK(inp); 1258 #ifdef INET6 1259 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1260 error = ip6_ctloutput(so, sopt); 1261 else 1262 #endif /* INET6 */ 1263 error = ip_ctloutput(so, sopt); 1264 return (error); 1265 } 1266 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1267 error = ECONNRESET; 1268 goto out; 1269 } 1270 tp = intotcpcb(inp); 1271 1272 switch (sopt->sopt_dir) { 1273 case SOPT_SET: 1274 switch (sopt->sopt_name) { 1275 #ifdef TCP_SIGNATURE 1276 case TCP_MD5SIG: 1277 error = sooptcopyin(sopt, &optval, sizeof optval, 1278 sizeof optval); 1279 if (error) 1280 break; 1281 1282 if (optval > 0) 1283 tp->t_flags |= TF_SIGNATURE; 1284 else 1285 tp->t_flags &= ~TF_SIGNATURE; 1286 break; 1287 #endif /* TCP_SIGNATURE */ 1288 case TCP_NODELAY: 1289 case TCP_NOOPT: 1290 error = sooptcopyin(sopt, &optval, sizeof optval, 1291 sizeof optval); 1292 if (error) 1293 break; 1294 1295 switch (sopt->sopt_name) { 1296 case TCP_NODELAY: 1297 opt = TF_NODELAY; 1298 break; 1299 case TCP_NOOPT: 1300 opt = TF_NOOPT; 1301 break; 1302 default: 1303 opt = 0; /* dead code to fool gcc */ 1304 break; 1305 } 1306 1307 if (optval) 1308 tp->t_flags |= opt; 1309 else 1310 tp->t_flags &= ~opt; 1311 break; 1312 1313 case TCP_NOPUSH: 1314 error = sooptcopyin(sopt, &optval, sizeof optval, 1315 sizeof optval); 1316 if (error) 1317 break; 1318 1319 if (optval) 1320 tp->t_flags |= TF_NOPUSH; 1321 else { 1322 tp->t_flags &= ~TF_NOPUSH; 1323 error = tcp_output(tp); 1324 } 1325 break; 1326 1327 case TCP_MAXSEG: 1328 error = sooptcopyin(sopt, &optval, sizeof optval, 1329 sizeof optval); 1330 if (error) 1331 break; 1332 1333 if (optval > 0 && optval <= tp->t_maxseg && 1334 optval + 40 >= tcp_minmss) 1335 tp->t_maxseg = optval; 1336 else 1337 error = EINVAL; 1338 break; 1339 1340 case TCP_INFO: 1341 error = EINVAL; 1342 break; 1343 1344 default: 1345 error = ENOPROTOOPT; 1346 break; 1347 } 1348 break; 1349 1350 case SOPT_GET: 1351 switch (sopt->sopt_name) { 1352 #ifdef TCP_SIGNATURE 1353 case TCP_MD5SIG: 1354 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1355 error = sooptcopyout(sopt, &optval, sizeof optval); 1356 break; 1357 #endif 1358 case TCP_NODELAY: 1359 optval = tp->t_flags & TF_NODELAY; 1360 error = sooptcopyout(sopt, &optval, sizeof optval); 1361 break; 1362 case TCP_MAXSEG: 1363 optval = tp->t_maxseg; 1364 error = sooptcopyout(sopt, &optval, sizeof optval); 1365 break; 1366 case TCP_NOOPT: 1367 optval = tp->t_flags & TF_NOOPT; 1368 error = sooptcopyout(sopt, &optval, sizeof optval); 1369 break; 1370 case TCP_NOPUSH: 1371 optval = tp->t_flags & TF_NOPUSH; 1372 error = sooptcopyout(sopt, &optval, sizeof optval); 1373 break; 1374 case TCP_INFO: 1375 tcp_fill_info(tp, &ti); 1376 error = sooptcopyout(sopt, &ti, sizeof ti); 1377 break; 1378 default: 1379 error = ENOPROTOOPT; 1380 break; 1381 } 1382 break; 1383 } 1384 out: 1385 INP_UNLOCK(inp); 1386 return (error); 1387 } 1388 1389 /* 1390 * tcp_sendspace and tcp_recvspace are the default send and receive window 1391 * sizes, respectively. These are obsolescent (this information should 1392 * be set by the route). 1393 */ 1394 u_long tcp_sendspace = 1024*32; 1395 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1396 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1397 u_long tcp_recvspace = 1024*64; 1398 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1399 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1400 1401 /* 1402 * Attach TCP protocol to socket, allocating 1403 * internet protocol control block, tcp control block, 1404 * bufer space, and entering LISTEN state if to accept connections. 1405 */ 1406 static int 1407 tcp_attach(struct socket *so) 1408 { 1409 struct tcpcb *tp; 1410 struct inpcb *inp; 1411 int error; 1412 #ifdef INET6 1413 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1414 #endif 1415 1416 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1417 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1418 if (error) 1419 return (error); 1420 } 1421 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1422 so->so_snd.sb_flags |= SB_AUTOSIZE; 1423 INP_INFO_WLOCK(&tcbinfo); 1424 error = in_pcballoc(so, &tcbinfo); 1425 if (error) { 1426 INP_INFO_WUNLOCK(&tcbinfo); 1427 return (error); 1428 } 1429 inp = sotoinpcb(so); 1430 #ifdef INET6 1431 if (isipv6) { 1432 inp->inp_vflag |= INP_IPV6; 1433 inp->in6p_hops = -1; /* use kernel default */ 1434 } 1435 else 1436 #endif 1437 inp->inp_vflag |= INP_IPV4; 1438 tp = tcp_newtcpcb(inp); 1439 if (tp == NULL) { 1440 #ifdef INET6 1441 if (isipv6) { 1442 in6_pcbdetach(inp); 1443 in6_pcbfree(inp); 1444 } else { 1445 #endif 1446 in_pcbdetach(inp); 1447 in_pcbfree(inp); 1448 #ifdef INET6 1449 } 1450 #endif 1451 INP_INFO_WUNLOCK(&tcbinfo); 1452 return (ENOBUFS); 1453 } 1454 tp->t_state = TCPS_CLOSED; 1455 INP_UNLOCK(inp); 1456 INP_INFO_WUNLOCK(&tcbinfo); 1457 return (0); 1458 } 1459 1460 /* 1461 * Initiate (or continue) disconnect. 1462 * If embryonic state, just send reset (once). 1463 * If in ``let data drain'' option and linger null, just drop. 1464 * Otherwise (hard), mark socket disconnecting and drop 1465 * current input data; switch states based on user close, and 1466 * send segment to peer (with FIN). 1467 */ 1468 static void 1469 tcp_disconnect(struct tcpcb *tp) 1470 { 1471 struct inpcb *inp = tp->t_inpcb; 1472 struct socket *so = inp->inp_socket; 1473 1474 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1475 INP_LOCK_ASSERT(inp); 1476 1477 /* 1478 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1479 * socket is still open. 1480 */ 1481 if (tp->t_state < TCPS_ESTABLISHED) { 1482 tp = tcp_close(tp); 1483 KASSERT(tp != NULL, 1484 ("tcp_disconnect: tcp_close() returned NULL")); 1485 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1486 tp = tcp_drop(tp, 0); 1487 KASSERT(tp != NULL, 1488 ("tcp_disconnect: tcp_drop() returned NULL")); 1489 } else { 1490 soisdisconnecting(so); 1491 sbflush(&so->so_rcv); 1492 tcp_usrclosed(tp); 1493 if (!(inp->inp_vflag & INP_DROPPED)) 1494 tcp_output_disconnect(tp); 1495 } 1496 } 1497 1498 /* 1499 * User issued close, and wish to trail through shutdown states: 1500 * if never received SYN, just forget it. If got a SYN from peer, 1501 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1502 * If already got a FIN from peer, then almost done; go to LAST_ACK 1503 * state. In all other cases, have already sent FIN to peer (e.g. 1504 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1505 * for peer to send FIN or not respond to keep-alives, etc. 1506 * We can let the user exit from the close as soon as the FIN is acked. 1507 */ 1508 static void 1509 tcp_usrclosed(struct tcpcb *tp) 1510 { 1511 1512 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1513 INP_LOCK_ASSERT(tp->t_inpcb); 1514 1515 switch (tp->t_state) { 1516 case TCPS_LISTEN: 1517 tcp_offload_listen_close(tp); 1518 /* FALLTHROUGH */ 1519 case TCPS_CLOSED: 1520 tp->t_state = TCPS_CLOSED; 1521 tp = tcp_close(tp); 1522 /* 1523 * tcp_close() should never return NULL here as the socket is 1524 * still open. 1525 */ 1526 KASSERT(tp != NULL, 1527 ("tcp_usrclosed: tcp_close() returned NULL")); 1528 break; 1529 1530 case TCPS_SYN_SENT: 1531 case TCPS_SYN_RECEIVED: 1532 tp->t_flags |= TF_NEEDFIN; 1533 break; 1534 1535 case TCPS_ESTABLISHED: 1536 tp->t_state = TCPS_FIN_WAIT_1; 1537 break; 1538 1539 case TCPS_CLOSE_WAIT: 1540 tp->t_state = TCPS_LAST_ACK; 1541 break; 1542 } 1543 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1544 soisdisconnected(tp->t_inpcb->inp_socket); 1545 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1546 if (tp->t_state == TCPS_FIN_WAIT_2) { 1547 int timeout; 1548 1549 timeout = (tcp_fast_finwait2_recycle) ? 1550 tcp_finwait2_timeout : tcp_maxidle; 1551 tcp_timer_activate(tp, TT_2MSL, timeout); 1552 } 1553 } 1554 } 1555 1556 #ifdef DDB 1557 static void 1558 db_print_indent(int indent) 1559 { 1560 int i; 1561 1562 for (i = 0; i < indent; i++) 1563 db_printf(" "); 1564 } 1565 1566 static void 1567 db_print_tstate(int t_state) 1568 { 1569 1570 switch (t_state) { 1571 case TCPS_CLOSED: 1572 db_printf("TCPS_CLOSED"); 1573 return; 1574 1575 case TCPS_LISTEN: 1576 db_printf("TCPS_LISTEN"); 1577 return; 1578 1579 case TCPS_SYN_SENT: 1580 db_printf("TCPS_SYN_SENT"); 1581 return; 1582 1583 case TCPS_SYN_RECEIVED: 1584 db_printf("TCPS_SYN_RECEIVED"); 1585 return; 1586 1587 case TCPS_ESTABLISHED: 1588 db_printf("TCPS_ESTABLISHED"); 1589 return; 1590 1591 case TCPS_CLOSE_WAIT: 1592 db_printf("TCPS_CLOSE_WAIT"); 1593 return; 1594 1595 case TCPS_FIN_WAIT_1: 1596 db_printf("TCPS_FIN_WAIT_1"); 1597 return; 1598 1599 case TCPS_CLOSING: 1600 db_printf("TCPS_CLOSING"); 1601 return; 1602 1603 case TCPS_LAST_ACK: 1604 db_printf("TCPS_LAST_ACK"); 1605 return; 1606 1607 case TCPS_FIN_WAIT_2: 1608 db_printf("TCPS_FIN_WAIT_2"); 1609 return; 1610 1611 case TCPS_TIME_WAIT: 1612 db_printf("TCPS_TIME_WAIT"); 1613 return; 1614 1615 default: 1616 db_printf("unknown"); 1617 return; 1618 } 1619 } 1620 1621 static void 1622 db_print_tflags(u_int t_flags) 1623 { 1624 int comma; 1625 1626 comma = 0; 1627 if (t_flags & TF_ACKNOW) { 1628 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1629 comma = 1; 1630 } 1631 if (t_flags & TF_DELACK) { 1632 db_printf("%sTF_DELACK", comma ? ", " : ""); 1633 comma = 1; 1634 } 1635 if (t_flags & TF_NODELAY) { 1636 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1637 comma = 1; 1638 } 1639 if (t_flags & TF_NOOPT) { 1640 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1641 comma = 1; 1642 } 1643 if (t_flags & TF_SENTFIN) { 1644 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1645 comma = 1; 1646 } 1647 if (t_flags & TF_REQ_SCALE) { 1648 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1649 comma = 1; 1650 } 1651 if (t_flags & TF_RCVD_SCALE) { 1652 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1653 comma = 1; 1654 } 1655 if (t_flags & TF_REQ_TSTMP) { 1656 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1657 comma = 1; 1658 } 1659 if (t_flags & TF_RCVD_TSTMP) { 1660 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1661 comma = 1; 1662 } 1663 if (t_flags & TF_SACK_PERMIT) { 1664 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1665 comma = 1; 1666 } 1667 if (t_flags & TF_NEEDSYN) { 1668 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1669 comma = 1; 1670 } 1671 if (t_flags & TF_NEEDFIN) { 1672 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1673 comma = 1; 1674 } 1675 if (t_flags & TF_NOPUSH) { 1676 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1677 comma = 1; 1678 } 1679 if (t_flags & TF_NOPUSH) { 1680 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1681 comma = 1; 1682 } 1683 if (t_flags & TF_MORETOCOME) { 1684 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1685 comma = 1; 1686 } 1687 if (t_flags & TF_LQ_OVERFLOW) { 1688 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1689 comma = 1; 1690 } 1691 if (t_flags & TF_LASTIDLE) { 1692 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1693 comma = 1; 1694 } 1695 if (t_flags & TF_RXWIN0SENT) { 1696 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1697 comma = 1; 1698 } 1699 if (t_flags & TF_FASTRECOVERY) { 1700 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1701 comma = 1; 1702 } 1703 if (t_flags & TF_WASFRECOVERY) { 1704 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1705 comma = 1; 1706 } 1707 if (t_flags & TF_SIGNATURE) { 1708 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1709 comma = 1; 1710 } 1711 if (t_flags & TF_FORCEDATA) { 1712 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1713 comma = 1; 1714 } 1715 if (t_flags & TF_TSO) { 1716 db_printf("%sTF_TSO", comma ? ", " : ""); 1717 comma = 1; 1718 } 1719 } 1720 1721 static void 1722 db_print_toobflags(char t_oobflags) 1723 { 1724 int comma; 1725 1726 comma = 0; 1727 if (t_oobflags & TCPOOB_HAVEDATA) { 1728 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1729 comma = 1; 1730 } 1731 if (t_oobflags & TCPOOB_HADDATA) { 1732 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1733 comma = 1; 1734 } 1735 } 1736 1737 static void 1738 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1739 { 1740 1741 db_print_indent(indent); 1742 db_printf("%s at %p\n", name, tp); 1743 1744 indent += 2; 1745 1746 db_print_indent(indent); 1747 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1748 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1749 1750 db_print_indent(indent); 1751 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1752 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1753 1754 db_print_indent(indent); 1755 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1756 &tp->t_timers->tt_delack, tp->t_inpcb); 1757 1758 db_print_indent(indent); 1759 db_printf("t_state: %d (", tp->t_state); 1760 db_print_tstate(tp->t_state); 1761 db_printf(")\n"); 1762 1763 db_print_indent(indent); 1764 db_printf("t_flags: 0x%x (", tp->t_flags); 1765 db_print_tflags(tp->t_flags); 1766 db_printf(")\n"); 1767 1768 db_print_indent(indent); 1769 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1770 tp->snd_una, tp->snd_max, tp->snd_nxt); 1771 1772 db_print_indent(indent); 1773 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1774 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1775 1776 db_print_indent(indent); 1777 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1778 tp->iss, tp->irs, tp->rcv_nxt); 1779 1780 db_print_indent(indent); 1781 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1782 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1783 1784 db_print_indent(indent); 1785 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1786 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1787 1788 db_print_indent(indent); 1789 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1790 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1791 tp->snd_recover); 1792 1793 db_print_indent(indent); 1794 db_printf("t_maxopd: %u t_rcvtime: %lu t_startime: %lu\n", 1795 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1796 1797 db_print_indent(indent); 1798 db_printf("t_rttime: %d t_rtsq: 0x%08x t_bw_rtttime: %d\n", 1799 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1800 1801 db_print_indent(indent); 1802 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1803 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1804 tp->t_srtt); 1805 1806 db_print_indent(indent); 1807 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1808 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1809 tp->t_rttbest); 1810 1811 db_print_indent(indent); 1812 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1813 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1814 1815 db_print_indent(indent); 1816 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1817 db_print_toobflags(tp->t_oobflags); 1818 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1819 1820 db_print_indent(indent); 1821 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1822 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1823 1824 db_print_indent(indent); 1825 db_printf("ts_recent: %u ts_recent_age: %lu\n", 1826 tp->ts_recent, tp->ts_recent_age); 1827 1828 db_print_indent(indent); 1829 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1830 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1831 1832 db_print_indent(indent); 1833 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1834 "t_badrxtwin: %lu\n", tp->snd_ssthresh_prev, 1835 tp->snd_recover_prev, tp->t_badrxtwin); 1836 1837 db_print_indent(indent); 1838 db_printf("snd_numholes: %d snd_holes first: %p\n", 1839 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1840 1841 db_print_indent(indent); 1842 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1843 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1844 1845 /* Skip sackblks, sackhint. */ 1846 1847 db_print_indent(indent); 1848 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1849 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1850 } 1851 1852 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1853 { 1854 struct tcpcb *tp; 1855 1856 if (!have_addr) { 1857 db_printf("usage: show tcpcb <addr>\n"); 1858 return; 1859 } 1860 tp = (struct tcpcb *)addr; 1861 1862 db_print_tcpcb(tp, "tcpcb", 0); 1863 } 1864 #endif 1865