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 89 /* 90 * TCP protocol interface to socket abstraction. 91 */ 92 static int tcp_attach(struct socket *); 93 static int tcp_connect(struct tcpcb *, struct sockaddr *, 94 struct thread *td); 95 #ifdef INET6 96 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 97 struct thread *td); 98 #endif /* INET6 */ 99 static void tcp_disconnect(struct tcpcb *); 100 static void tcp_usrclosed(struct tcpcb *); 101 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 102 103 #ifdef TCPDEBUG 104 #define TCPDEBUG0 int ostate = 0 105 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 106 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 107 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 108 #else 109 #define TCPDEBUG0 110 #define TCPDEBUG1() 111 #define TCPDEBUG2(req) 112 #endif 113 114 /* 115 * TCP attaches to socket via pru_attach(), reserving space, 116 * and an internet control block. 117 */ 118 static int 119 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 120 { 121 struct inpcb *inp; 122 struct tcpcb *tp = NULL; 123 int error; 124 TCPDEBUG0; 125 126 inp = sotoinpcb(so); 127 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 128 TCPDEBUG1(); 129 130 error = tcp_attach(so); 131 if (error) 132 goto out; 133 134 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 135 so->so_linger = TCP_LINGERTIME; 136 137 inp = sotoinpcb(so); 138 tp = intotcpcb(inp); 139 out: 140 TCPDEBUG2(PRU_ATTACH); 141 return error; 142 } 143 144 /* 145 * tcp_detach is called when the socket layer loses its final reference 146 * to the socket, be it a file descriptor reference, a reference from TCP, 147 * etc. At this point, there is only one case in which we will keep around 148 * inpcb state: time wait. 149 * 150 * This function can probably be re-absorbed back into tcp_usr_detach() now 151 * that there is a single detach path. 152 */ 153 static void 154 tcp_detach(struct socket *so, struct inpcb *inp) 155 { 156 struct tcpcb *tp; 157 #ifdef INET6 158 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 159 #endif 160 161 INP_INFO_WLOCK_ASSERT(&tcbinfo); 162 INP_LOCK_ASSERT(inp); 163 164 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 165 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 166 167 tp = intotcpcb(inp); 168 169 if (inp->inp_vflag & INP_TIMEWAIT) { 170 /* 171 * There are two cases to handle: one in which the time wait 172 * state is being discarded (INP_DROPPED), and one in which 173 * this connection will remain in timewait. In the former, 174 * it is time to discard all state (except tcptw, which has 175 * already been discarded by the timewait close code, which 176 * should be further up the call stack somewhere). In the 177 * latter case, we detach from the socket, but leave the pcb 178 * present until timewait ends. 179 * 180 * XXXRW: Would it be cleaner to free the tcptw here? 181 */ 182 if (inp->inp_vflag & INP_DROPPED) { 183 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 184 "INP_DROPPED && tp != NULL")); 185 #ifdef INET6 186 if (isipv6) { 187 in6_pcbdetach(inp); 188 in6_pcbfree(inp); 189 } else { 190 #endif 191 in_pcbdetach(inp); 192 in_pcbfree(inp); 193 #ifdef INET6 194 } 195 #endif 196 } else { 197 #ifdef INET6 198 if (isipv6) 199 in6_pcbdetach(inp); 200 else 201 #endif 202 in_pcbdetach(inp); 203 INP_UNLOCK(inp); 204 } 205 } else { 206 /* 207 * If the connection is not in timewait, we consider two 208 * two conditions: one in which no further processing is 209 * necessary (dropped || embryonic), and one in which TCP is 210 * not yet done, but no longer requires the socket, so the 211 * pcb will persist for the time being. 212 * 213 * XXXRW: Does the second case still occur? 214 */ 215 if (inp->inp_vflag & INP_DROPPED || 216 tp->t_state < TCPS_SYN_SENT) { 217 tcp_discardcb(tp); 218 #ifdef INET6 219 if (isipv6) { 220 in6_pcbdetach(inp); 221 in6_pcbfree(inp); 222 } else { 223 #endif 224 in_pcbdetach(inp); 225 in_pcbfree(inp); 226 #ifdef INET6 227 } 228 #endif 229 } else { 230 #ifdef INET6 231 if (isipv6) 232 in6_pcbdetach(inp); 233 else 234 #endif 235 in_pcbdetach(inp); 236 } 237 } 238 } 239 240 /* 241 * pru_detach() detaches the TCP protocol from the socket. 242 * If the protocol state is non-embryonic, then can't 243 * do this directly: have to initiate a pru_disconnect(), 244 * which may finish later; embryonic TCB's can just 245 * be discarded here. 246 */ 247 static void 248 tcp_usr_detach(struct socket *so) 249 { 250 struct inpcb *inp; 251 252 inp = sotoinpcb(so); 253 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 254 INP_INFO_WLOCK(&tcbinfo); 255 INP_LOCK(inp); 256 KASSERT(inp->inp_socket != NULL, 257 ("tcp_usr_detach: inp_socket == NULL")); 258 tcp_detach(so, inp); 259 INP_INFO_WUNLOCK(&tcbinfo); 260 } 261 262 /* 263 * Give the socket an address. 264 */ 265 static int 266 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 267 { 268 int error = 0; 269 struct inpcb *inp; 270 struct tcpcb *tp = NULL; 271 struct sockaddr_in *sinp; 272 273 sinp = (struct sockaddr_in *)nam; 274 if (nam->sa_len != sizeof (*sinp)) 275 return (EINVAL); 276 /* 277 * Must check for multicast addresses and disallow binding 278 * to them. 279 */ 280 if (sinp->sin_family == AF_INET && 281 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 282 return (EAFNOSUPPORT); 283 284 TCPDEBUG0; 285 INP_INFO_WLOCK(&tcbinfo); 286 inp = sotoinpcb(so); 287 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 288 INP_LOCK(inp); 289 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 290 error = EINVAL; 291 goto out; 292 } 293 tp = intotcpcb(inp); 294 TCPDEBUG1(); 295 error = in_pcbbind(inp, nam, td->td_ucred); 296 out: 297 TCPDEBUG2(PRU_BIND); 298 INP_UNLOCK(inp); 299 INP_INFO_WUNLOCK(&tcbinfo); 300 301 return (error); 302 } 303 304 #ifdef INET6 305 static int 306 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 307 { 308 int error = 0; 309 struct inpcb *inp; 310 struct tcpcb *tp = NULL; 311 struct sockaddr_in6 *sin6p; 312 313 sin6p = (struct sockaddr_in6 *)nam; 314 if (nam->sa_len != sizeof (*sin6p)) 315 return (EINVAL); 316 /* 317 * Must check for multicast addresses and disallow binding 318 * to them. 319 */ 320 if (sin6p->sin6_family == AF_INET6 && 321 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 322 return (EAFNOSUPPORT); 323 324 TCPDEBUG0; 325 INP_INFO_WLOCK(&tcbinfo); 326 inp = sotoinpcb(so); 327 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 328 INP_LOCK(inp); 329 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 330 error = EINVAL; 331 goto out; 332 } 333 tp = intotcpcb(inp); 334 TCPDEBUG1(); 335 inp->inp_vflag &= ~INP_IPV4; 336 inp->inp_vflag |= INP_IPV6; 337 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 338 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 339 inp->inp_vflag |= INP_IPV4; 340 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 341 struct sockaddr_in sin; 342 343 in6_sin6_2_sin(&sin, sin6p); 344 inp->inp_vflag |= INP_IPV4; 345 inp->inp_vflag &= ~INP_IPV6; 346 error = in_pcbbind(inp, (struct sockaddr *)&sin, 347 td->td_ucred); 348 goto out; 349 } 350 } 351 error = in6_pcbbind(inp, nam, td->td_ucred); 352 out: 353 TCPDEBUG2(PRU_BIND); 354 INP_UNLOCK(inp); 355 INP_INFO_WUNLOCK(&tcbinfo); 356 return (error); 357 } 358 #endif /* INET6 */ 359 360 /* 361 * Prepare to accept connections. 362 */ 363 static int 364 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 365 { 366 int error = 0; 367 struct inpcb *inp; 368 struct tcpcb *tp = NULL; 369 370 TCPDEBUG0; 371 INP_INFO_WLOCK(&tcbinfo); 372 inp = sotoinpcb(so); 373 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 374 INP_LOCK(inp); 375 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 376 error = EINVAL; 377 goto out; 378 } 379 tp = intotcpcb(inp); 380 TCPDEBUG1(); 381 SOCK_LOCK(so); 382 error = solisten_proto_check(so); 383 if (error == 0 && inp->inp_lport == 0) 384 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 385 if (error == 0) { 386 tp->t_state = TCPS_LISTEN; 387 solisten_proto(so, backlog); 388 } 389 SOCK_UNLOCK(so); 390 391 out: 392 TCPDEBUG2(PRU_LISTEN); 393 INP_UNLOCK(inp); 394 INP_INFO_WUNLOCK(&tcbinfo); 395 return (error); 396 } 397 398 #ifdef INET6 399 static int 400 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 401 { 402 int error = 0; 403 struct inpcb *inp; 404 struct tcpcb *tp = NULL; 405 406 TCPDEBUG0; 407 INP_INFO_WLOCK(&tcbinfo); 408 inp = sotoinpcb(so); 409 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 410 INP_LOCK(inp); 411 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 412 error = EINVAL; 413 goto out; 414 } 415 tp = intotcpcb(inp); 416 TCPDEBUG1(); 417 SOCK_LOCK(so); 418 error = solisten_proto_check(so); 419 if (error == 0 && inp->inp_lport == 0) { 420 inp->inp_vflag &= ~INP_IPV4; 421 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 422 inp->inp_vflag |= INP_IPV4; 423 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 424 } 425 if (error == 0) { 426 tp->t_state = TCPS_LISTEN; 427 solisten_proto(so, backlog); 428 } 429 SOCK_UNLOCK(so); 430 431 out: 432 TCPDEBUG2(PRU_LISTEN); 433 INP_UNLOCK(inp); 434 INP_INFO_WUNLOCK(&tcbinfo); 435 return (error); 436 } 437 #endif /* INET6 */ 438 439 /* 440 * Initiate connection to peer. 441 * Create a template for use in transmissions on this connection. 442 * Enter SYN_SENT state, and mark socket as connecting. 443 * Start keep-alive timer, and seed output sequence space. 444 * Send initial segment on connection. 445 */ 446 static int 447 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 448 { 449 int error = 0; 450 struct inpcb *inp; 451 struct tcpcb *tp = NULL; 452 struct sockaddr_in *sinp; 453 454 sinp = (struct sockaddr_in *)nam; 455 if (nam->sa_len != sizeof (*sinp)) 456 return (EINVAL); 457 /* 458 * Must disallow TCP ``connections'' to multicast addresses. 459 */ 460 if (sinp->sin_family == AF_INET 461 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 462 return (EAFNOSUPPORT); 463 if (jailed(td->td_ucred)) 464 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 465 466 TCPDEBUG0; 467 INP_INFO_WLOCK(&tcbinfo); 468 inp = sotoinpcb(so); 469 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 470 INP_LOCK(inp); 471 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 472 error = EINVAL; 473 goto out; 474 } 475 tp = intotcpcb(inp); 476 TCPDEBUG1(); 477 if ((error = tcp_connect(tp, nam, td)) != 0) 478 goto out; 479 error = tcp_output(tp); 480 out: 481 TCPDEBUG2(PRU_CONNECT); 482 INP_UNLOCK(inp); 483 INP_INFO_WUNLOCK(&tcbinfo); 484 return (error); 485 } 486 487 #ifdef INET6 488 static int 489 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 490 { 491 int error = 0; 492 struct inpcb *inp; 493 struct tcpcb *tp = NULL; 494 struct sockaddr_in6 *sin6p; 495 496 TCPDEBUG0; 497 498 sin6p = (struct sockaddr_in6 *)nam; 499 if (nam->sa_len != sizeof (*sin6p)) 500 return (EINVAL); 501 /* 502 * Must disallow TCP ``connections'' to multicast addresses. 503 */ 504 if (sin6p->sin6_family == AF_INET6 505 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 506 return (EAFNOSUPPORT); 507 508 INP_INFO_WLOCK(&tcbinfo); 509 inp = sotoinpcb(so); 510 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 511 INP_LOCK(inp); 512 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 513 error = EINVAL; 514 goto out; 515 } 516 tp = intotcpcb(inp); 517 TCPDEBUG1(); 518 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 519 struct sockaddr_in sin; 520 521 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 522 error = EINVAL; 523 goto out; 524 } 525 526 in6_sin6_2_sin(&sin, sin6p); 527 inp->inp_vflag |= INP_IPV4; 528 inp->inp_vflag &= ~INP_IPV6; 529 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 530 goto out; 531 error = tcp_output(tp); 532 goto out; 533 } 534 inp->inp_vflag &= ~INP_IPV4; 535 inp->inp_vflag |= INP_IPV6; 536 inp->inp_inc.inc_isipv6 = 1; 537 if ((error = tcp6_connect(tp, nam, td)) != 0) 538 goto out; 539 error = tcp_output(tp); 540 541 out: 542 TCPDEBUG2(PRU_CONNECT); 543 INP_UNLOCK(inp); 544 INP_INFO_WUNLOCK(&tcbinfo); 545 return (error); 546 } 547 #endif /* INET6 */ 548 549 /* 550 * Initiate disconnect from peer. 551 * If connection never passed embryonic stage, just drop; 552 * else if don't need to let data drain, then can just drop anyways, 553 * else have to begin TCP shutdown process: mark socket disconnecting, 554 * drain unread data, state switch to reflect user close, and 555 * send segment (e.g. FIN) to peer. Socket will be really disconnected 556 * when peer sends FIN and acks ours. 557 * 558 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 559 */ 560 static int 561 tcp_usr_disconnect(struct socket *so) 562 { 563 struct inpcb *inp; 564 struct tcpcb *tp = NULL; 565 int error = 0; 566 567 TCPDEBUG0; 568 INP_INFO_WLOCK(&tcbinfo); 569 inp = sotoinpcb(so); 570 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 571 INP_LOCK(inp); 572 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 573 error = ECONNRESET; 574 goto out; 575 } 576 tp = intotcpcb(inp); 577 TCPDEBUG1(); 578 tcp_disconnect(tp); 579 out: 580 TCPDEBUG2(PRU_DISCONNECT); 581 INP_UNLOCK(inp); 582 INP_INFO_WUNLOCK(&tcbinfo); 583 return (error); 584 } 585 586 /* 587 * Accept a connection. Essentially all the work is 588 * done at higher levels; just return the address 589 * of the peer, storing through addr. 590 */ 591 static int 592 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 593 { 594 int error = 0; 595 struct inpcb *inp = NULL; 596 struct tcpcb *tp = NULL; 597 struct in_addr addr; 598 in_port_t port = 0; 599 TCPDEBUG0; 600 601 if (so->so_state & SS_ISDISCONNECTED) 602 return (ECONNABORTED); 603 604 inp = sotoinpcb(so); 605 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 606 INP_LOCK(inp); 607 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 608 error = ECONNABORTED; 609 goto out; 610 } 611 tp = intotcpcb(inp); 612 TCPDEBUG1(); 613 614 /* 615 * We inline in_getpeeraddr and COMMON_END here, so that we can 616 * copy the data of interest and defer the malloc until after we 617 * release the lock. 618 */ 619 port = inp->inp_fport; 620 addr = inp->inp_faddr; 621 622 out: 623 TCPDEBUG2(PRU_ACCEPT); 624 INP_UNLOCK(inp); 625 if (error == 0) 626 *nam = in_sockaddr(port, &addr); 627 return error; 628 } 629 630 #ifdef INET6 631 static int 632 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 633 { 634 struct inpcb *inp = NULL; 635 int error = 0; 636 struct tcpcb *tp = NULL; 637 struct in_addr addr; 638 struct in6_addr addr6; 639 in_port_t port = 0; 640 int v4 = 0; 641 TCPDEBUG0; 642 643 if (so->so_state & SS_ISDISCONNECTED) 644 return (ECONNABORTED); 645 646 inp = sotoinpcb(so); 647 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 648 INP_LOCK(inp); 649 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 650 error = ECONNABORTED; 651 goto out; 652 } 653 tp = intotcpcb(inp); 654 TCPDEBUG1(); 655 656 /* 657 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 658 * copy the data of interest and defer the malloc until after we 659 * release the lock. 660 */ 661 if (inp->inp_vflag & INP_IPV4) { 662 v4 = 1; 663 port = inp->inp_fport; 664 addr = inp->inp_faddr; 665 } else { 666 port = inp->inp_fport; 667 addr6 = inp->in6p_faddr; 668 } 669 670 out: 671 TCPDEBUG2(PRU_ACCEPT); 672 INP_UNLOCK(inp); 673 if (error == 0) { 674 if (v4) 675 *nam = in6_v4mapsin6_sockaddr(port, &addr); 676 else 677 *nam = in6_sockaddr(port, &addr6); 678 } 679 return error; 680 } 681 #endif /* INET6 */ 682 683 /* 684 * Mark the connection as being incapable of further output. 685 */ 686 static int 687 tcp_usr_shutdown(struct socket *so) 688 { 689 int error = 0; 690 struct inpcb *inp; 691 struct tcpcb *tp = NULL; 692 693 TCPDEBUG0; 694 INP_INFO_WLOCK(&tcbinfo); 695 inp = sotoinpcb(so); 696 KASSERT(inp != NULL, ("inp == NULL")); 697 INP_LOCK(inp); 698 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 699 error = ECONNRESET; 700 goto out; 701 } 702 tp = intotcpcb(inp); 703 TCPDEBUG1(); 704 socantsendmore(so); 705 tcp_usrclosed(tp); 706 error = tcp_output(tp); 707 708 out: 709 TCPDEBUG2(PRU_SHUTDOWN); 710 INP_UNLOCK(inp); 711 INP_INFO_WUNLOCK(&tcbinfo); 712 713 return (error); 714 } 715 716 /* 717 * After a receive, possibly send window update to peer. 718 */ 719 static int 720 tcp_usr_rcvd(struct socket *so, int flags) 721 { 722 struct inpcb *inp; 723 struct tcpcb *tp = NULL; 724 int error = 0; 725 726 TCPDEBUG0; 727 inp = sotoinpcb(so); 728 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 729 INP_LOCK(inp); 730 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 731 error = ECONNRESET; 732 goto out; 733 } 734 tp = intotcpcb(inp); 735 TCPDEBUG1(); 736 tcp_output(tp); 737 738 out: 739 TCPDEBUG2(PRU_RCVD); 740 INP_UNLOCK(inp); 741 return (error); 742 } 743 744 /* 745 * Do a send by putting data in output queue and updating urgent 746 * marker if URG set. Possibly send more data. Unlike the other 747 * pru_*() routines, the mbuf chains are our responsibility. We 748 * must either enqueue them or free them. The other pru_* routines 749 * generally are caller-frees. 750 */ 751 static int 752 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 753 struct sockaddr *nam, struct mbuf *control, struct thread *td) 754 { 755 int error = 0; 756 struct inpcb *inp; 757 struct tcpcb *tp = NULL; 758 int headlocked = 0; 759 #ifdef INET6 760 int isipv6; 761 #endif 762 TCPDEBUG0; 763 764 /* 765 * We require the pcbinfo lock in two cases: 766 * 767 * (1) An implied connect is taking place, which can result in 768 * binding IPs and ports and hence modification of the pcb hash 769 * chains. 770 * 771 * (2) PRUS_EOF is set, resulting in explicit close on the send. 772 */ 773 if ((nam != NULL) || (flags & PRUS_EOF)) { 774 INP_INFO_WLOCK(&tcbinfo); 775 headlocked = 1; 776 } 777 inp = sotoinpcb(so); 778 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 779 INP_LOCK(inp); 780 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 781 if (control) 782 m_freem(control); 783 if (m) 784 m_freem(m); 785 error = ECONNRESET; 786 goto out; 787 } 788 #ifdef INET6 789 isipv6 = nam && nam->sa_family == AF_INET6; 790 #endif /* INET6 */ 791 tp = intotcpcb(inp); 792 TCPDEBUG1(); 793 if (control) { 794 /* TCP doesn't do control messages (rights, creds, etc) */ 795 if (control->m_len) { 796 m_freem(control); 797 if (m) 798 m_freem(m); 799 error = EINVAL; 800 goto out; 801 } 802 m_freem(control); /* empty control, just free it */ 803 } 804 if (!(flags & PRUS_OOB)) { 805 sbappendstream(&so->so_snd, m); 806 if (nam && tp->t_state < TCPS_SYN_SENT) { 807 /* 808 * Do implied connect if not yet connected, 809 * initialize window to default value, and 810 * initialize maxseg/maxopd using peer's cached 811 * MSS. 812 */ 813 INP_INFO_WLOCK_ASSERT(&tcbinfo); 814 #ifdef INET6 815 if (isipv6) 816 error = tcp6_connect(tp, nam, td); 817 else 818 #endif /* INET6 */ 819 error = tcp_connect(tp, nam, td); 820 if (error) 821 goto out; 822 tp->snd_wnd = TTCP_CLIENT_SND_WND; 823 tcp_mss(tp, -1); 824 } 825 if (flags & PRUS_EOF) { 826 /* 827 * Close the send side of the connection after 828 * the data is sent. 829 */ 830 INP_INFO_WLOCK_ASSERT(&tcbinfo); 831 socantsendmore(so); 832 tcp_usrclosed(tp); 833 } 834 if (headlocked) { 835 INP_INFO_WUNLOCK(&tcbinfo); 836 headlocked = 0; 837 } 838 if (tp != NULL) { 839 if (flags & PRUS_MORETOCOME) 840 tp->t_flags |= TF_MORETOCOME; 841 error = tcp_output(tp); 842 if (flags & PRUS_MORETOCOME) 843 tp->t_flags &= ~TF_MORETOCOME; 844 } 845 } else { 846 /* 847 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 848 */ 849 SOCKBUF_LOCK(&so->so_snd); 850 if (sbspace(&so->so_snd) < -512) { 851 SOCKBUF_UNLOCK(&so->so_snd); 852 m_freem(m); 853 error = ENOBUFS; 854 goto out; 855 } 856 /* 857 * According to RFC961 (Assigned Protocols), 858 * the urgent pointer points to the last octet 859 * of urgent data. We continue, however, 860 * to consider it to indicate the first octet 861 * of data past the urgent section. 862 * Otherwise, snd_up should be one lower. 863 */ 864 sbappendstream_locked(&so->so_snd, m); 865 SOCKBUF_UNLOCK(&so->so_snd); 866 if (nam && tp->t_state < TCPS_SYN_SENT) { 867 /* 868 * Do implied connect if not yet connected, 869 * initialize window to default value, and 870 * initialize maxseg/maxopd using peer's cached 871 * MSS. 872 */ 873 INP_INFO_WLOCK_ASSERT(&tcbinfo); 874 #ifdef INET6 875 if (isipv6) 876 error = tcp6_connect(tp, nam, td); 877 else 878 #endif /* INET6 */ 879 error = tcp_connect(tp, nam, td); 880 if (error) 881 goto out; 882 tp->snd_wnd = TTCP_CLIENT_SND_WND; 883 tcp_mss(tp, -1); 884 INP_INFO_WUNLOCK(&tcbinfo); 885 headlocked = 0; 886 } else if (nam) { 887 INP_INFO_WUNLOCK(&tcbinfo); 888 headlocked = 0; 889 } 890 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 891 tp->t_flags |= TF_FORCEDATA; 892 error = tcp_output(tp); 893 tp->t_flags &= ~TF_FORCEDATA; 894 } 895 out: 896 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 897 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 898 INP_UNLOCK(inp); 899 if (headlocked) 900 INP_INFO_WUNLOCK(&tcbinfo); 901 return (error); 902 } 903 904 /* 905 * Abort the TCP. Drop the connection abruptly. 906 */ 907 static void 908 tcp_usr_abort(struct socket *so) 909 { 910 struct inpcb *inp; 911 struct tcpcb *tp = NULL; 912 TCPDEBUG0; 913 914 inp = sotoinpcb(so); 915 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 916 917 INP_INFO_WLOCK(&tcbinfo); 918 INP_LOCK(inp); 919 KASSERT(inp->inp_socket != NULL, 920 ("tcp_usr_abort: inp_socket == NULL")); 921 922 /* 923 * If we still have full TCP state, and we're not dropped, drop. 924 */ 925 if (!(inp->inp_vflag & INP_TIMEWAIT) && 926 !(inp->inp_vflag & INP_DROPPED)) { 927 tp = intotcpcb(inp); 928 TCPDEBUG1(); 929 tcp_drop(tp, ECONNABORTED); 930 TCPDEBUG2(PRU_ABORT); 931 } 932 if (!(inp->inp_vflag & INP_DROPPED)) { 933 SOCK_LOCK(so); 934 so->so_state |= SS_PROTOREF; 935 SOCK_UNLOCK(so); 936 inp->inp_vflag |= INP_SOCKREF; 937 } 938 INP_UNLOCK(inp); 939 INP_INFO_WUNLOCK(&tcbinfo); 940 } 941 942 /* 943 * TCP socket is closed. Start friendly disconnect. 944 */ 945 static void 946 tcp_usr_close(struct socket *so) 947 { 948 struct inpcb *inp; 949 struct tcpcb *tp = NULL; 950 TCPDEBUG0; 951 952 inp = sotoinpcb(so); 953 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 954 955 INP_INFO_WLOCK(&tcbinfo); 956 INP_LOCK(inp); 957 KASSERT(inp->inp_socket != NULL, 958 ("tcp_usr_close: inp_socket == NULL")); 959 960 /* 961 * If we still have full TCP state, and we're not dropped, initiate 962 * a disconnect. 963 */ 964 if (!(inp->inp_vflag & INP_TIMEWAIT) && 965 !(inp->inp_vflag & INP_DROPPED)) { 966 tp = intotcpcb(inp); 967 TCPDEBUG1(); 968 tcp_disconnect(tp); 969 TCPDEBUG2(PRU_CLOSE); 970 } 971 if (!(inp->inp_vflag & INP_DROPPED)) { 972 SOCK_LOCK(so); 973 so->so_state |= SS_PROTOREF; 974 SOCK_UNLOCK(so); 975 inp->inp_vflag |= INP_SOCKREF; 976 } 977 INP_UNLOCK(inp); 978 INP_INFO_WUNLOCK(&tcbinfo); 979 } 980 981 /* 982 * Receive out-of-band data. 983 */ 984 static int 985 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 986 { 987 int error = 0; 988 struct inpcb *inp; 989 struct tcpcb *tp = NULL; 990 991 TCPDEBUG0; 992 inp = sotoinpcb(so); 993 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 994 INP_LOCK(inp); 995 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 996 error = ECONNRESET; 997 goto out; 998 } 999 tp = intotcpcb(inp); 1000 TCPDEBUG1(); 1001 if ((so->so_oobmark == 0 && 1002 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1003 so->so_options & SO_OOBINLINE || 1004 tp->t_oobflags & TCPOOB_HADDATA) { 1005 error = EINVAL; 1006 goto out; 1007 } 1008 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1009 error = EWOULDBLOCK; 1010 goto out; 1011 } 1012 m->m_len = 1; 1013 *mtod(m, caddr_t) = tp->t_iobc; 1014 if ((flags & MSG_PEEK) == 0) 1015 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1016 1017 out: 1018 TCPDEBUG2(PRU_RCVOOB); 1019 INP_UNLOCK(inp); 1020 return (error); 1021 } 1022 1023 struct pr_usrreqs tcp_usrreqs = { 1024 .pru_abort = tcp_usr_abort, 1025 .pru_accept = tcp_usr_accept, 1026 .pru_attach = tcp_usr_attach, 1027 .pru_bind = tcp_usr_bind, 1028 .pru_connect = tcp_usr_connect, 1029 .pru_control = in_control, 1030 .pru_detach = tcp_usr_detach, 1031 .pru_disconnect = tcp_usr_disconnect, 1032 .pru_listen = tcp_usr_listen, 1033 .pru_peeraddr = in_getpeeraddr, 1034 .pru_rcvd = tcp_usr_rcvd, 1035 .pru_rcvoob = tcp_usr_rcvoob, 1036 .pru_send = tcp_usr_send, 1037 .pru_shutdown = tcp_usr_shutdown, 1038 .pru_sockaddr = in_getsockaddr, 1039 .pru_sosetlabel = in_pcbsosetlabel, 1040 .pru_close = tcp_usr_close, 1041 }; 1042 1043 #ifdef INET6 1044 struct pr_usrreqs tcp6_usrreqs = { 1045 .pru_abort = tcp_usr_abort, 1046 .pru_accept = tcp6_usr_accept, 1047 .pru_attach = tcp_usr_attach, 1048 .pru_bind = tcp6_usr_bind, 1049 .pru_connect = tcp6_usr_connect, 1050 .pru_control = in6_control, 1051 .pru_detach = tcp_usr_detach, 1052 .pru_disconnect = tcp_usr_disconnect, 1053 .pru_listen = tcp6_usr_listen, 1054 .pru_peeraddr = in6_mapped_peeraddr, 1055 .pru_rcvd = tcp_usr_rcvd, 1056 .pru_rcvoob = tcp_usr_rcvoob, 1057 .pru_send = tcp_usr_send, 1058 .pru_shutdown = tcp_usr_shutdown, 1059 .pru_sockaddr = in6_mapped_sockaddr, 1060 .pru_sosetlabel = in_pcbsosetlabel, 1061 .pru_close = tcp_usr_close, 1062 }; 1063 #endif /* INET6 */ 1064 1065 /* 1066 * Common subroutine to open a TCP connection to remote host specified 1067 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1068 * port number if needed. Call in_pcbconnect_setup to do the routing and 1069 * to choose a local host address (interface). If there is an existing 1070 * incarnation of the same connection in TIME-WAIT state and if the remote 1071 * host was sending CC options and if the connection duration was < MSL, then 1072 * truncate the previous TIME-WAIT state and proceed. 1073 * Initialize connection parameters and enter SYN-SENT state. 1074 */ 1075 static int 1076 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1077 { 1078 struct inpcb *inp = tp->t_inpcb, *oinp; 1079 struct socket *so = inp->inp_socket; 1080 struct in_addr laddr; 1081 u_short lport; 1082 int error; 1083 1084 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1085 INP_LOCK_ASSERT(inp); 1086 1087 if (inp->inp_lport == 0) { 1088 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1089 if (error) 1090 return error; 1091 } 1092 1093 /* 1094 * Cannot simply call in_pcbconnect, because there might be an 1095 * earlier incarnation of this same connection still in 1096 * TIME_WAIT state, creating an ADDRINUSE error. 1097 */ 1098 laddr = inp->inp_laddr; 1099 lport = inp->inp_lport; 1100 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1101 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1102 if (error && oinp == NULL) 1103 return error; 1104 if (oinp) 1105 return EADDRINUSE; 1106 inp->inp_laddr = laddr; 1107 in_pcbrehash(inp); 1108 1109 /* 1110 * Compute window scaling to request: 1111 * Scale to fit into sweet spot. See tcp_syncache.c. 1112 * XXX: This should move to tcp_output(). 1113 */ 1114 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1115 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1116 tp->request_r_scale++; 1117 1118 soisconnecting(so); 1119 tcpstat.tcps_connattempt++; 1120 tp->t_state = TCPS_SYN_SENT; 1121 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1122 tp->iss = tcp_new_isn(tp); 1123 tp->t_bw_rtseq = tp->iss; 1124 tcp_sendseqinit(tp); 1125 1126 return 0; 1127 } 1128 1129 #ifdef INET6 1130 static int 1131 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1132 { 1133 struct inpcb *inp = tp->t_inpcb, *oinp; 1134 struct socket *so = inp->inp_socket; 1135 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1136 struct in6_addr *addr6; 1137 int error; 1138 1139 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1140 INP_LOCK_ASSERT(inp); 1141 1142 if (inp->inp_lport == 0) { 1143 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1144 if (error) 1145 return error; 1146 } 1147 1148 /* 1149 * Cannot simply call in_pcbconnect, because there might be an 1150 * earlier incarnation of this same connection still in 1151 * TIME_WAIT state, creating an ADDRINUSE error. 1152 * in6_pcbladdr() also handles scope zone IDs. 1153 */ 1154 error = in6_pcbladdr(inp, nam, &addr6); 1155 if (error) 1156 return error; 1157 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1158 &sin6->sin6_addr, sin6->sin6_port, 1159 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1160 ? addr6 1161 : &inp->in6p_laddr, 1162 inp->inp_lport, 0, NULL); 1163 if (oinp) 1164 return EADDRINUSE; 1165 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1166 inp->in6p_laddr = *addr6; 1167 inp->in6p_faddr = sin6->sin6_addr; 1168 inp->inp_fport = sin6->sin6_port; 1169 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1170 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 1171 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 1172 inp->in6p_flowinfo |= 1173 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1174 in_pcbrehash(inp); 1175 1176 /* Compute window scaling to request. */ 1177 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1178 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 1179 tp->request_r_scale++; 1180 1181 soisconnecting(so); 1182 tcpstat.tcps_connattempt++; 1183 tp->t_state = TCPS_SYN_SENT; 1184 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1185 tp->iss = tcp_new_isn(tp); 1186 tp->t_bw_rtseq = tp->iss; 1187 tcp_sendseqinit(tp); 1188 1189 return 0; 1190 } 1191 #endif /* INET6 */ 1192 1193 /* 1194 * Export TCP internal state information via a struct tcp_info, based on the 1195 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1196 * (TCP state machine, etc). We export all information using FreeBSD-native 1197 * constants -- for example, the numeric values for tcpi_state will differ 1198 * from Linux. 1199 */ 1200 static void 1201 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1202 { 1203 1204 INP_LOCK_ASSERT(tp->t_inpcb); 1205 bzero(ti, sizeof(*ti)); 1206 1207 ti->tcpi_state = tp->t_state; 1208 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1209 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1210 if (tp->t_flags & TF_SACK_PERMIT) 1211 ti->tcpi_options |= TCPI_OPT_SACK; 1212 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1213 ti->tcpi_options |= TCPI_OPT_WSCALE; 1214 ti->tcpi_snd_wscale = tp->snd_scale; 1215 ti->tcpi_rcv_wscale = tp->rcv_scale; 1216 } 1217 1218 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1219 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1220 1221 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1222 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1223 1224 /* 1225 * FreeBSD-specific extension fields for tcp_info. 1226 */ 1227 ti->tcpi_rcv_space = tp->rcv_wnd; 1228 ti->tcpi_snd_wnd = tp->snd_wnd; 1229 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1230 } 1231 1232 /* 1233 * The new sockopt interface makes it possible for us to block in the 1234 * copyin/out step (if we take a page fault). Taking a page fault at 1235 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1236 * use TSM, there probably isn't any need for this function to run at 1237 * splnet() any more. This needs more examination.) 1238 * 1239 * XXXRW: The locking here is wrong; we may take a page fault while holding 1240 * the inpcb lock. 1241 */ 1242 int 1243 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1244 { 1245 int error, opt, optval; 1246 struct inpcb *inp; 1247 struct tcpcb *tp; 1248 struct tcp_info ti; 1249 1250 error = 0; 1251 inp = sotoinpcb(so); 1252 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1253 INP_LOCK(inp); 1254 if (sopt->sopt_level != IPPROTO_TCP) { 1255 INP_UNLOCK(inp); 1256 #ifdef INET6 1257 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1258 error = ip6_ctloutput(so, sopt); 1259 else 1260 #endif /* INET6 */ 1261 error = ip_ctloutput(so, sopt); 1262 return (error); 1263 } 1264 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1265 error = ECONNRESET; 1266 goto out; 1267 } 1268 tp = intotcpcb(inp); 1269 1270 switch (sopt->sopt_dir) { 1271 case SOPT_SET: 1272 switch (sopt->sopt_name) { 1273 #ifdef TCP_SIGNATURE 1274 case TCP_MD5SIG: 1275 error = sooptcopyin(sopt, &optval, sizeof optval, 1276 sizeof optval); 1277 if (error) 1278 break; 1279 1280 if (optval > 0) 1281 tp->t_flags |= TF_SIGNATURE; 1282 else 1283 tp->t_flags &= ~TF_SIGNATURE; 1284 break; 1285 #endif /* TCP_SIGNATURE */ 1286 case TCP_NODELAY: 1287 case TCP_NOOPT: 1288 error = sooptcopyin(sopt, &optval, sizeof optval, 1289 sizeof optval); 1290 if (error) 1291 break; 1292 1293 switch (sopt->sopt_name) { 1294 case TCP_NODELAY: 1295 opt = TF_NODELAY; 1296 break; 1297 case TCP_NOOPT: 1298 opt = TF_NOOPT; 1299 break; 1300 default: 1301 opt = 0; /* dead code to fool gcc */ 1302 break; 1303 } 1304 1305 if (optval) 1306 tp->t_flags |= opt; 1307 else 1308 tp->t_flags &= ~opt; 1309 break; 1310 1311 case TCP_NOPUSH: 1312 error = sooptcopyin(sopt, &optval, sizeof optval, 1313 sizeof optval); 1314 if (error) 1315 break; 1316 1317 if (optval) 1318 tp->t_flags |= TF_NOPUSH; 1319 else { 1320 tp->t_flags &= ~TF_NOPUSH; 1321 error = tcp_output(tp); 1322 } 1323 break; 1324 1325 case TCP_MAXSEG: 1326 error = sooptcopyin(sopt, &optval, sizeof optval, 1327 sizeof optval); 1328 if (error) 1329 break; 1330 1331 if (optval > 0 && optval <= tp->t_maxseg && 1332 optval + 40 >= tcp_minmss) 1333 tp->t_maxseg = optval; 1334 else 1335 error = EINVAL; 1336 break; 1337 1338 case TCP_INFO: 1339 error = EINVAL; 1340 break; 1341 1342 default: 1343 error = ENOPROTOOPT; 1344 break; 1345 } 1346 break; 1347 1348 case SOPT_GET: 1349 switch (sopt->sopt_name) { 1350 #ifdef TCP_SIGNATURE 1351 case TCP_MD5SIG: 1352 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1353 error = sooptcopyout(sopt, &optval, sizeof optval); 1354 break; 1355 #endif 1356 case TCP_NODELAY: 1357 optval = tp->t_flags & TF_NODELAY; 1358 error = sooptcopyout(sopt, &optval, sizeof optval); 1359 break; 1360 case TCP_MAXSEG: 1361 optval = tp->t_maxseg; 1362 error = sooptcopyout(sopt, &optval, sizeof optval); 1363 break; 1364 case TCP_NOOPT: 1365 optval = tp->t_flags & TF_NOOPT; 1366 error = sooptcopyout(sopt, &optval, sizeof optval); 1367 break; 1368 case TCP_NOPUSH: 1369 optval = tp->t_flags & TF_NOPUSH; 1370 error = sooptcopyout(sopt, &optval, sizeof optval); 1371 break; 1372 case TCP_INFO: 1373 tcp_fill_info(tp, &ti); 1374 error = sooptcopyout(sopt, &ti, sizeof ti); 1375 break; 1376 default: 1377 error = ENOPROTOOPT; 1378 break; 1379 } 1380 break; 1381 } 1382 out: 1383 INP_UNLOCK(inp); 1384 return (error); 1385 } 1386 1387 /* 1388 * tcp_sendspace and tcp_recvspace are the default send and receive window 1389 * sizes, respectively. These are obsolescent (this information should 1390 * be set by the route). 1391 */ 1392 u_long tcp_sendspace = 1024*32; 1393 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1394 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1395 u_long tcp_recvspace = 1024*64; 1396 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1397 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1398 1399 /* 1400 * Attach TCP protocol to socket, allocating 1401 * internet protocol control block, tcp control block, 1402 * bufer space, and entering LISTEN state if to accept connections. 1403 */ 1404 static int 1405 tcp_attach(struct socket *so) 1406 { 1407 struct tcpcb *tp; 1408 struct inpcb *inp; 1409 int error; 1410 #ifdef INET6 1411 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1412 #endif 1413 1414 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1415 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1416 if (error) 1417 return (error); 1418 } 1419 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1420 so->so_snd.sb_flags |= SB_AUTOSIZE; 1421 INP_INFO_WLOCK(&tcbinfo); 1422 error = in_pcballoc(so, &tcbinfo); 1423 if (error) { 1424 INP_INFO_WUNLOCK(&tcbinfo); 1425 return (error); 1426 } 1427 inp = sotoinpcb(so); 1428 #ifdef INET6 1429 if (isipv6) { 1430 inp->inp_vflag |= INP_IPV6; 1431 inp->in6p_hops = -1; /* use kernel default */ 1432 } 1433 else 1434 #endif 1435 inp->inp_vflag |= INP_IPV4; 1436 tp = tcp_newtcpcb(inp); 1437 if (tp == NULL) { 1438 #ifdef INET6 1439 if (isipv6) { 1440 in6_pcbdetach(inp); 1441 in6_pcbfree(inp); 1442 } else { 1443 #endif 1444 in_pcbdetach(inp); 1445 in_pcbfree(inp); 1446 #ifdef INET6 1447 } 1448 #endif 1449 INP_INFO_WUNLOCK(&tcbinfo); 1450 return (ENOBUFS); 1451 } 1452 tp->t_state = TCPS_CLOSED; 1453 INP_UNLOCK(inp); 1454 INP_INFO_WUNLOCK(&tcbinfo); 1455 return (0); 1456 } 1457 1458 /* 1459 * Initiate (or continue) disconnect. 1460 * If embryonic state, just send reset (once). 1461 * If in ``let data drain'' option and linger null, just drop. 1462 * Otherwise (hard), mark socket disconnecting and drop 1463 * current input data; switch states based on user close, and 1464 * send segment to peer (with FIN). 1465 */ 1466 static void 1467 tcp_disconnect(struct tcpcb *tp) 1468 { 1469 struct inpcb *inp = tp->t_inpcb; 1470 struct socket *so = inp->inp_socket; 1471 1472 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1473 INP_LOCK_ASSERT(inp); 1474 1475 /* 1476 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1477 * socket is still open. 1478 */ 1479 if (tp->t_state < TCPS_ESTABLISHED) { 1480 tp = tcp_close(tp); 1481 KASSERT(tp != NULL, 1482 ("tcp_disconnect: tcp_close() returned NULL")); 1483 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1484 tp = tcp_drop(tp, 0); 1485 KASSERT(tp != NULL, 1486 ("tcp_disconnect: tcp_drop() returned NULL")); 1487 } else { 1488 soisdisconnecting(so); 1489 sbflush(&so->so_rcv); 1490 tcp_usrclosed(tp); 1491 if (!(inp->inp_vflag & INP_DROPPED)) 1492 tcp_output(tp); 1493 } 1494 } 1495 1496 /* 1497 * User issued close, and wish to trail through shutdown states: 1498 * if never received SYN, just forget it. If got a SYN from peer, 1499 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1500 * If already got a FIN from peer, then almost done; go to LAST_ACK 1501 * state. In all other cases, have already sent FIN to peer (e.g. 1502 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1503 * for peer to send FIN or not respond to keep-alives, etc. 1504 * We can let the user exit from the close as soon as the FIN is acked. 1505 */ 1506 static void 1507 tcp_usrclosed(struct tcpcb *tp) 1508 { 1509 1510 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1511 INP_LOCK_ASSERT(tp->t_inpcb); 1512 1513 switch (tp->t_state) { 1514 case TCPS_CLOSED: 1515 case TCPS_LISTEN: 1516 tp->t_state = TCPS_CLOSED; 1517 tp = tcp_close(tp); 1518 /* 1519 * tcp_close() should never return NULL here as the socket is 1520 * still open. 1521 */ 1522 KASSERT(tp != NULL, 1523 ("tcp_usrclosed: tcp_close() returned NULL")); 1524 break; 1525 1526 case TCPS_SYN_SENT: 1527 case TCPS_SYN_RECEIVED: 1528 tp->t_flags |= TF_NEEDFIN; 1529 break; 1530 1531 case TCPS_ESTABLISHED: 1532 tp->t_state = TCPS_FIN_WAIT_1; 1533 break; 1534 1535 case TCPS_CLOSE_WAIT: 1536 tp->t_state = TCPS_LAST_ACK; 1537 break; 1538 } 1539 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1540 soisdisconnected(tp->t_inpcb->inp_socket); 1541 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1542 if (tp->t_state == TCPS_FIN_WAIT_2) { 1543 int timeout; 1544 1545 timeout = (tcp_fast_finwait2_recycle) ? 1546 tcp_finwait2_timeout : tcp_maxidle; 1547 tcp_timer_activate(tp, TT_2MSL, timeout); 1548 } 1549 } 1550 } 1551 1552 #ifdef DDB 1553 static void 1554 db_print_indent(int indent) 1555 { 1556 int i; 1557 1558 for (i = 0; i < indent; i++) 1559 db_printf(" "); 1560 } 1561 1562 static void 1563 db_print_tstate(int t_state) 1564 { 1565 1566 switch (t_state) { 1567 case TCPS_CLOSED: 1568 db_printf("TCPS_CLOSED"); 1569 return; 1570 1571 case TCPS_LISTEN: 1572 db_printf("TCPS_LISTEN"); 1573 return; 1574 1575 case TCPS_SYN_SENT: 1576 db_printf("TCPS_SYN_SENT"); 1577 return; 1578 1579 case TCPS_SYN_RECEIVED: 1580 db_printf("TCPS_SYN_RECEIVED"); 1581 return; 1582 1583 case TCPS_ESTABLISHED: 1584 db_printf("TCPS_ESTABLISHED"); 1585 return; 1586 1587 case TCPS_CLOSE_WAIT: 1588 db_printf("TCPS_CLOSE_WAIT"); 1589 return; 1590 1591 case TCPS_FIN_WAIT_1: 1592 db_printf("TCPS_FIN_WAIT_1"); 1593 return; 1594 1595 case TCPS_CLOSING: 1596 db_printf("TCPS_CLOSING"); 1597 return; 1598 1599 case TCPS_LAST_ACK: 1600 db_printf("TCPS_LAST_ACK"); 1601 return; 1602 1603 case TCPS_FIN_WAIT_2: 1604 db_printf("TCPS_FIN_WAIT_2"); 1605 return; 1606 1607 case TCPS_TIME_WAIT: 1608 db_printf("TCPS_TIME_WAIT"); 1609 return; 1610 1611 default: 1612 db_printf("unknown"); 1613 return; 1614 } 1615 } 1616 1617 static void 1618 db_print_tflags(u_int t_flags) 1619 { 1620 int comma; 1621 1622 comma = 0; 1623 if (t_flags & TF_ACKNOW) { 1624 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1625 comma = 1; 1626 } 1627 if (t_flags & TF_DELACK) { 1628 db_printf("%sTF_DELACK", comma ? ", " : ""); 1629 comma = 1; 1630 } 1631 if (t_flags & TF_NODELAY) { 1632 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1633 comma = 1; 1634 } 1635 if (t_flags & TF_NOOPT) { 1636 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1637 comma = 1; 1638 } 1639 if (t_flags & TF_SENTFIN) { 1640 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1641 comma = 1; 1642 } 1643 if (t_flags & TF_REQ_SCALE) { 1644 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1645 comma = 1; 1646 } 1647 if (t_flags & TF_RCVD_SCALE) { 1648 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1649 comma = 1; 1650 } 1651 if (t_flags & TF_REQ_TSTMP) { 1652 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1653 comma = 1; 1654 } 1655 if (t_flags & TF_RCVD_TSTMP) { 1656 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1657 comma = 1; 1658 } 1659 if (t_flags & TF_SACK_PERMIT) { 1660 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1661 comma = 1; 1662 } 1663 if (t_flags & TF_NEEDSYN) { 1664 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1665 comma = 1; 1666 } 1667 if (t_flags & TF_NEEDFIN) { 1668 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1669 comma = 1; 1670 } 1671 if (t_flags & TF_NOPUSH) { 1672 db_printf("%sTF_NOPUSH", 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_MORETOCOME) { 1680 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1681 comma = 1; 1682 } 1683 if (t_flags & TF_LQ_OVERFLOW) { 1684 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1685 comma = 1; 1686 } 1687 if (t_flags & TF_LASTIDLE) { 1688 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1689 comma = 1; 1690 } 1691 if (t_flags & TF_RXWIN0SENT) { 1692 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1693 comma = 1; 1694 } 1695 if (t_flags & TF_FASTRECOVERY) { 1696 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1697 comma = 1; 1698 } 1699 if (t_flags & TF_WASFRECOVERY) { 1700 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1701 comma = 1; 1702 } 1703 if (t_flags & TF_SIGNATURE) { 1704 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1705 comma = 1; 1706 } 1707 if (t_flags & TF_FORCEDATA) { 1708 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1709 comma = 1; 1710 } 1711 if (t_flags & TF_TSO) { 1712 db_printf("%sTF_TSO", comma ? ", " : ""); 1713 comma = 1; 1714 } 1715 } 1716 1717 static void 1718 db_print_toobflags(char t_oobflags) 1719 { 1720 int comma; 1721 1722 comma = 0; 1723 if (t_oobflags & TCPOOB_HAVEDATA) { 1724 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1725 comma = 1; 1726 } 1727 if (t_oobflags & TCPOOB_HADDATA) { 1728 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1729 comma = 1; 1730 } 1731 } 1732 1733 static void 1734 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1735 { 1736 1737 db_print_indent(indent); 1738 db_printf("%s at %p\n", name, tp); 1739 1740 indent += 2; 1741 1742 db_print_indent(indent); 1743 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1744 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1745 1746 db_print_indent(indent); 1747 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1748 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1749 1750 db_print_indent(indent); 1751 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1752 &tp->t_timers->tt_delack, tp->t_inpcb); 1753 1754 db_print_indent(indent); 1755 db_printf("t_state: %d (", tp->t_state); 1756 db_print_tstate(tp->t_state); 1757 db_printf(")\n"); 1758 1759 db_print_indent(indent); 1760 db_printf("t_flags: 0x%x (", tp->t_flags); 1761 db_print_tflags(tp->t_flags); 1762 db_printf(")\n"); 1763 1764 db_print_indent(indent); 1765 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1766 tp->snd_una, tp->snd_max, tp->snd_nxt); 1767 1768 db_print_indent(indent); 1769 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1770 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1771 1772 db_print_indent(indent); 1773 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1774 tp->iss, tp->irs, tp->rcv_nxt); 1775 1776 db_print_indent(indent); 1777 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1778 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1779 1780 db_print_indent(indent); 1781 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1782 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1783 1784 db_print_indent(indent); 1785 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1786 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1787 tp->snd_recover); 1788 1789 db_print_indent(indent); 1790 db_printf("t_maxopd: %u t_rcvtime: %lu t_startime: %lu\n", 1791 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1792 1793 db_print_indent(indent); 1794 db_printf("t_rttime: %d t_rtsq: 0x%08x t_bw_rtttime: %d\n", 1795 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1796 1797 db_print_indent(indent); 1798 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1799 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1800 tp->t_srtt); 1801 1802 db_print_indent(indent); 1803 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1804 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1805 tp->t_rttbest); 1806 1807 db_print_indent(indent); 1808 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1809 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1810 1811 db_print_indent(indent); 1812 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1813 db_print_toobflags(tp->t_oobflags); 1814 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1815 1816 db_print_indent(indent); 1817 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1818 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1819 1820 db_print_indent(indent); 1821 db_printf("ts_recent: %u ts_recent_age: %lu\n", 1822 tp->ts_recent, tp->ts_recent_age); 1823 1824 db_print_indent(indent); 1825 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1826 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1827 1828 db_print_indent(indent); 1829 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1830 "t_badrxtwin: %lu\n", tp->snd_ssthresh_prev, 1831 tp->snd_recover_prev, tp->t_badrxtwin); 1832 1833 db_print_indent(indent); 1834 db_printf("snd_numholes: %d snd_holes first: %p\n", 1835 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1836 1837 db_print_indent(indent); 1838 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1839 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1840 1841 /* Skip sackblks, sackhint. */ 1842 1843 db_print_indent(indent); 1844 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1845 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1846 } 1847 1848 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1849 { 1850 struct tcpcb *tp; 1851 1852 if (!have_addr) { 1853 db_printf("usage: show tcpcb <addr>\n"); 1854 return; 1855 } 1856 tp = (struct tcpcb *)addr; 1857 1858 db_print_tcpcb(tp, "tcpcb", 0); 1859 } 1860 #endif 1861