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 * $FreeBSD$ 33 */ 34 35 #include "opt_ddb.h" 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/malloc.h> 43 #include <sys/kernel.h> 44 #include <sys/sysctl.h> 45 #include <sys/mbuf.h> 46 #ifdef INET6 47 #include <sys/domain.h> 48 #endif /* INET6 */ 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/protosw.h> 52 #include <sys/proc.h> 53 #include <sys/jail.h> 54 55 #ifdef DDB 56 #include <ddb/ddb.h> 57 #endif 58 59 #include <net/if.h> 60 #include <net/route.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_systm.h> 64 #ifdef INET6 65 #include <netinet/ip6.h> 66 #endif 67 #include <netinet/in_pcb.h> 68 #ifdef INET6 69 #include <netinet6/in6_pcb.h> 70 #endif 71 #include <netinet/in_var.h> 72 #include <netinet/ip_var.h> 73 #ifdef INET6 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/scope6_var.h> 76 #endif 77 #include <netinet/tcp.h> 78 #include <netinet/tcp_fsm.h> 79 #include <netinet/tcp_seq.h> 80 #include <netinet/tcp_timer.h> 81 #include <netinet/tcp_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif 86 87 /* 88 * TCP protocol interface to socket abstraction. 89 */ 90 static int tcp_attach(struct socket *); 91 static int tcp_connect(struct tcpcb *, struct sockaddr *, 92 struct thread *td); 93 #ifdef INET6 94 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 95 struct thread *td); 96 #endif /* INET6 */ 97 static void tcp_disconnect(struct tcpcb *); 98 static void tcp_usrclosed(struct tcpcb *); 99 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 100 101 #ifdef TCPDEBUG 102 #define TCPDEBUG0 int ostate = 0 103 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 104 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 105 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 106 #else 107 #define TCPDEBUG0 108 #define TCPDEBUG1() 109 #define TCPDEBUG2(req) 110 #endif 111 112 /* 113 * TCP attaches to socket via pru_attach(), reserving space, 114 * and an internet control block. 115 */ 116 static int 117 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 118 { 119 struct inpcb *inp; 120 struct tcpcb *tp = NULL; 121 int error; 122 TCPDEBUG0; 123 124 inp = sotoinpcb(so); 125 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 126 TCPDEBUG1(); 127 128 error = tcp_attach(so); 129 if (error) 130 goto out; 131 132 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 133 so->so_linger = TCP_LINGERTIME; 134 135 inp = sotoinpcb(so); 136 tp = intotcpcb(inp); 137 out: 138 TCPDEBUG2(PRU_ATTACH); 139 return error; 140 } 141 142 /* 143 * tcp_detach is called when the socket layer loses its final reference 144 * to the socket, be it a file descriptor reference, a reference from TCP, 145 * etc. At this point, there is only one case in which we will keep around 146 * inpcb state: time wait. 147 * 148 * This function can probably be re-absorbed back into tcp_usr_detach() now 149 * that there is a single detach path. 150 */ 151 static void 152 tcp_detach(struct socket *so, struct inpcb *inp) 153 { 154 struct tcpcb *tp; 155 #ifdef INET6 156 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 157 #endif 158 159 INP_INFO_WLOCK_ASSERT(&tcbinfo); 160 INP_LOCK_ASSERT(inp); 161 162 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 163 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 164 165 tp = intotcpcb(inp); 166 167 if (inp->inp_vflag & INP_TIMEWAIT) { 168 /* 169 * There are two cases to handle: one in which the time wait 170 * state is being discarded (INP_DROPPED), and one in which 171 * this connection will remain in timewait. In the former, 172 * it is time to discard all state (except tcptw, which has 173 * already been discarded by the timewait close code, which 174 * should be further up the call stack somewhere). In the 175 * latter case, we detach from the socket, but leave the pcb 176 * present until timewait ends. 177 * 178 * XXXRW: Would it be cleaner to free the tcptw here? 179 */ 180 if (inp->inp_vflag & INP_DROPPED) { 181 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 182 "INP_DROPPED && tp != NULL")); 183 #ifdef INET6 184 if (isipv6) { 185 in6_pcbdetach(inp); 186 in6_pcbfree(inp); 187 } else { 188 #endif 189 in_pcbdetach(inp); 190 in_pcbfree(inp); 191 #ifdef INET6 192 } 193 #endif 194 } else { 195 #ifdef INET6 196 if (isipv6) 197 in6_pcbdetach(inp); 198 else 199 #endif 200 in_pcbdetach(inp); 201 INP_UNLOCK(inp); 202 } 203 } else { 204 /* 205 * If the connection is not in timewait, we consider two 206 * two conditions: one in which no further processing is 207 * necessary (dropped || embryonic), and one in which TCP is 208 * not yet done, but no longer requires the socket, so the 209 * pcb will persist for the time being. 210 * 211 * XXXRW: Does the second case still occur? 212 */ 213 if (inp->inp_vflag & INP_DROPPED || 214 tp->t_state < TCPS_SYN_SENT) { 215 tcp_discardcb(tp); 216 #ifdef INET6 217 if (isipv6) { 218 in6_pcbdetach(inp); 219 in6_pcbfree(inp); 220 } else { 221 #endif 222 in_pcbdetach(inp); 223 in_pcbfree(inp); 224 #ifdef INET6 225 } 226 #endif 227 } else { 228 #ifdef INET6 229 if (isipv6) 230 in6_pcbdetach(inp); 231 else 232 #endif 233 in_pcbdetach(inp); 234 } 235 } 236 } 237 238 /* 239 * pru_detach() detaches the TCP protocol from the socket. 240 * If the protocol state is non-embryonic, then can't 241 * do this directly: have to initiate a pru_disconnect(), 242 * which may finish later; embryonic TCB's can just 243 * be discarded here. 244 */ 245 static void 246 tcp_usr_detach(struct socket *so) 247 { 248 struct inpcb *inp; 249 250 inp = sotoinpcb(so); 251 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 252 INP_INFO_WLOCK(&tcbinfo); 253 INP_LOCK(inp); 254 KASSERT(inp->inp_socket != NULL, 255 ("tcp_usr_detach: inp_socket == NULL")); 256 tcp_detach(so, inp); 257 INP_INFO_WUNLOCK(&tcbinfo); 258 } 259 260 /* 261 * Give the socket an address. 262 */ 263 static int 264 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 265 { 266 int error = 0; 267 struct inpcb *inp; 268 struct tcpcb *tp = NULL; 269 struct sockaddr_in *sinp; 270 271 sinp = (struct sockaddr_in *)nam; 272 if (nam->sa_len != sizeof (*sinp)) 273 return (EINVAL); 274 /* 275 * Must check for multicast addresses and disallow binding 276 * to them. 277 */ 278 if (sinp->sin_family == AF_INET && 279 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 280 return (EAFNOSUPPORT); 281 282 TCPDEBUG0; 283 INP_INFO_WLOCK(&tcbinfo); 284 inp = sotoinpcb(so); 285 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 286 INP_LOCK(inp); 287 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 288 error = EINVAL; 289 goto out; 290 } 291 tp = intotcpcb(inp); 292 TCPDEBUG1(); 293 error = in_pcbbind(inp, nam, td->td_ucred); 294 out: 295 TCPDEBUG2(PRU_BIND); 296 INP_UNLOCK(inp); 297 INP_INFO_WUNLOCK(&tcbinfo); 298 299 return (error); 300 } 301 302 #ifdef INET6 303 static int 304 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 305 { 306 int error = 0; 307 struct inpcb *inp; 308 struct tcpcb *tp = NULL; 309 struct sockaddr_in6 *sin6p; 310 311 sin6p = (struct sockaddr_in6 *)nam; 312 if (nam->sa_len != sizeof (*sin6p)) 313 return (EINVAL); 314 /* 315 * Must check for multicast addresses and disallow binding 316 * to them. 317 */ 318 if (sin6p->sin6_family == AF_INET6 && 319 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 320 return (EAFNOSUPPORT); 321 322 TCPDEBUG0; 323 INP_INFO_WLOCK(&tcbinfo); 324 inp = sotoinpcb(so); 325 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 326 INP_LOCK(inp); 327 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 328 error = EINVAL; 329 goto out; 330 } 331 tp = intotcpcb(inp); 332 TCPDEBUG1(); 333 inp->inp_vflag &= ~INP_IPV4; 334 inp->inp_vflag |= INP_IPV6; 335 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 336 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 337 inp->inp_vflag |= INP_IPV4; 338 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 339 struct sockaddr_in sin; 340 341 in6_sin6_2_sin(&sin, sin6p); 342 inp->inp_vflag |= INP_IPV4; 343 inp->inp_vflag &= ~INP_IPV6; 344 error = in_pcbbind(inp, (struct sockaddr *)&sin, 345 td->td_ucred); 346 goto out; 347 } 348 } 349 error = in6_pcbbind(inp, nam, td->td_ucred); 350 out: 351 TCPDEBUG2(PRU_BIND); 352 INP_UNLOCK(inp); 353 INP_INFO_WUNLOCK(&tcbinfo); 354 return (error); 355 } 356 #endif /* INET6 */ 357 358 /* 359 * Prepare to accept connections. 360 */ 361 static int 362 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 363 { 364 int error = 0; 365 struct inpcb *inp; 366 struct tcpcb *tp = NULL; 367 368 TCPDEBUG0; 369 INP_INFO_WLOCK(&tcbinfo); 370 inp = sotoinpcb(so); 371 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 372 INP_LOCK(inp); 373 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 374 error = EINVAL; 375 goto out; 376 } 377 tp = intotcpcb(inp); 378 TCPDEBUG1(); 379 SOCK_LOCK(so); 380 error = solisten_proto_check(so); 381 if (error == 0 && inp->inp_lport == 0) 382 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 383 if (error == 0) { 384 tp->t_state = TCPS_LISTEN; 385 solisten_proto(so, backlog); 386 } 387 SOCK_UNLOCK(so); 388 389 out: 390 TCPDEBUG2(PRU_LISTEN); 391 INP_UNLOCK(inp); 392 INP_INFO_WUNLOCK(&tcbinfo); 393 return (error); 394 } 395 396 #ifdef INET6 397 static int 398 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 399 { 400 int error = 0; 401 struct inpcb *inp; 402 struct tcpcb *tp = NULL; 403 404 TCPDEBUG0; 405 INP_INFO_WLOCK(&tcbinfo); 406 inp = sotoinpcb(so); 407 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 408 INP_LOCK(inp); 409 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 410 error = EINVAL; 411 goto out; 412 } 413 tp = intotcpcb(inp); 414 TCPDEBUG1(); 415 SOCK_LOCK(so); 416 error = solisten_proto_check(so); 417 if (error == 0 && inp->inp_lport == 0) { 418 inp->inp_vflag &= ~INP_IPV4; 419 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 420 inp->inp_vflag |= INP_IPV4; 421 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 422 } 423 if (error == 0) { 424 tp->t_state = TCPS_LISTEN; 425 solisten_proto(so, backlog); 426 } 427 SOCK_UNLOCK(so); 428 429 out: 430 TCPDEBUG2(PRU_LISTEN); 431 INP_UNLOCK(inp); 432 INP_INFO_WUNLOCK(&tcbinfo); 433 return (error); 434 } 435 #endif /* INET6 */ 436 437 /* 438 * Initiate connection to peer. 439 * Create a template for use in transmissions on this connection. 440 * Enter SYN_SENT state, and mark socket as connecting. 441 * Start keep-alive timer, and seed output sequence space. 442 * Send initial segment on connection. 443 */ 444 static int 445 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 446 { 447 int error = 0; 448 struct inpcb *inp; 449 struct tcpcb *tp = NULL; 450 struct sockaddr_in *sinp; 451 452 sinp = (struct sockaddr_in *)nam; 453 if (nam->sa_len != sizeof (*sinp)) 454 return (EINVAL); 455 /* 456 * Must disallow TCP ``connections'' to multicast addresses. 457 */ 458 if (sinp->sin_family == AF_INET 459 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 460 return (EAFNOSUPPORT); 461 if (jailed(td->td_ucred)) 462 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 463 464 TCPDEBUG0; 465 INP_INFO_WLOCK(&tcbinfo); 466 inp = sotoinpcb(so); 467 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 468 INP_LOCK(inp); 469 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 470 error = EINVAL; 471 goto out; 472 } 473 tp = intotcpcb(inp); 474 TCPDEBUG1(); 475 if ((error = tcp_connect(tp, nam, td)) != 0) 476 goto out; 477 error = tcp_output(tp); 478 out: 479 TCPDEBUG2(PRU_CONNECT); 480 INP_UNLOCK(inp); 481 INP_INFO_WUNLOCK(&tcbinfo); 482 return (error); 483 } 484 485 #ifdef INET6 486 static int 487 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 488 { 489 int error = 0; 490 struct inpcb *inp; 491 struct tcpcb *tp = NULL; 492 struct sockaddr_in6 *sin6p; 493 494 TCPDEBUG0; 495 496 sin6p = (struct sockaddr_in6 *)nam; 497 if (nam->sa_len != sizeof (*sin6p)) 498 return (EINVAL); 499 /* 500 * Must disallow TCP ``connections'' to multicast addresses. 501 */ 502 if (sin6p->sin6_family == AF_INET6 503 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 504 return (EAFNOSUPPORT); 505 506 INP_INFO_WLOCK(&tcbinfo); 507 inp = sotoinpcb(so); 508 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 509 INP_LOCK(inp); 510 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 511 error = EINVAL; 512 goto out; 513 } 514 tp = intotcpcb(inp); 515 TCPDEBUG1(); 516 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 517 struct sockaddr_in sin; 518 519 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 520 error = EINVAL; 521 goto out; 522 } 523 524 in6_sin6_2_sin(&sin, sin6p); 525 inp->inp_vflag |= INP_IPV4; 526 inp->inp_vflag &= ~INP_IPV6; 527 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 528 goto out; 529 error = tcp_output(tp); 530 goto out; 531 } 532 inp->inp_vflag &= ~INP_IPV4; 533 inp->inp_vflag |= INP_IPV6; 534 inp->inp_inc.inc_isipv6 = 1; 535 if ((error = tcp6_connect(tp, nam, td)) != 0) 536 goto out; 537 error = tcp_output(tp); 538 539 out: 540 TCPDEBUG2(PRU_CONNECT); 541 INP_UNLOCK(inp); 542 INP_INFO_WUNLOCK(&tcbinfo); 543 return (error); 544 } 545 #endif /* INET6 */ 546 547 /* 548 * Initiate disconnect from peer. 549 * If connection never passed embryonic stage, just drop; 550 * else if don't need to let data drain, then can just drop anyways, 551 * else have to begin TCP shutdown process: mark socket disconnecting, 552 * drain unread data, state switch to reflect user close, and 553 * send segment (e.g. FIN) to peer. Socket will be really disconnected 554 * when peer sends FIN and acks ours. 555 * 556 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 557 */ 558 static int 559 tcp_usr_disconnect(struct socket *so) 560 { 561 struct inpcb *inp; 562 struct tcpcb *tp = NULL; 563 int error = 0; 564 565 TCPDEBUG0; 566 INP_INFO_WLOCK(&tcbinfo); 567 inp = sotoinpcb(so); 568 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 569 INP_LOCK(inp); 570 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 571 error = ECONNRESET; 572 goto out; 573 } 574 tp = intotcpcb(inp); 575 TCPDEBUG1(); 576 tcp_disconnect(tp); 577 out: 578 TCPDEBUG2(PRU_DISCONNECT); 579 INP_UNLOCK(inp); 580 INP_INFO_WUNLOCK(&tcbinfo); 581 return (error); 582 } 583 584 /* 585 * Accept a connection. Essentially all the work is 586 * done at higher levels; just return the address 587 * of the peer, storing through addr. 588 */ 589 static int 590 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 591 { 592 int error = 0; 593 struct inpcb *inp = NULL; 594 struct tcpcb *tp = NULL; 595 struct in_addr addr; 596 in_port_t port = 0; 597 TCPDEBUG0; 598 599 if (so->so_state & SS_ISDISCONNECTED) 600 return (ECONNABORTED); 601 602 inp = sotoinpcb(so); 603 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 604 INP_LOCK(inp); 605 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 606 error = ECONNABORTED; 607 goto out; 608 } 609 tp = intotcpcb(inp); 610 TCPDEBUG1(); 611 612 /* 613 * We inline in_getpeeraddr and COMMON_END here, so that we can 614 * copy the data of interest and defer the malloc until after we 615 * release the lock. 616 */ 617 port = inp->inp_fport; 618 addr = inp->inp_faddr; 619 620 out: 621 TCPDEBUG2(PRU_ACCEPT); 622 INP_UNLOCK(inp); 623 if (error == 0) 624 *nam = in_sockaddr(port, &addr); 625 return error; 626 } 627 628 #ifdef INET6 629 static int 630 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 631 { 632 struct inpcb *inp = NULL; 633 int error = 0; 634 struct tcpcb *tp = NULL; 635 struct in_addr addr; 636 struct in6_addr addr6; 637 in_port_t port = 0; 638 int v4 = 0; 639 TCPDEBUG0; 640 641 if (so->so_state & SS_ISDISCONNECTED) 642 return (ECONNABORTED); 643 644 inp = sotoinpcb(so); 645 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 646 INP_LOCK(inp); 647 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 648 error = ECONNABORTED; 649 goto out; 650 } 651 tp = intotcpcb(inp); 652 TCPDEBUG1(); 653 654 /* 655 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 656 * copy the data of interest and defer the malloc until after we 657 * release the lock. 658 */ 659 if (inp->inp_vflag & INP_IPV4) { 660 v4 = 1; 661 port = inp->inp_fport; 662 addr = inp->inp_faddr; 663 } else { 664 port = inp->inp_fport; 665 addr6 = inp->in6p_faddr; 666 } 667 668 out: 669 TCPDEBUG2(PRU_ACCEPT); 670 INP_UNLOCK(inp); 671 if (error == 0) { 672 if (v4) 673 *nam = in6_v4mapsin6_sockaddr(port, &addr); 674 else 675 *nam = in6_sockaddr(port, &addr6); 676 } 677 return error; 678 } 679 #endif /* INET6 */ 680 681 /* 682 * Mark the connection as being incapable of further output. 683 */ 684 static int 685 tcp_usr_shutdown(struct socket *so) 686 { 687 int error = 0; 688 struct inpcb *inp; 689 struct tcpcb *tp = NULL; 690 691 TCPDEBUG0; 692 INP_INFO_WLOCK(&tcbinfo); 693 inp = sotoinpcb(so); 694 KASSERT(inp != NULL, ("inp == NULL")); 695 INP_LOCK(inp); 696 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 697 error = ECONNRESET; 698 goto out; 699 } 700 tp = intotcpcb(inp); 701 TCPDEBUG1(); 702 socantsendmore(so); 703 tcp_usrclosed(tp); 704 error = tcp_output(tp); 705 706 out: 707 TCPDEBUG2(PRU_SHUTDOWN); 708 INP_UNLOCK(inp); 709 INP_INFO_WUNLOCK(&tcbinfo); 710 711 return (error); 712 } 713 714 /* 715 * After a receive, possibly send window update to peer. 716 */ 717 static int 718 tcp_usr_rcvd(struct socket *so, int flags) 719 { 720 struct inpcb *inp; 721 struct tcpcb *tp = NULL; 722 int error = 0; 723 724 TCPDEBUG0; 725 inp = sotoinpcb(so); 726 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 727 INP_LOCK(inp); 728 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 729 error = ECONNRESET; 730 goto out; 731 } 732 tp = intotcpcb(inp); 733 TCPDEBUG1(); 734 tcp_output(tp); 735 736 out: 737 TCPDEBUG2(PRU_RCVD); 738 INP_UNLOCK(inp); 739 return (error); 740 } 741 742 /* 743 * Do a send by putting data in output queue and updating urgent 744 * marker if URG set. Possibly send more data. Unlike the other 745 * pru_*() routines, the mbuf chains are our responsibility. We 746 * must either enqueue them or free them. The other pru_* routines 747 * generally are caller-frees. 748 */ 749 static int 750 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 751 struct sockaddr *nam, struct mbuf *control, struct thread *td) 752 { 753 int error = 0; 754 struct inpcb *inp; 755 struct tcpcb *tp = NULL; 756 int headlocked = 0; 757 #ifdef INET6 758 int isipv6; 759 #endif 760 TCPDEBUG0; 761 762 /* 763 * We require the pcbinfo lock in two cases: 764 * 765 * (1) An implied connect is taking place, which can result in 766 * binding IPs and ports and hence modification of the pcb hash 767 * chains. 768 * 769 * (2) PRUS_EOF is set, resulting in explicit close on the send. 770 */ 771 if ((nam != NULL) || (flags & PRUS_EOF)) { 772 INP_INFO_WLOCK(&tcbinfo); 773 headlocked = 1; 774 } 775 inp = sotoinpcb(so); 776 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 777 INP_LOCK(inp); 778 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 779 if (control) 780 m_freem(control); 781 if (m) 782 m_freem(m); 783 error = ECONNRESET; 784 goto out; 785 } 786 #ifdef INET6 787 isipv6 = nam && nam->sa_family == AF_INET6; 788 #endif /* INET6 */ 789 tp = intotcpcb(inp); 790 TCPDEBUG1(); 791 if (control) { 792 /* TCP doesn't do control messages (rights, creds, etc) */ 793 if (control->m_len) { 794 m_freem(control); 795 if (m) 796 m_freem(m); 797 error = EINVAL; 798 goto out; 799 } 800 m_freem(control); /* empty control, just free it */ 801 } 802 if (!(flags & PRUS_OOB)) { 803 sbappendstream(&so->so_snd, m); 804 if (nam && tp->t_state < TCPS_SYN_SENT) { 805 /* 806 * Do implied connect if not yet connected, 807 * initialize window to default value, and 808 * initialize maxseg/maxopd using peer's cached 809 * MSS. 810 */ 811 INP_INFO_WLOCK_ASSERT(&tcbinfo); 812 #ifdef INET6 813 if (isipv6) 814 error = tcp6_connect(tp, nam, td); 815 else 816 #endif /* INET6 */ 817 error = tcp_connect(tp, nam, td); 818 if (error) 819 goto out; 820 tp->snd_wnd = TTCP_CLIENT_SND_WND; 821 tcp_mss(tp, -1); 822 } 823 if (flags & PRUS_EOF) { 824 /* 825 * Close the send side of the connection after 826 * the data is sent. 827 */ 828 INP_INFO_WLOCK_ASSERT(&tcbinfo); 829 socantsendmore(so); 830 tcp_usrclosed(tp); 831 } 832 if (headlocked) { 833 INP_INFO_WUNLOCK(&tcbinfo); 834 headlocked = 0; 835 } 836 if (tp != NULL) { 837 if (flags & PRUS_MORETOCOME) 838 tp->t_flags |= TF_MORETOCOME; 839 error = tcp_output(tp); 840 if (flags & PRUS_MORETOCOME) 841 tp->t_flags &= ~TF_MORETOCOME; 842 } 843 } else { 844 /* 845 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 846 */ 847 SOCKBUF_LOCK(&so->so_snd); 848 if (sbspace(&so->so_snd) < -512) { 849 SOCKBUF_UNLOCK(&so->so_snd); 850 m_freem(m); 851 error = ENOBUFS; 852 goto out; 853 } 854 /* 855 * According to RFC961 (Assigned Protocols), 856 * the urgent pointer points to the last octet 857 * of urgent data. We continue, however, 858 * to consider it to indicate the first octet 859 * of data past the urgent section. 860 * Otherwise, snd_up should be one lower. 861 */ 862 sbappendstream_locked(&so->so_snd, m); 863 SOCKBUF_UNLOCK(&so->so_snd); 864 if (nam && tp->t_state < TCPS_SYN_SENT) { 865 /* 866 * Do implied connect if not yet connected, 867 * initialize window to default value, and 868 * initialize maxseg/maxopd using peer's cached 869 * MSS. 870 */ 871 INP_INFO_WLOCK_ASSERT(&tcbinfo); 872 #ifdef INET6 873 if (isipv6) 874 error = tcp6_connect(tp, nam, td); 875 else 876 #endif /* INET6 */ 877 error = tcp_connect(tp, nam, td); 878 if (error) 879 goto out; 880 tp->snd_wnd = TTCP_CLIENT_SND_WND; 881 tcp_mss(tp, -1); 882 INP_INFO_WUNLOCK(&tcbinfo); 883 headlocked = 0; 884 } else if (nam) { 885 INP_INFO_WUNLOCK(&tcbinfo); 886 headlocked = 0; 887 } 888 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 889 tp->t_flags |= TF_FORCEDATA; 890 error = tcp_output(tp); 891 tp->t_flags &= ~TF_FORCEDATA; 892 } 893 out: 894 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 895 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 896 INP_UNLOCK(inp); 897 if (headlocked) 898 INP_INFO_WUNLOCK(&tcbinfo); 899 return (error); 900 } 901 902 /* 903 * Abort the TCP. Drop the connection abruptly. 904 */ 905 static void 906 tcp_usr_abort(struct socket *so) 907 { 908 struct inpcb *inp; 909 struct tcpcb *tp = NULL; 910 TCPDEBUG0; 911 912 inp = sotoinpcb(so); 913 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 914 915 INP_INFO_WLOCK(&tcbinfo); 916 INP_LOCK(inp); 917 KASSERT(inp->inp_socket != NULL, 918 ("tcp_usr_abort: inp_socket == NULL")); 919 920 /* 921 * If we still have full TCP state, and we're not dropped, drop. 922 */ 923 if (!(inp->inp_vflag & INP_TIMEWAIT) && 924 !(inp->inp_vflag & INP_DROPPED)) { 925 tp = intotcpcb(inp); 926 TCPDEBUG1(); 927 tcp_drop(tp, ECONNABORTED); 928 TCPDEBUG2(PRU_ABORT); 929 } 930 if (!(inp->inp_vflag & INP_DROPPED)) { 931 SOCK_LOCK(so); 932 so->so_state |= SS_PROTOREF; 933 SOCK_UNLOCK(so); 934 inp->inp_vflag |= INP_SOCKREF; 935 } 936 INP_UNLOCK(inp); 937 INP_INFO_WUNLOCK(&tcbinfo); 938 } 939 940 /* 941 * TCP socket is closed. Start friendly disconnect. 942 */ 943 static void 944 tcp_usr_close(struct socket *so) 945 { 946 struct inpcb *inp; 947 struct tcpcb *tp = NULL; 948 TCPDEBUG0; 949 950 inp = sotoinpcb(so); 951 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 952 953 INP_INFO_WLOCK(&tcbinfo); 954 INP_LOCK(inp); 955 KASSERT(inp->inp_socket != NULL, 956 ("tcp_usr_close: inp_socket == NULL")); 957 958 /* 959 * If we still have full TCP state, and we're not dropped, initiate 960 * a disconnect. 961 */ 962 if (!(inp->inp_vflag & INP_TIMEWAIT) && 963 !(inp->inp_vflag & INP_DROPPED)) { 964 tp = intotcpcb(inp); 965 TCPDEBUG1(); 966 tcp_disconnect(tp); 967 TCPDEBUG2(PRU_CLOSE); 968 } 969 if (!(inp->inp_vflag & INP_DROPPED)) { 970 SOCK_LOCK(so); 971 so->so_state |= SS_PROTOREF; 972 SOCK_UNLOCK(so); 973 inp->inp_vflag |= INP_SOCKREF; 974 } 975 INP_UNLOCK(inp); 976 INP_INFO_WUNLOCK(&tcbinfo); 977 } 978 979 /* 980 * Receive out-of-band data. 981 */ 982 static int 983 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 984 { 985 int error = 0; 986 struct inpcb *inp; 987 struct tcpcb *tp = NULL; 988 989 TCPDEBUG0; 990 inp = sotoinpcb(so); 991 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 992 INP_LOCK(inp); 993 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 994 error = ECONNRESET; 995 goto out; 996 } 997 tp = intotcpcb(inp); 998 TCPDEBUG1(); 999 if ((so->so_oobmark == 0 && 1000 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1001 so->so_options & SO_OOBINLINE || 1002 tp->t_oobflags & TCPOOB_HADDATA) { 1003 error = EINVAL; 1004 goto out; 1005 } 1006 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1007 error = EWOULDBLOCK; 1008 goto out; 1009 } 1010 m->m_len = 1; 1011 *mtod(m, caddr_t) = tp->t_iobc; 1012 if ((flags & MSG_PEEK) == 0) 1013 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1014 1015 out: 1016 TCPDEBUG2(PRU_RCVOOB); 1017 INP_UNLOCK(inp); 1018 return (error); 1019 } 1020 1021 struct pr_usrreqs tcp_usrreqs = { 1022 .pru_abort = tcp_usr_abort, 1023 .pru_accept = tcp_usr_accept, 1024 .pru_attach = tcp_usr_attach, 1025 .pru_bind = tcp_usr_bind, 1026 .pru_connect = tcp_usr_connect, 1027 .pru_control = in_control, 1028 .pru_detach = tcp_usr_detach, 1029 .pru_disconnect = tcp_usr_disconnect, 1030 .pru_listen = tcp_usr_listen, 1031 .pru_peeraddr = in_getpeeraddr, 1032 .pru_rcvd = tcp_usr_rcvd, 1033 .pru_rcvoob = tcp_usr_rcvoob, 1034 .pru_send = tcp_usr_send, 1035 .pru_shutdown = tcp_usr_shutdown, 1036 .pru_sockaddr = in_getsockaddr, 1037 .pru_sosetlabel = in_pcbsosetlabel, 1038 .pru_close = tcp_usr_close, 1039 }; 1040 1041 #ifdef INET6 1042 struct pr_usrreqs tcp6_usrreqs = { 1043 .pru_abort = tcp_usr_abort, 1044 .pru_accept = tcp6_usr_accept, 1045 .pru_attach = tcp_usr_attach, 1046 .pru_bind = tcp6_usr_bind, 1047 .pru_connect = tcp6_usr_connect, 1048 .pru_control = in6_control, 1049 .pru_detach = tcp_usr_detach, 1050 .pru_disconnect = tcp_usr_disconnect, 1051 .pru_listen = tcp6_usr_listen, 1052 .pru_peeraddr = in6_mapped_peeraddr, 1053 .pru_rcvd = tcp_usr_rcvd, 1054 .pru_rcvoob = tcp_usr_rcvoob, 1055 .pru_send = tcp_usr_send, 1056 .pru_shutdown = tcp_usr_shutdown, 1057 .pru_sockaddr = in6_mapped_sockaddr, 1058 .pru_sosetlabel = in_pcbsosetlabel, 1059 .pru_close = tcp_usr_close, 1060 }; 1061 #endif /* INET6 */ 1062 1063 /* 1064 * Common subroutine to open a TCP connection to remote host specified 1065 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1066 * port number if needed. Call in_pcbconnect_setup to do the routing and 1067 * to choose a local host address (interface). If there is an existing 1068 * incarnation of the same connection in TIME-WAIT state and if the remote 1069 * host was sending CC options and if the connection duration was < MSL, then 1070 * truncate the previous TIME-WAIT state and proceed. 1071 * Initialize connection parameters and enter SYN-SENT state. 1072 */ 1073 static int 1074 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1075 { 1076 struct inpcb *inp = tp->t_inpcb, *oinp; 1077 struct socket *so = inp->inp_socket; 1078 struct in_addr laddr; 1079 u_short lport; 1080 int error; 1081 1082 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1083 INP_LOCK_ASSERT(inp); 1084 1085 if (inp->inp_lport == 0) { 1086 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1087 if (error) 1088 return error; 1089 } 1090 1091 /* 1092 * Cannot simply call in_pcbconnect, because there might be an 1093 * earlier incarnation of this same connection still in 1094 * TIME_WAIT state, creating an ADDRINUSE error. 1095 */ 1096 laddr = inp->inp_laddr; 1097 lport = inp->inp_lport; 1098 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1099 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1100 if (error && oinp == NULL) 1101 return error; 1102 if (oinp) 1103 return EADDRINUSE; 1104 inp->inp_laddr = laddr; 1105 in_pcbrehash(inp); 1106 1107 /* 1108 * Compute window scaling to request: 1109 * Scale to fit into sweet spot. See tcp_syncache.c. 1110 * XXX: This should move to tcp_output(). 1111 * XXX: This should be based on the actual MSS. 1112 */ 1113 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1114 (0x1 << tp->request_r_scale) < tcp_minmss) 1115 tp->request_r_scale++; 1116 1117 soisconnecting(so); 1118 tcpstat.tcps_connattempt++; 1119 tp->t_state = TCPS_SYN_SENT; 1120 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1121 tp->iss = tcp_new_isn(tp); 1122 tp->t_bw_rtseq = tp->iss; 1123 tcp_sendseqinit(tp); 1124 1125 return 0; 1126 } 1127 1128 #ifdef INET6 1129 static int 1130 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1131 { 1132 struct inpcb *inp = tp->t_inpcb, *oinp; 1133 struct socket *so = inp->inp_socket; 1134 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1135 struct in6_addr *addr6; 1136 int error; 1137 1138 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1139 INP_LOCK_ASSERT(inp); 1140 1141 if (inp->inp_lport == 0) { 1142 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1143 if (error) 1144 return error; 1145 } 1146 1147 /* 1148 * Cannot simply call in_pcbconnect, because there might be an 1149 * earlier incarnation of this same connection still in 1150 * TIME_WAIT state, creating an ADDRINUSE error. 1151 * in6_pcbladdr() also handles scope zone IDs. 1152 */ 1153 error = in6_pcbladdr(inp, nam, &addr6); 1154 if (error) 1155 return error; 1156 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1157 &sin6->sin6_addr, sin6->sin6_port, 1158 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1159 ? addr6 1160 : &inp->in6p_laddr, 1161 inp->inp_lport, 0, NULL); 1162 if (oinp) 1163 return EADDRINUSE; 1164 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1165 inp->in6p_laddr = *addr6; 1166 inp->in6p_faddr = sin6->sin6_addr; 1167 inp->inp_fport = sin6->sin6_port; 1168 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1169 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 1170 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 1171 inp->in6p_flowinfo |= 1172 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1173 in_pcbrehash(inp); 1174 1175 /* Compute window scaling to request. */ 1176 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1177 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 1178 tp->request_r_scale++; 1179 1180 soisconnecting(so); 1181 tcpstat.tcps_connattempt++; 1182 tp->t_state = TCPS_SYN_SENT; 1183 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1184 tp->iss = tcp_new_isn(tp); 1185 tp->t_bw_rtseq = tp->iss; 1186 tcp_sendseqinit(tp); 1187 1188 return 0; 1189 } 1190 #endif /* INET6 */ 1191 1192 /* 1193 * Export TCP internal state information via a struct tcp_info, based on the 1194 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1195 * (TCP state machine, etc). We export all information using FreeBSD-native 1196 * constants -- for example, the numeric values for tcpi_state will differ 1197 * from Linux. 1198 */ 1199 static void 1200 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1201 { 1202 1203 INP_LOCK_ASSERT(tp->t_inpcb); 1204 bzero(ti, sizeof(*ti)); 1205 1206 ti->tcpi_state = tp->t_state; 1207 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1208 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1209 if (tp->t_flags & TF_SACK_PERMIT) 1210 ti->tcpi_options |= TCPI_OPT_SACK; 1211 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1212 ti->tcpi_options |= TCPI_OPT_WSCALE; 1213 ti->tcpi_snd_wscale = tp->snd_scale; 1214 ti->tcpi_rcv_wscale = tp->rcv_scale; 1215 } 1216 1217 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1218 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1219 1220 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1221 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1222 1223 /* 1224 * FreeBSD-specific extension fields for tcp_info. 1225 */ 1226 ti->tcpi_rcv_space = tp->rcv_wnd; 1227 ti->tcpi_snd_wnd = tp->snd_wnd; 1228 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1229 } 1230 1231 /* 1232 * The new sockopt interface makes it possible for us to block in the 1233 * copyin/out step (if we take a page fault). Taking a page fault at 1234 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now 1235 * use TSM, there probably isn't any need for this function to run at 1236 * splnet() any more. This needs more examination.) 1237 * 1238 * XXXRW: The locking here is wrong; we may take a page fault while holding 1239 * the inpcb lock. 1240 */ 1241 int 1242 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1243 { 1244 int error, opt, optval; 1245 struct inpcb *inp; 1246 struct tcpcb *tp; 1247 struct tcp_info ti; 1248 1249 error = 0; 1250 inp = sotoinpcb(so); 1251 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1252 INP_LOCK(inp); 1253 if (sopt->sopt_level != IPPROTO_TCP) { 1254 INP_UNLOCK(inp); 1255 #ifdef INET6 1256 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1257 error = ip6_ctloutput(so, sopt); 1258 else 1259 #endif /* INET6 */ 1260 error = ip_ctloutput(so, sopt); 1261 return (error); 1262 } 1263 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1264 error = ECONNRESET; 1265 goto out; 1266 } 1267 tp = intotcpcb(inp); 1268 1269 switch (sopt->sopt_dir) { 1270 case SOPT_SET: 1271 switch (sopt->sopt_name) { 1272 #ifdef TCP_SIGNATURE 1273 case TCP_MD5SIG: 1274 error = sooptcopyin(sopt, &optval, sizeof optval, 1275 sizeof optval); 1276 if (error) 1277 break; 1278 1279 if (optval > 0) 1280 tp->t_flags |= TF_SIGNATURE; 1281 else 1282 tp->t_flags &= ~TF_SIGNATURE; 1283 break; 1284 #endif /* TCP_SIGNATURE */ 1285 case TCP_NODELAY: 1286 case TCP_NOOPT: 1287 error = sooptcopyin(sopt, &optval, sizeof optval, 1288 sizeof optval); 1289 if (error) 1290 break; 1291 1292 switch (sopt->sopt_name) { 1293 case TCP_NODELAY: 1294 opt = TF_NODELAY; 1295 break; 1296 case TCP_NOOPT: 1297 opt = TF_NOOPT; 1298 break; 1299 default: 1300 opt = 0; /* dead code to fool gcc */ 1301 break; 1302 } 1303 1304 if (optval) 1305 tp->t_flags |= opt; 1306 else 1307 tp->t_flags &= ~opt; 1308 break; 1309 1310 case TCP_NOPUSH: 1311 error = sooptcopyin(sopt, &optval, sizeof optval, 1312 sizeof optval); 1313 if (error) 1314 break; 1315 1316 if (optval) 1317 tp->t_flags |= TF_NOPUSH; 1318 else { 1319 tp->t_flags &= ~TF_NOPUSH; 1320 error = tcp_output(tp); 1321 } 1322 break; 1323 1324 case TCP_MAXSEG: 1325 error = sooptcopyin(sopt, &optval, sizeof optval, 1326 sizeof optval); 1327 if (error) 1328 break; 1329 1330 if (optval > 0 && optval <= tp->t_maxseg && 1331 optval + 40 >= tcp_minmss) 1332 tp->t_maxseg = optval; 1333 else 1334 error = EINVAL; 1335 break; 1336 1337 case TCP_INFO: 1338 error = EINVAL; 1339 break; 1340 1341 default: 1342 error = ENOPROTOOPT; 1343 break; 1344 } 1345 break; 1346 1347 case SOPT_GET: 1348 switch (sopt->sopt_name) { 1349 #ifdef TCP_SIGNATURE 1350 case TCP_MD5SIG: 1351 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1352 error = sooptcopyout(sopt, &optval, sizeof optval); 1353 break; 1354 #endif 1355 case TCP_NODELAY: 1356 optval = tp->t_flags & TF_NODELAY; 1357 error = sooptcopyout(sopt, &optval, sizeof optval); 1358 break; 1359 case TCP_MAXSEG: 1360 optval = tp->t_maxseg; 1361 error = sooptcopyout(sopt, &optval, sizeof optval); 1362 break; 1363 case TCP_NOOPT: 1364 optval = tp->t_flags & TF_NOOPT; 1365 error = sooptcopyout(sopt, &optval, sizeof optval); 1366 break; 1367 case TCP_NOPUSH: 1368 optval = tp->t_flags & TF_NOPUSH; 1369 error = sooptcopyout(sopt, &optval, sizeof optval); 1370 break; 1371 case TCP_INFO: 1372 tcp_fill_info(tp, &ti); 1373 error = sooptcopyout(sopt, &ti, sizeof ti); 1374 break; 1375 default: 1376 error = ENOPROTOOPT; 1377 break; 1378 } 1379 break; 1380 } 1381 out: 1382 INP_UNLOCK(inp); 1383 return (error); 1384 } 1385 1386 /* 1387 * tcp_sendspace and tcp_recvspace are the default send and receive window 1388 * sizes, respectively. These are obsolescent (this information should 1389 * be set by the route). 1390 */ 1391 u_long tcp_sendspace = 1024*32; 1392 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1393 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1394 u_long tcp_recvspace = 1024*64; 1395 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1396 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1397 1398 /* 1399 * Attach TCP protocol to socket, allocating 1400 * internet protocol control block, tcp control block, 1401 * bufer space, and entering LISTEN state if to accept connections. 1402 */ 1403 static int 1404 tcp_attach(struct socket *so) 1405 { 1406 struct tcpcb *tp; 1407 struct inpcb *inp; 1408 int error; 1409 #ifdef INET6 1410 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1411 #endif 1412 1413 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1414 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1415 if (error) 1416 return (error); 1417 } 1418 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1419 so->so_snd.sb_flags |= SB_AUTOSIZE; 1420 INP_INFO_WLOCK(&tcbinfo); 1421 error = in_pcballoc(so, &tcbinfo); 1422 if (error) { 1423 INP_INFO_WUNLOCK(&tcbinfo); 1424 return (error); 1425 } 1426 inp = sotoinpcb(so); 1427 #ifdef INET6 1428 if (isipv6) { 1429 inp->inp_vflag |= INP_IPV6; 1430 inp->in6p_hops = -1; /* use kernel default */ 1431 } 1432 else 1433 #endif 1434 inp->inp_vflag |= INP_IPV4; 1435 tp = tcp_newtcpcb(inp); 1436 if (tp == NULL) { 1437 #ifdef INET6 1438 if (isipv6) { 1439 in6_pcbdetach(inp); 1440 in6_pcbfree(inp); 1441 } else { 1442 #endif 1443 in_pcbdetach(inp); 1444 in_pcbfree(inp); 1445 #ifdef INET6 1446 } 1447 #endif 1448 INP_INFO_WUNLOCK(&tcbinfo); 1449 return (ENOBUFS); 1450 } 1451 tp->t_state = TCPS_CLOSED; 1452 INP_UNLOCK(inp); 1453 INP_INFO_WUNLOCK(&tcbinfo); 1454 return (0); 1455 } 1456 1457 /* 1458 * Initiate (or continue) disconnect. 1459 * If embryonic state, just send reset (once). 1460 * If in ``let data drain'' option and linger null, just drop. 1461 * Otherwise (hard), mark socket disconnecting and drop 1462 * current input data; switch states based on user close, and 1463 * send segment to peer (with FIN). 1464 */ 1465 static void 1466 tcp_disconnect(struct tcpcb *tp) 1467 { 1468 struct inpcb *inp = tp->t_inpcb; 1469 struct socket *so = inp->inp_socket; 1470 1471 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1472 INP_LOCK_ASSERT(inp); 1473 1474 /* 1475 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1476 * socket is still open. 1477 */ 1478 if (tp->t_state < TCPS_ESTABLISHED) { 1479 tp = tcp_close(tp); 1480 KASSERT(tp != NULL, 1481 ("tcp_disconnect: tcp_close() returned NULL")); 1482 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1483 tp = tcp_drop(tp, 0); 1484 KASSERT(tp != NULL, 1485 ("tcp_disconnect: tcp_drop() returned NULL")); 1486 } else { 1487 soisdisconnecting(so); 1488 sbflush(&so->so_rcv); 1489 tcp_usrclosed(tp); 1490 if (!(inp->inp_vflag & INP_DROPPED)) 1491 tcp_output(tp); 1492 } 1493 } 1494 1495 /* 1496 * User issued close, and wish to trail through shutdown states: 1497 * if never received SYN, just forget it. If got a SYN from peer, 1498 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1499 * If already got a FIN from peer, then almost done; go to LAST_ACK 1500 * state. In all other cases, have already sent FIN to peer (e.g. 1501 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1502 * for peer to send FIN or not respond to keep-alives, etc. 1503 * We can let the user exit from the close as soon as the FIN is acked. 1504 */ 1505 static void 1506 tcp_usrclosed(struct tcpcb *tp) 1507 { 1508 1509 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1510 INP_LOCK_ASSERT(tp->t_inpcb); 1511 1512 switch (tp->t_state) { 1513 case TCPS_CLOSED: 1514 case TCPS_LISTEN: 1515 tp->t_state = TCPS_CLOSED; 1516 tp = tcp_close(tp); 1517 /* 1518 * tcp_close() should never return NULL here as the socket is 1519 * still open. 1520 */ 1521 KASSERT(tp != NULL, 1522 ("tcp_usrclosed: tcp_close() returned NULL")); 1523 break; 1524 1525 case TCPS_SYN_SENT: 1526 case TCPS_SYN_RECEIVED: 1527 tp->t_flags |= TF_NEEDFIN; 1528 break; 1529 1530 case TCPS_ESTABLISHED: 1531 tp->t_state = TCPS_FIN_WAIT_1; 1532 break; 1533 1534 case TCPS_CLOSE_WAIT: 1535 tp->t_state = TCPS_LAST_ACK; 1536 break; 1537 } 1538 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1539 soisdisconnected(tp->t_inpcb->inp_socket); 1540 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1541 if (tp->t_state == TCPS_FIN_WAIT_2) { 1542 int timeout; 1543 1544 timeout = (tcp_fast_finwait2_recycle) ? 1545 tcp_finwait2_timeout : tcp_maxidle; 1546 tcp_timer_activate(tp, TT_2MSL, timeout); 1547 } 1548 } 1549 } 1550 1551 #ifdef DDB 1552 static void 1553 db_print_indent(int indent) 1554 { 1555 int i; 1556 1557 for (i = 0; i < indent; i++) 1558 db_printf(" "); 1559 } 1560 1561 static void 1562 db_print_tstate(int t_state) 1563 { 1564 1565 switch (t_state) { 1566 case TCPS_CLOSED: 1567 db_printf("TCPS_CLOSED"); 1568 return; 1569 1570 case TCPS_LISTEN: 1571 db_printf("TCPS_LISTEN"); 1572 return; 1573 1574 case TCPS_SYN_SENT: 1575 db_printf("TCPS_SYN_SENT"); 1576 return; 1577 1578 case TCPS_SYN_RECEIVED: 1579 db_printf("TCPS_SYN_RECEIVED"); 1580 return; 1581 1582 case TCPS_ESTABLISHED: 1583 db_printf("TCPS_ESTABLISHED"); 1584 return; 1585 1586 case TCPS_CLOSE_WAIT: 1587 db_printf("TCPS_CLOSE_WAIT"); 1588 return; 1589 1590 case TCPS_FIN_WAIT_1: 1591 db_printf("TCPS_FIN_WAIT_1"); 1592 return; 1593 1594 case TCPS_CLOSING: 1595 db_printf("TCPS_CLOSING"); 1596 return; 1597 1598 case TCPS_LAST_ACK: 1599 db_printf("TCPS_LAST_ACK"); 1600 return; 1601 1602 case TCPS_FIN_WAIT_2: 1603 db_printf("TCPS_FIN_WAIT_2"); 1604 return; 1605 1606 case TCPS_TIME_WAIT: 1607 db_printf("TCPS_TIME_WAIT"); 1608 return; 1609 1610 default: 1611 db_printf("unknown"); 1612 return; 1613 } 1614 } 1615 1616 static void 1617 db_print_tflags(u_int t_flags) 1618 { 1619 int comma; 1620 1621 comma = 0; 1622 if (t_flags & TF_ACKNOW) { 1623 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1624 comma = 1; 1625 } 1626 if (t_flags & TF_DELACK) { 1627 db_printf("%sTF_DELACK", comma ? ", " : ""); 1628 comma = 1; 1629 } 1630 if (t_flags & TF_NODELAY) { 1631 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1632 comma = 1; 1633 } 1634 if (t_flags & TF_NOOPT) { 1635 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1636 comma = 1; 1637 } 1638 if (t_flags & TF_SENTFIN) { 1639 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1640 comma = 1; 1641 } 1642 if (t_flags & TF_REQ_SCALE) { 1643 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1644 comma = 1; 1645 } 1646 if (t_flags & TF_RCVD_SCALE) { 1647 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1648 comma = 1; 1649 } 1650 if (t_flags & TF_REQ_TSTMP) { 1651 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1652 comma = 1; 1653 } 1654 if (t_flags & TF_RCVD_TSTMP) { 1655 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1656 comma = 1; 1657 } 1658 if (t_flags & TF_SACK_PERMIT) { 1659 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1660 comma = 1; 1661 } 1662 if (t_flags & TF_NEEDSYN) { 1663 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1664 comma = 1; 1665 } 1666 if (t_flags & TF_NEEDFIN) { 1667 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1668 comma = 1; 1669 } 1670 if (t_flags & TF_NOPUSH) { 1671 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1672 comma = 1; 1673 } 1674 if (t_flags & TF_NOPUSH) { 1675 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1676 comma = 1; 1677 } 1678 if (t_flags & TF_MORETOCOME) { 1679 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1680 comma = 1; 1681 } 1682 if (t_flags & TF_LQ_OVERFLOW) { 1683 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1684 comma = 1; 1685 } 1686 if (t_flags & TF_LASTIDLE) { 1687 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1688 comma = 1; 1689 } 1690 if (t_flags & TF_RXWIN0SENT) { 1691 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1692 comma = 1; 1693 } 1694 if (t_flags & TF_FASTRECOVERY) { 1695 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1696 comma = 1; 1697 } 1698 if (t_flags & TF_WASFRECOVERY) { 1699 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1700 comma = 1; 1701 } 1702 if (t_flags & TF_SIGNATURE) { 1703 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1704 comma = 1; 1705 } 1706 if (t_flags & TF_FORCEDATA) { 1707 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1708 comma = 1; 1709 } 1710 if (t_flags & TF_TSO) { 1711 db_printf("%sTF_TSO", comma ? ", " : ""); 1712 comma = 1; 1713 } 1714 } 1715 1716 static void 1717 db_print_toobflags(char t_oobflags) 1718 { 1719 int comma; 1720 1721 comma = 0; 1722 if (t_oobflags & TCPOOB_HAVEDATA) { 1723 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1724 comma = 1; 1725 } 1726 if (t_oobflags & TCPOOB_HADDATA) { 1727 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1728 comma = 1; 1729 } 1730 } 1731 1732 static void 1733 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1734 { 1735 1736 db_print_indent(indent); 1737 db_printf("%s at %p\n", name, tp); 1738 1739 indent += 2; 1740 1741 db_print_indent(indent); 1742 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1743 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1744 1745 db_print_indent(indent); 1746 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1747 tp->tt_rexmt, tp->tt_persist, tp->tt_keep); 1748 1749 db_print_indent(indent); 1750 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", tp->tt_2msl, 1751 tp->tt_delack, tp->t_inpcb); 1752 1753 db_print_indent(indent); 1754 db_printf("t_state: %d (", tp->t_state); 1755 db_print_tstate(tp->t_state); 1756 db_printf(")\n"); 1757 1758 db_print_indent(indent); 1759 db_printf("t_flags: 0x%x (", tp->t_flags); 1760 db_print_tflags(tp->t_flags); 1761 db_printf(")\n"); 1762 1763 db_print_indent(indent); 1764 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1765 tp->snd_una, tp->snd_max, tp->snd_nxt); 1766 1767 db_print_indent(indent); 1768 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1769 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1770 1771 db_print_indent(indent); 1772 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1773 tp->iss, tp->irs, tp->rcv_nxt); 1774 1775 db_print_indent(indent); 1776 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1777 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1778 1779 db_print_indent(indent); 1780 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1781 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1782 1783 db_print_indent(indent); 1784 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1785 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1786 tp->snd_recover); 1787 1788 db_print_indent(indent); 1789 db_printf("t_maxopd: %u t_rcvtime: %lu t_startime: %lu\n", 1790 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1791 1792 db_print_indent(indent); 1793 db_printf("t_rttime: %d t_rtsq: 0x%08x t_bw_rtttime: %d\n", 1794 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1795 1796 db_print_indent(indent); 1797 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1798 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1799 tp->t_srtt); 1800 1801 db_print_indent(indent); 1802 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1803 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1804 tp->t_rttbest); 1805 1806 db_print_indent(indent); 1807 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1808 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1809 1810 db_print_indent(indent); 1811 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1812 db_print_toobflags(tp->t_oobflags); 1813 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1814 1815 db_print_indent(indent); 1816 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1817 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1818 1819 db_print_indent(indent); 1820 db_printf("ts_recent: %u ts_recent_age: %lu\n", 1821 tp->ts_recent, tp->ts_recent_age); 1822 1823 db_print_indent(indent); 1824 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1825 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1826 1827 db_print_indent(indent); 1828 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1829 "t_badrxtwin: %lu\n", tp->snd_ssthresh_prev, 1830 tp->snd_recover_prev, tp->t_badrxtwin); 1831 1832 db_print_indent(indent); 1833 db_printf("snd_numholes: %d snd_holes first: %p\n", 1834 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1835 1836 db_print_indent(indent); 1837 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1838 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1839 1840 /* Skip sackblks, sackhint. */ 1841 1842 db_print_indent(indent); 1843 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1844 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1845 } 1846 1847 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1848 { 1849 struct tcpcb *tp; 1850 1851 if (!have_addr) { 1852 db_printf("usage: show tcpcb <addr>\n"); 1853 return; 1854 } 1855 tp = (struct tcpcb *)addr; 1856 1857 db_print_tcpcb(tp, "tcpcb", 0); 1858 } 1859 #endif 1860