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 #include <sys/vimage.h> 57 58 #ifdef DDB 59 #include <ddb/ddb.h> 60 #endif 61 62 #include <net/if.h> 63 #include <net/route.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #ifdef INET6 68 #include <netinet/ip6.h> 69 #endif 70 #include <netinet/in_pcb.h> 71 #ifdef INET6 72 #include <netinet6/in6_pcb.h> 73 #endif 74 #include <netinet/in_var.h> 75 #include <netinet/ip_var.h> 76 #ifdef INET6 77 #include <netinet6/ip6_var.h> 78 #include <netinet6/scope6_var.h> 79 #endif 80 #include <netinet/tcp.h> 81 #include <netinet/tcp_fsm.h> 82 #include <netinet/tcp_seq.h> 83 #include <netinet/tcp_timer.h> 84 #include <netinet/tcp_var.h> 85 #include <netinet/tcpip.h> 86 #ifdef TCPDEBUG 87 #include <netinet/tcp_debug.h> 88 #endif 89 #include <netinet/tcp_offload.h> 90 91 /* 92 * TCP protocol interface to socket abstraction. 93 */ 94 static int tcp_attach(struct socket *); 95 static int tcp_connect(struct tcpcb *, struct sockaddr *, 96 struct thread *td); 97 #ifdef INET6 98 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 99 struct thread *td); 100 #endif /* INET6 */ 101 static void tcp_disconnect(struct tcpcb *); 102 static void tcp_usrclosed(struct tcpcb *); 103 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 104 105 #ifdef TCPDEBUG 106 #define TCPDEBUG0 int ostate = 0 107 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 108 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 109 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 110 #else 111 #define TCPDEBUG0 112 #define TCPDEBUG1() 113 #define TCPDEBUG2(req) 114 #endif 115 116 /* 117 * TCP attaches to socket via pru_attach(), reserving space, 118 * and an internet control block. 119 */ 120 static int 121 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 122 { 123 struct inpcb *inp; 124 struct tcpcb *tp = NULL; 125 int error; 126 TCPDEBUG0; 127 128 inp = sotoinpcb(so); 129 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 130 TCPDEBUG1(); 131 132 error = tcp_attach(so); 133 if (error) 134 goto out; 135 136 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 137 so->so_linger = TCP_LINGERTIME; 138 139 inp = sotoinpcb(so); 140 tp = intotcpcb(inp); 141 out: 142 TCPDEBUG2(PRU_ATTACH); 143 return error; 144 } 145 146 /* 147 * tcp_detach is called when the socket layer loses its final reference 148 * to the socket, be it a file descriptor reference, a reference from TCP, 149 * etc. At this point, there is only one case in which we will keep around 150 * inpcb state: time wait. 151 * 152 * This function can probably be re-absorbed back into tcp_usr_detach() now 153 * that there is a single detach path. 154 */ 155 static void 156 tcp_detach(struct socket *so, struct inpcb *inp) 157 { 158 struct tcpcb *tp; 159 #ifdef INET6 160 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 161 #endif 162 163 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 164 INP_WLOCK_ASSERT(inp); 165 166 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 167 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 168 169 tp = intotcpcb(inp); 170 171 if (inp->inp_vflag & INP_TIMEWAIT) { 172 /* 173 * There are two cases to handle: one in which the time wait 174 * state is being discarded (INP_DROPPED), and one in which 175 * this connection will remain in timewait. In the former, 176 * it is time to discard all state (except tcptw, which has 177 * already been discarded by the timewait close code, which 178 * should be further up the call stack somewhere). In the 179 * latter case, we detach from the socket, but leave the pcb 180 * present until timewait ends. 181 * 182 * XXXRW: Would it be cleaner to free the tcptw here? 183 */ 184 if (inp->inp_vflag & INP_DROPPED) { 185 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 186 "INP_DROPPED && tp != NULL")); 187 #ifdef INET6 188 if (isipv6) { 189 in6_pcbdetach(inp); 190 in6_pcbfree(inp); 191 } else { 192 #endif 193 in_pcbdetach(inp); 194 in_pcbfree(inp); 195 #ifdef INET6 196 } 197 #endif 198 } else { 199 #ifdef INET6 200 if (isipv6) 201 in6_pcbdetach(inp); 202 else 203 #endif 204 in_pcbdetach(inp); 205 INP_WUNLOCK(inp); 206 } 207 } else { 208 /* 209 * If the connection is not in timewait, we consider two 210 * two conditions: one in which no further processing is 211 * necessary (dropped || embryonic), and one in which TCP is 212 * not yet done, but no longer requires the socket, so the 213 * pcb will persist for the time being. 214 * 215 * XXXRW: Does the second case still occur? 216 */ 217 if (inp->inp_vflag & INP_DROPPED || 218 tp->t_state < TCPS_SYN_SENT) { 219 tcp_discardcb(tp); 220 #ifdef INET6 221 if (isipv6) { 222 in6_pcbdetach(inp); 223 in6_pcbfree(inp); 224 } else { 225 #endif 226 in_pcbdetach(inp); 227 in_pcbfree(inp); 228 #ifdef INET6 229 } 230 #endif 231 } else { 232 #ifdef INET6 233 if (isipv6) 234 in6_pcbdetach(inp); 235 else 236 #endif 237 in_pcbdetach(inp); 238 } 239 } 240 } 241 242 /* 243 * pru_detach() detaches the TCP protocol from the socket. 244 * If the protocol state is non-embryonic, then can't 245 * do this directly: have to initiate a pru_disconnect(), 246 * which may finish later; embryonic TCB's can just 247 * be discarded here. 248 */ 249 static void 250 tcp_usr_detach(struct socket *so) 251 { 252 struct inpcb *inp; 253 254 inp = sotoinpcb(so); 255 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 256 INP_INFO_WLOCK(&V_tcbinfo); 257 INP_WLOCK(inp); 258 KASSERT(inp->inp_socket != NULL, 259 ("tcp_usr_detach: inp_socket == NULL")); 260 tcp_detach(so, inp); 261 INP_INFO_WUNLOCK(&V_tcbinfo); 262 } 263 264 /* 265 * Give the socket an address. 266 */ 267 static int 268 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 269 { 270 int error = 0; 271 struct inpcb *inp; 272 struct tcpcb *tp = NULL; 273 struct sockaddr_in *sinp; 274 275 sinp = (struct sockaddr_in *)nam; 276 if (nam->sa_len != sizeof (*sinp)) 277 return (EINVAL); 278 /* 279 * Must check for multicast addresses and disallow binding 280 * to them. 281 */ 282 if (sinp->sin_family == AF_INET && 283 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 284 return (EAFNOSUPPORT); 285 286 TCPDEBUG0; 287 INP_INFO_WLOCK(&V_tcbinfo); 288 inp = sotoinpcb(so); 289 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 290 INP_WLOCK(inp); 291 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 292 error = EINVAL; 293 goto out; 294 } 295 tp = intotcpcb(inp); 296 TCPDEBUG1(); 297 error = in_pcbbind(inp, nam, td->td_ucred); 298 out: 299 TCPDEBUG2(PRU_BIND); 300 INP_WUNLOCK(inp); 301 INP_INFO_WUNLOCK(&V_tcbinfo); 302 303 return (error); 304 } 305 306 #ifdef INET6 307 static int 308 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 309 { 310 int error = 0; 311 struct inpcb *inp; 312 struct tcpcb *tp = NULL; 313 struct sockaddr_in6 *sin6p; 314 315 sin6p = (struct sockaddr_in6 *)nam; 316 if (nam->sa_len != sizeof (*sin6p)) 317 return (EINVAL); 318 /* 319 * Must check for multicast addresses and disallow binding 320 * to them. 321 */ 322 if (sin6p->sin6_family == AF_INET6 && 323 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 324 return (EAFNOSUPPORT); 325 326 TCPDEBUG0; 327 INP_INFO_WLOCK(&V_tcbinfo); 328 inp = sotoinpcb(so); 329 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 330 INP_WLOCK(inp); 331 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 332 error = EINVAL; 333 goto out; 334 } 335 tp = intotcpcb(inp); 336 TCPDEBUG1(); 337 inp->inp_vflag &= ~INP_IPV4; 338 inp->inp_vflag |= INP_IPV6; 339 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 340 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 341 inp->inp_vflag |= INP_IPV4; 342 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 343 struct sockaddr_in sin; 344 345 in6_sin6_2_sin(&sin, sin6p); 346 inp->inp_vflag |= INP_IPV4; 347 inp->inp_vflag &= ~INP_IPV6; 348 error = in_pcbbind(inp, (struct sockaddr *)&sin, 349 td->td_ucred); 350 goto out; 351 } 352 } 353 error = in6_pcbbind(inp, nam, td->td_ucred); 354 out: 355 TCPDEBUG2(PRU_BIND); 356 INP_WUNLOCK(inp); 357 INP_INFO_WUNLOCK(&V_tcbinfo); 358 return (error); 359 } 360 #endif /* INET6 */ 361 362 /* 363 * Prepare to accept connections. 364 */ 365 static int 366 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 367 { 368 int error = 0; 369 struct inpcb *inp; 370 struct tcpcb *tp = NULL; 371 372 TCPDEBUG0; 373 INP_INFO_WLOCK(&V_tcbinfo); 374 inp = sotoinpcb(so); 375 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 376 INP_WLOCK(inp); 377 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 378 error = EINVAL; 379 goto out; 380 } 381 tp = intotcpcb(inp); 382 TCPDEBUG1(); 383 SOCK_LOCK(so); 384 error = solisten_proto_check(so); 385 if (error == 0 && inp->inp_lport == 0) 386 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 387 if (error == 0) { 388 tp->t_state = TCPS_LISTEN; 389 solisten_proto(so, backlog); 390 tcp_offload_listen_open(tp); 391 } 392 SOCK_UNLOCK(so); 393 394 out: 395 TCPDEBUG2(PRU_LISTEN); 396 INP_WUNLOCK(inp); 397 INP_INFO_WUNLOCK(&V_tcbinfo); 398 return (error); 399 } 400 401 #ifdef INET6 402 static int 403 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 404 { 405 int error = 0; 406 struct inpcb *inp; 407 struct tcpcb *tp = NULL; 408 409 TCPDEBUG0; 410 INP_INFO_WLOCK(&V_tcbinfo); 411 inp = sotoinpcb(so); 412 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 413 INP_WLOCK(inp); 414 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 415 error = EINVAL; 416 goto out; 417 } 418 tp = intotcpcb(inp); 419 TCPDEBUG1(); 420 SOCK_LOCK(so); 421 error = solisten_proto_check(so); 422 if (error == 0 && inp->inp_lport == 0) { 423 inp->inp_vflag &= ~INP_IPV4; 424 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 425 inp->inp_vflag |= INP_IPV4; 426 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 427 } 428 if (error == 0) { 429 tp->t_state = TCPS_LISTEN; 430 solisten_proto(so, backlog); 431 } 432 SOCK_UNLOCK(so); 433 434 out: 435 TCPDEBUG2(PRU_LISTEN); 436 INP_WUNLOCK(inp); 437 INP_INFO_WUNLOCK(&V_tcbinfo); 438 return (error); 439 } 440 #endif /* INET6 */ 441 442 /* 443 * Initiate connection to peer. 444 * Create a template for use in transmissions on this connection. 445 * Enter SYN_SENT state, and mark socket as connecting. 446 * Start keep-alive timer, and seed output sequence space. 447 * Send initial segment on connection. 448 */ 449 static int 450 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 451 { 452 int error = 0; 453 struct inpcb *inp; 454 struct tcpcb *tp = NULL; 455 struct sockaddr_in *sinp; 456 457 sinp = (struct sockaddr_in *)nam; 458 if (nam->sa_len != sizeof (*sinp)) 459 return (EINVAL); 460 /* 461 * Must disallow TCP ``connections'' to multicast addresses. 462 */ 463 if (sinp->sin_family == AF_INET 464 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 465 return (EAFNOSUPPORT); 466 if (jailed(td->td_ucred)) 467 prison_remote_ip(td->td_ucred, 0, &sinp->sin_addr.s_addr); 468 469 TCPDEBUG0; 470 INP_INFO_WLOCK(&V_tcbinfo); 471 inp = sotoinpcb(so); 472 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 473 INP_WLOCK(inp); 474 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 475 error = EINVAL; 476 goto out; 477 } 478 tp = intotcpcb(inp); 479 TCPDEBUG1(); 480 if ((error = tcp_connect(tp, nam, td)) != 0) 481 goto out; 482 error = tcp_output_connect(so, nam); 483 out: 484 TCPDEBUG2(PRU_CONNECT); 485 INP_WUNLOCK(inp); 486 INP_INFO_WUNLOCK(&V_tcbinfo); 487 return (error); 488 } 489 490 #ifdef INET6 491 static int 492 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 493 { 494 int error = 0; 495 struct inpcb *inp; 496 struct tcpcb *tp = NULL; 497 struct sockaddr_in6 *sin6p; 498 499 TCPDEBUG0; 500 501 sin6p = (struct sockaddr_in6 *)nam; 502 if (nam->sa_len != sizeof (*sin6p)) 503 return (EINVAL); 504 /* 505 * Must disallow TCP ``connections'' to multicast addresses. 506 */ 507 if (sin6p->sin6_family == AF_INET6 508 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 509 return (EAFNOSUPPORT); 510 511 INP_INFO_WLOCK(&V_tcbinfo); 512 inp = sotoinpcb(so); 513 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 514 INP_WLOCK(inp); 515 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 516 error = EINVAL; 517 goto out; 518 } 519 tp = intotcpcb(inp); 520 TCPDEBUG1(); 521 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 522 struct sockaddr_in sin; 523 524 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 525 error = EINVAL; 526 goto out; 527 } 528 529 in6_sin6_2_sin(&sin, sin6p); 530 inp->inp_vflag |= INP_IPV4; 531 inp->inp_vflag &= ~INP_IPV6; 532 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 533 goto out; 534 error = tcp_output_connect(so, nam); 535 goto out; 536 } 537 inp->inp_vflag &= ~INP_IPV4; 538 inp->inp_vflag |= INP_IPV6; 539 inp->inp_inc.inc_isipv6 = 1; 540 if ((error = tcp6_connect(tp, nam, td)) != 0) 541 goto out; 542 error = tcp_output_connect(so, nam); 543 544 out: 545 TCPDEBUG2(PRU_CONNECT); 546 INP_WUNLOCK(inp); 547 INP_INFO_WUNLOCK(&V_tcbinfo); 548 return (error); 549 } 550 #endif /* INET6 */ 551 552 /* 553 * Initiate disconnect from peer. 554 * If connection never passed embryonic stage, just drop; 555 * else if don't need to let data drain, then can just drop anyways, 556 * else have to begin TCP shutdown process: mark socket disconnecting, 557 * drain unread data, state switch to reflect user close, and 558 * send segment (e.g. FIN) to peer. Socket will be really disconnected 559 * when peer sends FIN and acks ours. 560 * 561 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 562 */ 563 static int 564 tcp_usr_disconnect(struct socket *so) 565 { 566 struct inpcb *inp; 567 struct tcpcb *tp = NULL; 568 int error = 0; 569 570 TCPDEBUG0; 571 INP_INFO_WLOCK(&V_tcbinfo); 572 inp = sotoinpcb(so); 573 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 574 INP_WLOCK(inp); 575 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 576 error = ECONNRESET; 577 goto out; 578 } 579 tp = intotcpcb(inp); 580 TCPDEBUG1(); 581 tcp_disconnect(tp); 582 out: 583 TCPDEBUG2(PRU_DISCONNECT); 584 INP_WUNLOCK(inp); 585 INP_INFO_WUNLOCK(&V_tcbinfo); 586 return (error); 587 } 588 589 /* 590 * Accept a connection. Essentially all the work is 591 * done at higher levels; just return the address 592 * of the peer, storing through addr. 593 */ 594 static int 595 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 596 { 597 int error = 0; 598 struct inpcb *inp = NULL; 599 struct tcpcb *tp = NULL; 600 struct in_addr addr; 601 in_port_t port = 0; 602 TCPDEBUG0; 603 604 if (so->so_state & SS_ISDISCONNECTED) 605 return (ECONNABORTED); 606 607 inp = sotoinpcb(so); 608 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 609 INP_INFO_RLOCK(&V_tcbinfo); 610 INP_WLOCK(inp); 611 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 612 error = ECONNABORTED; 613 goto out; 614 } 615 tp = intotcpcb(inp); 616 TCPDEBUG1(); 617 618 /* 619 * We inline in_getpeeraddr and COMMON_END here, so that we can 620 * copy the data of interest and defer the malloc until after we 621 * release the lock. 622 */ 623 port = inp->inp_fport; 624 addr = inp->inp_faddr; 625 626 out: 627 TCPDEBUG2(PRU_ACCEPT); 628 INP_WUNLOCK(inp); 629 INP_INFO_RUNLOCK(&V_tcbinfo); 630 if (error == 0) 631 *nam = in_sockaddr(port, &addr); 632 return error; 633 } 634 635 #ifdef INET6 636 static int 637 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 638 { 639 struct inpcb *inp = NULL; 640 int error = 0; 641 struct tcpcb *tp = NULL; 642 struct in_addr addr; 643 struct in6_addr addr6; 644 in_port_t port = 0; 645 int v4 = 0; 646 TCPDEBUG0; 647 648 if (so->so_state & SS_ISDISCONNECTED) 649 return (ECONNABORTED); 650 651 inp = sotoinpcb(so); 652 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 653 INP_WLOCK(inp); 654 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 655 error = ECONNABORTED; 656 goto out; 657 } 658 tp = intotcpcb(inp); 659 TCPDEBUG1(); 660 661 /* 662 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 663 * copy the data of interest and defer the malloc until after we 664 * release the lock. 665 */ 666 if (inp->inp_vflag & INP_IPV4) { 667 v4 = 1; 668 port = inp->inp_fport; 669 addr = inp->inp_faddr; 670 } else { 671 port = inp->inp_fport; 672 addr6 = inp->in6p_faddr; 673 } 674 675 out: 676 TCPDEBUG2(PRU_ACCEPT); 677 INP_WUNLOCK(inp); 678 if (error == 0) { 679 if (v4) 680 *nam = in6_v4mapsin6_sockaddr(port, &addr); 681 else 682 *nam = in6_sockaddr(port, &addr6); 683 } 684 return error; 685 } 686 #endif /* INET6 */ 687 688 /* 689 * Mark the connection as being incapable of further output. 690 */ 691 static int 692 tcp_usr_shutdown(struct socket *so) 693 { 694 int error = 0; 695 struct inpcb *inp; 696 struct tcpcb *tp = NULL; 697 698 TCPDEBUG0; 699 INP_INFO_WLOCK(&V_tcbinfo); 700 inp = sotoinpcb(so); 701 KASSERT(inp != NULL, ("inp == NULL")); 702 INP_WLOCK(inp); 703 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 704 error = ECONNRESET; 705 goto out; 706 } 707 tp = intotcpcb(inp); 708 TCPDEBUG1(); 709 socantsendmore(so); 710 tcp_usrclosed(tp); 711 error = tcp_output_disconnect(tp); 712 713 out: 714 TCPDEBUG2(PRU_SHUTDOWN); 715 INP_WUNLOCK(inp); 716 INP_INFO_WUNLOCK(&V_tcbinfo); 717 718 return (error); 719 } 720 721 /* 722 * After a receive, possibly send window update to peer. 723 */ 724 static int 725 tcp_usr_rcvd(struct socket *so, int flags) 726 { 727 struct inpcb *inp; 728 struct tcpcb *tp = NULL; 729 int error = 0; 730 731 TCPDEBUG0; 732 inp = sotoinpcb(so); 733 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 734 INP_WLOCK(inp); 735 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 736 error = ECONNRESET; 737 goto out; 738 } 739 tp = intotcpcb(inp); 740 TCPDEBUG1(); 741 tcp_output_rcvd(tp); 742 743 out: 744 TCPDEBUG2(PRU_RCVD); 745 INP_WUNLOCK(inp); 746 return (error); 747 } 748 749 /* 750 * Do a send by putting data in output queue and updating urgent 751 * marker if URG set. Possibly send more data. Unlike the other 752 * pru_*() routines, the mbuf chains are our responsibility. We 753 * must either enqueue them or free them. The other pru_* routines 754 * generally are caller-frees. 755 */ 756 static int 757 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 758 struct sockaddr *nam, struct mbuf *control, struct thread *td) 759 { 760 int error = 0; 761 struct inpcb *inp; 762 struct tcpcb *tp = NULL; 763 int headlocked = 0; 764 #ifdef INET6 765 int isipv6; 766 #endif 767 TCPDEBUG0; 768 769 /* 770 * We require the pcbinfo lock in two cases: 771 * 772 * (1) An implied connect is taking place, which can result in 773 * binding IPs and ports and hence modification of the pcb hash 774 * chains. 775 * 776 * (2) PRUS_EOF is set, resulting in explicit close on the send. 777 */ 778 if ((nam != NULL) || (flags & PRUS_EOF)) { 779 INP_INFO_WLOCK(&V_tcbinfo); 780 headlocked = 1; 781 } 782 inp = sotoinpcb(so); 783 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 784 INP_WLOCK(inp); 785 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 786 if (control) 787 m_freem(control); 788 if (m) 789 m_freem(m); 790 error = ECONNRESET; 791 goto out; 792 } 793 #ifdef INET6 794 isipv6 = nam && nam->sa_family == AF_INET6; 795 #endif /* INET6 */ 796 tp = intotcpcb(inp); 797 TCPDEBUG1(); 798 if (control) { 799 /* TCP doesn't do control messages (rights, creds, etc) */ 800 if (control->m_len) { 801 m_freem(control); 802 if (m) 803 m_freem(m); 804 error = EINVAL; 805 goto out; 806 } 807 m_freem(control); /* empty control, just free it */ 808 } 809 if (!(flags & PRUS_OOB)) { 810 sbappendstream(&so->so_snd, m); 811 if (nam && tp->t_state < TCPS_SYN_SENT) { 812 /* 813 * Do implied connect if not yet connected, 814 * initialize window to default value, and 815 * initialize maxseg/maxopd using peer's cached 816 * MSS. 817 */ 818 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 819 #ifdef INET6 820 if (isipv6) 821 error = tcp6_connect(tp, nam, td); 822 else 823 #endif /* INET6 */ 824 error = tcp_connect(tp, nam, td); 825 if (error) 826 goto out; 827 tp->snd_wnd = TTCP_CLIENT_SND_WND; 828 tcp_mss(tp, -1); 829 } 830 if (flags & PRUS_EOF) { 831 /* 832 * Close the send side of the connection after 833 * the data is sent. 834 */ 835 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 836 socantsendmore(so); 837 tcp_usrclosed(tp); 838 } 839 if (headlocked) { 840 INP_INFO_WUNLOCK(&V_tcbinfo); 841 headlocked = 0; 842 } 843 if (tp != NULL) { 844 if (flags & PRUS_MORETOCOME) 845 tp->t_flags |= TF_MORETOCOME; 846 error = tcp_output_send(tp); 847 if (flags & PRUS_MORETOCOME) 848 tp->t_flags &= ~TF_MORETOCOME; 849 } 850 } else { 851 /* 852 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 853 */ 854 SOCKBUF_LOCK(&so->so_snd); 855 if (sbspace(&so->so_snd) < -512) { 856 SOCKBUF_UNLOCK(&so->so_snd); 857 m_freem(m); 858 error = ENOBUFS; 859 goto out; 860 } 861 /* 862 * According to RFC961 (Assigned Protocols), 863 * the urgent pointer points to the last octet 864 * of urgent data. We continue, however, 865 * to consider it to indicate the first octet 866 * of data past the urgent section. 867 * Otherwise, snd_up should be one lower. 868 */ 869 sbappendstream_locked(&so->so_snd, m); 870 SOCKBUF_UNLOCK(&so->so_snd); 871 if (nam && tp->t_state < TCPS_SYN_SENT) { 872 /* 873 * Do implied connect if not yet connected, 874 * initialize window to default value, and 875 * initialize maxseg/maxopd using peer's cached 876 * MSS. 877 */ 878 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 879 #ifdef INET6 880 if (isipv6) 881 error = tcp6_connect(tp, nam, td); 882 else 883 #endif /* INET6 */ 884 error = tcp_connect(tp, nam, td); 885 if (error) 886 goto out; 887 tp->snd_wnd = TTCP_CLIENT_SND_WND; 888 tcp_mss(tp, -1); 889 INP_INFO_WUNLOCK(&V_tcbinfo); 890 headlocked = 0; 891 } else if (nam) { 892 INP_INFO_WUNLOCK(&V_tcbinfo); 893 headlocked = 0; 894 } 895 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 896 tp->t_flags |= TF_FORCEDATA; 897 error = tcp_output_send(tp); 898 tp->t_flags &= ~TF_FORCEDATA; 899 } 900 out: 901 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 902 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 903 INP_WUNLOCK(inp); 904 if (headlocked) 905 INP_INFO_WUNLOCK(&V_tcbinfo); 906 return (error); 907 } 908 909 /* 910 * Abort the TCP. Drop the connection abruptly. 911 */ 912 static void 913 tcp_usr_abort(struct socket *so) 914 { 915 struct inpcb *inp; 916 struct tcpcb *tp = NULL; 917 TCPDEBUG0; 918 919 inp = sotoinpcb(so); 920 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 921 922 INP_INFO_WLOCK(&V_tcbinfo); 923 INP_WLOCK(inp); 924 KASSERT(inp->inp_socket != NULL, 925 ("tcp_usr_abort: inp_socket == NULL")); 926 927 /* 928 * If we still have full TCP state, and we're not dropped, drop. 929 */ 930 if (!(inp->inp_vflag & INP_TIMEWAIT) && 931 !(inp->inp_vflag & INP_DROPPED)) { 932 tp = intotcpcb(inp); 933 TCPDEBUG1(); 934 tcp_drop(tp, ECONNABORTED); 935 TCPDEBUG2(PRU_ABORT); 936 } 937 if (!(inp->inp_vflag & INP_DROPPED)) { 938 SOCK_LOCK(so); 939 so->so_state |= SS_PROTOREF; 940 SOCK_UNLOCK(so); 941 inp->inp_vflag |= INP_SOCKREF; 942 } 943 INP_WUNLOCK(inp); 944 INP_INFO_WUNLOCK(&V_tcbinfo); 945 } 946 947 /* 948 * TCP socket is closed. Start friendly disconnect. 949 */ 950 static void 951 tcp_usr_close(struct socket *so) 952 { 953 struct inpcb *inp; 954 struct tcpcb *tp = NULL; 955 TCPDEBUG0; 956 957 inp = sotoinpcb(so); 958 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 959 960 INP_INFO_WLOCK(&V_tcbinfo); 961 INP_WLOCK(inp); 962 KASSERT(inp->inp_socket != NULL, 963 ("tcp_usr_close: inp_socket == NULL")); 964 965 /* 966 * If we still have full TCP state, and we're not dropped, initiate 967 * a disconnect. 968 */ 969 if (!(inp->inp_vflag & INP_TIMEWAIT) && 970 !(inp->inp_vflag & INP_DROPPED)) { 971 tp = intotcpcb(inp); 972 TCPDEBUG1(); 973 tcp_disconnect(tp); 974 TCPDEBUG2(PRU_CLOSE); 975 } 976 if (!(inp->inp_vflag & INP_DROPPED)) { 977 SOCK_LOCK(so); 978 so->so_state |= SS_PROTOREF; 979 SOCK_UNLOCK(so); 980 inp->inp_vflag |= INP_SOCKREF; 981 } 982 INP_WUNLOCK(inp); 983 INP_INFO_WUNLOCK(&V_tcbinfo); 984 } 985 986 /* 987 * Receive out-of-band data. 988 */ 989 static int 990 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 991 { 992 int error = 0; 993 struct inpcb *inp; 994 struct tcpcb *tp = NULL; 995 996 TCPDEBUG0; 997 inp = sotoinpcb(so); 998 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 999 INP_WLOCK(inp); 1000 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1001 error = ECONNRESET; 1002 goto out; 1003 } 1004 tp = intotcpcb(inp); 1005 TCPDEBUG1(); 1006 if ((so->so_oobmark == 0 && 1007 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1008 so->so_options & SO_OOBINLINE || 1009 tp->t_oobflags & TCPOOB_HADDATA) { 1010 error = EINVAL; 1011 goto out; 1012 } 1013 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1014 error = EWOULDBLOCK; 1015 goto out; 1016 } 1017 m->m_len = 1; 1018 *mtod(m, caddr_t) = tp->t_iobc; 1019 if ((flags & MSG_PEEK) == 0) 1020 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1021 1022 out: 1023 TCPDEBUG2(PRU_RCVOOB); 1024 INP_WUNLOCK(inp); 1025 return (error); 1026 } 1027 1028 struct pr_usrreqs tcp_usrreqs = { 1029 .pru_abort = tcp_usr_abort, 1030 .pru_accept = tcp_usr_accept, 1031 .pru_attach = tcp_usr_attach, 1032 .pru_bind = tcp_usr_bind, 1033 .pru_connect = tcp_usr_connect, 1034 .pru_control = in_control, 1035 .pru_detach = tcp_usr_detach, 1036 .pru_disconnect = tcp_usr_disconnect, 1037 .pru_listen = tcp_usr_listen, 1038 .pru_peeraddr = in_getpeeraddr, 1039 .pru_rcvd = tcp_usr_rcvd, 1040 .pru_rcvoob = tcp_usr_rcvoob, 1041 .pru_send = tcp_usr_send, 1042 .pru_shutdown = tcp_usr_shutdown, 1043 .pru_sockaddr = in_getsockaddr, 1044 .pru_sosetlabel = in_pcbsosetlabel, 1045 .pru_close = tcp_usr_close, 1046 }; 1047 1048 #ifdef INET6 1049 struct pr_usrreqs tcp6_usrreqs = { 1050 .pru_abort = tcp_usr_abort, 1051 .pru_accept = tcp6_usr_accept, 1052 .pru_attach = tcp_usr_attach, 1053 .pru_bind = tcp6_usr_bind, 1054 .pru_connect = tcp6_usr_connect, 1055 .pru_control = in6_control, 1056 .pru_detach = tcp_usr_detach, 1057 .pru_disconnect = tcp_usr_disconnect, 1058 .pru_listen = tcp6_usr_listen, 1059 .pru_peeraddr = in6_mapped_peeraddr, 1060 .pru_rcvd = tcp_usr_rcvd, 1061 .pru_rcvoob = tcp_usr_rcvoob, 1062 .pru_send = tcp_usr_send, 1063 .pru_shutdown = tcp_usr_shutdown, 1064 .pru_sockaddr = in6_mapped_sockaddr, 1065 .pru_sosetlabel = in_pcbsosetlabel, 1066 .pru_close = tcp_usr_close, 1067 }; 1068 #endif /* INET6 */ 1069 1070 /* 1071 * Common subroutine to open a TCP connection to remote host specified 1072 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1073 * port number if needed. Call in_pcbconnect_setup to do the routing and 1074 * to choose a local host address (interface). If there is an existing 1075 * incarnation of the same connection in TIME-WAIT state and if the remote 1076 * host was sending CC options and if the connection duration was < MSL, then 1077 * truncate the previous TIME-WAIT state and proceed. 1078 * Initialize connection parameters and enter SYN-SENT state. 1079 */ 1080 static int 1081 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1082 { 1083 struct inpcb *inp = tp->t_inpcb, *oinp; 1084 struct socket *so = inp->inp_socket; 1085 struct in_addr laddr; 1086 u_short lport; 1087 int error; 1088 1089 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1090 INP_WLOCK_ASSERT(inp); 1091 1092 if (inp->inp_lport == 0) { 1093 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1094 if (error) 1095 return error; 1096 } 1097 1098 /* 1099 * Cannot simply call in_pcbconnect, because there might be an 1100 * earlier incarnation of this same connection still in 1101 * TIME_WAIT state, creating an ADDRINUSE error. 1102 */ 1103 laddr = inp->inp_laddr; 1104 lport = inp->inp_lport; 1105 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1106 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1107 if (error && oinp == NULL) 1108 return error; 1109 if (oinp) 1110 return EADDRINUSE; 1111 inp->inp_laddr = laddr; 1112 in_pcbrehash(inp); 1113 1114 /* 1115 * Compute window scaling to request: 1116 * Scale to fit into sweet spot. See tcp_syncache.c. 1117 * XXX: This should move to tcp_output(). 1118 */ 1119 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1120 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1121 tp->request_r_scale++; 1122 1123 soisconnecting(so); 1124 V_tcpstat.tcps_connattempt++; 1125 tp->t_state = TCPS_SYN_SENT; 1126 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1127 tp->iss = tcp_new_isn(tp); 1128 tp->t_bw_rtseq = tp->iss; 1129 tcp_sendseqinit(tp); 1130 1131 return 0; 1132 } 1133 1134 #ifdef INET6 1135 static int 1136 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1137 { 1138 struct inpcb *inp = tp->t_inpcb, *oinp; 1139 struct socket *so = inp->inp_socket; 1140 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1141 struct in6_addr *addr6; 1142 int error; 1143 1144 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1145 INP_WLOCK_ASSERT(inp); 1146 1147 if (inp->inp_lport == 0) { 1148 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1149 if (error) 1150 return error; 1151 } 1152 1153 /* 1154 * Cannot simply call in_pcbconnect, because there might be an 1155 * earlier incarnation of this same connection still in 1156 * TIME_WAIT state, creating an ADDRINUSE error. 1157 * in6_pcbladdr() also handles scope zone IDs. 1158 */ 1159 error = in6_pcbladdr(inp, nam, &addr6); 1160 if (error) 1161 return error; 1162 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1163 &sin6->sin6_addr, sin6->sin6_port, 1164 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1165 ? addr6 1166 : &inp->in6p_laddr, 1167 inp->inp_lport, 0, NULL); 1168 if (oinp) 1169 return EADDRINUSE; 1170 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1171 inp->in6p_laddr = *addr6; 1172 inp->in6p_faddr = sin6->sin6_addr; 1173 inp->inp_fport = sin6->sin6_port; 1174 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1175 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 1176 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 1177 inp->in6p_flowinfo |= 1178 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1179 in_pcbrehash(inp); 1180 1181 /* Compute window scaling to request. */ 1182 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1183 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) 1184 tp->request_r_scale++; 1185 1186 soisconnecting(so); 1187 V_tcpstat.tcps_connattempt++; 1188 tp->t_state = TCPS_SYN_SENT; 1189 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1190 tp->iss = tcp_new_isn(tp); 1191 tp->t_bw_rtseq = tp->iss; 1192 tcp_sendseqinit(tp); 1193 1194 return 0; 1195 } 1196 #endif /* INET6 */ 1197 1198 /* 1199 * Export TCP internal state information via a struct tcp_info, based on the 1200 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1201 * (TCP state machine, etc). We export all information using FreeBSD-native 1202 * constants -- for example, the numeric values for tcpi_state will differ 1203 * from Linux. 1204 */ 1205 static void 1206 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1207 { 1208 1209 INP_WLOCK_ASSERT(tp->t_inpcb); 1210 bzero(ti, sizeof(*ti)); 1211 1212 ti->tcpi_state = tp->t_state; 1213 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1214 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1215 if (tp->t_flags & TF_SACK_PERMIT) 1216 ti->tcpi_options |= TCPI_OPT_SACK; 1217 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1218 ti->tcpi_options |= TCPI_OPT_WSCALE; 1219 ti->tcpi_snd_wscale = tp->snd_scale; 1220 ti->tcpi_rcv_wscale = tp->rcv_scale; 1221 } 1222 1223 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1224 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1225 1226 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1227 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1228 1229 /* 1230 * FreeBSD-specific extension fields for tcp_info. 1231 */ 1232 ti->tcpi_rcv_space = tp->rcv_wnd; 1233 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1234 ti->tcpi_snd_wnd = tp->snd_wnd; 1235 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1236 ti->tcpi_snd_nxt = tp->snd_nxt; 1237 ti->__tcpi_snd_mss = tp->t_maxseg; 1238 ti->__tcpi_rcv_mss = tp->t_maxseg; 1239 if (tp->t_flags & TF_TOE) 1240 ti->tcpi_options |= TCPI_OPT_TOE; 1241 } 1242 1243 /* 1244 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1245 * socket option arguments. When it re-acquires the lock after the copy, it 1246 * has to revalidate that the connection is still valid for the socket 1247 * option. 1248 */ 1249 #define INP_WLOCK_RECHECK(inp) do { \ 1250 INP_WLOCK(inp); \ 1251 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { \ 1252 INP_WUNLOCK(inp); \ 1253 return (ECONNRESET); \ 1254 } \ 1255 tp = intotcpcb(inp); \ 1256 } while(0) 1257 1258 int 1259 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1260 { 1261 int error, opt, optval; 1262 struct inpcb *inp; 1263 struct tcpcb *tp; 1264 struct tcp_info ti; 1265 1266 error = 0; 1267 inp = sotoinpcb(so); 1268 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1269 INP_WLOCK(inp); 1270 if (sopt->sopt_level != IPPROTO_TCP) { 1271 #ifdef INET6 1272 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 1273 INP_WUNLOCK(inp); 1274 error = ip6_ctloutput(so, sopt); 1275 } else { 1276 #endif /* INET6 */ 1277 INP_WUNLOCK(inp); 1278 error = ip_ctloutput(so, sopt); 1279 #ifdef INET6 1280 } 1281 #endif 1282 return (error); 1283 } 1284 if (inp->inp_vflag & (INP_TIMEWAIT | INP_DROPPED)) { 1285 INP_WUNLOCK(inp); 1286 return (ECONNRESET); 1287 } 1288 1289 switch (sopt->sopt_dir) { 1290 case SOPT_SET: 1291 switch (sopt->sopt_name) { 1292 #ifdef TCP_SIGNATURE 1293 case TCP_MD5SIG: 1294 INP_WUNLOCK(inp); 1295 error = sooptcopyin(sopt, &optval, sizeof optval, 1296 sizeof optval); 1297 if (error) 1298 return (error); 1299 1300 INP_WLOCK_RECHECK(inp); 1301 if (optval > 0) 1302 tp->t_flags |= TF_SIGNATURE; 1303 else 1304 tp->t_flags &= ~TF_SIGNATURE; 1305 INP_WUNLOCK(inp); 1306 break; 1307 #endif /* TCP_SIGNATURE */ 1308 case TCP_NODELAY: 1309 case TCP_NOOPT: 1310 INP_WUNLOCK(inp); 1311 error = sooptcopyin(sopt, &optval, sizeof optval, 1312 sizeof optval); 1313 if (error) 1314 return (error); 1315 1316 INP_WLOCK_RECHECK(inp); 1317 switch (sopt->sopt_name) { 1318 case TCP_NODELAY: 1319 opt = TF_NODELAY; 1320 break; 1321 case TCP_NOOPT: 1322 opt = TF_NOOPT; 1323 break; 1324 default: 1325 opt = 0; /* dead code to fool gcc */ 1326 break; 1327 } 1328 1329 if (optval) 1330 tp->t_flags |= opt; 1331 else 1332 tp->t_flags &= ~opt; 1333 INP_WUNLOCK(inp); 1334 break; 1335 1336 case TCP_NOPUSH: 1337 INP_WUNLOCK(inp); 1338 error = sooptcopyin(sopt, &optval, sizeof optval, 1339 sizeof optval); 1340 if (error) 1341 return (error); 1342 1343 INP_WLOCK_RECHECK(inp); 1344 if (optval) 1345 tp->t_flags |= TF_NOPUSH; 1346 else { 1347 tp->t_flags &= ~TF_NOPUSH; 1348 error = tcp_output(tp); 1349 } 1350 INP_WUNLOCK(inp); 1351 break; 1352 1353 case TCP_MAXSEG: 1354 INP_WUNLOCK(inp); 1355 error = sooptcopyin(sopt, &optval, sizeof optval, 1356 sizeof optval); 1357 if (error) 1358 return (error); 1359 1360 INP_WLOCK_RECHECK(inp); 1361 if (optval > 0 && optval <= tp->t_maxseg && 1362 optval + 40 >= V_tcp_minmss) 1363 tp->t_maxseg = optval; 1364 else 1365 error = EINVAL; 1366 INP_WUNLOCK(inp); 1367 break; 1368 1369 case TCP_INFO: 1370 INP_WUNLOCK(inp); 1371 error = EINVAL; 1372 break; 1373 1374 default: 1375 INP_WUNLOCK(inp); 1376 error = ENOPROTOOPT; 1377 break; 1378 } 1379 break; 1380 1381 case SOPT_GET: 1382 tp = intotcpcb(inp); 1383 switch (sopt->sopt_name) { 1384 #ifdef TCP_SIGNATURE 1385 case TCP_MD5SIG: 1386 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1387 INP_WUNLOCK(inp); 1388 error = sooptcopyout(sopt, &optval, sizeof optval); 1389 break; 1390 #endif 1391 1392 case TCP_NODELAY: 1393 optval = tp->t_flags & TF_NODELAY; 1394 INP_WUNLOCK(inp); 1395 error = sooptcopyout(sopt, &optval, sizeof optval); 1396 break; 1397 case TCP_MAXSEG: 1398 optval = tp->t_maxseg; 1399 INP_WUNLOCK(inp); 1400 error = sooptcopyout(sopt, &optval, sizeof optval); 1401 break; 1402 case TCP_NOOPT: 1403 optval = tp->t_flags & TF_NOOPT; 1404 INP_WUNLOCK(inp); 1405 error = sooptcopyout(sopt, &optval, sizeof optval); 1406 break; 1407 case TCP_NOPUSH: 1408 optval = tp->t_flags & TF_NOPUSH; 1409 INP_WUNLOCK(inp); 1410 error = sooptcopyout(sopt, &optval, sizeof optval); 1411 break; 1412 case TCP_INFO: 1413 tcp_fill_info(tp, &ti); 1414 INP_WUNLOCK(inp); 1415 error = sooptcopyout(sopt, &ti, sizeof ti); 1416 break; 1417 default: 1418 INP_WUNLOCK(inp); 1419 error = ENOPROTOOPT; 1420 break; 1421 } 1422 break; 1423 } 1424 return (error); 1425 } 1426 #undef INP_WLOCK_RECHECK 1427 1428 /* 1429 * tcp_sendspace and tcp_recvspace are the default send and receive window 1430 * sizes, respectively. These are obsolescent (this information should 1431 * be set by the route). 1432 */ 1433 u_long tcp_sendspace = 1024*32; 1434 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1435 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1436 u_long tcp_recvspace = 1024*64; 1437 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1438 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1439 1440 /* 1441 * Attach TCP protocol to socket, allocating 1442 * internet protocol control block, tcp control block, 1443 * bufer space, and entering LISTEN state if to accept connections. 1444 */ 1445 static int 1446 tcp_attach(struct socket *so) 1447 { 1448 struct tcpcb *tp; 1449 struct inpcb *inp; 1450 int error; 1451 #ifdef INET6 1452 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1453 #endif 1454 1455 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1456 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1457 if (error) 1458 return (error); 1459 } 1460 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1461 so->so_snd.sb_flags |= SB_AUTOSIZE; 1462 INP_INFO_WLOCK(&V_tcbinfo); 1463 error = in_pcballoc(so, &V_tcbinfo); 1464 if (error) { 1465 INP_INFO_WUNLOCK(&V_tcbinfo); 1466 return (error); 1467 } 1468 inp = sotoinpcb(so); 1469 #ifdef INET6 1470 if (isipv6) { 1471 inp->inp_vflag |= INP_IPV6; 1472 inp->in6p_hops = -1; /* use kernel default */ 1473 } 1474 else 1475 #endif 1476 inp->inp_vflag |= INP_IPV4; 1477 tp = tcp_newtcpcb(inp); 1478 if (tp == NULL) { 1479 #ifdef INET6 1480 if (isipv6) { 1481 in6_pcbdetach(inp); 1482 in6_pcbfree(inp); 1483 } else { 1484 #endif 1485 in_pcbdetach(inp); 1486 in_pcbfree(inp); 1487 #ifdef INET6 1488 } 1489 #endif 1490 INP_INFO_WUNLOCK(&V_tcbinfo); 1491 return (ENOBUFS); 1492 } 1493 tp->t_state = TCPS_CLOSED; 1494 INP_WUNLOCK(inp); 1495 INP_INFO_WUNLOCK(&V_tcbinfo); 1496 return (0); 1497 } 1498 1499 /* 1500 * Initiate (or continue) disconnect. 1501 * If embryonic state, just send reset (once). 1502 * If in ``let data drain'' option and linger null, just drop. 1503 * Otherwise (hard), mark socket disconnecting and drop 1504 * current input data; switch states based on user close, and 1505 * send segment to peer (with FIN). 1506 */ 1507 static void 1508 tcp_disconnect(struct tcpcb *tp) 1509 { 1510 struct inpcb *inp = tp->t_inpcb; 1511 struct socket *so = inp->inp_socket; 1512 1513 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1514 INP_WLOCK_ASSERT(inp); 1515 1516 /* 1517 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1518 * socket is still open. 1519 */ 1520 if (tp->t_state < TCPS_ESTABLISHED) { 1521 tp = tcp_close(tp); 1522 KASSERT(tp != NULL, 1523 ("tcp_disconnect: tcp_close() returned NULL")); 1524 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1525 tp = tcp_drop(tp, 0); 1526 KASSERT(tp != NULL, 1527 ("tcp_disconnect: tcp_drop() returned NULL")); 1528 } else { 1529 soisdisconnecting(so); 1530 sbflush(&so->so_rcv); 1531 tcp_usrclosed(tp); 1532 if (!(inp->inp_vflag & INP_DROPPED)) 1533 tcp_output_disconnect(tp); 1534 } 1535 } 1536 1537 /* 1538 * User issued close, and wish to trail through shutdown states: 1539 * if never received SYN, just forget it. If got a SYN from peer, 1540 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1541 * If already got a FIN from peer, then almost done; go to LAST_ACK 1542 * state. In all other cases, have already sent FIN to peer (e.g. 1543 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1544 * for peer to send FIN or not respond to keep-alives, etc. 1545 * We can let the user exit from the close as soon as the FIN is acked. 1546 */ 1547 static void 1548 tcp_usrclosed(struct tcpcb *tp) 1549 { 1550 1551 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1552 INP_WLOCK_ASSERT(tp->t_inpcb); 1553 1554 switch (tp->t_state) { 1555 case TCPS_LISTEN: 1556 tcp_offload_listen_close(tp); 1557 /* FALLTHROUGH */ 1558 case TCPS_CLOSED: 1559 tp->t_state = TCPS_CLOSED; 1560 tp = tcp_close(tp); 1561 /* 1562 * tcp_close() should never return NULL here as the socket is 1563 * still open. 1564 */ 1565 KASSERT(tp != NULL, 1566 ("tcp_usrclosed: tcp_close() returned NULL")); 1567 break; 1568 1569 case TCPS_SYN_SENT: 1570 case TCPS_SYN_RECEIVED: 1571 tp->t_flags |= TF_NEEDFIN; 1572 break; 1573 1574 case TCPS_ESTABLISHED: 1575 tp->t_state = TCPS_FIN_WAIT_1; 1576 break; 1577 1578 case TCPS_CLOSE_WAIT: 1579 tp->t_state = TCPS_LAST_ACK; 1580 break; 1581 } 1582 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1583 soisdisconnected(tp->t_inpcb->inp_socket); 1584 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1585 if (tp->t_state == TCPS_FIN_WAIT_2) { 1586 int timeout; 1587 1588 timeout = (tcp_fast_finwait2_recycle) ? 1589 tcp_finwait2_timeout : tcp_maxidle; 1590 tcp_timer_activate(tp, TT_2MSL, timeout); 1591 } 1592 } 1593 } 1594 1595 #ifdef DDB 1596 static void 1597 db_print_indent(int indent) 1598 { 1599 int i; 1600 1601 for (i = 0; i < indent; i++) 1602 db_printf(" "); 1603 } 1604 1605 static void 1606 db_print_tstate(int t_state) 1607 { 1608 1609 switch (t_state) { 1610 case TCPS_CLOSED: 1611 db_printf("TCPS_CLOSED"); 1612 return; 1613 1614 case TCPS_LISTEN: 1615 db_printf("TCPS_LISTEN"); 1616 return; 1617 1618 case TCPS_SYN_SENT: 1619 db_printf("TCPS_SYN_SENT"); 1620 return; 1621 1622 case TCPS_SYN_RECEIVED: 1623 db_printf("TCPS_SYN_RECEIVED"); 1624 return; 1625 1626 case TCPS_ESTABLISHED: 1627 db_printf("TCPS_ESTABLISHED"); 1628 return; 1629 1630 case TCPS_CLOSE_WAIT: 1631 db_printf("TCPS_CLOSE_WAIT"); 1632 return; 1633 1634 case TCPS_FIN_WAIT_1: 1635 db_printf("TCPS_FIN_WAIT_1"); 1636 return; 1637 1638 case TCPS_CLOSING: 1639 db_printf("TCPS_CLOSING"); 1640 return; 1641 1642 case TCPS_LAST_ACK: 1643 db_printf("TCPS_LAST_ACK"); 1644 return; 1645 1646 case TCPS_FIN_WAIT_2: 1647 db_printf("TCPS_FIN_WAIT_2"); 1648 return; 1649 1650 case TCPS_TIME_WAIT: 1651 db_printf("TCPS_TIME_WAIT"); 1652 return; 1653 1654 default: 1655 db_printf("unknown"); 1656 return; 1657 } 1658 } 1659 1660 static void 1661 db_print_tflags(u_int t_flags) 1662 { 1663 int comma; 1664 1665 comma = 0; 1666 if (t_flags & TF_ACKNOW) { 1667 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1668 comma = 1; 1669 } 1670 if (t_flags & TF_DELACK) { 1671 db_printf("%sTF_DELACK", comma ? ", " : ""); 1672 comma = 1; 1673 } 1674 if (t_flags & TF_NODELAY) { 1675 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1676 comma = 1; 1677 } 1678 if (t_flags & TF_NOOPT) { 1679 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1680 comma = 1; 1681 } 1682 if (t_flags & TF_SENTFIN) { 1683 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1684 comma = 1; 1685 } 1686 if (t_flags & TF_REQ_SCALE) { 1687 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1688 comma = 1; 1689 } 1690 if (t_flags & TF_RCVD_SCALE) { 1691 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1692 comma = 1; 1693 } 1694 if (t_flags & TF_REQ_TSTMP) { 1695 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1696 comma = 1; 1697 } 1698 if (t_flags & TF_RCVD_TSTMP) { 1699 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1700 comma = 1; 1701 } 1702 if (t_flags & TF_SACK_PERMIT) { 1703 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1704 comma = 1; 1705 } 1706 if (t_flags & TF_NEEDSYN) { 1707 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1708 comma = 1; 1709 } 1710 if (t_flags & TF_NEEDFIN) { 1711 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1712 comma = 1; 1713 } 1714 if (t_flags & TF_NOPUSH) { 1715 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1716 comma = 1; 1717 } 1718 if (t_flags & TF_NOPUSH) { 1719 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1720 comma = 1; 1721 } 1722 if (t_flags & TF_MORETOCOME) { 1723 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1724 comma = 1; 1725 } 1726 if (t_flags & TF_LQ_OVERFLOW) { 1727 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1728 comma = 1; 1729 } 1730 if (t_flags & TF_LASTIDLE) { 1731 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1732 comma = 1; 1733 } 1734 if (t_flags & TF_RXWIN0SENT) { 1735 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1736 comma = 1; 1737 } 1738 if (t_flags & TF_FASTRECOVERY) { 1739 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1740 comma = 1; 1741 } 1742 if (t_flags & TF_WASFRECOVERY) { 1743 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1744 comma = 1; 1745 } 1746 if (t_flags & TF_SIGNATURE) { 1747 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1748 comma = 1; 1749 } 1750 if (t_flags & TF_FORCEDATA) { 1751 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1752 comma = 1; 1753 } 1754 if (t_flags & TF_TSO) { 1755 db_printf("%sTF_TSO", comma ? ", " : ""); 1756 comma = 1; 1757 } 1758 if (t_flags & TF_ECN_PERMIT) { 1759 db_printf("%sTF_ECN_PERMIT", comma ? ", " : ""); 1760 comma = 1; 1761 } 1762 } 1763 1764 static void 1765 db_print_toobflags(char t_oobflags) 1766 { 1767 int comma; 1768 1769 comma = 0; 1770 if (t_oobflags & TCPOOB_HAVEDATA) { 1771 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1772 comma = 1; 1773 } 1774 if (t_oobflags & TCPOOB_HADDATA) { 1775 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1776 comma = 1; 1777 } 1778 } 1779 1780 static void 1781 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1782 { 1783 1784 db_print_indent(indent); 1785 db_printf("%s at %p\n", name, tp); 1786 1787 indent += 2; 1788 1789 db_print_indent(indent); 1790 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1791 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1792 1793 db_print_indent(indent); 1794 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1795 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1796 1797 db_print_indent(indent); 1798 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1799 &tp->t_timers->tt_delack, tp->t_inpcb); 1800 1801 db_print_indent(indent); 1802 db_printf("t_state: %d (", tp->t_state); 1803 db_print_tstate(tp->t_state); 1804 db_printf(")\n"); 1805 1806 db_print_indent(indent); 1807 db_printf("t_flags: 0x%x (", tp->t_flags); 1808 db_print_tflags(tp->t_flags); 1809 db_printf(")\n"); 1810 1811 db_print_indent(indent); 1812 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1813 tp->snd_una, tp->snd_max, tp->snd_nxt); 1814 1815 db_print_indent(indent); 1816 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1817 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1818 1819 db_print_indent(indent); 1820 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1821 tp->iss, tp->irs, tp->rcv_nxt); 1822 1823 db_print_indent(indent); 1824 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1825 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1826 1827 db_print_indent(indent); 1828 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1829 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1830 1831 db_print_indent(indent); 1832 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1833 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1834 tp->snd_recover); 1835 1836 db_print_indent(indent); 1837 db_printf("t_maxopd: %u t_rcvtime: %lu t_startime: %lu\n", 1838 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1839 1840 db_print_indent(indent); 1841 db_printf("t_rttime: %d t_rtsq: 0x%08x t_bw_rtttime: %d\n", 1842 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1843 1844 db_print_indent(indent); 1845 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1846 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1847 tp->t_srtt); 1848 1849 db_print_indent(indent); 1850 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1851 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1852 tp->t_rttbest); 1853 1854 db_print_indent(indent); 1855 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1856 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1857 1858 db_print_indent(indent); 1859 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1860 db_print_toobflags(tp->t_oobflags); 1861 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1862 1863 db_print_indent(indent); 1864 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1865 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1866 1867 db_print_indent(indent); 1868 db_printf("ts_recent: %u ts_recent_age: %lu\n", 1869 tp->ts_recent, tp->ts_recent_age); 1870 1871 db_print_indent(indent); 1872 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1873 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1874 1875 db_print_indent(indent); 1876 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1877 "t_badrxtwin: %lu\n", tp->snd_ssthresh_prev, 1878 tp->snd_recover_prev, tp->t_badrxtwin); 1879 1880 db_print_indent(indent); 1881 db_printf("snd_numholes: %d snd_holes first: %p\n", 1882 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1883 1884 db_print_indent(indent); 1885 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1886 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1887 1888 /* Skip sackblks, sackhint. */ 1889 1890 db_print_indent(indent); 1891 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1892 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1893 } 1894 1895 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1896 { 1897 struct tcpcb *tp; 1898 1899 if (!have_addr) { 1900 db_printf("usage: show tcpcb <addr>\n"); 1901 return; 1902 } 1903 tp = (struct tcpcb *)addr; 1904 1905 db_print_tcpcb(tp, "tcpcb", 0); 1906 } 1907 #endif 1908