1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2006-2007 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_ddb.h" 38 #include "opt_inet.h" 39 #include "opt_inet6.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/malloc.h> 45 #include <sys/kernel.h> 46 #include <sys/sysctl.h> 47 #include <sys/mbuf.h> 48 #ifdef INET6 49 #include <sys/domain.h> 50 #endif /* INET6 */ 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/protosw.h> 54 #include <sys/proc.h> 55 #include <sys/jail.h> 56 57 #ifdef DDB 58 #include <ddb/ddb.h> 59 #endif 60 61 #include <net/if.h> 62 #include <net/route.h> 63 #include <net/vnet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #ifdef INET6 68 #include <netinet/ip6.h> 69 #endif 70 #include <netinet/in_pcb.h> 71 #ifdef INET6 72 #include <netinet6/in6_pcb.h> 73 #endif 74 #include <netinet/in_var.h> 75 #include <netinet/ip_var.h> 76 #ifdef INET6 77 #include <netinet6/ip6_var.h> 78 #include <netinet6/scope6_var.h> 79 #endif 80 #include <netinet/tcp.h> 81 #include <netinet/tcp_fsm.h> 82 #include <netinet/tcp_seq.h> 83 #include <netinet/tcp_timer.h> 84 #include <netinet/tcp_var.h> 85 #include <netinet/tcpip.h> 86 #ifdef TCPDEBUG 87 #include <netinet/tcp_debug.h> 88 #endif 89 #include <netinet/tcp_offload.h> 90 91 /* 92 * TCP protocol interface to socket abstraction. 93 */ 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; 159 160 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 161 INP_WLOCK_ASSERT(inp); 162 163 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 164 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 165 166 tp = intotcpcb(inp); 167 168 if (inp->inp_flags & INP_TIMEWAIT) { 169 /* 170 * There are two cases to handle: one in which the time wait 171 * state is being discarded (INP_DROPPED), and one in which 172 * this connection will remain in timewait. In the former, 173 * it is time to discard all state (except tcptw, which has 174 * already been discarded by the timewait close code, which 175 * should be further up the call stack somewhere). In the 176 * latter case, we detach from the socket, but leave the pcb 177 * present until timewait ends. 178 * 179 * XXXRW: Would it be cleaner to free the tcptw here? 180 */ 181 if (inp->inp_flags & INP_DROPPED) { 182 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 183 "INP_DROPPED && tp != NULL")); 184 in_pcbdetach(inp); 185 in_pcbfree(inp); 186 } else { 187 in_pcbdetach(inp); 188 INP_WUNLOCK(inp); 189 } 190 } else { 191 /* 192 * If the connection is not in timewait, we consider two 193 * two conditions: one in which no further processing is 194 * necessary (dropped || embryonic), and one in which TCP is 195 * not yet done, but no longer requires the socket, so the 196 * pcb will persist for the time being. 197 * 198 * XXXRW: Does the second case still occur? 199 */ 200 if (inp->inp_flags & INP_DROPPED || 201 tp->t_state < TCPS_SYN_SENT) { 202 tcp_discardcb(tp); 203 in_pcbdetach(inp); 204 in_pcbfree(inp); 205 } else 206 in_pcbdetach(inp); 207 } 208 } 209 210 /* 211 * pru_detach() detaches the TCP protocol from the socket. 212 * If the protocol state is non-embryonic, then can't 213 * do this directly: have to initiate a pru_disconnect(), 214 * which may finish later; embryonic TCB's can just 215 * be discarded here. 216 */ 217 static void 218 tcp_usr_detach(struct socket *so) 219 { 220 struct inpcb *inp; 221 222 inp = sotoinpcb(so); 223 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 224 INP_INFO_WLOCK(&V_tcbinfo); 225 INP_WLOCK(inp); 226 KASSERT(inp->inp_socket != NULL, 227 ("tcp_usr_detach: inp_socket == NULL")); 228 tcp_detach(so, inp); 229 INP_INFO_WUNLOCK(&V_tcbinfo); 230 } 231 232 /* 233 * Give the socket an address. 234 */ 235 static int 236 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 237 { 238 int error = 0; 239 struct inpcb *inp; 240 struct tcpcb *tp = NULL; 241 struct sockaddr_in *sinp; 242 243 sinp = (struct sockaddr_in *)nam; 244 if (nam->sa_len != sizeof (*sinp)) 245 return (EINVAL); 246 /* 247 * Must check for multicast addresses and disallow binding 248 * to them. 249 */ 250 if (sinp->sin_family == AF_INET && 251 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 252 return (EAFNOSUPPORT); 253 254 TCPDEBUG0; 255 INP_INFO_WLOCK(&V_tcbinfo); 256 inp = sotoinpcb(so); 257 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 258 INP_WLOCK(inp); 259 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 260 error = EINVAL; 261 goto out; 262 } 263 tp = intotcpcb(inp); 264 TCPDEBUG1(); 265 error = in_pcbbind(inp, nam, td->td_ucred); 266 out: 267 TCPDEBUG2(PRU_BIND); 268 INP_WUNLOCK(inp); 269 INP_INFO_WUNLOCK(&V_tcbinfo); 270 271 return (error); 272 } 273 274 #ifdef INET6 275 static int 276 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 277 { 278 int error = 0; 279 struct inpcb *inp; 280 struct tcpcb *tp = NULL; 281 struct sockaddr_in6 *sin6p; 282 283 sin6p = (struct sockaddr_in6 *)nam; 284 if (nam->sa_len != sizeof (*sin6p)) 285 return (EINVAL); 286 /* 287 * Must check for multicast addresses and disallow binding 288 * to them. 289 */ 290 if (sin6p->sin6_family == AF_INET6 && 291 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 292 return (EAFNOSUPPORT); 293 294 TCPDEBUG0; 295 INP_INFO_WLOCK(&V_tcbinfo); 296 inp = sotoinpcb(so); 297 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 298 INP_WLOCK(inp); 299 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 300 error = EINVAL; 301 goto out; 302 } 303 tp = intotcpcb(inp); 304 TCPDEBUG1(); 305 inp->inp_vflag &= ~INP_IPV4; 306 inp->inp_vflag |= INP_IPV6; 307 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 308 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 309 inp->inp_vflag |= INP_IPV4; 310 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 311 struct sockaddr_in sin; 312 313 in6_sin6_2_sin(&sin, sin6p); 314 inp->inp_vflag |= INP_IPV4; 315 inp->inp_vflag &= ~INP_IPV6; 316 error = in_pcbbind(inp, (struct sockaddr *)&sin, 317 td->td_ucred); 318 goto out; 319 } 320 } 321 error = in6_pcbbind(inp, nam, td->td_ucred); 322 out: 323 TCPDEBUG2(PRU_BIND); 324 INP_WUNLOCK(inp); 325 INP_INFO_WUNLOCK(&V_tcbinfo); 326 return (error); 327 } 328 #endif /* INET6 */ 329 330 /* 331 * Prepare to accept connections. 332 */ 333 static int 334 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 335 { 336 int error = 0; 337 struct inpcb *inp; 338 struct tcpcb *tp = NULL; 339 340 TCPDEBUG0; 341 INP_INFO_WLOCK(&V_tcbinfo); 342 inp = sotoinpcb(so); 343 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 344 INP_WLOCK(inp); 345 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 346 error = EINVAL; 347 goto out; 348 } 349 tp = intotcpcb(inp); 350 TCPDEBUG1(); 351 SOCK_LOCK(so); 352 error = solisten_proto_check(so); 353 if (error == 0 && inp->inp_lport == 0) 354 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 355 if (error == 0) { 356 tp->t_state = TCPS_LISTEN; 357 solisten_proto(so, backlog); 358 tcp_offload_listen_open(tp); 359 } 360 SOCK_UNLOCK(so); 361 362 out: 363 TCPDEBUG2(PRU_LISTEN); 364 INP_WUNLOCK(inp); 365 INP_INFO_WUNLOCK(&V_tcbinfo); 366 return (error); 367 } 368 369 #ifdef INET6 370 static int 371 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 372 { 373 int error = 0; 374 struct inpcb *inp; 375 struct tcpcb *tp = NULL; 376 377 TCPDEBUG0; 378 INP_INFO_WLOCK(&V_tcbinfo); 379 inp = sotoinpcb(so); 380 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 381 INP_WLOCK(inp); 382 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 383 error = EINVAL; 384 goto out; 385 } 386 tp = intotcpcb(inp); 387 TCPDEBUG1(); 388 SOCK_LOCK(so); 389 error = solisten_proto_check(so); 390 if (error == 0 && inp->inp_lport == 0) { 391 inp->inp_vflag &= ~INP_IPV4; 392 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 393 inp->inp_vflag |= INP_IPV4; 394 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 395 } 396 if (error == 0) { 397 tp->t_state = TCPS_LISTEN; 398 solisten_proto(so, backlog); 399 } 400 SOCK_UNLOCK(so); 401 402 out: 403 TCPDEBUG2(PRU_LISTEN); 404 INP_WUNLOCK(inp); 405 INP_INFO_WUNLOCK(&V_tcbinfo); 406 return (error); 407 } 408 #endif /* INET6 */ 409 410 /* 411 * Initiate connection to peer. 412 * Create a template for use in transmissions on this connection. 413 * Enter SYN_SENT state, and mark socket as connecting. 414 * Start keep-alive timer, and seed output sequence space. 415 * Send initial segment on connection. 416 */ 417 static int 418 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 419 { 420 int error = 0; 421 struct inpcb *inp; 422 struct tcpcb *tp = NULL; 423 struct sockaddr_in *sinp; 424 425 sinp = (struct sockaddr_in *)nam; 426 if (nam->sa_len != sizeof (*sinp)) 427 return (EINVAL); 428 /* 429 * Must disallow TCP ``connections'' to multicast addresses. 430 */ 431 if (sinp->sin_family == AF_INET 432 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 433 return (EAFNOSUPPORT); 434 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 435 return (error); 436 437 TCPDEBUG0; 438 INP_INFO_WLOCK(&V_tcbinfo); 439 inp = sotoinpcb(so); 440 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 441 INP_WLOCK(inp); 442 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 443 error = EINVAL; 444 goto out; 445 } 446 tp = intotcpcb(inp); 447 TCPDEBUG1(); 448 if ((error = tcp_connect(tp, nam, td)) != 0) 449 goto out; 450 error = tcp_output_connect(so, nam); 451 out: 452 TCPDEBUG2(PRU_CONNECT); 453 INP_WUNLOCK(inp); 454 INP_INFO_WUNLOCK(&V_tcbinfo); 455 return (error); 456 } 457 458 #ifdef INET6 459 static int 460 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 461 { 462 int error = 0; 463 struct inpcb *inp; 464 struct tcpcb *tp = NULL; 465 struct sockaddr_in6 *sin6p; 466 467 TCPDEBUG0; 468 469 sin6p = (struct sockaddr_in6 *)nam; 470 if (nam->sa_len != sizeof (*sin6p)) 471 return (EINVAL); 472 /* 473 * Must disallow TCP ``connections'' to multicast addresses. 474 */ 475 if (sin6p->sin6_family == AF_INET6 476 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 477 return (EAFNOSUPPORT); 478 479 INP_INFO_WLOCK(&V_tcbinfo); 480 inp = sotoinpcb(so); 481 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 482 INP_WLOCK(inp); 483 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 484 error = EINVAL; 485 goto out; 486 } 487 tp = intotcpcb(inp); 488 TCPDEBUG1(); 489 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 490 struct sockaddr_in sin; 491 492 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 493 error = EINVAL; 494 goto out; 495 } 496 497 in6_sin6_2_sin(&sin, sin6p); 498 inp->inp_vflag |= INP_IPV4; 499 inp->inp_vflag &= ~INP_IPV6; 500 if ((error = prison_remote_ip4(td->td_ucred, 501 &sin.sin_addr)) != 0) 502 goto out; 503 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 504 goto out; 505 error = tcp_output_connect(so, nam); 506 goto out; 507 } 508 inp->inp_vflag &= ~INP_IPV4; 509 inp->inp_vflag |= INP_IPV6; 510 inp->inp_inc.inc_flags |= INC_ISIPV6; 511 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0) 512 goto out; 513 if ((error = tcp6_connect(tp, nam, td)) != 0) 514 goto out; 515 error = tcp_output_connect(so, nam); 516 517 out: 518 TCPDEBUG2(PRU_CONNECT); 519 INP_WUNLOCK(inp); 520 INP_INFO_WUNLOCK(&V_tcbinfo); 521 return (error); 522 } 523 #endif /* INET6 */ 524 525 /* 526 * Initiate disconnect from peer. 527 * If connection never passed embryonic stage, just drop; 528 * else if don't need to let data drain, then can just drop anyways, 529 * else have to begin TCP shutdown process: mark socket disconnecting, 530 * drain unread data, state switch to reflect user close, and 531 * send segment (e.g. FIN) to peer. Socket will be really disconnected 532 * when peer sends FIN and acks ours. 533 * 534 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 535 */ 536 static int 537 tcp_usr_disconnect(struct socket *so) 538 { 539 struct inpcb *inp; 540 struct tcpcb *tp = NULL; 541 int error = 0; 542 543 TCPDEBUG0; 544 INP_INFO_WLOCK(&V_tcbinfo); 545 inp = sotoinpcb(so); 546 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 547 INP_WLOCK(inp); 548 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 549 error = ECONNRESET; 550 goto out; 551 } 552 tp = intotcpcb(inp); 553 TCPDEBUG1(); 554 tcp_disconnect(tp); 555 out: 556 TCPDEBUG2(PRU_DISCONNECT); 557 INP_WUNLOCK(inp); 558 INP_INFO_WUNLOCK(&V_tcbinfo); 559 return (error); 560 } 561 562 /* 563 * Accept a connection. Essentially all the work is done at higher levels; 564 * just return the address of the peer, storing through addr. 565 * 566 * The rationale for acquiring the tcbinfo lock here is somewhat complicated, 567 * and is described in detail in the commit log entry for r175612. Acquiring 568 * it delays an accept(2) racing with sonewconn(), which inserts the socket 569 * before the inpcb address/port fields are initialized. A better fix would 570 * prevent the socket from being placed in the listen queue until all fields 571 * are fully initialized. 572 */ 573 static int 574 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 575 { 576 int error = 0; 577 struct inpcb *inp = NULL; 578 struct tcpcb *tp = NULL; 579 struct in_addr addr; 580 in_port_t port = 0; 581 TCPDEBUG0; 582 583 if (so->so_state & SS_ISDISCONNECTED) 584 return (ECONNABORTED); 585 586 inp = sotoinpcb(so); 587 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 588 INP_INFO_RLOCK(&V_tcbinfo); 589 INP_WLOCK(inp); 590 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 591 error = ECONNABORTED; 592 goto out; 593 } 594 tp = intotcpcb(inp); 595 TCPDEBUG1(); 596 597 /* 598 * We inline in_getpeeraddr and COMMON_END here, so that we can 599 * copy the data of interest and defer the malloc until after we 600 * release the lock. 601 */ 602 port = inp->inp_fport; 603 addr = inp->inp_faddr; 604 605 out: 606 TCPDEBUG2(PRU_ACCEPT); 607 INP_WUNLOCK(inp); 608 INP_INFO_RUNLOCK(&V_tcbinfo); 609 if (error == 0) 610 *nam = in_sockaddr(port, &addr); 611 return error; 612 } 613 614 #ifdef INET6 615 static int 616 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 617 { 618 struct inpcb *inp = NULL; 619 int error = 0; 620 struct tcpcb *tp = NULL; 621 struct in_addr addr; 622 struct in6_addr addr6; 623 in_port_t port = 0; 624 int v4 = 0; 625 TCPDEBUG0; 626 627 if (so->so_state & SS_ISDISCONNECTED) 628 return (ECONNABORTED); 629 630 inp = sotoinpcb(so); 631 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 632 INP_WLOCK(inp); 633 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 634 error = ECONNABORTED; 635 goto out; 636 } 637 tp = intotcpcb(inp); 638 TCPDEBUG1(); 639 640 /* 641 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 642 * copy the data of interest and defer the malloc until after we 643 * release the lock. 644 */ 645 if (inp->inp_vflag & INP_IPV4) { 646 v4 = 1; 647 port = inp->inp_fport; 648 addr = inp->inp_faddr; 649 } else { 650 port = inp->inp_fport; 651 addr6 = inp->in6p_faddr; 652 } 653 654 out: 655 TCPDEBUG2(PRU_ACCEPT); 656 INP_WUNLOCK(inp); 657 if (error == 0) { 658 if (v4) 659 *nam = in6_v4mapsin6_sockaddr(port, &addr); 660 else 661 *nam = in6_sockaddr(port, &addr6); 662 } 663 return error; 664 } 665 #endif /* INET6 */ 666 667 /* 668 * Mark the connection as being incapable of further output. 669 */ 670 static int 671 tcp_usr_shutdown(struct socket *so) 672 { 673 int error = 0; 674 struct inpcb *inp; 675 struct tcpcb *tp = NULL; 676 677 TCPDEBUG0; 678 INP_INFO_WLOCK(&V_tcbinfo); 679 inp = sotoinpcb(so); 680 KASSERT(inp != NULL, ("inp == NULL")); 681 INP_WLOCK(inp); 682 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 683 error = ECONNRESET; 684 goto out; 685 } 686 tp = intotcpcb(inp); 687 TCPDEBUG1(); 688 socantsendmore(so); 689 tcp_usrclosed(tp); 690 if (!(inp->inp_flags & INP_DROPPED)) 691 error = tcp_output_disconnect(tp); 692 693 out: 694 TCPDEBUG2(PRU_SHUTDOWN); 695 INP_WUNLOCK(inp); 696 INP_INFO_WUNLOCK(&V_tcbinfo); 697 698 return (error); 699 } 700 701 /* 702 * After a receive, possibly send window update to peer. 703 */ 704 static int 705 tcp_usr_rcvd(struct socket *so, int flags) 706 { 707 struct inpcb *inp; 708 struct tcpcb *tp = NULL; 709 int error = 0; 710 711 TCPDEBUG0; 712 inp = sotoinpcb(so); 713 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 714 INP_WLOCK(inp); 715 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 716 error = ECONNRESET; 717 goto out; 718 } 719 tp = intotcpcb(inp); 720 TCPDEBUG1(); 721 tcp_output_rcvd(tp); 722 723 out: 724 TCPDEBUG2(PRU_RCVD); 725 INP_WUNLOCK(inp); 726 return (error); 727 } 728 729 /* 730 * Do a send by putting data in output queue and updating urgent 731 * marker if URG set. Possibly send more data. Unlike the other 732 * pru_*() routines, the mbuf chains are our responsibility. We 733 * must either enqueue them or free them. The other pru_* routines 734 * generally are caller-frees. 735 */ 736 static int 737 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 738 struct sockaddr *nam, struct mbuf *control, struct thread *td) 739 { 740 int error = 0; 741 struct inpcb *inp; 742 struct tcpcb *tp = NULL; 743 int headlocked = 0; 744 #ifdef INET6 745 int isipv6; 746 #endif 747 TCPDEBUG0; 748 749 /* 750 * We require the pcbinfo lock in two cases: 751 * 752 * (1) An implied connect is taking place, which can result in 753 * binding IPs and ports and hence modification of the pcb hash 754 * chains. 755 * 756 * (2) PRUS_EOF is set, resulting in explicit close on the send. 757 */ 758 if ((nam != NULL) || (flags & PRUS_EOF)) { 759 INP_INFO_WLOCK(&V_tcbinfo); 760 headlocked = 1; 761 } 762 inp = sotoinpcb(so); 763 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 764 INP_WLOCK(inp); 765 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 766 if (control) 767 m_freem(control); 768 if (m) 769 m_freem(m); 770 error = ECONNRESET; 771 goto out; 772 } 773 #ifdef INET6 774 isipv6 = nam && nam->sa_family == AF_INET6; 775 #endif /* INET6 */ 776 tp = intotcpcb(inp); 777 TCPDEBUG1(); 778 if (control) { 779 /* TCP doesn't do control messages (rights, creds, etc) */ 780 if (control->m_len) { 781 m_freem(control); 782 if (m) 783 m_freem(m); 784 error = EINVAL; 785 goto out; 786 } 787 m_freem(control); /* empty control, just free it */ 788 } 789 if (!(flags & PRUS_OOB)) { 790 sbappendstream(&so->so_snd, m); 791 if (nam && tp->t_state < TCPS_SYN_SENT) { 792 /* 793 * Do implied connect if not yet connected, 794 * initialize window to default value, and 795 * initialize maxseg/maxopd using peer's cached 796 * MSS. 797 */ 798 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 799 #ifdef INET6 800 if (isipv6) 801 error = tcp6_connect(tp, nam, td); 802 else 803 #endif /* INET6 */ 804 error = tcp_connect(tp, nam, td); 805 if (error) 806 goto out; 807 tp->snd_wnd = TTCP_CLIENT_SND_WND; 808 tcp_mss(tp, -1); 809 } 810 if (flags & PRUS_EOF) { 811 /* 812 * Close the send side of the connection after 813 * the data is sent. 814 */ 815 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 816 socantsendmore(so); 817 tcp_usrclosed(tp); 818 } 819 if (headlocked) { 820 INP_INFO_WUNLOCK(&V_tcbinfo); 821 headlocked = 0; 822 } 823 if (!(inp->inp_flags & INP_DROPPED)) { 824 if (flags & PRUS_MORETOCOME) 825 tp->t_flags |= TF_MORETOCOME; 826 error = tcp_output_send(tp); 827 if (flags & PRUS_MORETOCOME) 828 tp->t_flags &= ~TF_MORETOCOME; 829 } 830 } else { 831 /* 832 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 833 */ 834 SOCKBUF_LOCK(&so->so_snd); 835 if (sbspace(&so->so_snd) < -512) { 836 SOCKBUF_UNLOCK(&so->so_snd); 837 m_freem(m); 838 error = ENOBUFS; 839 goto out; 840 } 841 /* 842 * According to RFC961 (Assigned Protocols), 843 * the urgent pointer points to the last octet 844 * of urgent data. We continue, however, 845 * to consider it to indicate the first octet 846 * of data past the urgent section. 847 * Otherwise, snd_up should be one lower. 848 */ 849 sbappendstream_locked(&so->so_snd, m); 850 SOCKBUF_UNLOCK(&so->so_snd); 851 if (nam && tp->t_state < TCPS_SYN_SENT) { 852 /* 853 * Do implied connect if not yet connected, 854 * initialize window to default value, and 855 * initialize maxseg/maxopd using peer's cached 856 * MSS. 857 */ 858 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 859 #ifdef INET6 860 if (isipv6) 861 error = tcp6_connect(tp, nam, td); 862 else 863 #endif /* INET6 */ 864 error = tcp_connect(tp, nam, td); 865 if (error) 866 goto out; 867 tp->snd_wnd = TTCP_CLIENT_SND_WND; 868 tcp_mss(tp, -1); 869 INP_INFO_WUNLOCK(&V_tcbinfo); 870 headlocked = 0; 871 } else if (nam) { 872 INP_INFO_WUNLOCK(&V_tcbinfo); 873 headlocked = 0; 874 } 875 tp->snd_up = tp->snd_una + so->so_snd.sb_cc; 876 tp->t_flags |= TF_FORCEDATA; 877 error = tcp_output_send(tp); 878 tp->t_flags &= ~TF_FORCEDATA; 879 } 880 out: 881 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 882 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 883 INP_WUNLOCK(inp); 884 if (headlocked) 885 INP_INFO_WUNLOCK(&V_tcbinfo); 886 return (error); 887 } 888 889 /* 890 * Abort the TCP. Drop the connection abruptly. 891 */ 892 static void 893 tcp_usr_abort(struct socket *so) 894 { 895 struct inpcb *inp; 896 struct tcpcb *tp = NULL; 897 TCPDEBUG0; 898 899 inp = sotoinpcb(so); 900 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 901 902 INP_INFO_WLOCK(&V_tcbinfo); 903 INP_WLOCK(inp); 904 KASSERT(inp->inp_socket != NULL, 905 ("tcp_usr_abort: inp_socket == NULL")); 906 907 /* 908 * If we still have full TCP state, and we're not dropped, drop. 909 */ 910 if (!(inp->inp_flags & INP_TIMEWAIT) && 911 !(inp->inp_flags & INP_DROPPED)) { 912 tp = intotcpcb(inp); 913 TCPDEBUG1(); 914 tcp_drop(tp, ECONNABORTED); 915 TCPDEBUG2(PRU_ABORT); 916 } 917 if (!(inp->inp_flags & INP_DROPPED)) { 918 SOCK_LOCK(so); 919 so->so_state |= SS_PROTOREF; 920 SOCK_UNLOCK(so); 921 inp->inp_flags |= INP_SOCKREF; 922 } 923 INP_WUNLOCK(inp); 924 INP_INFO_WUNLOCK(&V_tcbinfo); 925 } 926 927 /* 928 * TCP socket is closed. Start friendly disconnect. 929 */ 930 static void 931 tcp_usr_close(struct socket *so) 932 { 933 struct inpcb *inp; 934 struct tcpcb *tp = NULL; 935 TCPDEBUG0; 936 937 inp = sotoinpcb(so); 938 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 939 940 INP_INFO_WLOCK(&V_tcbinfo); 941 INP_WLOCK(inp); 942 KASSERT(inp->inp_socket != NULL, 943 ("tcp_usr_close: inp_socket == NULL")); 944 945 /* 946 * If we still have full TCP state, and we're not dropped, initiate 947 * a disconnect. 948 */ 949 if (!(inp->inp_flags & INP_TIMEWAIT) && 950 !(inp->inp_flags & INP_DROPPED)) { 951 tp = intotcpcb(inp); 952 TCPDEBUG1(); 953 tcp_disconnect(tp); 954 TCPDEBUG2(PRU_CLOSE); 955 } 956 if (!(inp->inp_flags & INP_DROPPED)) { 957 SOCK_LOCK(so); 958 so->so_state |= SS_PROTOREF; 959 SOCK_UNLOCK(so); 960 inp->inp_flags |= INP_SOCKREF; 961 } 962 INP_WUNLOCK(inp); 963 INP_INFO_WUNLOCK(&V_tcbinfo); 964 } 965 966 /* 967 * Receive out-of-band data. 968 */ 969 static int 970 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 971 { 972 int error = 0; 973 struct inpcb *inp; 974 struct tcpcb *tp = NULL; 975 976 TCPDEBUG0; 977 inp = sotoinpcb(so); 978 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 979 INP_WLOCK(inp); 980 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 981 error = ECONNRESET; 982 goto out; 983 } 984 tp = intotcpcb(inp); 985 TCPDEBUG1(); 986 if ((so->so_oobmark == 0 && 987 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 988 so->so_options & SO_OOBINLINE || 989 tp->t_oobflags & TCPOOB_HADDATA) { 990 error = EINVAL; 991 goto out; 992 } 993 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 994 error = EWOULDBLOCK; 995 goto out; 996 } 997 m->m_len = 1; 998 *mtod(m, caddr_t) = tp->t_iobc; 999 if ((flags & MSG_PEEK) == 0) 1000 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1001 1002 out: 1003 TCPDEBUG2(PRU_RCVOOB); 1004 INP_WUNLOCK(inp); 1005 return (error); 1006 } 1007 1008 struct pr_usrreqs tcp_usrreqs = { 1009 .pru_abort = tcp_usr_abort, 1010 .pru_accept = tcp_usr_accept, 1011 .pru_attach = tcp_usr_attach, 1012 .pru_bind = tcp_usr_bind, 1013 .pru_connect = tcp_usr_connect, 1014 .pru_control = in_control, 1015 .pru_detach = tcp_usr_detach, 1016 .pru_disconnect = tcp_usr_disconnect, 1017 .pru_listen = tcp_usr_listen, 1018 .pru_peeraddr = in_getpeeraddr, 1019 .pru_rcvd = tcp_usr_rcvd, 1020 .pru_rcvoob = tcp_usr_rcvoob, 1021 .pru_send = tcp_usr_send, 1022 .pru_shutdown = tcp_usr_shutdown, 1023 .pru_sockaddr = in_getsockaddr, 1024 .pru_sosetlabel = in_pcbsosetlabel, 1025 .pru_close = tcp_usr_close, 1026 }; 1027 1028 #ifdef INET6 1029 struct pr_usrreqs tcp6_usrreqs = { 1030 .pru_abort = tcp_usr_abort, 1031 .pru_accept = tcp6_usr_accept, 1032 .pru_attach = tcp_usr_attach, 1033 .pru_bind = tcp6_usr_bind, 1034 .pru_connect = tcp6_usr_connect, 1035 .pru_control = in6_control, 1036 .pru_detach = tcp_usr_detach, 1037 .pru_disconnect = tcp_usr_disconnect, 1038 .pru_listen = tcp6_usr_listen, 1039 .pru_peeraddr = in6_mapped_peeraddr, 1040 .pru_rcvd = tcp_usr_rcvd, 1041 .pru_rcvoob = tcp_usr_rcvoob, 1042 .pru_send = tcp_usr_send, 1043 .pru_shutdown = tcp_usr_shutdown, 1044 .pru_sockaddr = in6_mapped_sockaddr, 1045 .pru_sosetlabel = in_pcbsosetlabel, 1046 .pru_close = tcp_usr_close, 1047 }; 1048 #endif /* INET6 */ 1049 1050 /* 1051 * Common subroutine to open a TCP connection to remote host specified 1052 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1053 * port number if needed. Call in_pcbconnect_setup to do the routing and 1054 * to choose a local host address (interface). If there is an existing 1055 * incarnation of the same connection in TIME-WAIT state and if the remote 1056 * host was sending CC options and if the connection duration was < MSL, then 1057 * truncate the previous TIME-WAIT state and proceed. 1058 * Initialize connection parameters and enter SYN-SENT state. 1059 */ 1060 static int 1061 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1062 { 1063 struct inpcb *inp = tp->t_inpcb, *oinp; 1064 struct socket *so = inp->inp_socket; 1065 struct in_addr laddr; 1066 u_short lport; 1067 int error; 1068 1069 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1070 INP_WLOCK_ASSERT(inp); 1071 1072 if (inp->inp_lport == 0) { 1073 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1074 if (error) 1075 return error; 1076 } 1077 1078 /* 1079 * Cannot simply call in_pcbconnect, because there might be an 1080 * earlier incarnation of this same connection still in 1081 * TIME_WAIT state, creating an ADDRINUSE error. 1082 */ 1083 laddr = inp->inp_laddr; 1084 lport = inp->inp_lport; 1085 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1086 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1087 if (error && oinp == NULL) 1088 return error; 1089 if (oinp) 1090 return EADDRINUSE; 1091 inp->inp_laddr = laddr; 1092 in_pcbrehash(inp); 1093 1094 /* 1095 * Compute window scaling to request: 1096 * Scale to fit into sweet spot. See tcp_syncache.c. 1097 * XXX: This should move to tcp_output(). 1098 */ 1099 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1100 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1101 tp->request_r_scale++; 1102 1103 soisconnecting(so); 1104 TCPSTAT_INC(tcps_connattempt); 1105 tp->t_state = TCPS_SYN_SENT; 1106 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1107 tp->iss = tcp_new_isn(tp); 1108 tp->t_bw_rtseq = tp->iss; 1109 tcp_sendseqinit(tp); 1110 1111 return 0; 1112 } 1113 1114 #ifdef INET6 1115 static int 1116 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1117 { 1118 struct inpcb *inp = tp->t_inpcb, *oinp; 1119 struct socket *so = inp->inp_socket; 1120 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1121 struct in6_addr addr6; 1122 int error; 1123 1124 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1125 INP_WLOCK_ASSERT(inp); 1126 1127 if (inp->inp_lport == 0) { 1128 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1129 if (error) 1130 return error; 1131 } 1132 1133 /* 1134 * Cannot simply call in_pcbconnect, because there might be an 1135 * earlier incarnation of this same connection still in 1136 * TIME_WAIT state, creating an ADDRINUSE error. 1137 * in6_pcbladdr() also handles scope zone IDs. 1138 */ 1139 error = in6_pcbladdr(inp, nam, &addr6); 1140 if (error) 1141 return error; 1142 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1143 &sin6->sin6_addr, sin6->sin6_port, 1144 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 1145 ? &addr6 1146 : &inp->in6p_laddr, 1147 inp->inp_lport, 0, NULL); 1148 if (oinp) 1149 return EADDRINUSE; 1150 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1151 inp->in6p_laddr = addr6; 1152 inp->in6p_faddr = sin6->sin6_addr; 1153 inp->inp_fport = sin6->sin6_port; 1154 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 1155 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; 1156 if (inp->inp_flags & IN6P_AUTOFLOWLABEL) 1157 inp->inp_flow |= 1158 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 1159 in_pcbrehash(inp); 1160 1161 /* Compute window scaling to request. */ 1162 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1163 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1164 tp->request_r_scale++; 1165 1166 soisconnecting(so); 1167 TCPSTAT_INC(tcps_connattempt); 1168 tp->t_state = TCPS_SYN_SENT; 1169 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit); 1170 tp->iss = tcp_new_isn(tp); 1171 tp->t_bw_rtseq = tp->iss; 1172 tcp_sendseqinit(tp); 1173 1174 return 0; 1175 } 1176 #endif /* INET6 */ 1177 1178 /* 1179 * Export TCP internal state information via a struct tcp_info, based on the 1180 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1181 * (TCP state machine, etc). We export all information using FreeBSD-native 1182 * constants -- for example, the numeric values for tcpi_state will differ 1183 * from Linux. 1184 */ 1185 static void 1186 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1187 { 1188 1189 INP_WLOCK_ASSERT(tp->t_inpcb); 1190 bzero(ti, sizeof(*ti)); 1191 1192 ti->tcpi_state = tp->t_state; 1193 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1194 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1195 if (tp->t_flags & TF_SACK_PERMIT) 1196 ti->tcpi_options |= TCPI_OPT_SACK; 1197 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1198 ti->tcpi_options |= TCPI_OPT_WSCALE; 1199 ti->tcpi_snd_wscale = tp->snd_scale; 1200 ti->tcpi_rcv_wscale = tp->rcv_scale; 1201 } 1202 1203 ti->tcpi_rto = tp->t_rxtcur * tick; 1204 ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick; 1205 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1206 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1207 1208 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1209 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1210 1211 /* 1212 * FreeBSD-specific extension fields for tcp_info. 1213 */ 1214 ti->tcpi_rcv_space = tp->rcv_wnd; 1215 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1216 ti->tcpi_snd_wnd = tp->snd_wnd; 1217 ti->tcpi_snd_bwnd = tp->snd_bwnd; 1218 ti->tcpi_snd_nxt = tp->snd_nxt; 1219 ti->tcpi_snd_mss = tp->t_maxseg; 1220 ti->tcpi_rcv_mss = tp->t_maxseg; 1221 if (tp->t_flags & TF_TOE) 1222 ti->tcpi_options |= TCPI_OPT_TOE; 1223 } 1224 1225 /* 1226 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1227 * socket option arguments. When it re-acquires the lock after the copy, it 1228 * has to revalidate that the connection is still valid for the socket 1229 * option. 1230 */ 1231 #define INP_WLOCK_RECHECK(inp) do { \ 1232 INP_WLOCK(inp); \ 1233 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1234 INP_WUNLOCK(inp); \ 1235 return (ECONNRESET); \ 1236 } \ 1237 tp = intotcpcb(inp); \ 1238 } while(0) 1239 1240 int 1241 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1242 { 1243 int error, opt, optval; 1244 struct inpcb *inp; 1245 struct tcpcb *tp; 1246 struct tcp_info ti; 1247 1248 error = 0; 1249 inp = sotoinpcb(so); 1250 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1251 INP_WLOCK(inp); 1252 if (sopt->sopt_level != IPPROTO_TCP) { 1253 #ifdef INET6 1254 if (inp->inp_vflag & INP_IPV6PROTO) { 1255 INP_WUNLOCK(inp); 1256 error = ip6_ctloutput(so, sopt); 1257 } else { 1258 #endif /* INET6 */ 1259 INP_WUNLOCK(inp); 1260 error = ip_ctloutput(so, sopt); 1261 #ifdef INET6 1262 } 1263 #endif 1264 return (error); 1265 } 1266 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1267 INP_WUNLOCK(inp); 1268 return (ECONNRESET); 1269 } 1270 1271 switch (sopt->sopt_dir) { 1272 case SOPT_SET: 1273 switch (sopt->sopt_name) { 1274 #ifdef TCP_SIGNATURE 1275 case TCP_MD5SIG: 1276 INP_WUNLOCK(inp); 1277 error = sooptcopyin(sopt, &optval, sizeof optval, 1278 sizeof optval); 1279 if (error) 1280 return (error); 1281 1282 INP_WLOCK_RECHECK(inp); 1283 if (optval > 0) 1284 tp->t_flags |= TF_SIGNATURE; 1285 else 1286 tp->t_flags &= ~TF_SIGNATURE; 1287 INP_WUNLOCK(inp); 1288 break; 1289 #endif /* TCP_SIGNATURE */ 1290 case TCP_NODELAY: 1291 case TCP_NOOPT: 1292 INP_WUNLOCK(inp); 1293 error = sooptcopyin(sopt, &optval, sizeof optval, 1294 sizeof optval); 1295 if (error) 1296 return (error); 1297 1298 INP_WLOCK_RECHECK(inp); 1299 switch (sopt->sopt_name) { 1300 case TCP_NODELAY: 1301 opt = TF_NODELAY; 1302 break; 1303 case TCP_NOOPT: 1304 opt = TF_NOOPT; 1305 break; 1306 default: 1307 opt = 0; /* dead code to fool gcc */ 1308 break; 1309 } 1310 1311 if (optval) 1312 tp->t_flags |= opt; 1313 else 1314 tp->t_flags &= ~opt; 1315 INP_WUNLOCK(inp); 1316 break; 1317 1318 case TCP_NOPUSH: 1319 INP_WUNLOCK(inp); 1320 error = sooptcopyin(sopt, &optval, sizeof optval, 1321 sizeof optval); 1322 if (error) 1323 return (error); 1324 1325 INP_WLOCK_RECHECK(inp); 1326 if (optval) 1327 tp->t_flags |= TF_NOPUSH; 1328 else { 1329 tp->t_flags &= ~TF_NOPUSH; 1330 error = tcp_output(tp); 1331 } 1332 INP_WUNLOCK(inp); 1333 break; 1334 1335 case TCP_MAXSEG: 1336 INP_WUNLOCK(inp); 1337 error = sooptcopyin(sopt, &optval, sizeof optval, 1338 sizeof optval); 1339 if (error) 1340 return (error); 1341 1342 INP_WLOCK_RECHECK(inp); 1343 if (optval > 0 && optval <= tp->t_maxseg && 1344 optval + 40 >= V_tcp_minmss) 1345 tp->t_maxseg = optval; 1346 else 1347 error = EINVAL; 1348 INP_WUNLOCK(inp); 1349 break; 1350 1351 case TCP_INFO: 1352 INP_WUNLOCK(inp); 1353 error = EINVAL; 1354 break; 1355 1356 default: 1357 INP_WUNLOCK(inp); 1358 error = ENOPROTOOPT; 1359 break; 1360 } 1361 break; 1362 1363 case SOPT_GET: 1364 tp = intotcpcb(inp); 1365 switch (sopt->sopt_name) { 1366 #ifdef TCP_SIGNATURE 1367 case TCP_MD5SIG: 1368 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1369 INP_WUNLOCK(inp); 1370 error = sooptcopyout(sopt, &optval, sizeof optval); 1371 break; 1372 #endif 1373 1374 case TCP_NODELAY: 1375 optval = tp->t_flags & TF_NODELAY; 1376 INP_WUNLOCK(inp); 1377 error = sooptcopyout(sopt, &optval, sizeof optval); 1378 break; 1379 case TCP_MAXSEG: 1380 optval = tp->t_maxseg; 1381 INP_WUNLOCK(inp); 1382 error = sooptcopyout(sopt, &optval, sizeof optval); 1383 break; 1384 case TCP_NOOPT: 1385 optval = tp->t_flags & TF_NOOPT; 1386 INP_WUNLOCK(inp); 1387 error = sooptcopyout(sopt, &optval, sizeof optval); 1388 break; 1389 case TCP_NOPUSH: 1390 optval = tp->t_flags & TF_NOPUSH; 1391 INP_WUNLOCK(inp); 1392 error = sooptcopyout(sopt, &optval, sizeof optval); 1393 break; 1394 case TCP_INFO: 1395 tcp_fill_info(tp, &ti); 1396 INP_WUNLOCK(inp); 1397 error = sooptcopyout(sopt, &ti, sizeof ti); 1398 break; 1399 default: 1400 INP_WUNLOCK(inp); 1401 error = ENOPROTOOPT; 1402 break; 1403 } 1404 break; 1405 } 1406 return (error); 1407 } 1408 #undef INP_WLOCK_RECHECK 1409 1410 /* 1411 * tcp_sendspace and tcp_recvspace are the default send and receive window 1412 * sizes, respectively. These are obsolescent (this information should 1413 * be set by the route). 1414 */ 1415 u_long tcp_sendspace = 1024*32; 1416 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1417 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1418 u_long tcp_recvspace = 1024*64; 1419 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1420 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1421 1422 /* 1423 * Attach TCP protocol to socket, allocating 1424 * internet protocol control block, tcp control block, 1425 * bufer space, and entering LISTEN state if to accept connections. 1426 */ 1427 static int 1428 tcp_attach(struct socket *so) 1429 { 1430 struct tcpcb *tp; 1431 struct inpcb *inp; 1432 int error; 1433 1434 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1435 error = soreserve(so, tcp_sendspace, tcp_recvspace); 1436 if (error) 1437 return (error); 1438 } 1439 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1440 so->so_snd.sb_flags |= SB_AUTOSIZE; 1441 INP_INFO_WLOCK(&V_tcbinfo); 1442 error = in_pcballoc(so, &V_tcbinfo); 1443 if (error) { 1444 INP_INFO_WUNLOCK(&V_tcbinfo); 1445 return (error); 1446 } 1447 inp = sotoinpcb(so); 1448 #ifdef INET6 1449 if (inp->inp_vflag & INP_IPV6PROTO) { 1450 inp->inp_vflag |= INP_IPV6; 1451 inp->in6p_hops = -1; /* use kernel default */ 1452 } 1453 else 1454 #endif 1455 inp->inp_vflag |= INP_IPV4; 1456 tp = tcp_newtcpcb(inp); 1457 if (tp == NULL) { 1458 in_pcbdetach(inp); 1459 in_pcbfree(inp); 1460 INP_INFO_WUNLOCK(&V_tcbinfo); 1461 return (ENOBUFS); 1462 } 1463 tp->t_state = TCPS_CLOSED; 1464 INP_WUNLOCK(inp); 1465 INP_INFO_WUNLOCK(&V_tcbinfo); 1466 return (0); 1467 } 1468 1469 /* 1470 * Initiate (or continue) disconnect. 1471 * If embryonic state, just send reset (once). 1472 * If in ``let data drain'' option and linger null, just drop. 1473 * Otherwise (hard), mark socket disconnecting and drop 1474 * current input data; switch states based on user close, and 1475 * send segment to peer (with FIN). 1476 */ 1477 static void 1478 tcp_disconnect(struct tcpcb *tp) 1479 { 1480 struct inpcb *inp = tp->t_inpcb; 1481 struct socket *so = inp->inp_socket; 1482 1483 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1484 INP_WLOCK_ASSERT(inp); 1485 1486 /* 1487 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1488 * socket is still open. 1489 */ 1490 if (tp->t_state < TCPS_ESTABLISHED) { 1491 tp = tcp_close(tp); 1492 KASSERT(tp != NULL, 1493 ("tcp_disconnect: tcp_close() returned NULL")); 1494 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1495 tp = tcp_drop(tp, 0); 1496 KASSERT(tp != NULL, 1497 ("tcp_disconnect: tcp_drop() returned NULL")); 1498 } else { 1499 soisdisconnecting(so); 1500 sbflush(&so->so_rcv); 1501 tcp_usrclosed(tp); 1502 if (!(inp->inp_flags & INP_DROPPED)) 1503 tcp_output_disconnect(tp); 1504 } 1505 } 1506 1507 /* 1508 * User issued close, and wish to trail through shutdown states: 1509 * if never received SYN, just forget it. If got a SYN from peer, 1510 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1511 * If already got a FIN from peer, then almost done; go to LAST_ACK 1512 * state. In all other cases, have already sent FIN to peer (e.g. 1513 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1514 * for peer to send FIN or not respond to keep-alives, etc. 1515 * We can let the user exit from the close as soon as the FIN is acked. 1516 */ 1517 static void 1518 tcp_usrclosed(struct tcpcb *tp) 1519 { 1520 1521 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1522 INP_WLOCK_ASSERT(tp->t_inpcb); 1523 1524 switch (tp->t_state) { 1525 case TCPS_LISTEN: 1526 tcp_offload_listen_close(tp); 1527 /* FALLTHROUGH */ 1528 case TCPS_CLOSED: 1529 tp->t_state = TCPS_CLOSED; 1530 tp = tcp_close(tp); 1531 /* 1532 * tcp_close() should never return NULL here as the socket is 1533 * still open. 1534 */ 1535 KASSERT(tp != NULL, 1536 ("tcp_usrclosed: tcp_close() returned NULL")); 1537 break; 1538 1539 case TCPS_SYN_SENT: 1540 case TCPS_SYN_RECEIVED: 1541 tp->t_flags |= TF_NEEDFIN; 1542 break; 1543 1544 case TCPS_ESTABLISHED: 1545 tp->t_state = TCPS_FIN_WAIT_1; 1546 break; 1547 1548 case TCPS_CLOSE_WAIT: 1549 tp->t_state = TCPS_LAST_ACK; 1550 break; 1551 } 1552 if (tp->t_state >= TCPS_FIN_WAIT_2) { 1553 soisdisconnected(tp->t_inpcb->inp_socket); 1554 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 1555 if (tp->t_state == TCPS_FIN_WAIT_2) { 1556 int timeout; 1557 1558 timeout = (tcp_fast_finwait2_recycle) ? 1559 tcp_finwait2_timeout : tcp_maxidle; 1560 tcp_timer_activate(tp, TT_2MSL, timeout); 1561 } 1562 } 1563 } 1564 1565 #ifdef DDB 1566 static void 1567 db_print_indent(int indent) 1568 { 1569 int i; 1570 1571 for (i = 0; i < indent; i++) 1572 db_printf(" "); 1573 } 1574 1575 static void 1576 db_print_tstate(int t_state) 1577 { 1578 1579 switch (t_state) { 1580 case TCPS_CLOSED: 1581 db_printf("TCPS_CLOSED"); 1582 return; 1583 1584 case TCPS_LISTEN: 1585 db_printf("TCPS_LISTEN"); 1586 return; 1587 1588 case TCPS_SYN_SENT: 1589 db_printf("TCPS_SYN_SENT"); 1590 return; 1591 1592 case TCPS_SYN_RECEIVED: 1593 db_printf("TCPS_SYN_RECEIVED"); 1594 return; 1595 1596 case TCPS_ESTABLISHED: 1597 db_printf("TCPS_ESTABLISHED"); 1598 return; 1599 1600 case TCPS_CLOSE_WAIT: 1601 db_printf("TCPS_CLOSE_WAIT"); 1602 return; 1603 1604 case TCPS_FIN_WAIT_1: 1605 db_printf("TCPS_FIN_WAIT_1"); 1606 return; 1607 1608 case TCPS_CLOSING: 1609 db_printf("TCPS_CLOSING"); 1610 return; 1611 1612 case TCPS_LAST_ACK: 1613 db_printf("TCPS_LAST_ACK"); 1614 return; 1615 1616 case TCPS_FIN_WAIT_2: 1617 db_printf("TCPS_FIN_WAIT_2"); 1618 return; 1619 1620 case TCPS_TIME_WAIT: 1621 db_printf("TCPS_TIME_WAIT"); 1622 return; 1623 1624 default: 1625 db_printf("unknown"); 1626 return; 1627 } 1628 } 1629 1630 static void 1631 db_print_tflags(u_int t_flags) 1632 { 1633 int comma; 1634 1635 comma = 0; 1636 if (t_flags & TF_ACKNOW) { 1637 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 1638 comma = 1; 1639 } 1640 if (t_flags & TF_DELACK) { 1641 db_printf("%sTF_DELACK", comma ? ", " : ""); 1642 comma = 1; 1643 } 1644 if (t_flags & TF_NODELAY) { 1645 db_printf("%sTF_NODELAY", comma ? ", " : ""); 1646 comma = 1; 1647 } 1648 if (t_flags & TF_NOOPT) { 1649 db_printf("%sTF_NOOPT", comma ? ", " : ""); 1650 comma = 1; 1651 } 1652 if (t_flags & TF_SENTFIN) { 1653 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 1654 comma = 1; 1655 } 1656 if (t_flags & TF_REQ_SCALE) { 1657 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 1658 comma = 1; 1659 } 1660 if (t_flags & TF_RCVD_SCALE) { 1661 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 1662 comma = 1; 1663 } 1664 if (t_flags & TF_REQ_TSTMP) { 1665 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 1666 comma = 1; 1667 } 1668 if (t_flags & TF_RCVD_TSTMP) { 1669 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 1670 comma = 1; 1671 } 1672 if (t_flags & TF_SACK_PERMIT) { 1673 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 1674 comma = 1; 1675 } 1676 if (t_flags & TF_NEEDSYN) { 1677 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 1678 comma = 1; 1679 } 1680 if (t_flags & TF_NEEDFIN) { 1681 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 1682 comma = 1; 1683 } 1684 if (t_flags & TF_NOPUSH) { 1685 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1686 comma = 1; 1687 } 1688 if (t_flags & TF_NOPUSH) { 1689 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 1690 comma = 1; 1691 } 1692 if (t_flags & TF_MORETOCOME) { 1693 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 1694 comma = 1; 1695 } 1696 if (t_flags & TF_LQ_OVERFLOW) { 1697 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 1698 comma = 1; 1699 } 1700 if (t_flags & TF_LASTIDLE) { 1701 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 1702 comma = 1; 1703 } 1704 if (t_flags & TF_RXWIN0SENT) { 1705 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 1706 comma = 1; 1707 } 1708 if (t_flags & TF_FASTRECOVERY) { 1709 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 1710 comma = 1; 1711 } 1712 if (t_flags & TF_WASFRECOVERY) { 1713 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 1714 comma = 1; 1715 } 1716 if (t_flags & TF_SIGNATURE) { 1717 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 1718 comma = 1; 1719 } 1720 if (t_flags & TF_FORCEDATA) { 1721 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 1722 comma = 1; 1723 } 1724 if (t_flags & TF_TSO) { 1725 db_printf("%sTF_TSO", comma ? ", " : ""); 1726 comma = 1; 1727 } 1728 if (t_flags & TF_ECN_PERMIT) { 1729 db_printf("%sTF_ECN_PERMIT", comma ? ", " : ""); 1730 comma = 1; 1731 } 1732 } 1733 1734 static void 1735 db_print_toobflags(char t_oobflags) 1736 { 1737 int comma; 1738 1739 comma = 0; 1740 if (t_oobflags & TCPOOB_HAVEDATA) { 1741 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 1742 comma = 1; 1743 } 1744 if (t_oobflags & TCPOOB_HADDATA) { 1745 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 1746 comma = 1; 1747 } 1748 } 1749 1750 static void 1751 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 1752 { 1753 1754 db_print_indent(indent); 1755 db_printf("%s at %p\n", name, tp); 1756 1757 indent += 2; 1758 1759 db_print_indent(indent); 1760 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 1761 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 1762 1763 db_print_indent(indent); 1764 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 1765 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 1766 1767 db_print_indent(indent); 1768 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 1769 &tp->t_timers->tt_delack, tp->t_inpcb); 1770 1771 db_print_indent(indent); 1772 db_printf("t_state: %d (", tp->t_state); 1773 db_print_tstate(tp->t_state); 1774 db_printf(")\n"); 1775 1776 db_print_indent(indent); 1777 db_printf("t_flags: 0x%x (", tp->t_flags); 1778 db_print_tflags(tp->t_flags); 1779 db_printf(")\n"); 1780 1781 db_print_indent(indent); 1782 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 1783 tp->snd_una, tp->snd_max, tp->snd_nxt); 1784 1785 db_print_indent(indent); 1786 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 1787 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 1788 1789 db_print_indent(indent); 1790 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 1791 tp->iss, tp->irs, tp->rcv_nxt); 1792 1793 db_print_indent(indent); 1794 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n", 1795 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 1796 1797 db_print_indent(indent); 1798 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n", 1799 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd); 1800 1801 db_print_indent(indent); 1802 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: " 1803 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth, 1804 tp->snd_recover); 1805 1806 db_print_indent(indent); 1807 db_printf("t_maxopd: %u t_rcvtime: %u t_startime: %u\n", 1808 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime); 1809 1810 db_print_indent(indent); 1811 db_printf("t_rttime: %u t_rtsq: 0x%08x t_bw_rtttime: %u\n", 1812 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime); 1813 1814 db_print_indent(indent); 1815 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u " 1816 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg, 1817 tp->t_srtt); 1818 1819 db_print_indent(indent); 1820 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 1821 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 1822 tp->t_rttbest); 1823 1824 db_print_indent(indent); 1825 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n", 1826 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 1827 1828 db_print_indent(indent); 1829 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 1830 db_print_toobflags(tp->t_oobflags); 1831 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 1832 1833 db_print_indent(indent); 1834 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 1835 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 1836 1837 db_print_indent(indent); 1838 db_printf("ts_recent: %u ts_recent_age: %u\n", 1839 tp->ts_recent, tp->ts_recent_age); 1840 1841 db_print_indent(indent); 1842 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 1843 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 1844 1845 db_print_indent(indent); 1846 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x " 1847 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 1848 tp->snd_recover_prev, tp->t_badrxtwin); 1849 1850 db_print_indent(indent); 1851 db_printf("snd_numholes: %d snd_holes first: %p\n", 1852 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 1853 1854 db_print_indent(indent); 1855 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 1856 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 1857 1858 /* Skip sackblks, sackhint. */ 1859 1860 db_print_indent(indent); 1861 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 1862 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 1863 } 1864 1865 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 1866 { 1867 struct tcpcb *tp; 1868 1869 if (!have_addr) { 1870 db_printf("usage: show tcpcb <addr>\n"); 1871 return; 1872 } 1873 tp = (struct tcpcb *)addr; 1874 1875 db_print_tcpcb(tp, "tcpcb", 0); 1876 } 1877 #endif 1878