1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_ipsec.h" 47 #include "opt_tcpdebug.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/limits.h> 52 #include <sys/malloc.h> 53 #include <sys/refcount.h> 54 #include <sys/kernel.h> 55 #include <sys/sysctl.h> 56 #include <sys/mbuf.h> 57 #ifdef INET6 58 #include <sys/domain.h> 59 #endif /* INET6 */ 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/protosw.h> 63 #include <sys/proc.h> 64 #include <sys/jail.h> 65 #include <sys/syslog.h> 66 67 #ifdef DDB 68 #include <ddb/ddb.h> 69 #endif 70 71 #include <net/if.h> 72 #include <net/if_var.h> 73 #include <net/route.h> 74 #include <net/vnet.h> 75 76 #include <netinet/in.h> 77 #include <netinet/in_kdtrace.h> 78 #include <netinet/in_pcb.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/in_var.h> 81 #include <netinet/ip_var.h> 82 #ifdef INET6 83 #include <netinet/ip6.h> 84 #include <netinet6/in6_pcb.h> 85 #include <netinet6/ip6_var.h> 86 #include <netinet6/scope6_var.h> 87 #endif 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/tcp_log_buf.h> 94 #include <netinet/tcpip.h> 95 #include <netinet/cc/cc.h> 96 #include <netinet/tcp_fastopen.h> 97 #ifdef TCPPCAP 98 #include <netinet/tcp_pcap.h> 99 #endif 100 #ifdef TCPDEBUG 101 #include <netinet/tcp_debug.h> 102 #endif 103 #ifdef TCP_OFFLOAD 104 #include <netinet/tcp_offload.h> 105 #endif 106 #include <netipsec/ipsec_support.h> 107 108 /* 109 * TCP protocol interface to socket abstraction. 110 */ 111 static int tcp_attach(struct socket *); 112 #ifdef INET 113 static int tcp_connect(struct tcpcb *, struct sockaddr *, 114 struct thread *td); 115 #endif /* INET */ 116 #ifdef INET6 117 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 118 struct thread *td); 119 #endif /* INET6 */ 120 static void tcp_disconnect(struct tcpcb *); 121 static void tcp_usrclosed(struct tcpcb *); 122 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 123 124 #ifdef TCPDEBUG 125 #define TCPDEBUG0 int ostate = 0 126 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 127 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 128 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 129 #else 130 #define TCPDEBUG0 131 #define TCPDEBUG1() 132 #define TCPDEBUG2(req) 133 #endif 134 135 /* 136 * TCP attaches to socket via pru_attach(), reserving space, 137 * and an internet control block. 138 */ 139 static int 140 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 141 { 142 struct inpcb *inp; 143 struct tcpcb *tp = NULL; 144 int error; 145 TCPDEBUG0; 146 147 inp = sotoinpcb(so); 148 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 149 TCPDEBUG1(); 150 151 error = tcp_attach(so); 152 if (error) 153 goto out; 154 155 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 156 so->so_linger = TCP_LINGERTIME; 157 158 inp = sotoinpcb(so); 159 tp = intotcpcb(inp); 160 out: 161 TCPDEBUG2(PRU_ATTACH); 162 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 163 return error; 164 } 165 166 /* 167 * tcp_detach is called when the socket layer loses its final reference 168 * to the socket, be it a file descriptor reference, a reference from TCP, 169 * etc. At this point, there is only one case in which we will keep around 170 * inpcb state: time wait. 171 * 172 * This function can probably be re-absorbed back into tcp_usr_detach() now 173 * that there is a single detach path. 174 */ 175 static void 176 tcp_detach(struct socket *so, struct inpcb *inp) 177 { 178 struct tcpcb *tp; 179 180 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 181 INP_WLOCK_ASSERT(inp); 182 183 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 184 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 185 186 tp = intotcpcb(inp); 187 188 if (inp->inp_flags & INP_TIMEWAIT) { 189 /* 190 * There are two cases to handle: one in which the time wait 191 * state is being discarded (INP_DROPPED), and one in which 192 * this connection will remain in timewait. In the former, 193 * it is time to discard all state (except tcptw, which has 194 * already been discarded by the timewait close code, which 195 * should be further up the call stack somewhere). In the 196 * latter case, we detach from the socket, but leave the pcb 197 * present until timewait ends. 198 * 199 * XXXRW: Would it be cleaner to free the tcptw here? 200 * 201 * Astute question indeed, from twtcp perspective there are 202 * four cases to consider: 203 * 204 * #1 tcp_detach is called at tcptw creation time by 205 * tcp_twstart, then do not discard the newly created tcptw 206 * and leave inpcb present until timewait ends 207 * #2 tcp_detach is called at tcptw creation time by 208 * tcp_twstart, but connection is local and tw will be 209 * discarded immediately 210 * #3 tcp_detach is called at timewait end (or reuse) by 211 * tcp_twclose, then the tcptw has already been discarded 212 * (or reused) and inpcb is freed here 213 * #4 tcp_detach is called() after timewait ends (or reuse) 214 * (e.g. by soclose), then tcptw has already been discarded 215 * (or reused) and inpcb is freed here 216 * 217 * In all three cases the tcptw should not be freed here. 218 */ 219 if (inp->inp_flags & INP_DROPPED) { 220 in_pcbdetach(inp); 221 if (__predict_true(tp == NULL)) { 222 in_pcbfree(inp); 223 } else { 224 /* 225 * This case should not happen as in TIMEWAIT 226 * state the inp should not be destroyed before 227 * its tcptw. If INVARIANTS is defined, panic. 228 */ 229 #ifdef INVARIANTS 230 panic("%s: Panic before an inp double-free: " 231 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 232 , __func__); 233 #else 234 log(LOG_ERR, "%s: Avoid an inp double-free: " 235 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 236 , __func__); 237 #endif 238 INP_WUNLOCK(inp); 239 } 240 } else { 241 in_pcbdetach(inp); 242 INP_WUNLOCK(inp); 243 } 244 } else { 245 /* 246 * If the connection is not in timewait, we consider two 247 * two conditions: one in which no further processing is 248 * necessary (dropped || embryonic), and one in which TCP is 249 * not yet done, but no longer requires the socket, so the 250 * pcb will persist for the time being. 251 * 252 * XXXRW: Does the second case still occur? 253 */ 254 if (inp->inp_flags & INP_DROPPED || 255 tp->t_state < TCPS_SYN_SENT) { 256 tcp_discardcb(tp); 257 in_pcbdetach(inp); 258 in_pcbfree(inp); 259 } else { 260 in_pcbdetach(inp); 261 INP_WUNLOCK(inp); 262 } 263 } 264 } 265 266 /* 267 * pru_detach() detaches the TCP protocol from the socket. 268 * If the protocol state is non-embryonic, then can't 269 * do this directly: have to initiate a pru_disconnect(), 270 * which may finish later; embryonic TCB's can just 271 * be discarded here. 272 */ 273 static void 274 tcp_usr_detach(struct socket *so) 275 { 276 struct inpcb *inp; 277 int rlock = 0; 278 279 inp = sotoinpcb(so); 280 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 281 if (!INP_INFO_WLOCKED(&V_tcbinfo)) { 282 INP_INFO_RLOCK(&V_tcbinfo); 283 rlock = 1; 284 } 285 INP_WLOCK(inp); 286 KASSERT(inp->inp_socket != NULL, 287 ("tcp_usr_detach: inp_socket == NULL")); 288 tcp_detach(so, inp); 289 if (rlock) 290 INP_INFO_RUNLOCK(&V_tcbinfo); 291 } 292 293 #ifdef INET 294 /* 295 * Give the socket an address. 296 */ 297 static int 298 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 299 { 300 int error = 0; 301 struct inpcb *inp; 302 struct tcpcb *tp = NULL; 303 struct sockaddr_in *sinp; 304 305 sinp = (struct sockaddr_in *)nam; 306 if (nam->sa_len != sizeof (*sinp)) 307 return (EINVAL); 308 /* 309 * Must check for multicast addresses and disallow binding 310 * to them. 311 */ 312 if (sinp->sin_family == AF_INET && 313 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 314 return (EAFNOSUPPORT); 315 316 TCPDEBUG0; 317 inp = sotoinpcb(so); 318 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 319 INP_WLOCK(inp); 320 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 321 error = EINVAL; 322 goto out; 323 } 324 tp = intotcpcb(inp); 325 TCPDEBUG1(); 326 INP_HASH_WLOCK(&V_tcbinfo); 327 error = in_pcbbind(inp, nam, td->td_ucred); 328 INP_HASH_WUNLOCK(&V_tcbinfo); 329 out: 330 TCPDEBUG2(PRU_BIND); 331 TCP_PROBE2(debug__user, tp, PRU_BIND); 332 INP_WUNLOCK(inp); 333 334 return (error); 335 } 336 #endif /* INET */ 337 338 #ifdef INET6 339 static int 340 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 341 { 342 int error = 0; 343 struct inpcb *inp; 344 struct tcpcb *tp = NULL; 345 struct sockaddr_in6 *sin6p; 346 347 sin6p = (struct sockaddr_in6 *)nam; 348 if (nam->sa_len != sizeof (*sin6p)) 349 return (EINVAL); 350 /* 351 * Must check for multicast addresses and disallow binding 352 * to them. 353 */ 354 if (sin6p->sin6_family == AF_INET6 && 355 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 356 return (EAFNOSUPPORT); 357 358 TCPDEBUG0; 359 inp = sotoinpcb(so); 360 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 361 INP_WLOCK(inp); 362 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 363 error = EINVAL; 364 goto out; 365 } 366 tp = intotcpcb(inp); 367 TCPDEBUG1(); 368 INP_HASH_WLOCK(&V_tcbinfo); 369 inp->inp_vflag &= ~INP_IPV4; 370 inp->inp_vflag |= INP_IPV6; 371 #ifdef INET 372 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 373 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 374 inp->inp_vflag |= INP_IPV4; 375 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 376 struct sockaddr_in sin; 377 378 in6_sin6_2_sin(&sin, sin6p); 379 inp->inp_vflag |= INP_IPV4; 380 inp->inp_vflag &= ~INP_IPV6; 381 error = in_pcbbind(inp, (struct sockaddr *)&sin, 382 td->td_ucred); 383 INP_HASH_WUNLOCK(&V_tcbinfo); 384 goto out; 385 } 386 } 387 #endif 388 error = in6_pcbbind(inp, nam, td->td_ucred); 389 INP_HASH_WUNLOCK(&V_tcbinfo); 390 out: 391 TCPDEBUG2(PRU_BIND); 392 TCP_PROBE2(debug__user, tp, PRU_BIND); 393 INP_WUNLOCK(inp); 394 return (error); 395 } 396 #endif /* INET6 */ 397 398 #ifdef INET 399 /* 400 * Prepare to accept connections. 401 */ 402 static int 403 tcp_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 = sotoinpcb(so); 411 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 412 INP_WLOCK(inp); 413 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 414 error = EINVAL; 415 goto out; 416 } 417 tp = intotcpcb(inp); 418 TCPDEBUG1(); 419 SOCK_LOCK(so); 420 error = solisten_proto_check(so); 421 INP_HASH_WLOCK(&V_tcbinfo); 422 if (error == 0 && inp->inp_lport == 0) 423 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 424 INP_HASH_WUNLOCK(&V_tcbinfo); 425 if (error == 0) { 426 tcp_state_change(tp, TCPS_LISTEN); 427 solisten_proto(so, backlog); 428 #ifdef TCP_OFFLOAD 429 if ((so->so_options & SO_NO_OFFLOAD) == 0) 430 tcp_offload_listen_start(tp); 431 #endif 432 } 433 SOCK_UNLOCK(so); 434 435 if (IS_FASTOPEN(tp->t_flags)) 436 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 437 438 out: 439 TCPDEBUG2(PRU_LISTEN); 440 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 441 INP_WUNLOCK(inp); 442 return (error); 443 } 444 #endif /* INET */ 445 446 #ifdef INET6 447 static int 448 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 449 { 450 int error = 0; 451 struct inpcb *inp; 452 struct tcpcb *tp = NULL; 453 454 TCPDEBUG0; 455 inp = sotoinpcb(so); 456 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 457 INP_WLOCK(inp); 458 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 459 error = EINVAL; 460 goto out; 461 } 462 tp = intotcpcb(inp); 463 TCPDEBUG1(); 464 SOCK_LOCK(so); 465 error = solisten_proto_check(so); 466 INP_HASH_WLOCK(&V_tcbinfo); 467 if (error == 0 && inp->inp_lport == 0) { 468 inp->inp_vflag &= ~INP_IPV4; 469 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 470 inp->inp_vflag |= INP_IPV4; 471 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 472 } 473 INP_HASH_WUNLOCK(&V_tcbinfo); 474 if (error == 0) { 475 tcp_state_change(tp, TCPS_LISTEN); 476 solisten_proto(so, backlog); 477 #ifdef TCP_OFFLOAD 478 if ((so->so_options & SO_NO_OFFLOAD) == 0) 479 tcp_offload_listen_start(tp); 480 #endif 481 } 482 SOCK_UNLOCK(so); 483 484 if (IS_FASTOPEN(tp->t_flags)) 485 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 486 487 out: 488 TCPDEBUG2(PRU_LISTEN); 489 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 490 INP_WUNLOCK(inp); 491 return (error); 492 } 493 #endif /* INET6 */ 494 495 #ifdef INET 496 /* 497 * Initiate connection to peer. 498 * Create a template for use in transmissions on this connection. 499 * Enter SYN_SENT state, and mark socket as connecting. 500 * Start keep-alive timer, and seed output sequence space. 501 * Send initial segment on connection. 502 */ 503 static int 504 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 505 { 506 int error = 0; 507 struct inpcb *inp; 508 struct tcpcb *tp = NULL; 509 struct sockaddr_in *sinp; 510 511 sinp = (struct sockaddr_in *)nam; 512 if (nam->sa_len != sizeof (*sinp)) 513 return (EINVAL); 514 /* 515 * Must disallow TCP ``connections'' to multicast addresses. 516 */ 517 if (sinp->sin_family == AF_INET 518 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 519 return (EAFNOSUPPORT); 520 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 521 return (error); 522 523 TCPDEBUG0; 524 inp = sotoinpcb(so); 525 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 526 INP_WLOCK(inp); 527 if (inp->inp_flags & INP_TIMEWAIT) { 528 error = EADDRINUSE; 529 goto out; 530 } 531 if (inp->inp_flags & INP_DROPPED) { 532 error = ECONNREFUSED; 533 goto out; 534 } 535 tp = intotcpcb(inp); 536 TCPDEBUG1(); 537 if ((error = tcp_connect(tp, nam, td)) != 0) 538 goto out; 539 #ifdef TCP_OFFLOAD 540 if (registered_toedevs > 0 && 541 (so->so_options & SO_NO_OFFLOAD) == 0 && 542 (error = tcp_offload_connect(so, nam)) == 0) 543 goto out; 544 #endif 545 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 546 error = tp->t_fb->tfb_tcp_output(tp); 547 out: 548 TCPDEBUG2(PRU_CONNECT); 549 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 550 INP_WUNLOCK(inp); 551 return (error); 552 } 553 #endif /* INET */ 554 555 #ifdef INET6 556 static int 557 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 558 { 559 int error = 0; 560 struct inpcb *inp; 561 struct tcpcb *tp = NULL; 562 struct sockaddr_in6 *sin6p; 563 564 TCPDEBUG0; 565 566 sin6p = (struct sockaddr_in6 *)nam; 567 if (nam->sa_len != sizeof (*sin6p)) 568 return (EINVAL); 569 /* 570 * Must disallow TCP ``connections'' to multicast addresses. 571 */ 572 if (sin6p->sin6_family == AF_INET6 573 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 574 return (EAFNOSUPPORT); 575 576 inp = sotoinpcb(so); 577 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 578 INP_WLOCK(inp); 579 if (inp->inp_flags & INP_TIMEWAIT) { 580 error = EADDRINUSE; 581 goto out; 582 } 583 if (inp->inp_flags & INP_DROPPED) { 584 error = ECONNREFUSED; 585 goto out; 586 } 587 tp = intotcpcb(inp); 588 TCPDEBUG1(); 589 #ifdef INET 590 /* 591 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 592 * therefore probably require the hash lock, which isn't held here. 593 * Is this a significant problem? 594 */ 595 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 596 struct sockaddr_in sin; 597 598 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 599 error = EINVAL; 600 goto out; 601 } 602 if ((inp->inp_vflag & INP_IPV4) == 0) { 603 error = EAFNOSUPPORT; 604 goto out; 605 } 606 607 in6_sin6_2_sin(&sin, sin6p); 608 inp->inp_vflag |= INP_IPV4; 609 inp->inp_vflag &= ~INP_IPV6; 610 if ((error = prison_remote_ip4(td->td_ucred, 611 &sin.sin_addr)) != 0) 612 goto out; 613 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 614 goto out; 615 #ifdef TCP_OFFLOAD 616 if (registered_toedevs > 0 && 617 (so->so_options & SO_NO_OFFLOAD) == 0 && 618 (error = tcp_offload_connect(so, nam)) == 0) 619 goto out; 620 #endif 621 error = tp->t_fb->tfb_tcp_output(tp); 622 goto out; 623 } else { 624 if ((inp->inp_vflag & INP_IPV6) == 0) { 625 error = EAFNOSUPPORT; 626 goto out; 627 } 628 } 629 #endif 630 inp->inp_vflag &= ~INP_IPV4; 631 inp->inp_vflag |= INP_IPV6; 632 inp->inp_inc.inc_flags |= INC_ISIPV6; 633 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0) 634 goto out; 635 if ((error = tcp6_connect(tp, nam, td)) != 0) 636 goto out; 637 #ifdef TCP_OFFLOAD 638 if (registered_toedevs > 0 && 639 (so->so_options & SO_NO_OFFLOAD) == 0 && 640 (error = tcp_offload_connect(so, nam)) == 0) 641 goto out; 642 #endif 643 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 644 error = tp->t_fb->tfb_tcp_output(tp); 645 646 out: 647 TCPDEBUG2(PRU_CONNECT); 648 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 649 INP_WUNLOCK(inp); 650 return (error); 651 } 652 #endif /* INET6 */ 653 654 /* 655 * Initiate disconnect from peer. 656 * If connection never passed embryonic stage, just drop; 657 * else if don't need to let data drain, then can just drop anyways, 658 * else have to begin TCP shutdown process: mark socket disconnecting, 659 * drain unread data, state switch to reflect user close, and 660 * send segment (e.g. FIN) to peer. Socket will be really disconnected 661 * when peer sends FIN and acks ours. 662 * 663 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 664 */ 665 static int 666 tcp_usr_disconnect(struct socket *so) 667 { 668 struct inpcb *inp; 669 struct tcpcb *tp = NULL; 670 int error = 0; 671 672 TCPDEBUG0; 673 INP_INFO_RLOCK(&V_tcbinfo); 674 inp = sotoinpcb(so); 675 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 676 INP_WLOCK(inp); 677 if (inp->inp_flags & INP_TIMEWAIT) 678 goto out; 679 if (inp->inp_flags & INP_DROPPED) { 680 error = ECONNRESET; 681 goto out; 682 } 683 tp = intotcpcb(inp); 684 TCPDEBUG1(); 685 tcp_disconnect(tp); 686 out: 687 TCPDEBUG2(PRU_DISCONNECT); 688 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 689 INP_WUNLOCK(inp); 690 INP_INFO_RUNLOCK(&V_tcbinfo); 691 return (error); 692 } 693 694 #ifdef INET 695 /* 696 * Accept a connection. Essentially all the work is done at higher levels; 697 * just return the address of the peer, storing through addr. 698 */ 699 static int 700 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 701 { 702 int error = 0; 703 struct inpcb *inp = NULL; 704 struct tcpcb *tp = NULL; 705 struct in_addr addr; 706 in_port_t port = 0; 707 TCPDEBUG0; 708 709 if (so->so_state & SS_ISDISCONNECTED) 710 return (ECONNABORTED); 711 712 inp = sotoinpcb(so); 713 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 714 INP_WLOCK(inp); 715 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 716 error = ECONNABORTED; 717 goto out; 718 } 719 tp = intotcpcb(inp); 720 TCPDEBUG1(); 721 722 /* 723 * We inline in_getpeeraddr and COMMON_END here, so that we can 724 * copy the data of interest and defer the malloc until after we 725 * release the lock. 726 */ 727 port = inp->inp_fport; 728 addr = inp->inp_faddr; 729 730 out: 731 TCPDEBUG2(PRU_ACCEPT); 732 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 733 INP_WUNLOCK(inp); 734 if (error == 0) 735 *nam = in_sockaddr(port, &addr); 736 return error; 737 } 738 #endif /* INET */ 739 740 #ifdef INET6 741 static int 742 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 743 { 744 struct inpcb *inp = NULL; 745 int error = 0; 746 struct tcpcb *tp = NULL; 747 struct in_addr addr; 748 struct in6_addr addr6; 749 in_port_t port = 0; 750 int v4 = 0; 751 TCPDEBUG0; 752 753 if (so->so_state & SS_ISDISCONNECTED) 754 return (ECONNABORTED); 755 756 inp = sotoinpcb(so); 757 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 758 INP_INFO_RLOCK(&V_tcbinfo); 759 INP_WLOCK(inp); 760 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 761 error = ECONNABORTED; 762 goto out; 763 } 764 tp = intotcpcb(inp); 765 TCPDEBUG1(); 766 767 /* 768 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 769 * copy the data of interest and defer the malloc until after we 770 * release the lock. 771 */ 772 if (inp->inp_vflag & INP_IPV4) { 773 v4 = 1; 774 port = inp->inp_fport; 775 addr = inp->inp_faddr; 776 } else { 777 port = inp->inp_fport; 778 addr6 = inp->in6p_faddr; 779 } 780 781 out: 782 TCPDEBUG2(PRU_ACCEPT); 783 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 784 INP_WUNLOCK(inp); 785 INP_INFO_RUNLOCK(&V_tcbinfo); 786 if (error == 0) { 787 if (v4) 788 *nam = in6_v4mapsin6_sockaddr(port, &addr); 789 else 790 *nam = in6_sockaddr(port, &addr6); 791 } 792 return error; 793 } 794 #endif /* INET6 */ 795 796 /* 797 * Mark the connection as being incapable of further output. 798 */ 799 static int 800 tcp_usr_shutdown(struct socket *so) 801 { 802 int error = 0; 803 struct inpcb *inp; 804 struct tcpcb *tp = NULL; 805 806 TCPDEBUG0; 807 INP_INFO_RLOCK(&V_tcbinfo); 808 inp = sotoinpcb(so); 809 KASSERT(inp != NULL, ("inp == NULL")); 810 INP_WLOCK(inp); 811 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 812 error = ECONNRESET; 813 goto out; 814 } 815 tp = intotcpcb(inp); 816 TCPDEBUG1(); 817 socantsendmore(so); 818 tcp_usrclosed(tp); 819 if (!(inp->inp_flags & INP_DROPPED)) 820 error = tp->t_fb->tfb_tcp_output(tp); 821 822 out: 823 TCPDEBUG2(PRU_SHUTDOWN); 824 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 825 INP_WUNLOCK(inp); 826 INP_INFO_RUNLOCK(&V_tcbinfo); 827 828 return (error); 829 } 830 831 /* 832 * After a receive, possibly send window update to peer. 833 */ 834 static int 835 tcp_usr_rcvd(struct socket *so, int flags) 836 { 837 struct inpcb *inp; 838 struct tcpcb *tp = NULL; 839 int error = 0; 840 841 TCPDEBUG0; 842 inp = sotoinpcb(so); 843 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 844 INP_WLOCK(inp); 845 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 846 error = ECONNRESET; 847 goto out; 848 } 849 tp = intotcpcb(inp); 850 TCPDEBUG1(); 851 /* 852 * For passively-created TFO connections, don't attempt a window 853 * update while still in SYN_RECEIVED as this may trigger an early 854 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 855 * application response data, or failing that, when the DELACK timer 856 * expires. 857 */ 858 if (IS_FASTOPEN(tp->t_flags) && 859 (tp->t_state == TCPS_SYN_RECEIVED)) 860 goto out; 861 #ifdef TCP_OFFLOAD 862 if (tp->t_flags & TF_TOE) 863 tcp_offload_rcvd(tp); 864 else 865 #endif 866 tp->t_fb->tfb_tcp_output(tp); 867 868 out: 869 TCPDEBUG2(PRU_RCVD); 870 TCP_PROBE2(debug__user, tp, PRU_RCVD); 871 INP_WUNLOCK(inp); 872 return (error); 873 } 874 875 /* 876 * Do a send by putting data in output queue and updating urgent 877 * marker if URG set. Possibly send more data. Unlike the other 878 * pru_*() routines, the mbuf chains are our responsibility. We 879 * must either enqueue them or free them. The other pru_* routines 880 * generally are caller-frees. 881 */ 882 static int 883 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 884 struct sockaddr *nam, struct mbuf *control, struct thread *td) 885 { 886 int error = 0; 887 struct inpcb *inp; 888 struct tcpcb *tp = NULL; 889 #ifdef INET6 890 int isipv6; 891 #endif 892 TCPDEBUG0; 893 894 /* 895 * We require the pcbinfo lock if we will close the socket as part of 896 * this call. 897 */ 898 if (flags & PRUS_EOF) 899 INP_INFO_RLOCK(&V_tcbinfo); 900 inp = sotoinpcb(so); 901 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 902 INP_WLOCK(inp); 903 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 904 if (control) 905 m_freem(control); 906 /* 907 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible 908 * for freeing memory. 909 */ 910 if (m && (flags & PRUS_NOTREADY) == 0) 911 m_freem(m); 912 error = ECONNRESET; 913 goto out; 914 } 915 #ifdef INET6 916 isipv6 = nam && nam->sa_family == AF_INET6; 917 #endif /* INET6 */ 918 tp = intotcpcb(inp); 919 TCPDEBUG1(); 920 if (control) { 921 /* TCP doesn't do control messages (rights, creds, etc) */ 922 if (control->m_len) { 923 m_freem(control); 924 if (m) 925 m_freem(m); 926 error = EINVAL; 927 goto out; 928 } 929 m_freem(control); /* empty control, just free it */ 930 } 931 if (!(flags & PRUS_OOB)) { 932 sbappendstream(&so->so_snd, m, flags); 933 if (nam && tp->t_state < TCPS_SYN_SENT) { 934 /* 935 * Do implied connect if not yet connected, 936 * initialize window to default value, and 937 * initialize maxseg using peer's cached MSS. 938 */ 939 #ifdef INET6 940 if (isipv6) 941 error = tcp6_connect(tp, nam, td); 942 #endif /* INET6 */ 943 #if defined(INET6) && defined(INET) 944 else 945 #endif 946 #ifdef INET 947 error = tcp_connect(tp, nam, td); 948 #endif 949 if (error) 950 goto out; 951 if (IS_FASTOPEN(tp->t_flags)) 952 tcp_fastopen_connect(tp); 953 else { 954 tp->snd_wnd = TTCP_CLIENT_SND_WND; 955 tcp_mss(tp, -1); 956 } 957 } 958 if (flags & PRUS_EOF) { 959 /* 960 * Close the send side of the connection after 961 * the data is sent. 962 */ 963 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 964 socantsendmore(so); 965 tcp_usrclosed(tp); 966 } 967 if (!(inp->inp_flags & INP_DROPPED) && 968 !(flags & PRUS_NOTREADY)) { 969 if (flags & PRUS_MORETOCOME) 970 tp->t_flags |= TF_MORETOCOME; 971 error = tp->t_fb->tfb_tcp_output(tp); 972 if (flags & PRUS_MORETOCOME) 973 tp->t_flags &= ~TF_MORETOCOME; 974 } 975 } else { 976 /* 977 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 978 */ 979 SOCKBUF_LOCK(&so->so_snd); 980 if (sbspace(&so->so_snd) < -512) { 981 SOCKBUF_UNLOCK(&so->so_snd); 982 m_freem(m); 983 error = ENOBUFS; 984 goto out; 985 } 986 /* 987 * According to RFC961 (Assigned Protocols), 988 * the urgent pointer points to the last octet 989 * of urgent data. We continue, however, 990 * to consider it to indicate the first octet 991 * of data past the urgent section. 992 * Otherwise, snd_up should be one lower. 993 */ 994 sbappendstream_locked(&so->so_snd, m, flags); 995 SOCKBUF_UNLOCK(&so->so_snd); 996 if (nam && tp->t_state < TCPS_SYN_SENT) { 997 /* 998 * Do implied connect if not yet connected, 999 * initialize window to default value, and 1000 * initialize maxseg using peer's cached MSS. 1001 */ 1002 1003 /* 1004 * Not going to contemplate SYN|URG 1005 */ 1006 if (IS_FASTOPEN(tp->t_flags)) 1007 tp->t_flags &= ~TF_FASTOPEN; 1008 #ifdef INET6 1009 if (isipv6) 1010 error = tcp6_connect(tp, nam, td); 1011 #endif /* INET6 */ 1012 #if defined(INET6) && defined(INET) 1013 else 1014 #endif 1015 #ifdef INET 1016 error = tcp_connect(tp, nam, td); 1017 #endif 1018 if (error) 1019 goto out; 1020 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1021 tcp_mss(tp, -1); 1022 } 1023 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1024 if (!(flags & PRUS_NOTREADY)) { 1025 tp->t_flags |= TF_FORCEDATA; 1026 error = tp->t_fb->tfb_tcp_output(tp); 1027 tp->t_flags &= ~TF_FORCEDATA; 1028 } 1029 } 1030 TCP_LOG_EVENT(tp, NULL, 1031 &inp->inp_socket->so_rcv, 1032 &inp->inp_socket->so_snd, 1033 TCP_LOG_USERSEND, error, 1034 0, NULL, false); 1035 out: 1036 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1037 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1038 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1039 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1040 INP_WUNLOCK(inp); 1041 if (flags & PRUS_EOF) 1042 INP_INFO_RUNLOCK(&V_tcbinfo); 1043 return (error); 1044 } 1045 1046 static int 1047 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1048 { 1049 struct inpcb *inp; 1050 struct tcpcb *tp; 1051 int error; 1052 1053 inp = sotoinpcb(so); 1054 INP_WLOCK(inp); 1055 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1056 INP_WUNLOCK(inp); 1057 for (int i = 0; i < count; i++) 1058 m = m_free(m); 1059 return (ECONNRESET); 1060 } 1061 tp = intotcpcb(inp); 1062 1063 SOCKBUF_LOCK(&so->so_snd); 1064 error = sbready(&so->so_snd, m, count); 1065 SOCKBUF_UNLOCK(&so->so_snd); 1066 if (error == 0) 1067 error = tp->t_fb->tfb_tcp_output(tp); 1068 INP_WUNLOCK(inp); 1069 1070 return (error); 1071 } 1072 1073 /* 1074 * Abort the TCP. Drop the connection abruptly. 1075 */ 1076 static void 1077 tcp_usr_abort(struct socket *so) 1078 { 1079 struct inpcb *inp; 1080 struct tcpcb *tp = NULL; 1081 TCPDEBUG0; 1082 1083 inp = sotoinpcb(so); 1084 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1085 1086 INP_INFO_RLOCK(&V_tcbinfo); 1087 INP_WLOCK(inp); 1088 KASSERT(inp->inp_socket != NULL, 1089 ("tcp_usr_abort: inp_socket == NULL")); 1090 1091 /* 1092 * If we still have full TCP state, and we're not dropped, drop. 1093 */ 1094 if (!(inp->inp_flags & INP_TIMEWAIT) && 1095 !(inp->inp_flags & INP_DROPPED)) { 1096 tp = intotcpcb(inp); 1097 TCPDEBUG1(); 1098 tcp_drop(tp, ECONNABORTED); 1099 TCPDEBUG2(PRU_ABORT); 1100 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1101 } 1102 if (!(inp->inp_flags & INP_DROPPED)) { 1103 SOCK_LOCK(so); 1104 so->so_state |= SS_PROTOREF; 1105 SOCK_UNLOCK(so); 1106 inp->inp_flags |= INP_SOCKREF; 1107 } 1108 INP_WUNLOCK(inp); 1109 INP_INFO_RUNLOCK(&V_tcbinfo); 1110 } 1111 1112 /* 1113 * TCP socket is closed. Start friendly disconnect. 1114 */ 1115 static void 1116 tcp_usr_close(struct socket *so) 1117 { 1118 struct inpcb *inp; 1119 struct tcpcb *tp = NULL; 1120 TCPDEBUG0; 1121 1122 inp = sotoinpcb(so); 1123 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1124 1125 INP_INFO_RLOCK(&V_tcbinfo); 1126 INP_WLOCK(inp); 1127 KASSERT(inp->inp_socket != NULL, 1128 ("tcp_usr_close: inp_socket == NULL")); 1129 1130 /* 1131 * If we still have full TCP state, and we're not dropped, initiate 1132 * a disconnect. 1133 */ 1134 if (!(inp->inp_flags & INP_TIMEWAIT) && 1135 !(inp->inp_flags & INP_DROPPED)) { 1136 tp = intotcpcb(inp); 1137 TCPDEBUG1(); 1138 tcp_disconnect(tp); 1139 TCPDEBUG2(PRU_CLOSE); 1140 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1141 } 1142 if (!(inp->inp_flags & INP_DROPPED)) { 1143 SOCK_LOCK(so); 1144 so->so_state |= SS_PROTOREF; 1145 SOCK_UNLOCK(so); 1146 inp->inp_flags |= INP_SOCKREF; 1147 } 1148 INP_WUNLOCK(inp); 1149 INP_INFO_RUNLOCK(&V_tcbinfo); 1150 } 1151 1152 /* 1153 * Receive out-of-band data. 1154 */ 1155 static int 1156 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1157 { 1158 int error = 0; 1159 struct inpcb *inp; 1160 struct tcpcb *tp = NULL; 1161 1162 TCPDEBUG0; 1163 inp = sotoinpcb(so); 1164 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1165 INP_WLOCK(inp); 1166 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1167 error = ECONNRESET; 1168 goto out; 1169 } 1170 tp = intotcpcb(inp); 1171 TCPDEBUG1(); 1172 if ((so->so_oobmark == 0 && 1173 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1174 so->so_options & SO_OOBINLINE || 1175 tp->t_oobflags & TCPOOB_HADDATA) { 1176 error = EINVAL; 1177 goto out; 1178 } 1179 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1180 error = EWOULDBLOCK; 1181 goto out; 1182 } 1183 m->m_len = 1; 1184 *mtod(m, caddr_t) = tp->t_iobc; 1185 if ((flags & MSG_PEEK) == 0) 1186 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1187 1188 out: 1189 TCPDEBUG2(PRU_RCVOOB); 1190 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1191 INP_WUNLOCK(inp); 1192 return (error); 1193 } 1194 1195 #ifdef INET 1196 struct pr_usrreqs tcp_usrreqs = { 1197 .pru_abort = tcp_usr_abort, 1198 .pru_accept = tcp_usr_accept, 1199 .pru_attach = tcp_usr_attach, 1200 .pru_bind = tcp_usr_bind, 1201 .pru_connect = tcp_usr_connect, 1202 .pru_control = in_control, 1203 .pru_detach = tcp_usr_detach, 1204 .pru_disconnect = tcp_usr_disconnect, 1205 .pru_listen = tcp_usr_listen, 1206 .pru_peeraddr = in_getpeeraddr, 1207 .pru_rcvd = tcp_usr_rcvd, 1208 .pru_rcvoob = tcp_usr_rcvoob, 1209 .pru_send = tcp_usr_send, 1210 .pru_ready = tcp_usr_ready, 1211 .pru_shutdown = tcp_usr_shutdown, 1212 .pru_sockaddr = in_getsockaddr, 1213 .pru_sosetlabel = in_pcbsosetlabel, 1214 .pru_close = tcp_usr_close, 1215 }; 1216 #endif /* INET */ 1217 1218 #ifdef INET6 1219 struct pr_usrreqs tcp6_usrreqs = { 1220 .pru_abort = tcp_usr_abort, 1221 .pru_accept = tcp6_usr_accept, 1222 .pru_attach = tcp_usr_attach, 1223 .pru_bind = tcp6_usr_bind, 1224 .pru_connect = tcp6_usr_connect, 1225 .pru_control = in6_control, 1226 .pru_detach = tcp_usr_detach, 1227 .pru_disconnect = tcp_usr_disconnect, 1228 .pru_listen = tcp6_usr_listen, 1229 .pru_peeraddr = in6_mapped_peeraddr, 1230 .pru_rcvd = tcp_usr_rcvd, 1231 .pru_rcvoob = tcp_usr_rcvoob, 1232 .pru_send = tcp_usr_send, 1233 .pru_ready = tcp_usr_ready, 1234 .pru_shutdown = tcp_usr_shutdown, 1235 .pru_sockaddr = in6_mapped_sockaddr, 1236 .pru_sosetlabel = in_pcbsosetlabel, 1237 .pru_close = tcp_usr_close, 1238 }; 1239 #endif /* INET6 */ 1240 1241 #ifdef INET 1242 /* 1243 * Common subroutine to open a TCP connection to remote host specified 1244 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1245 * port number if needed. Call in_pcbconnect_setup to do the routing and 1246 * to choose a local host address (interface). If there is an existing 1247 * incarnation of the same connection in TIME-WAIT state and if the remote 1248 * host was sending CC options and if the connection duration was < MSL, then 1249 * truncate the previous TIME-WAIT state and proceed. 1250 * Initialize connection parameters and enter SYN-SENT state. 1251 */ 1252 static int 1253 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1254 { 1255 struct inpcb *inp = tp->t_inpcb, *oinp; 1256 struct socket *so = inp->inp_socket; 1257 struct in_addr laddr; 1258 u_short lport; 1259 int error; 1260 1261 INP_WLOCK_ASSERT(inp); 1262 INP_HASH_WLOCK(&V_tcbinfo); 1263 1264 if (inp->inp_lport == 0) { 1265 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1266 if (error) 1267 goto out; 1268 } 1269 1270 /* 1271 * Cannot simply call in_pcbconnect, because there might be an 1272 * earlier incarnation of this same connection still in 1273 * TIME_WAIT state, creating an ADDRINUSE error. 1274 */ 1275 laddr = inp->inp_laddr; 1276 lport = inp->inp_lport; 1277 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1278 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1279 if (error && oinp == NULL) 1280 goto out; 1281 if (oinp) { 1282 error = EADDRINUSE; 1283 goto out; 1284 } 1285 inp->inp_laddr = laddr; 1286 in_pcbrehash(inp); 1287 INP_HASH_WUNLOCK(&V_tcbinfo); 1288 1289 /* 1290 * Compute window scaling to request: 1291 * Scale to fit into sweet spot. See tcp_syncache.c. 1292 * XXX: This should move to tcp_output(). 1293 */ 1294 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1295 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1296 tp->request_r_scale++; 1297 1298 soisconnecting(so); 1299 TCPSTAT_INC(tcps_connattempt); 1300 tcp_state_change(tp, TCPS_SYN_SENT); 1301 tp->iss = tcp_new_isn(tp); 1302 tcp_sendseqinit(tp); 1303 1304 return 0; 1305 1306 out: 1307 INP_HASH_WUNLOCK(&V_tcbinfo); 1308 return (error); 1309 } 1310 #endif /* INET */ 1311 1312 #ifdef INET6 1313 static int 1314 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1315 { 1316 struct inpcb *inp = tp->t_inpcb; 1317 int error; 1318 1319 INP_WLOCK_ASSERT(inp); 1320 INP_HASH_WLOCK(&V_tcbinfo); 1321 1322 if (inp->inp_lport == 0) { 1323 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1324 if (error) 1325 goto out; 1326 } 1327 error = in6_pcbconnect(inp, nam, td->td_ucred); 1328 if (error != 0) 1329 goto out; 1330 INP_HASH_WUNLOCK(&V_tcbinfo); 1331 1332 /* Compute window scaling to request. */ 1333 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1334 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1335 tp->request_r_scale++; 1336 1337 soisconnecting(inp->inp_socket); 1338 TCPSTAT_INC(tcps_connattempt); 1339 tcp_state_change(tp, TCPS_SYN_SENT); 1340 tp->iss = tcp_new_isn(tp); 1341 tcp_sendseqinit(tp); 1342 1343 return 0; 1344 1345 out: 1346 INP_HASH_WUNLOCK(&V_tcbinfo); 1347 return error; 1348 } 1349 #endif /* INET6 */ 1350 1351 /* 1352 * Export TCP internal state information via a struct tcp_info, based on the 1353 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1354 * (TCP state machine, etc). We export all information using FreeBSD-native 1355 * constants -- for example, the numeric values for tcpi_state will differ 1356 * from Linux. 1357 */ 1358 static void 1359 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1360 { 1361 1362 INP_WLOCK_ASSERT(tp->t_inpcb); 1363 bzero(ti, sizeof(*ti)); 1364 1365 ti->tcpi_state = tp->t_state; 1366 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1367 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1368 if (tp->t_flags & TF_SACK_PERMIT) 1369 ti->tcpi_options |= TCPI_OPT_SACK; 1370 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1371 ti->tcpi_options |= TCPI_OPT_WSCALE; 1372 ti->tcpi_snd_wscale = tp->snd_scale; 1373 ti->tcpi_rcv_wscale = tp->rcv_scale; 1374 } 1375 if (tp->t_flags & TF_ECN_PERMIT) 1376 ti->tcpi_options |= TCPI_OPT_ECN; 1377 1378 ti->tcpi_rto = tp->t_rxtcur * tick; 1379 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1380 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1381 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1382 1383 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1384 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1385 1386 /* 1387 * FreeBSD-specific extension fields for tcp_info. 1388 */ 1389 ti->tcpi_rcv_space = tp->rcv_wnd; 1390 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1391 ti->tcpi_snd_wnd = tp->snd_wnd; 1392 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1393 ti->tcpi_snd_nxt = tp->snd_nxt; 1394 ti->tcpi_snd_mss = tp->t_maxseg; 1395 ti->tcpi_rcv_mss = tp->t_maxseg; 1396 if (tp->t_flags & TF_TOE) 1397 ti->tcpi_options |= TCPI_OPT_TOE; 1398 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1399 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1400 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1401 } 1402 1403 /* 1404 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1405 * socket option arguments. When it re-acquires the lock after the copy, it 1406 * has to revalidate that the connection is still valid for the socket 1407 * option. 1408 */ 1409 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1410 INP_WLOCK(inp); \ 1411 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1412 INP_WUNLOCK(inp); \ 1413 cleanup; \ 1414 return (ECONNRESET); \ 1415 } \ 1416 tp = intotcpcb(inp); \ 1417 } while(0) 1418 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1419 1420 int 1421 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1422 { 1423 int error; 1424 struct inpcb *inp; 1425 struct tcpcb *tp; 1426 struct tcp_function_block *blk; 1427 struct tcp_function_set fsn; 1428 1429 error = 0; 1430 inp = sotoinpcb(so); 1431 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1432 INP_WLOCK(inp); 1433 if (sopt->sopt_level != IPPROTO_TCP) { 1434 #ifdef INET6 1435 if (inp->inp_vflag & INP_IPV6PROTO) { 1436 INP_WUNLOCK(inp); 1437 error = ip6_ctloutput(so, sopt); 1438 } 1439 #endif /* INET6 */ 1440 #if defined(INET6) && defined(INET) 1441 else 1442 #endif 1443 #ifdef INET 1444 { 1445 INP_WUNLOCK(inp); 1446 error = ip_ctloutput(so, sopt); 1447 } 1448 #endif 1449 return (error); 1450 } 1451 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1452 INP_WUNLOCK(inp); 1453 return (ECONNRESET); 1454 } 1455 tp = intotcpcb(inp); 1456 /* 1457 * Protect the TCP option TCP_FUNCTION_BLK so 1458 * that a sub-function can *never* overwrite this. 1459 */ 1460 if ((sopt->sopt_dir == SOPT_SET) && 1461 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1462 INP_WUNLOCK(inp); 1463 error = sooptcopyin(sopt, &fsn, sizeof fsn, 1464 sizeof fsn); 1465 if (error) 1466 return (error); 1467 INP_WLOCK_RECHECK(inp); 1468 blk = find_and_ref_tcp_functions(&fsn); 1469 if (blk == NULL) { 1470 INP_WUNLOCK(inp); 1471 return (ENOENT); 1472 } 1473 if (tp->t_fb == blk) { 1474 /* You already have this */ 1475 refcount_release(&blk->tfb_refcnt); 1476 INP_WUNLOCK(inp); 1477 return (0); 1478 } 1479 if (tp->t_state != TCPS_CLOSED) { 1480 int error=EINVAL; 1481 /* 1482 * The user has advanced the state 1483 * past the initial point, we may not 1484 * be able to switch. 1485 */ 1486 if (blk->tfb_tcp_handoff_ok != NULL) { 1487 /* 1488 * Does the stack provide a 1489 * query mechanism, if so it may 1490 * still be possible? 1491 */ 1492 error = (*blk->tfb_tcp_handoff_ok)(tp); 1493 } 1494 if (error) { 1495 refcount_release(&blk->tfb_refcnt); 1496 INP_WUNLOCK(inp); 1497 return(error); 1498 } 1499 } 1500 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1501 refcount_release(&blk->tfb_refcnt); 1502 INP_WUNLOCK(inp); 1503 return (ENOENT); 1504 } 1505 /* 1506 * Release the old refcnt, the 1507 * lookup acquired a ref on the 1508 * new one already. 1509 */ 1510 if (tp->t_fb->tfb_tcp_fb_fini) { 1511 /* 1512 * Tell the stack to cleanup with 0 i.e. 1513 * the tcb is not going away. 1514 */ 1515 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1516 } 1517 refcount_release(&tp->t_fb->tfb_refcnt); 1518 tp->t_fb = blk; 1519 if (tp->t_fb->tfb_tcp_fb_init) { 1520 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1521 } 1522 #ifdef TCP_OFFLOAD 1523 if (tp->t_flags & TF_TOE) { 1524 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1525 sopt->sopt_name); 1526 } 1527 #endif 1528 INP_WUNLOCK(inp); 1529 return (error); 1530 } else if ((sopt->sopt_dir == SOPT_GET) && 1531 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1532 strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name); 1533 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1534 INP_WUNLOCK(inp); 1535 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1536 return (error); 1537 } 1538 /* Pass in the INP locked, called must unlock it */ 1539 return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp)); 1540 } 1541 1542 /* 1543 * If this assert becomes untrue, we need to change the size of the buf 1544 * variable in tcp_default_ctloutput(). 1545 */ 1546 #ifdef CTASSERT 1547 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1548 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1549 #endif 1550 1551 int 1552 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 1553 { 1554 int error, opt, optval; 1555 u_int ui; 1556 struct tcp_info ti; 1557 struct cc_algo *algo; 1558 char *pbuf, buf[TCP_LOG_ID_LEN]; 1559 size_t len; 1560 1561 /* 1562 * For TCP_CCALGOOPT forward the control to CC module, for both 1563 * SOPT_SET and SOPT_GET. 1564 */ 1565 switch (sopt->sopt_name) { 1566 case TCP_CCALGOOPT: 1567 INP_WUNLOCK(inp); 1568 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 1569 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 1570 sopt->sopt_valsize); 1571 if (error) { 1572 free(pbuf, M_TEMP); 1573 return (error); 1574 } 1575 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 1576 if (CC_ALGO(tp)->ctl_output != NULL) 1577 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 1578 else 1579 error = ENOENT; 1580 INP_WUNLOCK(inp); 1581 if (error == 0 && sopt->sopt_dir == SOPT_GET) 1582 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 1583 free(pbuf, M_TEMP); 1584 return (error); 1585 } 1586 1587 switch (sopt->sopt_dir) { 1588 case SOPT_SET: 1589 switch (sopt->sopt_name) { 1590 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1591 case TCP_MD5SIG: 1592 if (!TCPMD5_ENABLED()) { 1593 INP_WUNLOCK(inp); 1594 return (ENOPROTOOPT); 1595 } 1596 error = TCPMD5_PCBCTL(inp, sopt); 1597 if (error) 1598 return (error); 1599 goto unlock_and_done; 1600 #endif /* IPSEC */ 1601 1602 case TCP_NODELAY: 1603 case TCP_NOOPT: 1604 INP_WUNLOCK(inp); 1605 error = sooptcopyin(sopt, &optval, sizeof optval, 1606 sizeof optval); 1607 if (error) 1608 return (error); 1609 1610 INP_WLOCK_RECHECK(inp); 1611 switch (sopt->sopt_name) { 1612 case TCP_NODELAY: 1613 opt = TF_NODELAY; 1614 break; 1615 case TCP_NOOPT: 1616 opt = TF_NOOPT; 1617 break; 1618 default: 1619 opt = 0; /* dead code to fool gcc */ 1620 break; 1621 } 1622 1623 if (optval) 1624 tp->t_flags |= opt; 1625 else 1626 tp->t_flags &= ~opt; 1627 unlock_and_done: 1628 #ifdef TCP_OFFLOAD 1629 if (tp->t_flags & TF_TOE) { 1630 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1631 sopt->sopt_name); 1632 } 1633 #endif 1634 INP_WUNLOCK(inp); 1635 break; 1636 1637 case TCP_NOPUSH: 1638 INP_WUNLOCK(inp); 1639 error = sooptcopyin(sopt, &optval, sizeof optval, 1640 sizeof optval); 1641 if (error) 1642 return (error); 1643 1644 INP_WLOCK_RECHECK(inp); 1645 if (optval) 1646 tp->t_flags |= TF_NOPUSH; 1647 else if (tp->t_flags & TF_NOPUSH) { 1648 tp->t_flags &= ~TF_NOPUSH; 1649 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1650 error = tp->t_fb->tfb_tcp_output(tp); 1651 } 1652 goto unlock_and_done; 1653 1654 case TCP_MAXSEG: 1655 INP_WUNLOCK(inp); 1656 error = sooptcopyin(sopt, &optval, sizeof optval, 1657 sizeof optval); 1658 if (error) 1659 return (error); 1660 1661 INP_WLOCK_RECHECK(inp); 1662 if (optval > 0 && optval <= tp->t_maxseg && 1663 optval + 40 >= V_tcp_minmss) 1664 tp->t_maxseg = optval; 1665 else 1666 error = EINVAL; 1667 goto unlock_and_done; 1668 1669 case TCP_INFO: 1670 INP_WUNLOCK(inp); 1671 error = EINVAL; 1672 break; 1673 1674 case TCP_CONGESTION: 1675 INP_WUNLOCK(inp); 1676 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1677 if (error) 1678 break; 1679 buf[sopt->sopt_valsize] = '\0'; 1680 INP_WLOCK_RECHECK(inp); 1681 CC_LIST_RLOCK(); 1682 STAILQ_FOREACH(algo, &cc_list, entries) 1683 if (strncmp(buf, algo->name, 1684 TCP_CA_NAME_MAX) == 0) 1685 break; 1686 CC_LIST_RUNLOCK(); 1687 if (algo == NULL) { 1688 INP_WUNLOCK(inp); 1689 error = EINVAL; 1690 break; 1691 } 1692 /* 1693 * We hold a write lock over the tcb so it's safe to 1694 * do these things without ordering concerns. 1695 */ 1696 if (CC_ALGO(tp)->cb_destroy != NULL) 1697 CC_ALGO(tp)->cb_destroy(tp->ccv); 1698 CC_ALGO(tp) = algo; 1699 /* 1700 * If something goes pear shaped initialising the new 1701 * algo, fall back to newreno (which does not 1702 * require initialisation). 1703 */ 1704 if (algo->cb_init != NULL && 1705 algo->cb_init(tp->ccv) != 0) { 1706 CC_ALGO(tp) = &newreno_cc_algo; 1707 /* 1708 * The only reason init should fail is 1709 * because of malloc. 1710 */ 1711 error = ENOMEM; 1712 } 1713 INP_WUNLOCK(inp); 1714 break; 1715 1716 case TCP_KEEPIDLE: 1717 case TCP_KEEPINTVL: 1718 case TCP_KEEPINIT: 1719 INP_WUNLOCK(inp); 1720 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 1721 if (error) 1722 return (error); 1723 1724 if (ui > (UINT_MAX / hz)) { 1725 error = EINVAL; 1726 break; 1727 } 1728 ui *= hz; 1729 1730 INP_WLOCK_RECHECK(inp); 1731 switch (sopt->sopt_name) { 1732 case TCP_KEEPIDLE: 1733 tp->t_keepidle = ui; 1734 /* 1735 * XXX: better check current remaining 1736 * timeout and "merge" it with new value. 1737 */ 1738 if ((tp->t_state > TCPS_LISTEN) && 1739 (tp->t_state <= TCPS_CLOSING)) 1740 tcp_timer_activate(tp, TT_KEEP, 1741 TP_KEEPIDLE(tp)); 1742 break; 1743 case TCP_KEEPINTVL: 1744 tp->t_keepintvl = ui; 1745 if ((tp->t_state == TCPS_FIN_WAIT_2) && 1746 (TP_MAXIDLE(tp) > 0)) 1747 tcp_timer_activate(tp, TT_2MSL, 1748 TP_MAXIDLE(tp)); 1749 break; 1750 case TCP_KEEPINIT: 1751 tp->t_keepinit = ui; 1752 if (tp->t_state == TCPS_SYN_RECEIVED || 1753 tp->t_state == TCPS_SYN_SENT) 1754 tcp_timer_activate(tp, TT_KEEP, 1755 TP_KEEPINIT(tp)); 1756 break; 1757 } 1758 goto unlock_and_done; 1759 1760 case TCP_KEEPCNT: 1761 INP_WUNLOCK(inp); 1762 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 1763 if (error) 1764 return (error); 1765 1766 INP_WLOCK_RECHECK(inp); 1767 tp->t_keepcnt = ui; 1768 if ((tp->t_state == TCPS_FIN_WAIT_2) && 1769 (TP_MAXIDLE(tp) > 0)) 1770 tcp_timer_activate(tp, TT_2MSL, 1771 TP_MAXIDLE(tp)); 1772 goto unlock_and_done; 1773 1774 #ifdef TCPPCAP 1775 case TCP_PCAP_OUT: 1776 case TCP_PCAP_IN: 1777 INP_WUNLOCK(inp); 1778 error = sooptcopyin(sopt, &optval, sizeof optval, 1779 sizeof optval); 1780 if (error) 1781 return (error); 1782 1783 INP_WLOCK_RECHECK(inp); 1784 if (optval >= 0) 1785 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 1786 &(tp->t_outpkts) : &(tp->t_inpkts), 1787 optval); 1788 else 1789 error = EINVAL; 1790 goto unlock_and_done; 1791 #endif 1792 1793 case TCP_FASTOPEN: { 1794 struct tcp_fastopen tfo_optval; 1795 1796 INP_WUNLOCK(inp); 1797 if (!V_tcp_fastopen_client_enable && 1798 !V_tcp_fastopen_server_enable) 1799 return (EPERM); 1800 1801 error = sooptcopyin(sopt, &tfo_optval, 1802 sizeof(tfo_optval), sizeof(int)); 1803 if (error) 1804 return (error); 1805 1806 INP_WLOCK_RECHECK(inp); 1807 if (tfo_optval.enable) { 1808 if (tp->t_state == TCPS_LISTEN) { 1809 if (!V_tcp_fastopen_server_enable) { 1810 error = EPERM; 1811 goto unlock_and_done; 1812 } 1813 1814 tp->t_flags |= TF_FASTOPEN; 1815 if (tp->t_tfo_pending == NULL) 1816 tp->t_tfo_pending = 1817 tcp_fastopen_alloc_counter(); 1818 } else { 1819 /* 1820 * If a pre-shared key was provided, 1821 * stash it in the client cookie 1822 * field of the tcpcb for use during 1823 * connect. 1824 */ 1825 if (sopt->sopt_valsize == 1826 sizeof(tfo_optval)) { 1827 memcpy(tp->t_tfo_cookie.client, 1828 tfo_optval.psk, 1829 TCP_FASTOPEN_PSK_LEN); 1830 tp->t_tfo_client_cookie_len = 1831 TCP_FASTOPEN_PSK_LEN; 1832 } 1833 tp->t_flags |= TF_FASTOPEN; 1834 } 1835 } else 1836 tp->t_flags &= ~TF_FASTOPEN; 1837 goto unlock_and_done; 1838 } 1839 1840 #ifdef TCP_BLACKBOX 1841 case TCP_LOG: 1842 INP_WUNLOCK(inp); 1843 error = sooptcopyin(sopt, &optval, sizeof optval, 1844 sizeof optval); 1845 if (error) 1846 return (error); 1847 1848 INP_WLOCK_RECHECK(inp); 1849 error = tcp_log_state_change(tp, optval); 1850 goto unlock_and_done; 1851 1852 case TCP_LOGBUF: 1853 INP_WUNLOCK(inp); 1854 error = EINVAL; 1855 break; 1856 1857 case TCP_LOGID: 1858 INP_WUNLOCK(inp); 1859 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 1860 if (error) 1861 break; 1862 buf[sopt->sopt_valsize] = '\0'; 1863 INP_WLOCK_RECHECK(inp); 1864 error = tcp_log_set_id(tp, buf); 1865 /* tcp_log_set_id() unlocks the INP. */ 1866 break; 1867 1868 case TCP_LOGDUMP: 1869 case TCP_LOGDUMPID: 1870 INP_WUNLOCK(inp); 1871 error = 1872 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 1873 if (error) 1874 break; 1875 buf[sopt->sopt_valsize] = '\0'; 1876 INP_WLOCK_RECHECK(inp); 1877 if (sopt->sopt_name == TCP_LOGDUMP) { 1878 error = tcp_log_dump_tp_logbuf(tp, buf, 1879 M_WAITOK, true); 1880 INP_WUNLOCK(inp); 1881 } else { 1882 tcp_log_dump_tp_bucket_logbufs(tp, buf); 1883 /* 1884 * tcp_log_dump_tp_bucket_logbufs() drops the 1885 * INP lock. 1886 */ 1887 } 1888 break; 1889 #endif 1890 1891 default: 1892 INP_WUNLOCK(inp); 1893 error = ENOPROTOOPT; 1894 break; 1895 } 1896 break; 1897 1898 case SOPT_GET: 1899 tp = intotcpcb(inp); 1900 switch (sopt->sopt_name) { 1901 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1902 case TCP_MD5SIG: 1903 if (!TCPMD5_ENABLED()) { 1904 INP_WUNLOCK(inp); 1905 return (ENOPROTOOPT); 1906 } 1907 error = TCPMD5_PCBCTL(inp, sopt); 1908 break; 1909 #endif 1910 1911 case TCP_NODELAY: 1912 optval = tp->t_flags & TF_NODELAY; 1913 INP_WUNLOCK(inp); 1914 error = sooptcopyout(sopt, &optval, sizeof optval); 1915 break; 1916 case TCP_MAXSEG: 1917 optval = tp->t_maxseg; 1918 INP_WUNLOCK(inp); 1919 error = sooptcopyout(sopt, &optval, sizeof optval); 1920 break; 1921 case TCP_NOOPT: 1922 optval = tp->t_flags & TF_NOOPT; 1923 INP_WUNLOCK(inp); 1924 error = sooptcopyout(sopt, &optval, sizeof optval); 1925 break; 1926 case TCP_NOPUSH: 1927 optval = tp->t_flags & TF_NOPUSH; 1928 INP_WUNLOCK(inp); 1929 error = sooptcopyout(sopt, &optval, sizeof optval); 1930 break; 1931 case TCP_INFO: 1932 tcp_fill_info(tp, &ti); 1933 INP_WUNLOCK(inp); 1934 error = sooptcopyout(sopt, &ti, sizeof ti); 1935 break; 1936 case TCP_CONGESTION: 1937 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 1938 INP_WUNLOCK(inp); 1939 error = sooptcopyout(sopt, buf, len + 1); 1940 break; 1941 case TCP_KEEPIDLE: 1942 case TCP_KEEPINTVL: 1943 case TCP_KEEPINIT: 1944 case TCP_KEEPCNT: 1945 switch (sopt->sopt_name) { 1946 case TCP_KEEPIDLE: 1947 ui = TP_KEEPIDLE(tp) / hz; 1948 break; 1949 case TCP_KEEPINTVL: 1950 ui = TP_KEEPINTVL(tp) / hz; 1951 break; 1952 case TCP_KEEPINIT: 1953 ui = TP_KEEPINIT(tp) / hz; 1954 break; 1955 case TCP_KEEPCNT: 1956 ui = TP_KEEPCNT(tp); 1957 break; 1958 } 1959 INP_WUNLOCK(inp); 1960 error = sooptcopyout(sopt, &ui, sizeof(ui)); 1961 break; 1962 #ifdef TCPPCAP 1963 case TCP_PCAP_OUT: 1964 case TCP_PCAP_IN: 1965 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 1966 &(tp->t_outpkts) : &(tp->t_inpkts)); 1967 INP_WUNLOCK(inp); 1968 error = sooptcopyout(sopt, &optval, sizeof optval); 1969 break; 1970 #endif 1971 case TCP_FASTOPEN: 1972 optval = tp->t_flags & TF_FASTOPEN; 1973 INP_WUNLOCK(inp); 1974 error = sooptcopyout(sopt, &optval, sizeof optval); 1975 break; 1976 #ifdef TCP_BLACKBOX 1977 case TCP_LOG: 1978 optval = tp->t_logstate; 1979 INP_WUNLOCK(inp); 1980 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1981 break; 1982 case TCP_LOGBUF: 1983 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 1984 error = tcp_log_getlogbuf(sopt, tp); 1985 break; 1986 case TCP_LOGID: 1987 len = tcp_log_get_id(tp, buf); 1988 INP_WUNLOCK(inp); 1989 error = sooptcopyout(sopt, buf, len + 1); 1990 break; 1991 case TCP_LOGDUMP: 1992 case TCP_LOGDUMPID: 1993 INP_WUNLOCK(inp); 1994 error = EINVAL; 1995 break; 1996 #endif 1997 default: 1998 INP_WUNLOCK(inp); 1999 error = ENOPROTOOPT; 2000 break; 2001 } 2002 break; 2003 } 2004 return (error); 2005 } 2006 #undef INP_WLOCK_RECHECK 2007 #undef INP_WLOCK_RECHECK_CLEANUP 2008 2009 /* 2010 * Attach TCP protocol to socket, allocating 2011 * internet protocol control block, tcp control block, 2012 * bufer space, and entering LISTEN state if to accept connections. 2013 */ 2014 static int 2015 tcp_attach(struct socket *so) 2016 { 2017 struct tcpcb *tp; 2018 struct inpcb *inp; 2019 int error; 2020 2021 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 2022 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 2023 if (error) 2024 return (error); 2025 } 2026 so->so_rcv.sb_flags |= SB_AUTOSIZE; 2027 so->so_snd.sb_flags |= SB_AUTOSIZE; 2028 INP_INFO_RLOCK(&V_tcbinfo); 2029 error = in_pcballoc(so, &V_tcbinfo); 2030 if (error) { 2031 INP_INFO_RUNLOCK(&V_tcbinfo); 2032 return (error); 2033 } 2034 inp = sotoinpcb(so); 2035 #ifdef INET6 2036 if (inp->inp_vflag & INP_IPV6PROTO) { 2037 inp->inp_vflag |= INP_IPV6; 2038 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 2039 inp->inp_vflag |= INP_IPV4; 2040 inp->in6p_hops = -1; /* use kernel default */ 2041 } 2042 else 2043 #endif 2044 inp->inp_vflag |= INP_IPV4; 2045 tp = tcp_newtcpcb(inp); 2046 if (tp == NULL) { 2047 in_pcbdetach(inp); 2048 in_pcbfree(inp); 2049 INP_INFO_RUNLOCK(&V_tcbinfo); 2050 return (ENOBUFS); 2051 } 2052 tp->t_state = TCPS_CLOSED; 2053 INP_WUNLOCK(inp); 2054 INP_INFO_RUNLOCK(&V_tcbinfo); 2055 TCPSTATES_INC(TCPS_CLOSED); 2056 return (0); 2057 } 2058 2059 /* 2060 * Initiate (or continue) disconnect. 2061 * If embryonic state, just send reset (once). 2062 * If in ``let data drain'' option and linger null, just drop. 2063 * Otherwise (hard), mark socket disconnecting and drop 2064 * current input data; switch states based on user close, and 2065 * send segment to peer (with FIN). 2066 */ 2067 static void 2068 tcp_disconnect(struct tcpcb *tp) 2069 { 2070 struct inpcb *inp = tp->t_inpcb; 2071 struct socket *so = inp->inp_socket; 2072 2073 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2074 INP_WLOCK_ASSERT(inp); 2075 2076 /* 2077 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2078 * socket is still open. 2079 */ 2080 if (tp->t_state < TCPS_ESTABLISHED) { 2081 tp = tcp_close(tp); 2082 KASSERT(tp != NULL, 2083 ("tcp_disconnect: tcp_close() returned NULL")); 2084 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2085 tp = tcp_drop(tp, 0); 2086 KASSERT(tp != NULL, 2087 ("tcp_disconnect: tcp_drop() returned NULL")); 2088 } else { 2089 soisdisconnecting(so); 2090 sbflush(&so->so_rcv); 2091 tcp_usrclosed(tp); 2092 if (!(inp->inp_flags & INP_DROPPED)) 2093 tp->t_fb->tfb_tcp_output(tp); 2094 } 2095 } 2096 2097 /* 2098 * User issued close, and wish to trail through shutdown states: 2099 * if never received SYN, just forget it. If got a SYN from peer, 2100 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2101 * If already got a FIN from peer, then almost done; go to LAST_ACK 2102 * state. In all other cases, have already sent FIN to peer (e.g. 2103 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2104 * for peer to send FIN or not respond to keep-alives, etc. 2105 * We can let the user exit from the close as soon as the FIN is acked. 2106 */ 2107 static void 2108 tcp_usrclosed(struct tcpcb *tp) 2109 { 2110 2111 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2112 INP_WLOCK_ASSERT(tp->t_inpcb); 2113 2114 switch (tp->t_state) { 2115 case TCPS_LISTEN: 2116 #ifdef TCP_OFFLOAD 2117 tcp_offload_listen_stop(tp); 2118 #endif 2119 tcp_state_change(tp, TCPS_CLOSED); 2120 /* FALLTHROUGH */ 2121 case TCPS_CLOSED: 2122 tp = tcp_close(tp); 2123 /* 2124 * tcp_close() should never return NULL here as the socket is 2125 * still open. 2126 */ 2127 KASSERT(tp != NULL, 2128 ("tcp_usrclosed: tcp_close() returned NULL")); 2129 break; 2130 2131 case TCPS_SYN_SENT: 2132 case TCPS_SYN_RECEIVED: 2133 tp->t_flags |= TF_NEEDFIN; 2134 break; 2135 2136 case TCPS_ESTABLISHED: 2137 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2138 break; 2139 2140 case TCPS_CLOSE_WAIT: 2141 tcp_state_change(tp, TCPS_LAST_ACK); 2142 break; 2143 } 2144 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2145 soisdisconnected(tp->t_inpcb->inp_socket); 2146 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2147 if (tp->t_state == TCPS_FIN_WAIT_2) { 2148 int timeout; 2149 2150 timeout = (tcp_fast_finwait2_recycle) ? 2151 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2152 tcp_timer_activate(tp, TT_2MSL, timeout); 2153 } 2154 } 2155 } 2156 2157 #ifdef DDB 2158 static void 2159 db_print_indent(int indent) 2160 { 2161 int i; 2162 2163 for (i = 0; i < indent; i++) 2164 db_printf(" "); 2165 } 2166 2167 static void 2168 db_print_tstate(int t_state) 2169 { 2170 2171 switch (t_state) { 2172 case TCPS_CLOSED: 2173 db_printf("TCPS_CLOSED"); 2174 return; 2175 2176 case TCPS_LISTEN: 2177 db_printf("TCPS_LISTEN"); 2178 return; 2179 2180 case TCPS_SYN_SENT: 2181 db_printf("TCPS_SYN_SENT"); 2182 return; 2183 2184 case TCPS_SYN_RECEIVED: 2185 db_printf("TCPS_SYN_RECEIVED"); 2186 return; 2187 2188 case TCPS_ESTABLISHED: 2189 db_printf("TCPS_ESTABLISHED"); 2190 return; 2191 2192 case TCPS_CLOSE_WAIT: 2193 db_printf("TCPS_CLOSE_WAIT"); 2194 return; 2195 2196 case TCPS_FIN_WAIT_1: 2197 db_printf("TCPS_FIN_WAIT_1"); 2198 return; 2199 2200 case TCPS_CLOSING: 2201 db_printf("TCPS_CLOSING"); 2202 return; 2203 2204 case TCPS_LAST_ACK: 2205 db_printf("TCPS_LAST_ACK"); 2206 return; 2207 2208 case TCPS_FIN_WAIT_2: 2209 db_printf("TCPS_FIN_WAIT_2"); 2210 return; 2211 2212 case TCPS_TIME_WAIT: 2213 db_printf("TCPS_TIME_WAIT"); 2214 return; 2215 2216 default: 2217 db_printf("unknown"); 2218 return; 2219 } 2220 } 2221 2222 static void 2223 db_print_tflags(u_int t_flags) 2224 { 2225 int comma; 2226 2227 comma = 0; 2228 if (t_flags & TF_ACKNOW) { 2229 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2230 comma = 1; 2231 } 2232 if (t_flags & TF_DELACK) { 2233 db_printf("%sTF_DELACK", comma ? ", " : ""); 2234 comma = 1; 2235 } 2236 if (t_flags & TF_NODELAY) { 2237 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2238 comma = 1; 2239 } 2240 if (t_flags & TF_NOOPT) { 2241 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2242 comma = 1; 2243 } 2244 if (t_flags & TF_SENTFIN) { 2245 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2246 comma = 1; 2247 } 2248 if (t_flags & TF_REQ_SCALE) { 2249 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2250 comma = 1; 2251 } 2252 if (t_flags & TF_RCVD_SCALE) { 2253 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2254 comma = 1; 2255 } 2256 if (t_flags & TF_REQ_TSTMP) { 2257 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2258 comma = 1; 2259 } 2260 if (t_flags & TF_RCVD_TSTMP) { 2261 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2262 comma = 1; 2263 } 2264 if (t_flags & TF_SACK_PERMIT) { 2265 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2266 comma = 1; 2267 } 2268 if (t_flags & TF_NEEDSYN) { 2269 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2270 comma = 1; 2271 } 2272 if (t_flags & TF_NEEDFIN) { 2273 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2274 comma = 1; 2275 } 2276 if (t_flags & TF_NOPUSH) { 2277 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2278 comma = 1; 2279 } 2280 if (t_flags & TF_MORETOCOME) { 2281 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2282 comma = 1; 2283 } 2284 if (t_flags & TF_LQ_OVERFLOW) { 2285 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2286 comma = 1; 2287 } 2288 if (t_flags & TF_LASTIDLE) { 2289 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2290 comma = 1; 2291 } 2292 if (t_flags & TF_RXWIN0SENT) { 2293 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2294 comma = 1; 2295 } 2296 if (t_flags & TF_FASTRECOVERY) { 2297 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2298 comma = 1; 2299 } 2300 if (t_flags & TF_CONGRECOVERY) { 2301 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2302 comma = 1; 2303 } 2304 if (t_flags & TF_WASFRECOVERY) { 2305 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2306 comma = 1; 2307 } 2308 if (t_flags & TF_SIGNATURE) { 2309 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2310 comma = 1; 2311 } 2312 if (t_flags & TF_FORCEDATA) { 2313 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2314 comma = 1; 2315 } 2316 if (t_flags & TF_TSO) { 2317 db_printf("%sTF_TSO", comma ? ", " : ""); 2318 comma = 1; 2319 } 2320 if (t_flags & TF_ECN_PERMIT) { 2321 db_printf("%sTF_ECN_PERMIT", comma ? ", " : ""); 2322 comma = 1; 2323 } 2324 if (t_flags & TF_FASTOPEN) { 2325 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2326 comma = 1; 2327 } 2328 } 2329 2330 static void 2331 db_print_toobflags(char t_oobflags) 2332 { 2333 int comma; 2334 2335 comma = 0; 2336 if (t_oobflags & TCPOOB_HAVEDATA) { 2337 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 2338 comma = 1; 2339 } 2340 if (t_oobflags & TCPOOB_HADDATA) { 2341 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 2342 comma = 1; 2343 } 2344 } 2345 2346 static void 2347 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 2348 { 2349 2350 db_print_indent(indent); 2351 db_printf("%s at %p\n", name, tp); 2352 2353 indent += 2; 2354 2355 db_print_indent(indent); 2356 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 2357 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 2358 2359 db_print_indent(indent); 2360 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 2361 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 2362 2363 db_print_indent(indent); 2364 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 2365 &tp->t_timers->tt_delack, tp->t_inpcb); 2366 2367 db_print_indent(indent); 2368 db_printf("t_state: %d (", tp->t_state); 2369 db_print_tstate(tp->t_state); 2370 db_printf(")\n"); 2371 2372 db_print_indent(indent); 2373 db_printf("t_flags: 0x%x (", tp->t_flags); 2374 db_print_tflags(tp->t_flags); 2375 db_printf(")\n"); 2376 2377 db_print_indent(indent); 2378 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 2379 tp->snd_una, tp->snd_max, tp->snd_nxt); 2380 2381 db_print_indent(indent); 2382 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 2383 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 2384 2385 db_print_indent(indent); 2386 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 2387 tp->iss, tp->irs, tp->rcv_nxt); 2388 2389 db_print_indent(indent); 2390 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 2391 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 2392 2393 db_print_indent(indent); 2394 db_printf("snd_wnd: %u snd_cwnd: %u\n", 2395 tp->snd_wnd, tp->snd_cwnd); 2396 2397 db_print_indent(indent); 2398 db_printf("snd_ssthresh: %u snd_recover: " 2399 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 2400 2401 db_print_indent(indent); 2402 db_printf("t_rcvtime: %u t_startime: %u\n", 2403 tp->t_rcvtime, tp->t_starttime); 2404 2405 db_print_indent(indent); 2406 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 2407 tp->t_rtttime, tp->t_rtseq); 2408 2409 db_print_indent(indent); 2410 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 2411 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 2412 2413 db_print_indent(indent); 2414 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 2415 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 2416 tp->t_rttbest); 2417 2418 db_print_indent(indent); 2419 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 2420 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 2421 2422 db_print_indent(indent); 2423 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 2424 db_print_toobflags(tp->t_oobflags); 2425 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 2426 2427 db_print_indent(indent); 2428 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 2429 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 2430 2431 db_print_indent(indent); 2432 db_printf("ts_recent: %u ts_recent_age: %u\n", 2433 tp->ts_recent, tp->ts_recent_age); 2434 2435 db_print_indent(indent); 2436 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 2437 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 2438 2439 db_print_indent(indent); 2440 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 2441 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 2442 tp->snd_recover_prev, tp->t_badrxtwin); 2443 2444 db_print_indent(indent); 2445 db_printf("snd_numholes: %d snd_holes first: %p\n", 2446 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 2447 2448 db_print_indent(indent); 2449 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 2450 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 2451 2452 /* Skip sackblks, sackhint. */ 2453 2454 db_print_indent(indent); 2455 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 2456 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 2457 } 2458 2459 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 2460 { 2461 struct tcpcb *tp; 2462 2463 if (!have_addr) { 2464 db_printf("usage: show tcpcb <addr>\n"); 2465 return; 2466 } 2467 tp = (struct tcpcb *)addr; 2468 2469 db_print_tcpcb(tp, "tcpcb", 0); 2470 } 2471 #endif 2472