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/tcpip.h> 94 #include <netinet/cc/cc.h> 95 #include <netinet/tcp_fastopen.h> 96 #ifdef TCPPCAP 97 #include <netinet/tcp_pcap.h> 98 #endif 99 #ifdef TCPDEBUG 100 #include <netinet/tcp_debug.h> 101 #endif 102 #ifdef TCP_OFFLOAD 103 #include <netinet/tcp_offload.h> 104 #endif 105 #include <netipsec/ipsec_support.h> 106 107 /* 108 * TCP protocol interface to socket abstraction. 109 */ 110 static int tcp_attach(struct socket *); 111 #ifdef INET 112 static int tcp_connect(struct tcpcb *, struct sockaddr *, 113 struct thread *td); 114 #endif /* INET */ 115 #ifdef INET6 116 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 117 struct thread *td); 118 #endif /* INET6 */ 119 static void tcp_disconnect(struct tcpcb *); 120 static void tcp_usrclosed(struct tcpcb *); 121 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 122 123 #ifdef TCPDEBUG 124 #define TCPDEBUG0 int ostate = 0 125 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 126 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 127 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 128 #else 129 #define TCPDEBUG0 130 #define TCPDEBUG1() 131 #define TCPDEBUG2(req) 132 #endif 133 134 /* 135 * TCP attaches to socket via pru_attach(), reserving space, 136 * and an internet control block. 137 */ 138 static int 139 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 140 { 141 struct inpcb *inp; 142 struct tcpcb *tp = NULL; 143 int error; 144 TCPDEBUG0; 145 146 inp = sotoinpcb(so); 147 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 148 TCPDEBUG1(); 149 150 error = tcp_attach(so); 151 if (error) 152 goto out; 153 154 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 155 so->so_linger = TCP_LINGERTIME; 156 157 inp = sotoinpcb(so); 158 tp = intotcpcb(inp); 159 out: 160 TCPDEBUG2(PRU_ATTACH); 161 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 162 return error; 163 } 164 165 /* 166 * tcp_detach is called when the socket layer loses its final reference 167 * to the socket, be it a file descriptor reference, a reference from TCP, 168 * etc. At this point, there is only one case in which we will keep around 169 * inpcb state: time wait. 170 * 171 * This function can probably be re-absorbed back into tcp_usr_detach() now 172 * that there is a single detach path. 173 */ 174 static void 175 tcp_detach(struct socket *so, struct inpcb *inp) 176 { 177 struct tcpcb *tp; 178 179 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 180 INP_WLOCK_ASSERT(inp); 181 182 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp")); 183 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so")); 184 185 tp = intotcpcb(inp); 186 187 if (inp->inp_flags & INP_TIMEWAIT) { 188 /* 189 * There are two cases to handle: one in which the time wait 190 * state is being discarded (INP_DROPPED), and one in which 191 * this connection will remain in timewait. In the former, 192 * it is time to discard all state (except tcptw, which has 193 * already been discarded by the timewait close code, which 194 * should be further up the call stack somewhere). In the 195 * latter case, we detach from the socket, but leave the pcb 196 * present until timewait ends. 197 * 198 * XXXRW: Would it be cleaner to free the tcptw here? 199 * 200 * Astute question indeed, from twtcp perspective there are 201 * four cases to consider: 202 * 203 * #1 tcp_detach is called at tcptw creation time by 204 * tcp_twstart, then do not discard the newly created tcptw 205 * and leave inpcb present until timewait ends 206 * #2 tcp_detach is called at tcptw creation time by 207 * tcp_twstart, but connection is local and tw will be 208 * discarded immediately 209 * #3 tcp_detach is called at timewait end (or reuse) by 210 * tcp_twclose, then the tcptw has already been discarded 211 * (or reused) and inpcb is freed here 212 * #4 tcp_detach is called() after timewait ends (or reuse) 213 * (e.g. by soclose), then tcptw has already been discarded 214 * (or reused) and inpcb is freed here 215 * 216 * In all three cases the tcptw should not be freed here. 217 */ 218 if (inp->inp_flags & INP_DROPPED) { 219 in_pcbdetach(inp); 220 if (__predict_true(tp == NULL)) { 221 in_pcbfree(inp); 222 } else { 223 /* 224 * This case should not happen as in TIMEWAIT 225 * state the inp should not be destroyed before 226 * its tcptw. If INVARIANTS is defined, panic. 227 */ 228 #ifdef INVARIANTS 229 panic("%s: Panic before an inp double-free: " 230 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 231 , __func__); 232 #else 233 log(LOG_ERR, "%s: Avoid an inp double-free: " 234 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 235 , __func__); 236 #endif 237 INP_WUNLOCK(inp); 238 } 239 } else { 240 in_pcbdetach(inp); 241 INP_WUNLOCK(inp); 242 } 243 } else { 244 /* 245 * If the connection is not in timewait, we consider two 246 * two conditions: one in which no further processing is 247 * necessary (dropped || embryonic), and one in which TCP is 248 * not yet done, but no longer requires the socket, so the 249 * pcb will persist for the time being. 250 * 251 * XXXRW: Does the second case still occur? 252 */ 253 if (inp->inp_flags & INP_DROPPED || 254 tp->t_state < TCPS_SYN_SENT) { 255 tcp_discardcb(tp); 256 in_pcbdetach(inp); 257 in_pcbfree(inp); 258 } else { 259 in_pcbdetach(inp); 260 INP_WUNLOCK(inp); 261 } 262 } 263 } 264 265 /* 266 * pru_detach() detaches the TCP protocol from the socket. 267 * If the protocol state is non-embryonic, then can't 268 * do this directly: have to initiate a pru_disconnect(), 269 * which may finish later; embryonic TCB's can just 270 * be discarded here. 271 */ 272 static void 273 tcp_usr_detach(struct socket *so) 274 { 275 struct inpcb *inp; 276 int rlock = 0; 277 278 inp = sotoinpcb(so); 279 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL")); 280 if (!INP_INFO_WLOCKED(&V_tcbinfo)) { 281 INP_INFO_RLOCK(&V_tcbinfo); 282 rlock = 1; 283 } 284 INP_WLOCK(inp); 285 KASSERT(inp->inp_socket != NULL, 286 ("tcp_usr_detach: inp_socket == NULL")); 287 tcp_detach(so, inp); 288 if (rlock) 289 INP_INFO_RUNLOCK(&V_tcbinfo); 290 } 291 292 #ifdef INET 293 /* 294 * Give the socket an address. 295 */ 296 static int 297 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 298 { 299 int error = 0; 300 struct inpcb *inp; 301 struct tcpcb *tp = NULL; 302 struct sockaddr_in *sinp; 303 304 sinp = (struct sockaddr_in *)nam; 305 if (nam->sa_len != sizeof (*sinp)) 306 return (EINVAL); 307 /* 308 * Must check for multicast addresses and disallow binding 309 * to them. 310 */ 311 if (sinp->sin_family == AF_INET && 312 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 313 return (EAFNOSUPPORT); 314 315 TCPDEBUG0; 316 inp = sotoinpcb(so); 317 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 318 INP_WLOCK(inp); 319 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 320 error = EINVAL; 321 goto out; 322 } 323 tp = intotcpcb(inp); 324 TCPDEBUG1(); 325 INP_HASH_WLOCK(&V_tcbinfo); 326 error = in_pcbbind(inp, nam, td->td_ucred); 327 INP_HASH_WUNLOCK(&V_tcbinfo); 328 out: 329 TCPDEBUG2(PRU_BIND); 330 TCP_PROBE2(debug__user, tp, PRU_BIND); 331 INP_WUNLOCK(inp); 332 333 return (error); 334 } 335 #endif /* INET */ 336 337 #ifdef INET6 338 static int 339 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 340 { 341 int error = 0; 342 struct inpcb *inp; 343 struct tcpcb *tp = NULL; 344 struct sockaddr_in6 *sin6p; 345 346 sin6p = (struct sockaddr_in6 *)nam; 347 if (nam->sa_len != sizeof (*sin6p)) 348 return (EINVAL); 349 /* 350 * Must check for multicast addresses and disallow binding 351 * to them. 352 */ 353 if (sin6p->sin6_family == AF_INET6 && 354 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 355 return (EAFNOSUPPORT); 356 357 TCPDEBUG0; 358 inp = sotoinpcb(so); 359 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 360 INP_WLOCK(inp); 361 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 362 error = EINVAL; 363 goto out; 364 } 365 tp = intotcpcb(inp); 366 TCPDEBUG1(); 367 INP_HASH_WLOCK(&V_tcbinfo); 368 inp->inp_vflag &= ~INP_IPV4; 369 inp->inp_vflag |= INP_IPV6; 370 #ifdef INET 371 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 372 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 373 inp->inp_vflag |= INP_IPV4; 374 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 375 struct sockaddr_in sin; 376 377 in6_sin6_2_sin(&sin, sin6p); 378 inp->inp_vflag |= INP_IPV4; 379 inp->inp_vflag &= ~INP_IPV6; 380 error = in_pcbbind(inp, (struct sockaddr *)&sin, 381 td->td_ucred); 382 INP_HASH_WUNLOCK(&V_tcbinfo); 383 goto out; 384 } 385 } 386 #endif 387 error = in6_pcbbind(inp, nam, td->td_ucred); 388 INP_HASH_WUNLOCK(&V_tcbinfo); 389 out: 390 TCPDEBUG2(PRU_BIND); 391 TCP_PROBE2(debug__user, tp, PRU_BIND); 392 INP_WUNLOCK(inp); 393 return (error); 394 } 395 #endif /* INET6 */ 396 397 #ifdef INET 398 /* 399 * Prepare to accept connections. 400 */ 401 static int 402 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 403 { 404 int error = 0; 405 struct inpcb *inp; 406 struct tcpcb *tp = NULL; 407 408 TCPDEBUG0; 409 inp = sotoinpcb(so); 410 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 411 INP_WLOCK(inp); 412 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 413 error = EINVAL; 414 goto out; 415 } 416 tp = intotcpcb(inp); 417 TCPDEBUG1(); 418 SOCK_LOCK(so); 419 error = solisten_proto_check(so); 420 INP_HASH_WLOCK(&V_tcbinfo); 421 if (error == 0 && inp->inp_lport == 0) 422 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 423 INP_HASH_WUNLOCK(&V_tcbinfo); 424 if (error == 0) { 425 tcp_state_change(tp, TCPS_LISTEN); 426 solisten_proto(so, backlog); 427 #ifdef TCP_OFFLOAD 428 if ((so->so_options & SO_NO_OFFLOAD) == 0) 429 tcp_offload_listen_start(tp); 430 #endif 431 } 432 SOCK_UNLOCK(so); 433 434 if (IS_FASTOPEN(tp->t_flags)) 435 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 436 437 out: 438 TCPDEBUG2(PRU_LISTEN); 439 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 440 INP_WUNLOCK(inp); 441 return (error); 442 } 443 #endif /* INET */ 444 445 #ifdef INET6 446 static int 447 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 448 { 449 int error = 0; 450 struct inpcb *inp; 451 struct tcpcb *tp = NULL; 452 453 TCPDEBUG0; 454 inp = sotoinpcb(so); 455 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 456 INP_WLOCK(inp); 457 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 458 error = EINVAL; 459 goto out; 460 } 461 tp = intotcpcb(inp); 462 TCPDEBUG1(); 463 SOCK_LOCK(so); 464 error = solisten_proto_check(so); 465 INP_HASH_WLOCK(&V_tcbinfo); 466 if (error == 0 && inp->inp_lport == 0) { 467 inp->inp_vflag &= ~INP_IPV4; 468 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 469 inp->inp_vflag |= INP_IPV4; 470 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 471 } 472 INP_HASH_WUNLOCK(&V_tcbinfo); 473 if (error == 0) { 474 tcp_state_change(tp, TCPS_LISTEN); 475 solisten_proto(so, backlog); 476 #ifdef TCP_OFFLOAD 477 if ((so->so_options & SO_NO_OFFLOAD) == 0) 478 tcp_offload_listen_start(tp); 479 #endif 480 } 481 SOCK_UNLOCK(so); 482 483 if (IS_FASTOPEN(tp->t_flags)) 484 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 485 486 out: 487 TCPDEBUG2(PRU_LISTEN); 488 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 489 INP_WUNLOCK(inp); 490 return (error); 491 } 492 #endif /* INET6 */ 493 494 #ifdef INET 495 /* 496 * Initiate connection to peer. 497 * Create a template for use in transmissions on this connection. 498 * Enter SYN_SENT state, and mark socket as connecting. 499 * Start keep-alive timer, and seed output sequence space. 500 * Send initial segment on connection. 501 */ 502 static int 503 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 504 { 505 int error = 0; 506 struct inpcb *inp; 507 struct tcpcb *tp = NULL; 508 struct sockaddr_in *sinp; 509 510 sinp = (struct sockaddr_in *)nam; 511 if (nam->sa_len != sizeof (*sinp)) 512 return (EINVAL); 513 /* 514 * Must disallow TCP ``connections'' to multicast addresses. 515 */ 516 if (sinp->sin_family == AF_INET 517 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 518 return (EAFNOSUPPORT); 519 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 520 return (error); 521 522 TCPDEBUG0; 523 inp = sotoinpcb(so); 524 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 525 INP_WLOCK(inp); 526 if (inp->inp_flags & INP_TIMEWAIT) { 527 error = EADDRINUSE; 528 goto out; 529 } 530 if (inp->inp_flags & INP_DROPPED) { 531 error = ECONNREFUSED; 532 goto out; 533 } 534 tp = intotcpcb(inp); 535 TCPDEBUG1(); 536 if ((error = tcp_connect(tp, nam, td)) != 0) 537 goto out; 538 #ifdef TCP_OFFLOAD 539 if (registered_toedevs > 0 && 540 (so->so_options & SO_NO_OFFLOAD) == 0 && 541 (error = tcp_offload_connect(so, nam)) == 0) 542 goto out; 543 #endif 544 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 545 error = tp->t_fb->tfb_tcp_output(tp); 546 out: 547 TCPDEBUG2(PRU_CONNECT); 548 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 549 INP_WUNLOCK(inp); 550 return (error); 551 } 552 #endif /* INET */ 553 554 #ifdef INET6 555 static int 556 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 557 { 558 int error = 0; 559 struct inpcb *inp; 560 struct tcpcb *tp = NULL; 561 struct sockaddr_in6 *sin6p; 562 563 TCPDEBUG0; 564 565 sin6p = (struct sockaddr_in6 *)nam; 566 if (nam->sa_len != sizeof (*sin6p)) 567 return (EINVAL); 568 /* 569 * Must disallow TCP ``connections'' to multicast addresses. 570 */ 571 if (sin6p->sin6_family == AF_INET6 572 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) 573 return (EAFNOSUPPORT); 574 575 inp = sotoinpcb(so); 576 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 577 INP_WLOCK(inp); 578 if (inp->inp_flags & INP_TIMEWAIT) { 579 error = EADDRINUSE; 580 goto out; 581 } 582 if (inp->inp_flags & INP_DROPPED) { 583 error = ECONNREFUSED; 584 goto out; 585 } 586 tp = intotcpcb(inp); 587 TCPDEBUG1(); 588 #ifdef INET 589 /* 590 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 591 * therefore probably require the hash lock, which isn't held here. 592 * Is this a significant problem? 593 */ 594 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 595 struct sockaddr_in sin; 596 597 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 598 error = EINVAL; 599 goto out; 600 } 601 if ((inp->inp_vflag & INP_IPV4) == 0) { 602 error = EAFNOSUPPORT; 603 goto out; 604 } 605 606 in6_sin6_2_sin(&sin, sin6p); 607 inp->inp_vflag |= INP_IPV4; 608 inp->inp_vflag &= ~INP_IPV6; 609 if ((error = prison_remote_ip4(td->td_ucred, 610 &sin.sin_addr)) != 0) 611 goto out; 612 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 613 goto out; 614 #ifdef TCP_OFFLOAD 615 if (registered_toedevs > 0 && 616 (so->so_options & SO_NO_OFFLOAD) == 0 && 617 (error = tcp_offload_connect(so, nam)) == 0) 618 goto out; 619 #endif 620 error = tp->t_fb->tfb_tcp_output(tp); 621 goto out; 622 } else { 623 if ((inp->inp_vflag & INP_IPV6) == 0) { 624 error = EAFNOSUPPORT; 625 goto out; 626 } 627 } 628 #endif 629 inp->inp_vflag &= ~INP_IPV4; 630 inp->inp_vflag |= INP_IPV6; 631 inp->inp_inc.inc_flags |= INC_ISIPV6; 632 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0) 633 goto out; 634 if ((error = tcp6_connect(tp, nam, td)) != 0) 635 goto out; 636 #ifdef TCP_OFFLOAD 637 if (registered_toedevs > 0 && 638 (so->so_options & SO_NO_OFFLOAD) == 0 && 639 (error = tcp_offload_connect(so, nam)) == 0) 640 goto out; 641 #endif 642 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 643 error = tp->t_fb->tfb_tcp_output(tp); 644 645 out: 646 TCPDEBUG2(PRU_CONNECT); 647 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 648 INP_WUNLOCK(inp); 649 return (error); 650 } 651 #endif /* INET6 */ 652 653 /* 654 * Initiate disconnect from peer. 655 * If connection never passed embryonic stage, just drop; 656 * else if don't need to let data drain, then can just drop anyways, 657 * else have to begin TCP shutdown process: mark socket disconnecting, 658 * drain unread data, state switch to reflect user close, and 659 * send segment (e.g. FIN) to peer. Socket will be really disconnected 660 * when peer sends FIN and acks ours. 661 * 662 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 663 */ 664 static int 665 tcp_usr_disconnect(struct socket *so) 666 { 667 struct inpcb *inp; 668 struct tcpcb *tp = NULL; 669 int error = 0; 670 671 TCPDEBUG0; 672 INP_INFO_RLOCK(&V_tcbinfo); 673 inp = sotoinpcb(so); 674 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 675 INP_WLOCK(inp); 676 if (inp->inp_flags & INP_TIMEWAIT) 677 goto out; 678 if (inp->inp_flags & INP_DROPPED) { 679 error = ECONNRESET; 680 goto out; 681 } 682 tp = intotcpcb(inp); 683 TCPDEBUG1(); 684 tcp_disconnect(tp); 685 out: 686 TCPDEBUG2(PRU_DISCONNECT); 687 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 688 INP_WUNLOCK(inp); 689 INP_INFO_RUNLOCK(&V_tcbinfo); 690 return (error); 691 } 692 693 #ifdef INET 694 /* 695 * Accept a connection. Essentially all the work is done at higher levels; 696 * just return the address of the peer, storing through addr. 697 */ 698 static int 699 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 700 { 701 int error = 0; 702 struct inpcb *inp = NULL; 703 struct tcpcb *tp = NULL; 704 struct in_addr addr; 705 in_port_t port = 0; 706 TCPDEBUG0; 707 708 if (so->so_state & SS_ISDISCONNECTED) 709 return (ECONNABORTED); 710 711 inp = sotoinpcb(so); 712 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 713 INP_WLOCK(inp); 714 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 715 error = ECONNABORTED; 716 goto out; 717 } 718 tp = intotcpcb(inp); 719 TCPDEBUG1(); 720 721 /* 722 * We inline in_getpeeraddr and COMMON_END here, so that we can 723 * copy the data of interest and defer the malloc until after we 724 * release the lock. 725 */ 726 port = inp->inp_fport; 727 addr = inp->inp_faddr; 728 729 out: 730 TCPDEBUG2(PRU_ACCEPT); 731 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 732 INP_WUNLOCK(inp); 733 if (error == 0) 734 *nam = in_sockaddr(port, &addr); 735 return error; 736 } 737 #endif /* INET */ 738 739 #ifdef INET6 740 static int 741 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 742 { 743 struct inpcb *inp = NULL; 744 int error = 0; 745 struct tcpcb *tp = NULL; 746 struct in_addr addr; 747 struct in6_addr addr6; 748 in_port_t port = 0; 749 int v4 = 0; 750 TCPDEBUG0; 751 752 if (so->so_state & SS_ISDISCONNECTED) 753 return (ECONNABORTED); 754 755 inp = sotoinpcb(so); 756 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 757 INP_INFO_RLOCK(&V_tcbinfo); 758 INP_WLOCK(inp); 759 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 760 error = ECONNABORTED; 761 goto out; 762 } 763 tp = intotcpcb(inp); 764 TCPDEBUG1(); 765 766 /* 767 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 768 * copy the data of interest and defer the malloc until after we 769 * release the lock. 770 */ 771 if (inp->inp_vflag & INP_IPV4) { 772 v4 = 1; 773 port = inp->inp_fport; 774 addr = inp->inp_faddr; 775 } else { 776 port = inp->inp_fport; 777 addr6 = inp->in6p_faddr; 778 } 779 780 out: 781 TCPDEBUG2(PRU_ACCEPT); 782 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 783 INP_WUNLOCK(inp); 784 INP_INFO_RUNLOCK(&V_tcbinfo); 785 if (error == 0) { 786 if (v4) 787 *nam = in6_v4mapsin6_sockaddr(port, &addr); 788 else 789 *nam = in6_sockaddr(port, &addr6); 790 } 791 return error; 792 } 793 #endif /* INET6 */ 794 795 /* 796 * Mark the connection as being incapable of further output. 797 */ 798 static int 799 tcp_usr_shutdown(struct socket *so) 800 { 801 int error = 0; 802 struct inpcb *inp; 803 struct tcpcb *tp = NULL; 804 805 TCPDEBUG0; 806 INP_INFO_RLOCK(&V_tcbinfo); 807 inp = sotoinpcb(so); 808 KASSERT(inp != NULL, ("inp == NULL")); 809 INP_WLOCK(inp); 810 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 811 error = ECONNRESET; 812 goto out; 813 } 814 tp = intotcpcb(inp); 815 TCPDEBUG1(); 816 socantsendmore(so); 817 tcp_usrclosed(tp); 818 if (!(inp->inp_flags & INP_DROPPED)) 819 error = tp->t_fb->tfb_tcp_output(tp); 820 821 out: 822 TCPDEBUG2(PRU_SHUTDOWN); 823 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 824 INP_WUNLOCK(inp); 825 INP_INFO_RUNLOCK(&V_tcbinfo); 826 827 return (error); 828 } 829 830 /* 831 * After a receive, possibly send window update to peer. 832 */ 833 static int 834 tcp_usr_rcvd(struct socket *so, int flags) 835 { 836 struct inpcb *inp; 837 struct tcpcb *tp = NULL; 838 int error = 0; 839 840 TCPDEBUG0; 841 inp = sotoinpcb(so); 842 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 843 INP_WLOCK(inp); 844 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 845 error = ECONNRESET; 846 goto out; 847 } 848 tp = intotcpcb(inp); 849 TCPDEBUG1(); 850 /* 851 * For passively-created TFO connections, don't attempt a window 852 * update while still in SYN_RECEIVED as this may trigger an early 853 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 854 * application response data, or failing that, when the DELACK timer 855 * expires. 856 */ 857 if (IS_FASTOPEN(tp->t_flags) && 858 (tp->t_state == TCPS_SYN_RECEIVED)) 859 goto out; 860 #ifdef TCP_OFFLOAD 861 if (tp->t_flags & TF_TOE) 862 tcp_offload_rcvd(tp); 863 else 864 #endif 865 tp->t_fb->tfb_tcp_output(tp); 866 867 out: 868 TCPDEBUG2(PRU_RCVD); 869 TCP_PROBE2(debug__user, tp, PRU_RCVD); 870 INP_WUNLOCK(inp); 871 return (error); 872 } 873 874 /* 875 * Do a send by putting data in output queue and updating urgent 876 * marker if URG set. Possibly send more data. Unlike the other 877 * pru_*() routines, the mbuf chains are our responsibility. We 878 * must either enqueue them or free them. The other pru_* routines 879 * generally are caller-frees. 880 */ 881 static int 882 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 883 struct sockaddr *nam, struct mbuf *control, struct thread *td) 884 { 885 int error = 0; 886 struct inpcb *inp; 887 struct tcpcb *tp = NULL; 888 #ifdef INET6 889 int isipv6; 890 #endif 891 TCPDEBUG0; 892 893 /* 894 * We require the pcbinfo lock if we will close the socket as part of 895 * this call. 896 */ 897 if (flags & PRUS_EOF) 898 INP_INFO_RLOCK(&V_tcbinfo); 899 inp = sotoinpcb(so); 900 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 901 INP_WLOCK(inp); 902 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 903 if (control) 904 m_freem(control); 905 /* 906 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible 907 * for freeing memory. 908 */ 909 if (m && (flags & PRUS_NOTREADY) == 0) 910 m_freem(m); 911 error = ECONNRESET; 912 goto out; 913 } 914 #ifdef INET6 915 isipv6 = nam && nam->sa_family == AF_INET6; 916 #endif /* INET6 */ 917 tp = intotcpcb(inp); 918 TCPDEBUG1(); 919 if (control) { 920 /* TCP doesn't do control messages (rights, creds, etc) */ 921 if (control->m_len) { 922 m_freem(control); 923 if (m) 924 m_freem(m); 925 error = EINVAL; 926 goto out; 927 } 928 m_freem(control); /* empty control, just free it */ 929 } 930 if (!(flags & PRUS_OOB)) { 931 sbappendstream(&so->so_snd, m, flags); 932 if (nam && tp->t_state < TCPS_SYN_SENT) { 933 /* 934 * Do implied connect if not yet connected, 935 * initialize window to default value, and 936 * initialize maxseg using peer's cached MSS. 937 */ 938 #ifdef INET6 939 if (isipv6) 940 error = tcp6_connect(tp, nam, td); 941 #endif /* INET6 */ 942 #if defined(INET6) && defined(INET) 943 else 944 #endif 945 #ifdef INET 946 error = tcp_connect(tp, nam, td); 947 #endif 948 if (error) 949 goto out; 950 if (IS_FASTOPEN(tp->t_flags)) 951 tcp_fastopen_connect(tp); 952 else { 953 tp->snd_wnd = TTCP_CLIENT_SND_WND; 954 tcp_mss(tp, -1); 955 } 956 } 957 if (flags & PRUS_EOF) { 958 /* 959 * Close the send side of the connection after 960 * the data is sent. 961 */ 962 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 963 socantsendmore(so); 964 tcp_usrclosed(tp); 965 } 966 if (!(inp->inp_flags & INP_DROPPED) && 967 !(flags & PRUS_NOTREADY)) { 968 if (flags & PRUS_MORETOCOME) 969 tp->t_flags |= TF_MORETOCOME; 970 error = tp->t_fb->tfb_tcp_output(tp); 971 if (flags & PRUS_MORETOCOME) 972 tp->t_flags &= ~TF_MORETOCOME; 973 } 974 } else { 975 /* 976 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 977 */ 978 SOCKBUF_LOCK(&so->so_snd); 979 if (sbspace(&so->so_snd) < -512) { 980 SOCKBUF_UNLOCK(&so->so_snd); 981 m_freem(m); 982 error = ENOBUFS; 983 goto out; 984 } 985 /* 986 * According to RFC961 (Assigned Protocols), 987 * the urgent pointer points to the last octet 988 * of urgent data. We continue, however, 989 * to consider it to indicate the first octet 990 * of data past the urgent section. 991 * Otherwise, snd_up should be one lower. 992 */ 993 sbappendstream_locked(&so->so_snd, m, flags); 994 SOCKBUF_UNLOCK(&so->so_snd); 995 if (nam && tp->t_state < TCPS_SYN_SENT) { 996 /* 997 * Do implied connect if not yet connected, 998 * initialize window to default value, and 999 * initialize maxseg using peer's cached MSS. 1000 */ 1001 1002 /* 1003 * Not going to contemplate SYN|URG 1004 */ 1005 if (IS_FASTOPEN(tp->t_flags)) 1006 tp->t_flags &= ~TF_FASTOPEN; 1007 #ifdef INET6 1008 if (isipv6) 1009 error = tcp6_connect(tp, nam, td); 1010 #endif /* INET6 */ 1011 #if defined(INET6) && defined(INET) 1012 else 1013 #endif 1014 #ifdef INET 1015 error = tcp_connect(tp, nam, td); 1016 #endif 1017 if (error) 1018 goto out; 1019 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1020 tcp_mss(tp, -1); 1021 } 1022 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1023 if (!(flags & PRUS_NOTREADY)) { 1024 tp->t_flags |= TF_FORCEDATA; 1025 error = tp->t_fb->tfb_tcp_output(tp); 1026 tp->t_flags &= ~TF_FORCEDATA; 1027 } 1028 } 1029 out: 1030 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1031 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1032 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1033 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1034 INP_WUNLOCK(inp); 1035 if (flags & PRUS_EOF) 1036 INP_INFO_RUNLOCK(&V_tcbinfo); 1037 return (error); 1038 } 1039 1040 static int 1041 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1042 { 1043 struct inpcb *inp; 1044 struct tcpcb *tp; 1045 int error; 1046 1047 inp = sotoinpcb(so); 1048 INP_WLOCK(inp); 1049 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1050 INP_WUNLOCK(inp); 1051 for (int i = 0; i < count; i++) 1052 m = m_free(m); 1053 return (ECONNRESET); 1054 } 1055 tp = intotcpcb(inp); 1056 1057 SOCKBUF_LOCK(&so->so_snd); 1058 error = sbready(&so->so_snd, m, count); 1059 SOCKBUF_UNLOCK(&so->so_snd); 1060 if (error == 0) 1061 error = tp->t_fb->tfb_tcp_output(tp); 1062 INP_WUNLOCK(inp); 1063 1064 return (error); 1065 } 1066 1067 /* 1068 * Abort the TCP. Drop the connection abruptly. 1069 */ 1070 static void 1071 tcp_usr_abort(struct socket *so) 1072 { 1073 struct inpcb *inp; 1074 struct tcpcb *tp = NULL; 1075 TCPDEBUG0; 1076 1077 inp = sotoinpcb(so); 1078 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1079 1080 INP_INFO_RLOCK(&V_tcbinfo); 1081 INP_WLOCK(inp); 1082 KASSERT(inp->inp_socket != NULL, 1083 ("tcp_usr_abort: inp_socket == NULL")); 1084 1085 /* 1086 * If we still have full TCP state, and we're not dropped, drop. 1087 */ 1088 if (!(inp->inp_flags & INP_TIMEWAIT) && 1089 !(inp->inp_flags & INP_DROPPED)) { 1090 tp = intotcpcb(inp); 1091 TCPDEBUG1(); 1092 tcp_drop(tp, ECONNABORTED); 1093 TCPDEBUG2(PRU_ABORT); 1094 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1095 } 1096 if (!(inp->inp_flags & INP_DROPPED)) { 1097 SOCK_LOCK(so); 1098 so->so_state |= SS_PROTOREF; 1099 SOCK_UNLOCK(so); 1100 inp->inp_flags |= INP_SOCKREF; 1101 } 1102 INP_WUNLOCK(inp); 1103 INP_INFO_RUNLOCK(&V_tcbinfo); 1104 } 1105 1106 /* 1107 * TCP socket is closed. Start friendly disconnect. 1108 */ 1109 static void 1110 tcp_usr_close(struct socket *so) 1111 { 1112 struct inpcb *inp; 1113 struct tcpcb *tp = NULL; 1114 TCPDEBUG0; 1115 1116 inp = sotoinpcb(so); 1117 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1118 1119 INP_INFO_RLOCK(&V_tcbinfo); 1120 INP_WLOCK(inp); 1121 KASSERT(inp->inp_socket != NULL, 1122 ("tcp_usr_close: inp_socket == NULL")); 1123 1124 /* 1125 * If we still have full TCP state, and we're not dropped, initiate 1126 * a disconnect. 1127 */ 1128 if (!(inp->inp_flags & INP_TIMEWAIT) && 1129 !(inp->inp_flags & INP_DROPPED)) { 1130 tp = intotcpcb(inp); 1131 TCPDEBUG1(); 1132 tcp_disconnect(tp); 1133 TCPDEBUG2(PRU_CLOSE); 1134 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1135 } 1136 if (!(inp->inp_flags & INP_DROPPED)) { 1137 SOCK_LOCK(so); 1138 so->so_state |= SS_PROTOREF; 1139 SOCK_UNLOCK(so); 1140 inp->inp_flags |= INP_SOCKREF; 1141 } 1142 INP_WUNLOCK(inp); 1143 INP_INFO_RUNLOCK(&V_tcbinfo); 1144 } 1145 1146 /* 1147 * Receive out-of-band data. 1148 */ 1149 static int 1150 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1151 { 1152 int error = 0; 1153 struct inpcb *inp; 1154 struct tcpcb *tp = NULL; 1155 1156 TCPDEBUG0; 1157 inp = sotoinpcb(so); 1158 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1159 INP_WLOCK(inp); 1160 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1161 error = ECONNRESET; 1162 goto out; 1163 } 1164 tp = intotcpcb(inp); 1165 TCPDEBUG1(); 1166 if ((so->so_oobmark == 0 && 1167 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1168 so->so_options & SO_OOBINLINE || 1169 tp->t_oobflags & TCPOOB_HADDATA) { 1170 error = EINVAL; 1171 goto out; 1172 } 1173 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1174 error = EWOULDBLOCK; 1175 goto out; 1176 } 1177 m->m_len = 1; 1178 *mtod(m, caddr_t) = tp->t_iobc; 1179 if ((flags & MSG_PEEK) == 0) 1180 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1181 1182 out: 1183 TCPDEBUG2(PRU_RCVOOB); 1184 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1185 INP_WUNLOCK(inp); 1186 return (error); 1187 } 1188 1189 #ifdef INET 1190 struct pr_usrreqs tcp_usrreqs = { 1191 .pru_abort = tcp_usr_abort, 1192 .pru_accept = tcp_usr_accept, 1193 .pru_attach = tcp_usr_attach, 1194 .pru_bind = tcp_usr_bind, 1195 .pru_connect = tcp_usr_connect, 1196 .pru_control = in_control, 1197 .pru_detach = tcp_usr_detach, 1198 .pru_disconnect = tcp_usr_disconnect, 1199 .pru_listen = tcp_usr_listen, 1200 .pru_peeraddr = in_getpeeraddr, 1201 .pru_rcvd = tcp_usr_rcvd, 1202 .pru_rcvoob = tcp_usr_rcvoob, 1203 .pru_send = tcp_usr_send, 1204 .pru_ready = tcp_usr_ready, 1205 .pru_shutdown = tcp_usr_shutdown, 1206 .pru_sockaddr = in_getsockaddr, 1207 .pru_sosetlabel = in_pcbsosetlabel, 1208 .pru_close = tcp_usr_close, 1209 }; 1210 #endif /* INET */ 1211 1212 #ifdef INET6 1213 struct pr_usrreqs tcp6_usrreqs = { 1214 .pru_abort = tcp_usr_abort, 1215 .pru_accept = tcp6_usr_accept, 1216 .pru_attach = tcp_usr_attach, 1217 .pru_bind = tcp6_usr_bind, 1218 .pru_connect = tcp6_usr_connect, 1219 .pru_control = in6_control, 1220 .pru_detach = tcp_usr_detach, 1221 .pru_disconnect = tcp_usr_disconnect, 1222 .pru_listen = tcp6_usr_listen, 1223 .pru_peeraddr = in6_mapped_peeraddr, 1224 .pru_rcvd = tcp_usr_rcvd, 1225 .pru_rcvoob = tcp_usr_rcvoob, 1226 .pru_send = tcp_usr_send, 1227 .pru_ready = tcp_usr_ready, 1228 .pru_shutdown = tcp_usr_shutdown, 1229 .pru_sockaddr = in6_mapped_sockaddr, 1230 .pru_sosetlabel = in_pcbsosetlabel, 1231 .pru_close = tcp_usr_close, 1232 }; 1233 #endif /* INET6 */ 1234 1235 #ifdef INET 1236 /* 1237 * Common subroutine to open a TCP connection to remote host specified 1238 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1239 * port number if needed. Call in_pcbconnect_setup to do the routing and 1240 * to choose a local host address (interface). If there is an existing 1241 * incarnation of the same connection in TIME-WAIT state and if the remote 1242 * host was sending CC options and if the connection duration was < MSL, then 1243 * truncate the previous TIME-WAIT state and proceed. 1244 * Initialize connection parameters and enter SYN-SENT state. 1245 */ 1246 static int 1247 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1248 { 1249 struct inpcb *inp = tp->t_inpcb, *oinp; 1250 struct socket *so = inp->inp_socket; 1251 struct in_addr laddr; 1252 u_short lport; 1253 int error; 1254 1255 INP_WLOCK_ASSERT(inp); 1256 INP_HASH_WLOCK(&V_tcbinfo); 1257 1258 if (inp->inp_lport == 0) { 1259 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1260 if (error) 1261 goto out; 1262 } 1263 1264 /* 1265 * Cannot simply call in_pcbconnect, because there might be an 1266 * earlier incarnation of this same connection still in 1267 * TIME_WAIT state, creating an ADDRINUSE error. 1268 */ 1269 laddr = inp->inp_laddr; 1270 lport = inp->inp_lport; 1271 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1272 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1273 if (error && oinp == NULL) 1274 goto out; 1275 if (oinp) { 1276 error = EADDRINUSE; 1277 goto out; 1278 } 1279 inp->inp_laddr = laddr; 1280 in_pcbrehash(inp); 1281 INP_HASH_WUNLOCK(&V_tcbinfo); 1282 1283 /* 1284 * Compute window scaling to request: 1285 * Scale to fit into sweet spot. See tcp_syncache.c. 1286 * XXX: This should move to tcp_output(). 1287 */ 1288 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1289 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1290 tp->request_r_scale++; 1291 1292 soisconnecting(so); 1293 TCPSTAT_INC(tcps_connattempt); 1294 tcp_state_change(tp, TCPS_SYN_SENT); 1295 tp->iss = tcp_new_isn(tp); 1296 tcp_sendseqinit(tp); 1297 1298 return 0; 1299 1300 out: 1301 INP_HASH_WUNLOCK(&V_tcbinfo); 1302 return (error); 1303 } 1304 #endif /* INET */ 1305 1306 #ifdef INET6 1307 static int 1308 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1309 { 1310 struct inpcb *inp = tp->t_inpcb; 1311 int error; 1312 1313 INP_WLOCK_ASSERT(inp); 1314 INP_HASH_WLOCK(&V_tcbinfo); 1315 1316 if (inp->inp_lport == 0) { 1317 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1318 if (error) 1319 goto out; 1320 } 1321 error = in6_pcbconnect(inp, nam, td->td_ucred); 1322 if (error != 0) 1323 goto out; 1324 INP_HASH_WUNLOCK(&V_tcbinfo); 1325 1326 /* Compute window scaling to request. */ 1327 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1328 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1329 tp->request_r_scale++; 1330 1331 soisconnecting(inp->inp_socket); 1332 TCPSTAT_INC(tcps_connattempt); 1333 tcp_state_change(tp, TCPS_SYN_SENT); 1334 tp->iss = tcp_new_isn(tp); 1335 tcp_sendseqinit(tp); 1336 1337 return 0; 1338 1339 out: 1340 INP_HASH_WUNLOCK(&V_tcbinfo); 1341 return error; 1342 } 1343 #endif /* INET6 */ 1344 1345 /* 1346 * Export TCP internal state information via a struct tcp_info, based on the 1347 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1348 * (TCP state machine, etc). We export all information using FreeBSD-native 1349 * constants -- for example, the numeric values for tcpi_state will differ 1350 * from Linux. 1351 */ 1352 static void 1353 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1354 { 1355 1356 INP_WLOCK_ASSERT(tp->t_inpcb); 1357 bzero(ti, sizeof(*ti)); 1358 1359 ti->tcpi_state = tp->t_state; 1360 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1361 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1362 if (tp->t_flags & TF_SACK_PERMIT) 1363 ti->tcpi_options |= TCPI_OPT_SACK; 1364 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1365 ti->tcpi_options |= TCPI_OPT_WSCALE; 1366 ti->tcpi_snd_wscale = tp->snd_scale; 1367 ti->tcpi_rcv_wscale = tp->rcv_scale; 1368 } 1369 if (tp->t_flags & TF_ECN_PERMIT) 1370 ti->tcpi_options |= TCPI_OPT_ECN; 1371 1372 ti->tcpi_rto = tp->t_rxtcur * tick; 1373 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1374 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1375 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1376 1377 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1378 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1379 1380 /* 1381 * FreeBSD-specific extension fields for tcp_info. 1382 */ 1383 ti->tcpi_rcv_space = tp->rcv_wnd; 1384 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1385 ti->tcpi_snd_wnd = tp->snd_wnd; 1386 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1387 ti->tcpi_snd_nxt = tp->snd_nxt; 1388 ti->tcpi_snd_mss = tp->t_maxseg; 1389 ti->tcpi_rcv_mss = tp->t_maxseg; 1390 if (tp->t_flags & TF_TOE) 1391 ti->tcpi_options |= TCPI_OPT_TOE; 1392 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1393 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1394 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1395 } 1396 1397 /* 1398 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1399 * socket option arguments. When it re-acquires the lock after the copy, it 1400 * has to revalidate that the connection is still valid for the socket 1401 * option. 1402 */ 1403 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1404 INP_WLOCK(inp); \ 1405 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1406 INP_WUNLOCK(inp); \ 1407 cleanup; \ 1408 return (ECONNRESET); \ 1409 } \ 1410 tp = intotcpcb(inp); \ 1411 } while(0) 1412 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1413 1414 int 1415 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1416 { 1417 int error; 1418 struct inpcb *inp; 1419 struct tcpcb *tp; 1420 struct tcp_function_block *blk; 1421 struct tcp_function_set fsn; 1422 1423 error = 0; 1424 inp = sotoinpcb(so); 1425 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1426 INP_WLOCK(inp); 1427 if (sopt->sopt_level != IPPROTO_TCP) { 1428 #ifdef INET6 1429 if (inp->inp_vflag & INP_IPV6PROTO) { 1430 INP_WUNLOCK(inp); 1431 error = ip6_ctloutput(so, sopt); 1432 } 1433 #endif /* INET6 */ 1434 #if defined(INET6) && defined(INET) 1435 else 1436 #endif 1437 #ifdef INET 1438 { 1439 INP_WUNLOCK(inp); 1440 error = ip_ctloutput(so, sopt); 1441 } 1442 #endif 1443 return (error); 1444 } 1445 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1446 INP_WUNLOCK(inp); 1447 return (ECONNRESET); 1448 } 1449 tp = intotcpcb(inp); 1450 /* 1451 * Protect the TCP option TCP_FUNCTION_BLK so 1452 * that a sub-function can *never* overwrite this. 1453 */ 1454 if ((sopt->sopt_dir == SOPT_SET) && 1455 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1456 INP_WUNLOCK(inp); 1457 error = sooptcopyin(sopt, &fsn, sizeof fsn, 1458 sizeof fsn); 1459 if (error) 1460 return (error); 1461 INP_WLOCK_RECHECK(inp); 1462 blk = find_and_ref_tcp_functions(&fsn); 1463 if (blk == NULL) { 1464 INP_WUNLOCK(inp); 1465 return (ENOENT); 1466 } 1467 if (tp->t_fb == blk) { 1468 /* You already have this */ 1469 refcount_release(&blk->tfb_refcnt); 1470 INP_WUNLOCK(inp); 1471 return (0); 1472 } 1473 if (tp->t_state != TCPS_CLOSED) { 1474 int error=EINVAL; 1475 /* 1476 * The user has advanced the state 1477 * past the initial point, we may not 1478 * be able to switch. 1479 */ 1480 if (blk->tfb_tcp_handoff_ok != NULL) { 1481 /* 1482 * Does the stack provide a 1483 * query mechanism, if so it may 1484 * still be possible? 1485 */ 1486 error = (*blk->tfb_tcp_handoff_ok)(tp); 1487 } 1488 if (error) { 1489 refcount_release(&blk->tfb_refcnt); 1490 INP_WUNLOCK(inp); 1491 return(error); 1492 } 1493 } 1494 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1495 refcount_release(&blk->tfb_refcnt); 1496 INP_WUNLOCK(inp); 1497 return (ENOENT); 1498 } 1499 /* 1500 * Release the old refcnt, the 1501 * lookup acquired a ref on the 1502 * new one already. 1503 */ 1504 if (tp->t_fb->tfb_tcp_fb_fini) { 1505 /* 1506 * Tell the stack to cleanup with 0 i.e. 1507 * the tcb is not going away. 1508 */ 1509 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1510 } 1511 refcount_release(&tp->t_fb->tfb_refcnt); 1512 tp->t_fb = blk; 1513 if (tp->t_fb->tfb_tcp_fb_init) { 1514 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1515 } 1516 #ifdef TCP_OFFLOAD 1517 if (tp->t_flags & TF_TOE) { 1518 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1519 sopt->sopt_name); 1520 } 1521 #endif 1522 INP_WUNLOCK(inp); 1523 return (error); 1524 } else if ((sopt->sopt_dir == SOPT_GET) && 1525 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1526 strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name); 1527 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1528 INP_WUNLOCK(inp); 1529 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1530 return (error); 1531 } 1532 /* Pass in the INP locked, called must unlock it */ 1533 return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp)); 1534 } 1535 1536 int 1537 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 1538 { 1539 int error, opt, optval; 1540 u_int ui; 1541 struct tcp_info ti; 1542 struct cc_algo *algo; 1543 char *pbuf, buf[TCP_CA_NAME_MAX]; 1544 size_t len; 1545 1546 /* 1547 * For TCP_CCALGOOPT forward the control to CC module, for both 1548 * SOPT_SET and SOPT_GET. 1549 */ 1550 switch (sopt->sopt_name) { 1551 case TCP_CCALGOOPT: 1552 INP_WUNLOCK(inp); 1553 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 1554 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 1555 sopt->sopt_valsize); 1556 if (error) { 1557 free(pbuf, M_TEMP); 1558 return (error); 1559 } 1560 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 1561 if (CC_ALGO(tp)->ctl_output != NULL) 1562 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 1563 else 1564 error = ENOENT; 1565 INP_WUNLOCK(inp); 1566 if (error == 0 && sopt->sopt_dir == SOPT_GET) 1567 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 1568 free(pbuf, M_TEMP); 1569 return (error); 1570 } 1571 1572 switch (sopt->sopt_dir) { 1573 case SOPT_SET: 1574 switch (sopt->sopt_name) { 1575 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1576 case TCP_MD5SIG: 1577 if (!TCPMD5_ENABLED()) { 1578 INP_WUNLOCK(inp); 1579 return (ENOPROTOOPT); 1580 } 1581 error = TCPMD5_PCBCTL(inp, sopt); 1582 if (error) 1583 return (error); 1584 goto unlock_and_done; 1585 #endif /* IPSEC */ 1586 1587 case TCP_NODELAY: 1588 case TCP_NOOPT: 1589 INP_WUNLOCK(inp); 1590 error = sooptcopyin(sopt, &optval, sizeof optval, 1591 sizeof optval); 1592 if (error) 1593 return (error); 1594 1595 INP_WLOCK_RECHECK(inp); 1596 switch (sopt->sopt_name) { 1597 case TCP_NODELAY: 1598 opt = TF_NODELAY; 1599 break; 1600 case TCP_NOOPT: 1601 opt = TF_NOOPT; 1602 break; 1603 default: 1604 opt = 0; /* dead code to fool gcc */ 1605 break; 1606 } 1607 1608 if (optval) 1609 tp->t_flags |= opt; 1610 else 1611 tp->t_flags &= ~opt; 1612 unlock_and_done: 1613 #ifdef TCP_OFFLOAD 1614 if (tp->t_flags & TF_TOE) { 1615 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1616 sopt->sopt_name); 1617 } 1618 #endif 1619 INP_WUNLOCK(inp); 1620 break; 1621 1622 case TCP_NOPUSH: 1623 INP_WUNLOCK(inp); 1624 error = sooptcopyin(sopt, &optval, sizeof optval, 1625 sizeof optval); 1626 if (error) 1627 return (error); 1628 1629 INP_WLOCK_RECHECK(inp); 1630 if (optval) 1631 tp->t_flags |= TF_NOPUSH; 1632 else if (tp->t_flags & TF_NOPUSH) { 1633 tp->t_flags &= ~TF_NOPUSH; 1634 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1635 error = tp->t_fb->tfb_tcp_output(tp); 1636 } 1637 goto unlock_and_done; 1638 1639 case TCP_MAXSEG: 1640 INP_WUNLOCK(inp); 1641 error = sooptcopyin(sopt, &optval, sizeof optval, 1642 sizeof optval); 1643 if (error) 1644 return (error); 1645 1646 INP_WLOCK_RECHECK(inp); 1647 if (optval > 0 && optval <= tp->t_maxseg && 1648 optval + 40 >= V_tcp_minmss) 1649 tp->t_maxseg = optval; 1650 else 1651 error = EINVAL; 1652 goto unlock_and_done; 1653 1654 case TCP_INFO: 1655 INP_WUNLOCK(inp); 1656 error = EINVAL; 1657 break; 1658 1659 case TCP_CONGESTION: 1660 INP_WUNLOCK(inp); 1661 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1662 if (error) 1663 break; 1664 buf[sopt->sopt_valsize] = '\0'; 1665 INP_WLOCK_RECHECK(inp); 1666 CC_LIST_RLOCK(); 1667 STAILQ_FOREACH(algo, &cc_list, entries) 1668 if (strncmp(buf, algo->name, 1669 TCP_CA_NAME_MAX) == 0) 1670 break; 1671 CC_LIST_RUNLOCK(); 1672 if (algo == NULL) { 1673 INP_WUNLOCK(inp); 1674 error = EINVAL; 1675 break; 1676 } 1677 /* 1678 * We hold a write lock over the tcb so it's safe to 1679 * do these things without ordering concerns. 1680 */ 1681 if (CC_ALGO(tp)->cb_destroy != NULL) 1682 CC_ALGO(tp)->cb_destroy(tp->ccv); 1683 CC_ALGO(tp) = algo; 1684 /* 1685 * If something goes pear shaped initialising the new 1686 * algo, fall back to newreno (which does not 1687 * require initialisation). 1688 */ 1689 if (algo->cb_init != NULL && 1690 algo->cb_init(tp->ccv) != 0) { 1691 CC_ALGO(tp) = &newreno_cc_algo; 1692 /* 1693 * The only reason init should fail is 1694 * because of malloc. 1695 */ 1696 error = ENOMEM; 1697 } 1698 INP_WUNLOCK(inp); 1699 break; 1700 1701 case TCP_KEEPIDLE: 1702 case TCP_KEEPINTVL: 1703 case TCP_KEEPINIT: 1704 INP_WUNLOCK(inp); 1705 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 1706 if (error) 1707 return (error); 1708 1709 if (ui > (UINT_MAX / hz)) { 1710 error = EINVAL; 1711 break; 1712 } 1713 ui *= hz; 1714 1715 INP_WLOCK_RECHECK(inp); 1716 switch (sopt->sopt_name) { 1717 case TCP_KEEPIDLE: 1718 tp->t_keepidle = ui; 1719 /* 1720 * XXX: better check current remaining 1721 * timeout and "merge" it with new value. 1722 */ 1723 if ((tp->t_state > TCPS_LISTEN) && 1724 (tp->t_state <= TCPS_CLOSING)) 1725 tcp_timer_activate(tp, TT_KEEP, 1726 TP_KEEPIDLE(tp)); 1727 break; 1728 case TCP_KEEPINTVL: 1729 tp->t_keepintvl = ui; 1730 if ((tp->t_state == TCPS_FIN_WAIT_2) && 1731 (TP_MAXIDLE(tp) > 0)) 1732 tcp_timer_activate(tp, TT_2MSL, 1733 TP_MAXIDLE(tp)); 1734 break; 1735 case TCP_KEEPINIT: 1736 tp->t_keepinit = ui; 1737 if (tp->t_state == TCPS_SYN_RECEIVED || 1738 tp->t_state == TCPS_SYN_SENT) 1739 tcp_timer_activate(tp, TT_KEEP, 1740 TP_KEEPINIT(tp)); 1741 break; 1742 } 1743 goto unlock_and_done; 1744 1745 case TCP_KEEPCNT: 1746 INP_WUNLOCK(inp); 1747 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 1748 if (error) 1749 return (error); 1750 1751 INP_WLOCK_RECHECK(inp); 1752 tp->t_keepcnt = ui; 1753 if ((tp->t_state == TCPS_FIN_WAIT_2) && 1754 (TP_MAXIDLE(tp) > 0)) 1755 tcp_timer_activate(tp, TT_2MSL, 1756 TP_MAXIDLE(tp)); 1757 goto unlock_and_done; 1758 1759 #ifdef TCPPCAP 1760 case TCP_PCAP_OUT: 1761 case TCP_PCAP_IN: 1762 INP_WUNLOCK(inp); 1763 error = sooptcopyin(sopt, &optval, sizeof optval, 1764 sizeof optval); 1765 if (error) 1766 return (error); 1767 1768 INP_WLOCK_RECHECK(inp); 1769 if (optval >= 0) 1770 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 1771 &(tp->t_outpkts) : &(tp->t_inpkts), 1772 optval); 1773 else 1774 error = EINVAL; 1775 goto unlock_and_done; 1776 #endif 1777 1778 case TCP_FASTOPEN: { 1779 struct tcp_fastopen tfo_optval; 1780 1781 INP_WUNLOCK(inp); 1782 if (!V_tcp_fastopen_client_enable && 1783 !V_tcp_fastopen_server_enable) 1784 return (EPERM); 1785 1786 error = sooptcopyin(sopt, &tfo_optval, 1787 sizeof(tfo_optval), sizeof(int)); 1788 if (error) 1789 return (error); 1790 1791 INP_WLOCK_RECHECK(inp); 1792 if (tfo_optval.enable) { 1793 if (tp->t_state == TCPS_LISTEN) { 1794 if (!V_tcp_fastopen_server_enable) { 1795 error = EPERM; 1796 goto unlock_and_done; 1797 } 1798 1799 tp->t_flags |= TF_FASTOPEN; 1800 if (tp->t_tfo_pending == NULL) 1801 tp->t_tfo_pending = 1802 tcp_fastopen_alloc_counter(); 1803 } else { 1804 /* 1805 * If a pre-shared key was provided, 1806 * stash it in the client cookie 1807 * field of the tcpcb for use during 1808 * connect. 1809 */ 1810 if (sopt->sopt_valsize == 1811 sizeof(tfo_optval)) { 1812 memcpy(tp->t_tfo_cookie.client, 1813 tfo_optval.psk, 1814 TCP_FASTOPEN_PSK_LEN); 1815 tp->t_tfo_client_cookie_len = 1816 TCP_FASTOPEN_PSK_LEN; 1817 } 1818 tp->t_flags |= TF_FASTOPEN; 1819 } 1820 } else 1821 tp->t_flags &= ~TF_FASTOPEN; 1822 goto unlock_and_done; 1823 } 1824 1825 default: 1826 INP_WUNLOCK(inp); 1827 error = ENOPROTOOPT; 1828 break; 1829 } 1830 break; 1831 1832 case SOPT_GET: 1833 tp = intotcpcb(inp); 1834 switch (sopt->sopt_name) { 1835 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1836 case TCP_MD5SIG: 1837 if (!TCPMD5_ENABLED()) { 1838 INP_WUNLOCK(inp); 1839 return (ENOPROTOOPT); 1840 } 1841 error = TCPMD5_PCBCTL(inp, sopt); 1842 break; 1843 #endif 1844 1845 case TCP_NODELAY: 1846 optval = tp->t_flags & TF_NODELAY; 1847 INP_WUNLOCK(inp); 1848 error = sooptcopyout(sopt, &optval, sizeof optval); 1849 break; 1850 case TCP_MAXSEG: 1851 optval = tp->t_maxseg; 1852 INP_WUNLOCK(inp); 1853 error = sooptcopyout(sopt, &optval, sizeof optval); 1854 break; 1855 case TCP_NOOPT: 1856 optval = tp->t_flags & TF_NOOPT; 1857 INP_WUNLOCK(inp); 1858 error = sooptcopyout(sopt, &optval, sizeof optval); 1859 break; 1860 case TCP_NOPUSH: 1861 optval = tp->t_flags & TF_NOPUSH; 1862 INP_WUNLOCK(inp); 1863 error = sooptcopyout(sopt, &optval, sizeof optval); 1864 break; 1865 case TCP_INFO: 1866 tcp_fill_info(tp, &ti); 1867 INP_WUNLOCK(inp); 1868 error = sooptcopyout(sopt, &ti, sizeof ti); 1869 break; 1870 case TCP_CONGESTION: 1871 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 1872 INP_WUNLOCK(inp); 1873 error = sooptcopyout(sopt, buf, len + 1); 1874 break; 1875 case TCP_KEEPIDLE: 1876 case TCP_KEEPINTVL: 1877 case TCP_KEEPINIT: 1878 case TCP_KEEPCNT: 1879 switch (sopt->sopt_name) { 1880 case TCP_KEEPIDLE: 1881 ui = TP_KEEPIDLE(tp) / hz; 1882 break; 1883 case TCP_KEEPINTVL: 1884 ui = TP_KEEPINTVL(tp) / hz; 1885 break; 1886 case TCP_KEEPINIT: 1887 ui = TP_KEEPINIT(tp) / hz; 1888 break; 1889 case TCP_KEEPCNT: 1890 ui = TP_KEEPCNT(tp); 1891 break; 1892 } 1893 INP_WUNLOCK(inp); 1894 error = sooptcopyout(sopt, &ui, sizeof(ui)); 1895 break; 1896 #ifdef TCPPCAP 1897 case TCP_PCAP_OUT: 1898 case TCP_PCAP_IN: 1899 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 1900 &(tp->t_outpkts) : &(tp->t_inpkts)); 1901 INP_WUNLOCK(inp); 1902 error = sooptcopyout(sopt, &optval, sizeof optval); 1903 break; 1904 #endif 1905 case TCP_FASTOPEN: 1906 optval = tp->t_flags & TF_FASTOPEN; 1907 INP_WUNLOCK(inp); 1908 error = sooptcopyout(sopt, &optval, sizeof optval); 1909 break; 1910 default: 1911 INP_WUNLOCK(inp); 1912 error = ENOPROTOOPT; 1913 break; 1914 } 1915 break; 1916 } 1917 return (error); 1918 } 1919 #undef INP_WLOCK_RECHECK 1920 #undef INP_WLOCK_RECHECK_CLEANUP 1921 1922 /* 1923 * Attach TCP protocol to socket, allocating 1924 * internet protocol control block, tcp control block, 1925 * bufer space, and entering LISTEN state if to accept connections. 1926 */ 1927 static int 1928 tcp_attach(struct socket *so) 1929 { 1930 struct tcpcb *tp; 1931 struct inpcb *inp; 1932 int error; 1933 1934 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 1935 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 1936 if (error) 1937 return (error); 1938 } 1939 so->so_rcv.sb_flags |= SB_AUTOSIZE; 1940 so->so_snd.sb_flags |= SB_AUTOSIZE; 1941 INP_INFO_RLOCK(&V_tcbinfo); 1942 error = in_pcballoc(so, &V_tcbinfo); 1943 if (error) { 1944 INP_INFO_RUNLOCK(&V_tcbinfo); 1945 return (error); 1946 } 1947 inp = sotoinpcb(so); 1948 #ifdef INET6 1949 if (inp->inp_vflag & INP_IPV6PROTO) { 1950 inp->inp_vflag |= INP_IPV6; 1951 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 1952 inp->inp_vflag |= INP_IPV4; 1953 inp->in6p_hops = -1; /* use kernel default */ 1954 } 1955 else 1956 #endif 1957 inp->inp_vflag |= INP_IPV4; 1958 tp = tcp_newtcpcb(inp); 1959 if (tp == NULL) { 1960 in_pcbdetach(inp); 1961 in_pcbfree(inp); 1962 INP_INFO_RUNLOCK(&V_tcbinfo); 1963 return (ENOBUFS); 1964 } 1965 tp->t_state = TCPS_CLOSED; 1966 INP_WUNLOCK(inp); 1967 INP_INFO_RUNLOCK(&V_tcbinfo); 1968 TCPSTATES_INC(TCPS_CLOSED); 1969 return (0); 1970 } 1971 1972 /* 1973 * Initiate (or continue) disconnect. 1974 * If embryonic state, just send reset (once). 1975 * If in ``let data drain'' option and linger null, just drop. 1976 * Otherwise (hard), mark socket disconnecting and drop 1977 * current input data; switch states based on user close, and 1978 * send segment to peer (with FIN). 1979 */ 1980 static void 1981 tcp_disconnect(struct tcpcb *tp) 1982 { 1983 struct inpcb *inp = tp->t_inpcb; 1984 struct socket *so = inp->inp_socket; 1985 1986 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1987 INP_WLOCK_ASSERT(inp); 1988 1989 /* 1990 * Neither tcp_close() nor tcp_drop() should return NULL, as the 1991 * socket is still open. 1992 */ 1993 if (tp->t_state < TCPS_ESTABLISHED) { 1994 tp = tcp_close(tp); 1995 KASSERT(tp != NULL, 1996 ("tcp_disconnect: tcp_close() returned NULL")); 1997 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1998 tp = tcp_drop(tp, 0); 1999 KASSERT(tp != NULL, 2000 ("tcp_disconnect: tcp_drop() returned NULL")); 2001 } else { 2002 soisdisconnecting(so); 2003 sbflush(&so->so_rcv); 2004 tcp_usrclosed(tp); 2005 if (!(inp->inp_flags & INP_DROPPED)) 2006 tp->t_fb->tfb_tcp_output(tp); 2007 } 2008 } 2009 2010 /* 2011 * User issued close, and wish to trail through shutdown states: 2012 * if never received SYN, just forget it. If got a SYN from peer, 2013 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2014 * If already got a FIN from peer, then almost done; go to LAST_ACK 2015 * state. In all other cases, have already sent FIN to peer (e.g. 2016 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2017 * for peer to send FIN or not respond to keep-alives, etc. 2018 * We can let the user exit from the close as soon as the FIN is acked. 2019 */ 2020 static void 2021 tcp_usrclosed(struct tcpcb *tp) 2022 { 2023 2024 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2025 INP_WLOCK_ASSERT(tp->t_inpcb); 2026 2027 switch (tp->t_state) { 2028 case TCPS_LISTEN: 2029 #ifdef TCP_OFFLOAD 2030 tcp_offload_listen_stop(tp); 2031 #endif 2032 tcp_state_change(tp, TCPS_CLOSED); 2033 /* FALLTHROUGH */ 2034 case TCPS_CLOSED: 2035 tp = tcp_close(tp); 2036 /* 2037 * tcp_close() should never return NULL here as the socket is 2038 * still open. 2039 */ 2040 KASSERT(tp != NULL, 2041 ("tcp_usrclosed: tcp_close() returned NULL")); 2042 break; 2043 2044 case TCPS_SYN_SENT: 2045 case TCPS_SYN_RECEIVED: 2046 tp->t_flags |= TF_NEEDFIN; 2047 break; 2048 2049 case TCPS_ESTABLISHED: 2050 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2051 break; 2052 2053 case TCPS_CLOSE_WAIT: 2054 tcp_state_change(tp, TCPS_LAST_ACK); 2055 break; 2056 } 2057 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2058 soisdisconnected(tp->t_inpcb->inp_socket); 2059 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2060 if (tp->t_state == TCPS_FIN_WAIT_2) { 2061 int timeout; 2062 2063 timeout = (tcp_fast_finwait2_recycle) ? 2064 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2065 tcp_timer_activate(tp, TT_2MSL, timeout); 2066 } 2067 } 2068 } 2069 2070 #ifdef DDB 2071 static void 2072 db_print_indent(int indent) 2073 { 2074 int i; 2075 2076 for (i = 0; i < indent; i++) 2077 db_printf(" "); 2078 } 2079 2080 static void 2081 db_print_tstate(int t_state) 2082 { 2083 2084 switch (t_state) { 2085 case TCPS_CLOSED: 2086 db_printf("TCPS_CLOSED"); 2087 return; 2088 2089 case TCPS_LISTEN: 2090 db_printf("TCPS_LISTEN"); 2091 return; 2092 2093 case TCPS_SYN_SENT: 2094 db_printf("TCPS_SYN_SENT"); 2095 return; 2096 2097 case TCPS_SYN_RECEIVED: 2098 db_printf("TCPS_SYN_RECEIVED"); 2099 return; 2100 2101 case TCPS_ESTABLISHED: 2102 db_printf("TCPS_ESTABLISHED"); 2103 return; 2104 2105 case TCPS_CLOSE_WAIT: 2106 db_printf("TCPS_CLOSE_WAIT"); 2107 return; 2108 2109 case TCPS_FIN_WAIT_1: 2110 db_printf("TCPS_FIN_WAIT_1"); 2111 return; 2112 2113 case TCPS_CLOSING: 2114 db_printf("TCPS_CLOSING"); 2115 return; 2116 2117 case TCPS_LAST_ACK: 2118 db_printf("TCPS_LAST_ACK"); 2119 return; 2120 2121 case TCPS_FIN_WAIT_2: 2122 db_printf("TCPS_FIN_WAIT_2"); 2123 return; 2124 2125 case TCPS_TIME_WAIT: 2126 db_printf("TCPS_TIME_WAIT"); 2127 return; 2128 2129 default: 2130 db_printf("unknown"); 2131 return; 2132 } 2133 } 2134 2135 static void 2136 db_print_tflags(u_int t_flags) 2137 { 2138 int comma; 2139 2140 comma = 0; 2141 if (t_flags & TF_ACKNOW) { 2142 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2143 comma = 1; 2144 } 2145 if (t_flags & TF_DELACK) { 2146 db_printf("%sTF_DELACK", comma ? ", " : ""); 2147 comma = 1; 2148 } 2149 if (t_flags & TF_NODELAY) { 2150 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2151 comma = 1; 2152 } 2153 if (t_flags & TF_NOOPT) { 2154 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2155 comma = 1; 2156 } 2157 if (t_flags & TF_SENTFIN) { 2158 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2159 comma = 1; 2160 } 2161 if (t_flags & TF_REQ_SCALE) { 2162 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2163 comma = 1; 2164 } 2165 if (t_flags & TF_RCVD_SCALE) { 2166 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2167 comma = 1; 2168 } 2169 if (t_flags & TF_REQ_TSTMP) { 2170 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2171 comma = 1; 2172 } 2173 if (t_flags & TF_RCVD_TSTMP) { 2174 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2175 comma = 1; 2176 } 2177 if (t_flags & TF_SACK_PERMIT) { 2178 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2179 comma = 1; 2180 } 2181 if (t_flags & TF_NEEDSYN) { 2182 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2183 comma = 1; 2184 } 2185 if (t_flags & TF_NEEDFIN) { 2186 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2187 comma = 1; 2188 } 2189 if (t_flags & TF_NOPUSH) { 2190 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2191 comma = 1; 2192 } 2193 if (t_flags & TF_MORETOCOME) { 2194 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2195 comma = 1; 2196 } 2197 if (t_flags & TF_LQ_OVERFLOW) { 2198 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2199 comma = 1; 2200 } 2201 if (t_flags & TF_LASTIDLE) { 2202 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2203 comma = 1; 2204 } 2205 if (t_flags & TF_RXWIN0SENT) { 2206 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2207 comma = 1; 2208 } 2209 if (t_flags & TF_FASTRECOVERY) { 2210 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2211 comma = 1; 2212 } 2213 if (t_flags & TF_CONGRECOVERY) { 2214 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2215 comma = 1; 2216 } 2217 if (t_flags & TF_WASFRECOVERY) { 2218 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2219 comma = 1; 2220 } 2221 if (t_flags & TF_SIGNATURE) { 2222 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2223 comma = 1; 2224 } 2225 if (t_flags & TF_FORCEDATA) { 2226 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2227 comma = 1; 2228 } 2229 if (t_flags & TF_TSO) { 2230 db_printf("%sTF_TSO", comma ? ", " : ""); 2231 comma = 1; 2232 } 2233 if (t_flags & TF_ECN_PERMIT) { 2234 db_printf("%sTF_ECN_PERMIT", comma ? ", " : ""); 2235 comma = 1; 2236 } 2237 if (t_flags & TF_FASTOPEN) { 2238 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2239 comma = 1; 2240 } 2241 } 2242 2243 static void 2244 db_print_toobflags(char t_oobflags) 2245 { 2246 int comma; 2247 2248 comma = 0; 2249 if (t_oobflags & TCPOOB_HAVEDATA) { 2250 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 2251 comma = 1; 2252 } 2253 if (t_oobflags & TCPOOB_HADDATA) { 2254 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 2255 comma = 1; 2256 } 2257 } 2258 2259 static void 2260 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 2261 { 2262 2263 db_print_indent(indent); 2264 db_printf("%s at %p\n", name, tp); 2265 2266 indent += 2; 2267 2268 db_print_indent(indent); 2269 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 2270 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 2271 2272 db_print_indent(indent); 2273 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 2274 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 2275 2276 db_print_indent(indent); 2277 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 2278 &tp->t_timers->tt_delack, tp->t_inpcb); 2279 2280 db_print_indent(indent); 2281 db_printf("t_state: %d (", tp->t_state); 2282 db_print_tstate(tp->t_state); 2283 db_printf(")\n"); 2284 2285 db_print_indent(indent); 2286 db_printf("t_flags: 0x%x (", tp->t_flags); 2287 db_print_tflags(tp->t_flags); 2288 db_printf(")\n"); 2289 2290 db_print_indent(indent); 2291 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 2292 tp->snd_una, tp->snd_max, tp->snd_nxt); 2293 2294 db_print_indent(indent); 2295 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 2296 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 2297 2298 db_print_indent(indent); 2299 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 2300 tp->iss, tp->irs, tp->rcv_nxt); 2301 2302 db_print_indent(indent); 2303 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 2304 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 2305 2306 db_print_indent(indent); 2307 db_printf("snd_wnd: %u snd_cwnd: %u\n", 2308 tp->snd_wnd, tp->snd_cwnd); 2309 2310 db_print_indent(indent); 2311 db_printf("snd_ssthresh: %u snd_recover: " 2312 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 2313 2314 db_print_indent(indent); 2315 db_printf("t_rcvtime: %u t_startime: %u\n", 2316 tp->t_rcvtime, tp->t_starttime); 2317 2318 db_print_indent(indent); 2319 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 2320 tp->t_rtttime, tp->t_rtseq); 2321 2322 db_print_indent(indent); 2323 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 2324 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 2325 2326 db_print_indent(indent); 2327 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 2328 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 2329 tp->t_rttbest); 2330 2331 db_print_indent(indent); 2332 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 2333 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 2334 2335 db_print_indent(indent); 2336 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 2337 db_print_toobflags(tp->t_oobflags); 2338 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 2339 2340 db_print_indent(indent); 2341 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 2342 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 2343 2344 db_print_indent(indent); 2345 db_printf("ts_recent: %u ts_recent_age: %u\n", 2346 tp->ts_recent, tp->ts_recent_age); 2347 2348 db_print_indent(indent); 2349 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 2350 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 2351 2352 db_print_indent(indent); 2353 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 2354 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 2355 tp->snd_recover_prev, tp->t_badrxtwin); 2356 2357 db_print_indent(indent); 2358 db_printf("snd_numholes: %d snd_holes first: %p\n", 2359 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 2360 2361 db_print_indent(indent); 2362 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: " 2363 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata); 2364 2365 /* Skip sackblks, sackhint. */ 2366 2367 db_print_indent(indent); 2368 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 2369 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 2370 } 2371 2372 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 2373 { 2374 struct tcpcb *tp; 2375 2376 if (!have_addr) { 2377 db_printf("usage: show tcpcb <addr>\n"); 2378 return; 2379 } 2380 tp = (struct tcpcb *)addr; 2381 2382 db_print_tcpcb(tp, "tcpcb", 0); 2383 } 2384 #endif 2385