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_kern_tls.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/arb.h> 52 #include <sys/limits.h> 53 #include <sys/malloc.h> 54 #include <sys/refcount.h> 55 #include <sys/kernel.h> 56 #include <sys/ktls.h> 57 #include <sys/qmath.h> 58 #include <sys/sysctl.h> 59 #include <sys/mbuf.h> 60 #ifdef INET6 61 #include <sys/domain.h> 62 #endif /* INET6 */ 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/protosw.h> 66 #include <sys/proc.h> 67 #include <sys/jail.h> 68 #include <sys/stats.h> 69 70 #ifdef DDB 71 #include <ddb/ddb.h> 72 #endif 73 74 #include <net/if.h> 75 #include <net/if_var.h> 76 #include <net/route.h> 77 #include <net/vnet.h> 78 79 #include <netinet/in.h> 80 #include <netinet/in_kdtrace.h> 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_systm.h> 83 #include <netinet/in_var.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_var.h> 86 #ifdef INET6 87 #include <netinet/ip6.h> 88 #include <netinet6/in6_pcb.h> 89 #include <netinet6/ip6_var.h> 90 #include <netinet6/scope6_var.h> 91 #endif 92 #include <netinet/tcp.h> 93 #include <netinet/tcp_fsm.h> 94 #include <netinet/tcp_seq.h> 95 #include <netinet/tcp_timer.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/tcp_log_buf.h> 98 #include <netinet/tcpip.h> 99 #include <netinet/cc/cc.h> 100 #include <netinet/tcp_fastopen.h> 101 #include <netinet/tcp_hpts.h> 102 #ifdef TCPPCAP 103 #include <netinet/tcp_pcap.h> 104 #endif 105 #ifdef TCP_OFFLOAD 106 #include <netinet/tcp_offload.h> 107 #endif 108 #include <netipsec/ipsec_support.h> 109 110 #include <vm/vm.h> 111 #include <vm/vm_param.h> 112 #include <vm/pmap.h> 113 #include <vm/vm_extern.h> 114 #include <vm/vm_map.h> 115 #include <vm/vm_page.h> 116 117 /* 118 * TCP protocol interface to socket abstraction. 119 */ 120 #ifdef INET 121 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 122 struct thread *td); 123 #endif /* INET */ 124 #ifdef INET6 125 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 126 struct thread *td); 127 #endif /* INET6 */ 128 static void tcp_disconnect(struct tcpcb *); 129 static void tcp_usrclosed(struct tcpcb *); 130 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 131 132 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 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; 144 145 inp = sotoinpcb(so); 146 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 147 148 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 149 if (error) 150 goto out; 151 152 so->so_rcv.sb_flags |= SB_AUTOSIZE; 153 so->so_snd.sb_flags |= SB_AUTOSIZE; 154 error = in_pcballoc(so, &V_tcbinfo); 155 if (error) 156 goto out; 157 inp = sotoinpcb(so); 158 tp = tcp_newtcpcb(inp); 159 if (tp == NULL) { 160 error = ENOBUFS; 161 in_pcbdetach(inp); 162 in_pcbfree(inp); 163 goto out; 164 } 165 tp->t_state = TCPS_CLOSED; 166 INP_WUNLOCK(inp); 167 TCPSTATES_INC(TCPS_CLOSED); 168 out: 169 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 170 return (error); 171 } 172 173 /* 174 * tcp_usr_detach is called when the socket layer loses its final reference 175 * to the socket, be it a file descriptor reference, a reference from TCP, 176 * etc. At this point, there is only one case in which we will keep around 177 * inpcb state: time wait. 178 */ 179 static void 180 tcp_usr_detach(struct socket *so) 181 { 182 struct inpcb *inp; 183 struct tcpcb *tp; 184 185 inp = sotoinpcb(so); 186 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 187 INP_WLOCK(inp); 188 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 189 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 190 191 tp = intotcpcb(inp); 192 193 KASSERT(inp->inp_flags & INP_DROPPED || 194 tp->t_state < TCPS_SYN_SENT, 195 ("%s: inp %p not dropped or embryonic", __func__, inp)); 196 197 tcp_discardcb(tp); 198 in_pcbdetach(inp); 199 in_pcbfree(inp); 200 } 201 202 #ifdef INET 203 /* 204 * Give the socket an address. 205 */ 206 static int 207 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 208 { 209 int error = 0; 210 struct inpcb *inp; 211 #ifdef KDTRACE_HOOKS 212 struct tcpcb *tp = NULL; 213 #endif 214 struct sockaddr_in *sinp; 215 216 sinp = (struct sockaddr_in *)nam; 217 if (nam->sa_family != AF_INET) { 218 /* 219 * Preserve compatibility with old programs. 220 */ 221 if (nam->sa_family != AF_UNSPEC || 222 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 223 sinp->sin_addr.s_addr != INADDR_ANY) 224 return (EAFNOSUPPORT); 225 nam->sa_family = AF_INET; 226 } 227 if (nam->sa_len != sizeof(*sinp)) 228 return (EINVAL); 229 230 /* 231 * Must check for multicast addresses and disallow binding 232 * to them. 233 */ 234 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 235 return (EAFNOSUPPORT); 236 237 inp = sotoinpcb(so); 238 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 239 INP_WLOCK(inp); 240 if (inp->inp_flags & INP_DROPPED) { 241 error = EINVAL; 242 goto out; 243 } 244 #ifdef KDTRACE_HOOKS 245 tp = intotcpcb(inp); 246 #endif 247 INP_HASH_WLOCK(&V_tcbinfo); 248 error = in_pcbbind(inp, sinp, td->td_ucred); 249 INP_HASH_WUNLOCK(&V_tcbinfo); 250 out: 251 TCP_PROBE2(debug__user, tp, PRU_BIND); 252 INP_WUNLOCK(inp); 253 254 return (error); 255 } 256 #endif /* INET */ 257 258 #ifdef INET6 259 static int 260 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 261 { 262 int error = 0; 263 struct inpcb *inp; 264 #ifdef KDTRACE_HOOKS 265 struct tcpcb *tp = NULL; 266 #endif 267 struct sockaddr_in6 *sin6; 268 u_char vflagsav; 269 270 sin6 = (struct sockaddr_in6 *)nam; 271 if (nam->sa_family != AF_INET6) 272 return (EAFNOSUPPORT); 273 if (nam->sa_len != sizeof(*sin6)) 274 return (EINVAL); 275 276 /* 277 * Must check for multicast addresses and disallow binding 278 * to them. 279 */ 280 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 281 return (EAFNOSUPPORT); 282 283 inp = sotoinpcb(so); 284 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 285 INP_WLOCK(inp); 286 vflagsav = inp->inp_vflag; 287 if (inp->inp_flags & INP_DROPPED) { 288 error = EINVAL; 289 goto out; 290 } 291 #ifdef KDTRACE_HOOKS 292 tp = intotcpcb(inp); 293 #endif 294 INP_HASH_WLOCK(&V_tcbinfo); 295 inp->inp_vflag &= ~INP_IPV4; 296 inp->inp_vflag |= INP_IPV6; 297 #ifdef INET 298 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 299 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 300 inp->inp_vflag |= INP_IPV4; 301 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 302 struct sockaddr_in sin; 303 304 in6_sin6_2_sin(&sin, sin6); 305 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 306 error = EAFNOSUPPORT; 307 INP_HASH_WUNLOCK(&V_tcbinfo); 308 goto out; 309 } 310 inp->inp_vflag |= INP_IPV4; 311 inp->inp_vflag &= ~INP_IPV6; 312 error = in_pcbbind(inp, &sin, td->td_ucred); 313 INP_HASH_WUNLOCK(&V_tcbinfo); 314 goto out; 315 } 316 } 317 #endif 318 error = in6_pcbbind(inp, sin6, td->td_ucred); 319 INP_HASH_WUNLOCK(&V_tcbinfo); 320 out: 321 if (error != 0) 322 inp->inp_vflag = vflagsav; 323 TCP_PROBE2(debug__user, tp, PRU_BIND); 324 INP_WUNLOCK(inp); 325 return (error); 326 } 327 #endif /* INET6 */ 328 329 #ifdef INET 330 /* 331 * Prepare to accept connections. 332 */ 333 static int 334 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 335 { 336 int error = 0; 337 struct inpcb *inp; 338 struct tcpcb *tp = NULL; 339 340 inp = sotoinpcb(so); 341 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 342 INP_WLOCK(inp); 343 if (inp->inp_flags & INP_DROPPED) { 344 error = EINVAL; 345 goto out; 346 } 347 tp = intotcpcb(inp); 348 SOCK_LOCK(so); 349 error = solisten_proto_check(so); 350 if (error != 0) { 351 SOCK_UNLOCK(so); 352 goto out; 353 } 354 if (inp->inp_lport == 0) { 355 INP_HASH_WLOCK(&V_tcbinfo); 356 error = in_pcbbind(inp, NULL, td->td_ucred); 357 INP_HASH_WUNLOCK(&V_tcbinfo); 358 } 359 if (error == 0) { 360 tcp_state_change(tp, TCPS_LISTEN); 361 solisten_proto(so, backlog); 362 #ifdef TCP_OFFLOAD 363 if ((so->so_options & SO_NO_OFFLOAD) == 0) 364 tcp_offload_listen_start(tp); 365 #endif 366 } else { 367 solisten_proto_abort(so); 368 } 369 SOCK_UNLOCK(so); 370 371 if (IS_FASTOPEN(tp->t_flags)) 372 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 373 374 out: 375 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 376 INP_WUNLOCK(inp); 377 return (error); 378 } 379 #endif /* INET */ 380 381 #ifdef INET6 382 static int 383 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 384 { 385 int error = 0; 386 struct inpcb *inp; 387 struct tcpcb *tp = NULL; 388 u_char vflagsav; 389 390 inp = sotoinpcb(so); 391 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 392 INP_WLOCK(inp); 393 if (inp->inp_flags & INP_DROPPED) { 394 error = EINVAL; 395 goto out; 396 } 397 vflagsav = inp->inp_vflag; 398 tp = intotcpcb(inp); 399 SOCK_LOCK(so); 400 error = solisten_proto_check(so); 401 if (error != 0) { 402 SOCK_UNLOCK(so); 403 goto out; 404 } 405 INP_HASH_WLOCK(&V_tcbinfo); 406 if (inp->inp_lport == 0) { 407 inp->inp_vflag &= ~INP_IPV4; 408 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 409 inp->inp_vflag |= INP_IPV4; 410 error = in6_pcbbind(inp, NULL, td->td_ucred); 411 } 412 INP_HASH_WUNLOCK(&V_tcbinfo); 413 if (error == 0) { 414 tcp_state_change(tp, TCPS_LISTEN); 415 solisten_proto(so, backlog); 416 #ifdef TCP_OFFLOAD 417 if ((so->so_options & SO_NO_OFFLOAD) == 0) 418 tcp_offload_listen_start(tp); 419 #endif 420 } else { 421 solisten_proto_abort(so); 422 } 423 SOCK_UNLOCK(so); 424 425 if (IS_FASTOPEN(tp->t_flags)) 426 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 427 428 if (error != 0) 429 inp->inp_vflag = vflagsav; 430 431 out: 432 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 433 INP_WUNLOCK(inp); 434 return (error); 435 } 436 #endif /* INET6 */ 437 438 #ifdef INET 439 /* 440 * Initiate connection to peer. 441 * Create a template for use in transmissions on this connection. 442 * Enter SYN_SENT state, and mark socket as connecting. 443 * Start keep-alive timer, and seed output sequence space. 444 * Send initial segment on connection. 445 */ 446 static int 447 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 448 { 449 struct epoch_tracker et; 450 int error = 0; 451 struct inpcb *inp; 452 struct tcpcb *tp = NULL; 453 struct sockaddr_in *sinp; 454 455 sinp = (struct sockaddr_in *)nam; 456 if (nam->sa_family != AF_INET) 457 return (EAFNOSUPPORT); 458 if (nam->sa_len != sizeof (*sinp)) 459 return (EINVAL); 460 461 /* 462 * Must disallow TCP ``connections'' to multicast addresses. 463 */ 464 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 465 return (EAFNOSUPPORT); 466 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) 467 return (EACCES); 468 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 469 return (error); 470 471 inp = sotoinpcb(so); 472 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 473 INP_WLOCK(inp); 474 if (inp->inp_flags & INP_DROPPED) { 475 error = ECONNREFUSED; 476 goto out; 477 } 478 if (SOLISTENING(so)) { 479 error = EOPNOTSUPP; 480 goto out; 481 } 482 tp = intotcpcb(inp); 483 NET_EPOCH_ENTER(et); 484 if ((error = tcp_connect(tp, sinp, td)) != 0) 485 goto out_in_epoch; 486 #ifdef TCP_OFFLOAD 487 if (registered_toedevs > 0 && 488 (so->so_options & SO_NO_OFFLOAD) == 0 && 489 (error = tcp_offload_connect(so, nam)) == 0) 490 goto out_in_epoch; 491 #endif 492 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 493 error = tcp_output(tp); 494 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 495 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 496 out_in_epoch: 497 NET_EPOCH_EXIT(et); 498 out: 499 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 500 INP_WUNLOCK(inp); 501 return (error); 502 } 503 #endif /* INET */ 504 505 #ifdef INET6 506 static int 507 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 508 { 509 struct epoch_tracker et; 510 int error = 0; 511 struct inpcb *inp; 512 struct tcpcb *tp = NULL; 513 struct sockaddr_in6 *sin6; 514 u_int8_t incflagsav; 515 u_char vflagsav; 516 517 sin6 = (struct sockaddr_in6 *)nam; 518 if (nam->sa_family != AF_INET6) 519 return (EAFNOSUPPORT); 520 if (nam->sa_len != sizeof (*sin6)) 521 return (EINVAL); 522 523 /* 524 * Must disallow TCP ``connections'' to multicast addresses. 525 */ 526 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 527 return (EAFNOSUPPORT); 528 529 inp = sotoinpcb(so); 530 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 531 INP_WLOCK(inp); 532 vflagsav = inp->inp_vflag; 533 incflagsav = inp->inp_inc.inc_flags; 534 if (inp->inp_flags & INP_DROPPED) { 535 error = ECONNREFUSED; 536 goto out; 537 } 538 if (SOLISTENING(so)) { 539 error = EINVAL; 540 goto out; 541 } 542 tp = intotcpcb(inp); 543 #ifdef INET 544 /* 545 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 546 * therefore probably require the hash lock, which isn't held here. 547 * Is this a significant problem? 548 */ 549 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 550 struct sockaddr_in sin; 551 552 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 553 error = EINVAL; 554 goto out; 555 } 556 if ((inp->inp_vflag & INP_IPV4) == 0) { 557 error = EAFNOSUPPORT; 558 goto out; 559 } 560 561 in6_sin6_2_sin(&sin, sin6); 562 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 563 error = EAFNOSUPPORT; 564 goto out; 565 } 566 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 567 error = EACCES; 568 goto out; 569 } 570 if ((error = prison_remote_ip4(td->td_ucred, 571 &sin.sin_addr)) != 0) 572 goto out; 573 inp->inp_vflag |= INP_IPV4; 574 inp->inp_vflag &= ~INP_IPV6; 575 NET_EPOCH_ENTER(et); 576 if ((error = tcp_connect(tp, &sin, td)) != 0) 577 goto out_in_epoch; 578 #ifdef TCP_OFFLOAD 579 if (registered_toedevs > 0 && 580 (so->so_options & SO_NO_OFFLOAD) == 0 && 581 (error = tcp_offload_connect(so, nam)) == 0) 582 goto out_in_epoch; 583 #endif 584 error = tcp_output(tp); 585 goto out_in_epoch; 586 } else { 587 if ((inp->inp_vflag & INP_IPV6) == 0) { 588 error = EAFNOSUPPORT; 589 goto out; 590 } 591 } 592 #endif 593 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 594 goto out; 595 inp->inp_vflag &= ~INP_IPV4; 596 inp->inp_vflag |= INP_IPV6; 597 inp->inp_inc.inc_flags |= INC_ISIPV6; 598 NET_EPOCH_ENTER(et); 599 if ((error = tcp6_connect(tp, sin6, td)) != 0) 600 goto out_in_epoch; 601 #ifdef TCP_OFFLOAD 602 if (registered_toedevs > 0 && 603 (so->so_options & SO_NO_OFFLOAD) == 0 && 604 (error = tcp_offload_connect(so, nam)) == 0) 605 goto out_in_epoch; 606 #endif 607 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 608 error = tcp_output(tp); 609 out_in_epoch: 610 NET_EPOCH_EXIT(et); 611 out: 612 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 613 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 614 /* 615 * If the implicit bind in the connect call fails, restore 616 * the flags we modified. 617 */ 618 if (error != 0 && inp->inp_lport == 0) { 619 inp->inp_vflag = vflagsav; 620 inp->inp_inc.inc_flags = incflagsav; 621 } 622 623 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 624 INP_WUNLOCK(inp); 625 return (error); 626 } 627 #endif /* INET6 */ 628 629 /* 630 * Initiate disconnect from peer. 631 * If connection never passed embryonic stage, just drop; 632 * else if don't need to let data drain, then can just drop anyways, 633 * else have to begin TCP shutdown process: mark socket disconnecting, 634 * drain unread data, state switch to reflect user close, and 635 * send segment (e.g. FIN) to peer. Socket will be really disconnected 636 * when peer sends FIN and acks ours. 637 * 638 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 639 */ 640 static int 641 tcp_usr_disconnect(struct socket *so) 642 { 643 struct inpcb *inp; 644 struct tcpcb *tp = NULL; 645 struct epoch_tracker et; 646 int error = 0; 647 648 NET_EPOCH_ENTER(et); 649 inp = sotoinpcb(so); 650 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 651 INP_WLOCK(inp); 652 if (inp->inp_flags & INP_DROPPED) { 653 error = ECONNRESET; 654 goto out; 655 } 656 tp = intotcpcb(inp); 657 if (tp->t_state == TCPS_TIME_WAIT) 658 goto out; 659 tcp_disconnect(tp); 660 out: 661 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 662 INP_WUNLOCK(inp); 663 NET_EPOCH_EXIT(et); 664 return (error); 665 } 666 667 #ifdef INET 668 /* 669 * Accept a connection. Essentially all the work is done at higher levels; 670 * just return the address of the peer, storing through addr. 671 */ 672 static int 673 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 674 { 675 int error = 0; 676 struct inpcb *inp = NULL; 677 #ifdef KDTRACE_HOOKS 678 struct tcpcb *tp = NULL; 679 #endif 680 struct in_addr addr; 681 in_port_t port = 0; 682 683 if (so->so_state & SS_ISDISCONNECTED) 684 return (ECONNABORTED); 685 686 inp = sotoinpcb(so); 687 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 688 INP_WLOCK(inp); 689 if (inp->inp_flags & INP_DROPPED) { 690 error = ECONNABORTED; 691 goto out; 692 } 693 #ifdef KDTRACE_HOOKS 694 tp = intotcpcb(inp); 695 #endif 696 697 /* 698 * We inline in_getpeeraddr and COMMON_END here, so that we can 699 * copy the data of interest and defer the malloc until after we 700 * release the lock. 701 */ 702 port = inp->inp_fport; 703 addr = inp->inp_faddr; 704 705 out: 706 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 707 INP_WUNLOCK(inp); 708 if (error == 0) 709 *nam = in_sockaddr(port, &addr); 710 return error; 711 } 712 #endif /* INET */ 713 714 #ifdef INET6 715 static int 716 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 717 { 718 struct inpcb *inp = NULL; 719 int error = 0; 720 #ifdef KDTRACE_HOOKS 721 struct tcpcb *tp = NULL; 722 #endif 723 struct in_addr addr; 724 struct in6_addr addr6; 725 struct epoch_tracker et; 726 in_port_t port = 0; 727 int v4 = 0; 728 729 if (so->so_state & SS_ISDISCONNECTED) 730 return (ECONNABORTED); 731 732 inp = sotoinpcb(so); 733 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 734 NET_EPOCH_ENTER(et); 735 INP_WLOCK(inp); 736 if (inp->inp_flags & INP_DROPPED) { 737 error = ECONNABORTED; 738 goto out; 739 } 740 #ifdef KDTRACE_HOOKS 741 tp = intotcpcb(inp); 742 #endif 743 744 /* 745 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 746 * copy the data of interest and defer the malloc until after we 747 * release the lock. 748 */ 749 if (inp->inp_vflag & INP_IPV4) { 750 v4 = 1; 751 port = inp->inp_fport; 752 addr = inp->inp_faddr; 753 } else { 754 port = inp->inp_fport; 755 addr6 = inp->in6p_faddr; 756 } 757 758 out: 759 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 760 INP_WUNLOCK(inp); 761 NET_EPOCH_EXIT(et); 762 if (error == 0) { 763 if (v4) 764 *nam = in6_v4mapsin6_sockaddr(port, &addr); 765 else 766 *nam = in6_sockaddr(port, &addr6); 767 } 768 return error; 769 } 770 #endif /* INET6 */ 771 772 /* 773 * Mark the connection as being incapable of further output. 774 */ 775 static int 776 tcp_usr_shutdown(struct socket *so) 777 { 778 int error = 0; 779 struct inpcb *inp; 780 struct tcpcb *tp = NULL; 781 struct epoch_tracker et; 782 783 inp = sotoinpcb(so); 784 KASSERT(inp != NULL, ("inp == NULL")); 785 INP_WLOCK(inp); 786 if (inp->inp_flags & INP_DROPPED) { 787 INP_WUNLOCK(inp); 788 return (ECONNRESET); 789 } 790 tp = intotcpcb(inp); 791 NET_EPOCH_ENTER(et); 792 socantsendmore(so); 793 tcp_usrclosed(tp); 794 if (!(inp->inp_flags & INP_DROPPED)) 795 error = tcp_output_nodrop(tp); 796 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 797 error = tcp_unlock_or_drop(tp, error); 798 NET_EPOCH_EXIT(et); 799 800 return (error); 801 } 802 803 /* 804 * After a receive, possibly send window update to peer. 805 */ 806 static int 807 tcp_usr_rcvd(struct socket *so, int flags) 808 { 809 struct epoch_tracker et; 810 struct inpcb *inp; 811 struct tcpcb *tp = NULL; 812 int outrv = 0, error = 0; 813 814 inp = sotoinpcb(so); 815 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 816 INP_WLOCK(inp); 817 if (inp->inp_flags & INP_DROPPED) { 818 INP_WUNLOCK(inp); 819 return (ECONNRESET); 820 } 821 tp = intotcpcb(inp); 822 NET_EPOCH_ENTER(et); 823 /* 824 * For passively-created TFO connections, don't attempt a window 825 * update while still in SYN_RECEIVED as this may trigger an early 826 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 827 * application response data, or failing that, when the DELACK timer 828 * expires. 829 */ 830 if (IS_FASTOPEN(tp->t_flags) && 831 (tp->t_state == TCPS_SYN_RECEIVED)) 832 goto out; 833 #ifdef TCP_OFFLOAD 834 if (tp->t_flags & TF_TOE) 835 tcp_offload_rcvd(tp); 836 else 837 #endif 838 outrv = tcp_output_nodrop(tp); 839 out: 840 TCP_PROBE2(debug__user, tp, PRU_RCVD); 841 (void) tcp_unlock_or_drop(tp, outrv); 842 NET_EPOCH_EXIT(et); 843 return (error); 844 } 845 846 /* 847 * Do a send by putting data in output queue and updating urgent 848 * marker if URG set. Possibly send more data. Unlike the other 849 * pru_*() routines, the mbuf chains are our responsibility. We 850 * must either enqueue them or free them. The other pru_* routines 851 * generally are caller-frees. 852 */ 853 static int 854 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 855 struct sockaddr *nam, struct mbuf *control, struct thread *td) 856 { 857 struct epoch_tracker et; 858 int error = 0; 859 struct inpcb *inp; 860 struct tcpcb *tp = NULL; 861 #ifdef INET 862 #ifdef INET6 863 struct sockaddr_in sin; 864 #endif 865 struct sockaddr_in *sinp; 866 #endif 867 #ifdef INET6 868 struct sockaddr_in6 *sin6; 869 int isipv6; 870 #endif 871 u_int8_t incflagsav; 872 u_char vflagsav; 873 bool restoreflags; 874 875 if (control != NULL) { 876 /* TCP doesn't do control messages (rights, creds, etc) */ 877 if (control->m_len) { 878 m_freem(control); 879 return (EINVAL); 880 } 881 m_freem(control); /* empty control, just free it */ 882 } 883 884 inp = sotoinpcb(so); 885 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 886 INP_WLOCK(inp); 887 if (inp->inp_flags & INP_DROPPED) { 888 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 889 m_freem(m); 890 INP_WUNLOCK(inp); 891 return (ECONNRESET); 892 } 893 894 vflagsav = inp->inp_vflag; 895 incflagsav = inp->inp_inc.inc_flags; 896 restoreflags = false; 897 tp = intotcpcb(inp); 898 899 NET_EPOCH_ENTER(et); 900 if ((flags & PRUS_OOB) != 0 && 901 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 902 goto out; 903 904 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 905 if (tp->t_state == TCPS_LISTEN) { 906 error = EINVAL; 907 goto out; 908 } 909 switch (nam->sa_family) { 910 #ifdef INET 911 case AF_INET: 912 sinp = (struct sockaddr_in *)nam; 913 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 914 error = EINVAL; 915 goto out; 916 } 917 if ((inp->inp_vflag & INP_IPV6) != 0) { 918 error = EAFNOSUPPORT; 919 goto out; 920 } 921 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 922 error = EAFNOSUPPORT; 923 goto out; 924 } 925 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 926 error = EACCES; 927 goto out; 928 } 929 if ((error = prison_remote_ip4(td->td_ucred, 930 &sinp->sin_addr))) 931 goto out; 932 #ifdef INET6 933 isipv6 = 0; 934 #endif 935 break; 936 #endif /* INET */ 937 #ifdef INET6 938 case AF_INET6: 939 sin6 = (struct sockaddr_in6 *)nam; 940 if (sin6->sin6_len != sizeof(*sin6)) { 941 error = EINVAL; 942 goto out; 943 } 944 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 945 error = EAFNOSUPPORT; 946 goto out; 947 } 948 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 949 error = EAFNOSUPPORT; 950 goto out; 951 } 952 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 953 #ifdef INET 954 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 955 error = EINVAL; 956 goto out; 957 } 958 if ((inp->inp_vflag & INP_IPV4) == 0) { 959 error = EAFNOSUPPORT; 960 goto out; 961 } 962 restoreflags = true; 963 inp->inp_vflag &= ~INP_IPV6; 964 sinp = &sin; 965 in6_sin6_2_sin(sinp, sin6); 966 if (IN_MULTICAST( 967 ntohl(sinp->sin_addr.s_addr))) { 968 error = EAFNOSUPPORT; 969 goto out; 970 } 971 if ((error = prison_remote_ip4(td->td_ucred, 972 &sinp->sin_addr))) 973 goto out; 974 isipv6 = 0; 975 #else /* !INET */ 976 error = EAFNOSUPPORT; 977 goto out; 978 #endif /* INET */ 979 } else { 980 if ((inp->inp_vflag & INP_IPV6) == 0) { 981 error = EAFNOSUPPORT; 982 goto out; 983 } 984 restoreflags = true; 985 inp->inp_vflag &= ~INP_IPV4; 986 inp->inp_inc.inc_flags |= INC_ISIPV6; 987 if ((error = prison_remote_ip6(td->td_ucred, 988 &sin6->sin6_addr))) 989 goto out; 990 isipv6 = 1; 991 } 992 break; 993 #endif /* INET6 */ 994 default: 995 error = EAFNOSUPPORT; 996 goto out; 997 } 998 } 999 if (!(flags & PRUS_OOB)) { 1000 if (tp->t_acktime == 0) 1001 tp->t_acktime = ticks; 1002 sbappendstream(&so->so_snd, m, flags); 1003 m = NULL; 1004 if (nam && tp->t_state < TCPS_SYN_SENT) { 1005 KASSERT(tp->t_state == TCPS_CLOSED, 1006 ("%s: tp %p is listening", __func__, tp)); 1007 1008 /* 1009 * Do implied connect if not yet connected, 1010 * initialize window to default value, and 1011 * initialize maxseg using peer's cached MSS. 1012 */ 1013 #ifdef INET6 1014 if (isipv6) 1015 error = tcp6_connect(tp, sin6, td); 1016 #endif /* INET6 */ 1017 #if defined(INET6) && defined(INET) 1018 else 1019 #endif 1020 #ifdef INET 1021 error = tcp_connect(tp, sinp, td); 1022 #endif 1023 /* 1024 * The bind operation in tcp_connect succeeded. We 1025 * no longer want to restore the flags if later 1026 * operations fail. 1027 */ 1028 if (error == 0 || inp->inp_lport != 0) 1029 restoreflags = false; 1030 1031 if (error) { 1032 /* m is freed if PRUS_NOTREADY is unset. */ 1033 sbflush(&so->so_snd); 1034 goto out; 1035 } 1036 if (IS_FASTOPEN(tp->t_flags)) 1037 tcp_fastopen_connect(tp); 1038 else { 1039 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1040 tcp_mss(tp, -1); 1041 } 1042 } 1043 if (flags & PRUS_EOF) { 1044 /* 1045 * Close the send side of the connection after 1046 * the data is sent. 1047 */ 1048 socantsendmore(so); 1049 tcp_usrclosed(tp); 1050 } 1051 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1052 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1053 (tp->t_fbyte_out == 0) && 1054 (so->so_snd.sb_ccc > 0)) { 1055 tp->t_fbyte_out = ticks; 1056 if (tp->t_fbyte_out == 0) 1057 tp->t_fbyte_out = 1; 1058 if (tp->t_fbyte_out && tp->t_fbyte_in) 1059 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1060 } 1061 if (!(inp->inp_flags & INP_DROPPED) && 1062 !(flags & PRUS_NOTREADY)) { 1063 if (flags & PRUS_MORETOCOME) 1064 tp->t_flags |= TF_MORETOCOME; 1065 error = tcp_output_nodrop(tp); 1066 if (flags & PRUS_MORETOCOME) 1067 tp->t_flags &= ~TF_MORETOCOME; 1068 } 1069 } else { 1070 /* 1071 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1072 */ 1073 SOCKBUF_LOCK(&so->so_snd); 1074 if (sbspace(&so->so_snd) < -512) { 1075 SOCKBUF_UNLOCK(&so->so_snd); 1076 error = ENOBUFS; 1077 goto out; 1078 } 1079 /* 1080 * According to RFC961 (Assigned Protocols), 1081 * the urgent pointer points to the last octet 1082 * of urgent data. We continue, however, 1083 * to consider it to indicate the first octet 1084 * of data past the urgent section. 1085 * Otherwise, snd_up should be one lower. 1086 */ 1087 if (tp->t_acktime == 0) 1088 tp->t_acktime = ticks; 1089 sbappendstream_locked(&so->so_snd, m, flags); 1090 SOCKBUF_UNLOCK(&so->so_snd); 1091 m = NULL; 1092 if (nam && tp->t_state < TCPS_SYN_SENT) { 1093 /* 1094 * Do implied connect if not yet connected, 1095 * initialize window to default value, and 1096 * initialize maxseg using peer's cached MSS. 1097 */ 1098 1099 /* 1100 * Not going to contemplate SYN|URG 1101 */ 1102 if (IS_FASTOPEN(tp->t_flags)) 1103 tp->t_flags &= ~TF_FASTOPEN; 1104 #ifdef INET6 1105 if (isipv6) 1106 error = tcp6_connect(tp, sin6, td); 1107 #endif /* INET6 */ 1108 #if defined(INET6) && defined(INET) 1109 else 1110 #endif 1111 #ifdef INET 1112 error = tcp_connect(tp, sinp, td); 1113 #endif 1114 /* 1115 * The bind operation in tcp_connect succeeded. We 1116 * no longer want to restore the flags if later 1117 * operations fail. 1118 */ 1119 if (error == 0 || inp->inp_lport != 0) 1120 restoreflags = false; 1121 1122 if (error != 0) { 1123 /* m is freed if PRUS_NOTREADY is unset. */ 1124 sbflush(&so->so_snd); 1125 goto out; 1126 } 1127 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1128 tcp_mss(tp, -1); 1129 } 1130 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1131 if ((flags & PRUS_NOTREADY) == 0) { 1132 tp->t_flags |= TF_FORCEDATA; 1133 error = tcp_output_nodrop(tp); 1134 tp->t_flags &= ~TF_FORCEDATA; 1135 } 1136 } 1137 TCP_LOG_EVENT(tp, NULL, 1138 &inp->inp_socket->so_rcv, 1139 &inp->inp_socket->so_snd, 1140 TCP_LOG_USERSEND, error, 1141 0, NULL, false); 1142 1143 out: 1144 /* 1145 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1146 * responsible for freeing memory. 1147 */ 1148 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1149 m_freem(m); 1150 1151 /* 1152 * If the request was unsuccessful and we changed flags, 1153 * restore the original flags. 1154 */ 1155 if (error != 0 && restoreflags) { 1156 inp->inp_vflag = vflagsav; 1157 inp->inp_inc.inc_flags = incflagsav; 1158 } 1159 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1160 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1161 error = tcp_unlock_or_drop(tp, error); 1162 NET_EPOCH_EXIT(et); 1163 return (error); 1164 } 1165 1166 static int 1167 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1168 { 1169 struct epoch_tracker et; 1170 struct inpcb *inp; 1171 struct tcpcb *tp; 1172 int error; 1173 1174 inp = sotoinpcb(so); 1175 INP_WLOCK(inp); 1176 if (inp->inp_flags & INP_DROPPED) { 1177 INP_WUNLOCK(inp); 1178 mb_free_notready(m, count); 1179 return (ECONNRESET); 1180 } 1181 tp = intotcpcb(inp); 1182 1183 SOCKBUF_LOCK(&so->so_snd); 1184 error = sbready(&so->so_snd, m, count); 1185 SOCKBUF_UNLOCK(&so->so_snd); 1186 if (error) { 1187 INP_WUNLOCK(inp); 1188 return (error); 1189 } 1190 NET_EPOCH_ENTER(et); 1191 error = tcp_output_unlock(tp); 1192 NET_EPOCH_EXIT(et); 1193 1194 return (error); 1195 } 1196 1197 /* 1198 * Abort the TCP. Drop the connection abruptly. 1199 */ 1200 static void 1201 tcp_usr_abort(struct socket *so) 1202 { 1203 struct inpcb *inp; 1204 struct tcpcb *tp = NULL; 1205 struct epoch_tracker et; 1206 1207 inp = sotoinpcb(so); 1208 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1209 1210 NET_EPOCH_ENTER(et); 1211 INP_WLOCK(inp); 1212 KASSERT(inp->inp_socket != NULL, 1213 ("tcp_usr_abort: inp_socket == NULL")); 1214 1215 /* 1216 * If we still have full TCP state, and we're not dropped, drop. 1217 */ 1218 if (!(inp->inp_flags & INP_DROPPED)) { 1219 tp = intotcpcb(inp); 1220 tp = tcp_drop(tp, ECONNABORTED); 1221 if (tp == NULL) 1222 goto dropped; 1223 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1224 } 1225 if (!(inp->inp_flags & INP_DROPPED)) { 1226 soref(so); 1227 inp->inp_flags |= INP_SOCKREF; 1228 } 1229 INP_WUNLOCK(inp); 1230 dropped: 1231 NET_EPOCH_EXIT(et); 1232 } 1233 1234 /* 1235 * TCP socket is closed. Start friendly disconnect. 1236 */ 1237 static void 1238 tcp_usr_close(struct socket *so) 1239 { 1240 struct inpcb *inp; 1241 struct tcpcb *tp = NULL; 1242 struct epoch_tracker et; 1243 1244 inp = sotoinpcb(so); 1245 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1246 1247 NET_EPOCH_ENTER(et); 1248 INP_WLOCK(inp); 1249 KASSERT(inp->inp_socket != NULL, 1250 ("tcp_usr_close: inp_socket == NULL")); 1251 1252 /* 1253 * If we are still connected and we're not dropped, initiate 1254 * a disconnect. 1255 */ 1256 if (!(inp->inp_flags & INP_DROPPED)) { 1257 tp = intotcpcb(inp); 1258 if (tp->t_state != TCPS_TIME_WAIT) { 1259 tp->t_flags |= TF_CLOSED; 1260 tcp_disconnect(tp); 1261 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1262 } 1263 } 1264 if (!(inp->inp_flags & INP_DROPPED)) { 1265 soref(so); 1266 inp->inp_flags |= INP_SOCKREF; 1267 } 1268 INP_WUNLOCK(inp); 1269 NET_EPOCH_EXIT(et); 1270 } 1271 1272 static int 1273 tcp_pru_options_support(struct tcpcb *tp, int flags) 1274 { 1275 /* 1276 * If the specific TCP stack has a pru_options 1277 * specified then it does not always support 1278 * all the PRU_XX options and we must ask it. 1279 * If the function is not specified then all 1280 * of the PRU_XX options are supported. 1281 */ 1282 int ret = 0; 1283 1284 if (tp->t_fb->tfb_pru_options) { 1285 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1286 } 1287 return (ret); 1288 } 1289 1290 /* 1291 * Receive out-of-band data. 1292 */ 1293 static int 1294 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1295 { 1296 int error = 0; 1297 struct inpcb *inp; 1298 struct tcpcb *tp = NULL; 1299 1300 inp = sotoinpcb(so); 1301 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1302 INP_WLOCK(inp); 1303 if (inp->inp_flags & INP_DROPPED) { 1304 error = ECONNRESET; 1305 goto out; 1306 } 1307 tp = intotcpcb(inp); 1308 error = tcp_pru_options_support(tp, PRUS_OOB); 1309 if (error) { 1310 goto out; 1311 } 1312 if ((so->so_oobmark == 0 && 1313 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1314 so->so_options & SO_OOBINLINE || 1315 tp->t_oobflags & TCPOOB_HADDATA) { 1316 error = EINVAL; 1317 goto out; 1318 } 1319 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1320 error = EWOULDBLOCK; 1321 goto out; 1322 } 1323 m->m_len = 1; 1324 *mtod(m, caddr_t) = tp->t_iobc; 1325 if ((flags & MSG_PEEK) == 0) 1326 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1327 1328 out: 1329 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1330 INP_WUNLOCK(inp); 1331 return (error); 1332 } 1333 1334 #ifdef INET 1335 struct protosw tcp_protosw = { 1336 .pr_type = SOCK_STREAM, 1337 .pr_protocol = IPPROTO_TCP, 1338 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1339 PR_CAPATTACH, 1340 .pr_ctloutput = tcp_ctloutput, 1341 .pr_abort = tcp_usr_abort, 1342 .pr_accept = tcp_usr_accept, 1343 .pr_attach = tcp_usr_attach, 1344 .pr_bind = tcp_usr_bind, 1345 .pr_connect = tcp_usr_connect, 1346 .pr_control = in_control, 1347 .pr_detach = tcp_usr_detach, 1348 .pr_disconnect = tcp_usr_disconnect, 1349 .pr_listen = tcp_usr_listen, 1350 .pr_peeraddr = in_getpeeraddr, 1351 .pr_rcvd = tcp_usr_rcvd, 1352 .pr_rcvoob = tcp_usr_rcvoob, 1353 .pr_send = tcp_usr_send, 1354 .pr_ready = tcp_usr_ready, 1355 .pr_shutdown = tcp_usr_shutdown, 1356 .pr_sockaddr = in_getsockaddr, 1357 .pr_sosetlabel = in_pcbsosetlabel, 1358 .pr_close = tcp_usr_close, 1359 }; 1360 #endif /* INET */ 1361 1362 #ifdef INET6 1363 struct protosw tcp6_protosw = { 1364 .pr_type = SOCK_STREAM, 1365 .pr_protocol = IPPROTO_TCP, 1366 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1367 PR_CAPATTACH, 1368 .pr_ctloutput = tcp_ctloutput, 1369 .pr_abort = tcp_usr_abort, 1370 .pr_accept = tcp6_usr_accept, 1371 .pr_attach = tcp_usr_attach, 1372 .pr_bind = tcp6_usr_bind, 1373 .pr_connect = tcp6_usr_connect, 1374 .pr_control = in6_control, 1375 .pr_detach = tcp_usr_detach, 1376 .pr_disconnect = tcp_usr_disconnect, 1377 .pr_listen = tcp6_usr_listen, 1378 .pr_peeraddr = in6_mapped_peeraddr, 1379 .pr_rcvd = tcp_usr_rcvd, 1380 .pr_rcvoob = tcp_usr_rcvoob, 1381 .pr_send = tcp_usr_send, 1382 .pr_ready = tcp_usr_ready, 1383 .pr_shutdown = tcp_usr_shutdown, 1384 .pr_sockaddr = in6_mapped_sockaddr, 1385 .pr_sosetlabel = in_pcbsosetlabel, 1386 .pr_close = tcp_usr_close, 1387 }; 1388 #endif /* INET6 */ 1389 1390 #ifdef INET 1391 /* 1392 * Common subroutine to open a TCP connection to remote host specified 1393 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1394 * and assign a local port number and install the inpcb into the hash. 1395 * Initialize connection parameters and enter SYN-SENT state. 1396 */ 1397 static int 1398 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1399 { 1400 struct inpcb *inp = tptoinpcb(tp); 1401 struct socket *so = tptosocket(tp); 1402 int error; 1403 1404 NET_EPOCH_ASSERT(); 1405 INP_WLOCK_ASSERT(inp); 1406 1407 if (__predict_false((so->so_state & 1408 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1409 return (EISCONN); 1410 1411 INP_HASH_WLOCK(&V_tcbinfo); 1412 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1413 INP_HASH_WUNLOCK(&V_tcbinfo); 1414 if (error != 0) 1415 return (error); 1416 1417 /* 1418 * Compute window scaling to request: 1419 * Scale to fit into sweet spot. See tcp_syncache.c. 1420 * XXX: This should move to tcp_output(). 1421 */ 1422 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1423 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1424 tp->request_r_scale++; 1425 1426 soisconnecting(so); 1427 TCPSTAT_INC(tcps_connattempt); 1428 tcp_state_change(tp, TCPS_SYN_SENT); 1429 tp->iss = tcp_new_isn(&inp->inp_inc); 1430 if (tp->t_flags & TF_REQ_TSTMP) 1431 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1432 tcp_sendseqinit(tp); 1433 1434 return (0); 1435 } 1436 #endif /* INET */ 1437 1438 #ifdef INET6 1439 static int 1440 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1441 { 1442 struct inpcb *inp = tptoinpcb(tp); 1443 struct socket *so = tptosocket(tp); 1444 int error; 1445 1446 NET_EPOCH_ASSERT(); 1447 INP_WLOCK_ASSERT(inp); 1448 1449 if (__predict_false((so->so_state & 1450 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1451 return (EISCONN); 1452 1453 INP_HASH_WLOCK(&V_tcbinfo); 1454 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1455 INP_HASH_WUNLOCK(&V_tcbinfo); 1456 if (error != 0) 1457 return (error); 1458 1459 /* Compute window scaling to request. */ 1460 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1461 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1462 tp->request_r_scale++; 1463 1464 soisconnecting(so); 1465 TCPSTAT_INC(tcps_connattempt); 1466 tcp_state_change(tp, TCPS_SYN_SENT); 1467 tp->iss = tcp_new_isn(&inp->inp_inc); 1468 if (tp->t_flags & TF_REQ_TSTMP) 1469 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1470 tcp_sendseqinit(tp); 1471 1472 return (0); 1473 } 1474 #endif /* INET6 */ 1475 1476 /* 1477 * Export TCP internal state information via a struct tcp_info, based on the 1478 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1479 * (TCP state machine, etc). We export all information using FreeBSD-native 1480 * constants -- for example, the numeric values for tcpi_state will differ 1481 * from Linux. 1482 */ 1483 static void 1484 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1485 { 1486 1487 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1488 bzero(ti, sizeof(*ti)); 1489 1490 ti->tcpi_state = tp->t_state; 1491 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1492 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1493 if (tp->t_flags & TF_SACK_PERMIT) 1494 ti->tcpi_options |= TCPI_OPT_SACK; 1495 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1496 ti->tcpi_options |= TCPI_OPT_WSCALE; 1497 ti->tcpi_snd_wscale = tp->snd_scale; 1498 ti->tcpi_rcv_wscale = tp->rcv_scale; 1499 } 1500 if (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1501 ti->tcpi_options |= TCPI_OPT_ECN; 1502 1503 ti->tcpi_rto = tp->t_rxtcur * tick; 1504 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1505 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1506 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1507 1508 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1509 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1510 1511 /* 1512 * FreeBSD-specific extension fields for tcp_info. 1513 */ 1514 ti->tcpi_rcv_space = tp->rcv_wnd; 1515 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1516 ti->tcpi_snd_wnd = tp->snd_wnd; 1517 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1518 ti->tcpi_snd_nxt = tp->snd_nxt; 1519 ti->tcpi_snd_mss = tp->t_maxseg; 1520 ti->tcpi_rcv_mss = tp->t_maxseg; 1521 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1522 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1523 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1524 #ifdef TCP_OFFLOAD 1525 if (tp->t_flags & TF_TOE) { 1526 ti->tcpi_options |= TCPI_OPT_TOE; 1527 tcp_offload_tcp_info(tp, ti); 1528 } 1529 #endif 1530 /* 1531 * AccECN related counters. 1532 */ 1533 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1534 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1535 /* 1536 * Internal counter starts at 5 for AccECN 1537 * but 0 for RFC3168 ECN. 1538 */ 1539 ti->tcpi_delivered_ce = tp->t_scep - 5; 1540 else 1541 ti->tcpi_delivered_ce = tp->t_scep; 1542 ti->tcpi_received_ce = tp->t_rcep; 1543 } 1544 1545 /* 1546 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1547 * socket option arguments. When it re-acquires the lock after the copy, it 1548 * has to revalidate that the connection is still valid for the socket 1549 * option. 1550 */ 1551 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1552 INP_WLOCK(inp); \ 1553 if (inp->inp_flags & INP_DROPPED) { \ 1554 INP_WUNLOCK(inp); \ 1555 cleanup; \ 1556 return (ECONNRESET); \ 1557 } \ 1558 tp = intotcpcb(inp); \ 1559 } while(0) 1560 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1561 1562 int 1563 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1564 { 1565 struct socket *so = inp->inp_socket; 1566 struct tcpcb *tp = intotcpcb(inp); 1567 int error = 0; 1568 1569 MPASS(sopt->sopt_dir == SOPT_SET); 1570 INP_WLOCK_ASSERT(inp); 1571 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1572 ("inp_flags == %x", inp->inp_flags)); 1573 KASSERT(so != NULL, ("inp_socket == NULL")); 1574 1575 if (sopt->sopt_level != IPPROTO_TCP) { 1576 INP_WUNLOCK(inp); 1577 #ifdef INET6 1578 if (inp->inp_vflag & INP_IPV6PROTO) 1579 error = ip6_ctloutput(so, sopt); 1580 #endif 1581 #if defined(INET6) && defined(INET) 1582 else 1583 #endif 1584 #ifdef INET 1585 error = ip_ctloutput(so, sopt); 1586 #endif 1587 /* 1588 * When an IP-level socket option affects TCP, pass control 1589 * down to stack tfb_tcp_ctloutput, otherwise return what 1590 * IP level returned. 1591 */ 1592 switch (sopt->sopt_level) { 1593 #ifdef INET6 1594 case IPPROTO_IPV6: 1595 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1596 return (error); 1597 switch (sopt->sopt_name) { 1598 case IPV6_TCLASS: 1599 /* Notify tcp stacks that care (e.g. RACK). */ 1600 break; 1601 case IPV6_USE_MIN_MTU: 1602 /* Update t_maxseg accordingly. */ 1603 break; 1604 default: 1605 return (error); 1606 } 1607 break; 1608 #endif 1609 #ifdef INET 1610 case IPPROTO_IP: 1611 switch (sopt->sopt_name) { 1612 case IP_TOS: 1613 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1614 break; 1615 case IP_TTL: 1616 /* Notify tcp stacks that care (e.g. RACK). */ 1617 break; 1618 default: 1619 return (error); 1620 } 1621 break; 1622 #endif 1623 default: 1624 return (error); 1625 } 1626 INP_WLOCK(inp); 1627 if (inp->inp_flags & INP_DROPPED) { 1628 INP_WUNLOCK(inp); 1629 return (ECONNRESET); 1630 } 1631 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1632 /* 1633 * Protect the TCP option TCP_FUNCTION_BLK so 1634 * that a sub-function can *never* overwrite this. 1635 */ 1636 struct tcp_function_set fsn; 1637 struct tcp_function_block *blk; 1638 1639 INP_WUNLOCK(inp); 1640 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1641 if (error) 1642 return (error); 1643 1644 INP_WLOCK(inp); 1645 if (inp->inp_flags & INP_DROPPED) { 1646 INP_WUNLOCK(inp); 1647 return (ECONNRESET); 1648 } 1649 tp = intotcpcb(inp); 1650 1651 blk = find_and_ref_tcp_functions(&fsn); 1652 if (blk == NULL) { 1653 INP_WUNLOCK(inp); 1654 return (ENOENT); 1655 } 1656 if (tp->t_fb == blk) { 1657 /* You already have this */ 1658 refcount_release(&blk->tfb_refcnt); 1659 INP_WUNLOCK(inp); 1660 return (0); 1661 } 1662 if (tp->t_state != TCPS_CLOSED) { 1663 /* 1664 * The user has advanced the state 1665 * past the initial point, we may not 1666 * be able to switch. 1667 */ 1668 if (blk->tfb_tcp_handoff_ok != NULL) { 1669 /* 1670 * Does the stack provide a 1671 * query mechanism, if so it may 1672 * still be possible? 1673 */ 1674 error = (*blk->tfb_tcp_handoff_ok)(tp); 1675 } else 1676 error = EINVAL; 1677 if (error) { 1678 refcount_release(&blk->tfb_refcnt); 1679 INP_WUNLOCK(inp); 1680 return(error); 1681 } 1682 } 1683 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1684 refcount_release(&blk->tfb_refcnt); 1685 INP_WUNLOCK(inp); 1686 return (ENOENT); 1687 } 1688 /* 1689 * Release the old refcnt, the 1690 * lookup acquired a ref on the 1691 * new one already. 1692 */ 1693 if (tp->t_fb->tfb_tcp_fb_fini) { 1694 struct epoch_tracker et; 1695 /* 1696 * Tell the stack to cleanup with 0 i.e. 1697 * the tcb is not going away. 1698 */ 1699 NET_EPOCH_ENTER(et); 1700 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1701 NET_EPOCH_EXIT(et); 1702 } 1703 #ifdef TCPHPTS 1704 /* Assure that we are not on any hpts */ 1705 tcp_hpts_remove(tptoinpcb(tp)); 1706 #endif 1707 if (blk->tfb_tcp_fb_init) { 1708 error = (*blk->tfb_tcp_fb_init)(tp); 1709 if (error) { 1710 refcount_release(&blk->tfb_refcnt); 1711 if (tp->t_fb->tfb_tcp_fb_init) { 1712 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1713 /* Fall back failed, drop the connection */ 1714 INP_WUNLOCK(inp); 1715 soabort(so); 1716 return (error); 1717 } 1718 } 1719 goto err_out; 1720 } 1721 } 1722 refcount_release(&tp->t_fb->tfb_refcnt); 1723 tp->t_fb = blk; 1724 #ifdef TCP_OFFLOAD 1725 if (tp->t_flags & TF_TOE) { 1726 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1727 sopt->sopt_name); 1728 } 1729 #endif 1730 err_out: 1731 INP_WUNLOCK(inp); 1732 return (error); 1733 } 1734 1735 /* Pass in the INP locked, callee must unlock it. */ 1736 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt)); 1737 } 1738 1739 static int 1740 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1741 { 1742 struct socket *so = inp->inp_socket; 1743 struct tcpcb *tp = intotcpcb(inp); 1744 int error = 0; 1745 1746 MPASS(sopt->sopt_dir == SOPT_GET); 1747 INP_WLOCK_ASSERT(inp); 1748 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1749 ("inp_flags == %x", inp->inp_flags)); 1750 KASSERT(so != NULL, ("inp_socket == NULL")); 1751 1752 if (sopt->sopt_level != IPPROTO_TCP) { 1753 INP_WUNLOCK(inp); 1754 #ifdef INET6 1755 if (inp->inp_vflag & INP_IPV6PROTO) 1756 error = ip6_ctloutput(so, sopt); 1757 #endif /* INET6 */ 1758 #if defined(INET6) && defined(INET) 1759 else 1760 #endif 1761 #ifdef INET 1762 error = ip_ctloutput(so, sopt); 1763 #endif 1764 return (error); 1765 } 1766 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1767 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1768 struct tcp_function_set fsn; 1769 1770 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1771 memset(&fsn, 0, sizeof(fsn)); 1772 find_tcp_function_alias(tp->t_fb, &fsn); 1773 } else { 1774 strncpy(fsn.function_set_name, 1775 tp->t_fb->tfb_tcp_block_name, 1776 TCP_FUNCTION_NAME_LEN_MAX); 1777 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1778 } 1779 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1780 INP_WUNLOCK(inp); 1781 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1782 return (error); 1783 } 1784 1785 /* Pass in the INP locked, callee must unlock it. */ 1786 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt)); 1787 } 1788 1789 int 1790 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1791 { 1792 struct inpcb *inp; 1793 1794 inp = sotoinpcb(so); 1795 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1796 1797 INP_WLOCK(inp); 1798 if (inp->inp_flags & INP_DROPPED) { 1799 INP_WUNLOCK(inp); 1800 return (ECONNRESET); 1801 } 1802 if (sopt->sopt_dir == SOPT_SET) 1803 return (tcp_ctloutput_set(inp, sopt)); 1804 else if (sopt->sopt_dir == SOPT_GET) 1805 return (tcp_ctloutput_get(inp, sopt)); 1806 else 1807 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1808 } 1809 1810 /* 1811 * If this assert becomes untrue, we need to change the size of the buf 1812 * variable in tcp_default_ctloutput(). 1813 */ 1814 #ifdef CTASSERT 1815 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1816 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1817 #endif 1818 1819 #ifdef KERN_TLS 1820 static int 1821 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1822 { 1823 struct tls_enable_v0 tls_v0; 1824 int error; 1825 1826 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1827 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1828 sizeof(tls_v0)); 1829 if (error) 1830 return (error); 1831 memset(tls, 0, sizeof(*tls)); 1832 tls->cipher_key = tls_v0.cipher_key; 1833 tls->iv = tls_v0.iv; 1834 tls->auth_key = tls_v0.auth_key; 1835 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1836 tls->cipher_key_len = tls_v0.cipher_key_len; 1837 tls->iv_len = tls_v0.iv_len; 1838 tls->auth_algorithm = tls_v0.auth_algorithm; 1839 tls->auth_key_len = tls_v0.auth_key_len; 1840 tls->flags = tls_v0.flags; 1841 tls->tls_vmajor = tls_v0.tls_vmajor; 1842 tls->tls_vminor = tls_v0.tls_vminor; 1843 return (0); 1844 } 1845 1846 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1847 } 1848 #endif 1849 1850 extern struct cc_algo newreno_cc_algo; 1851 1852 static int 1853 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1854 { 1855 struct cc_algo *algo; 1856 void *ptr = NULL; 1857 struct tcpcb *tp; 1858 struct cc_var cc_mem; 1859 char buf[TCP_CA_NAME_MAX]; 1860 size_t mem_sz; 1861 int error; 1862 1863 INP_WUNLOCK(inp); 1864 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1865 if (error) 1866 return(error); 1867 buf[sopt->sopt_valsize] = '\0'; 1868 CC_LIST_RLOCK(); 1869 STAILQ_FOREACH(algo, &cc_list, entries) { 1870 if (strncmp(buf, algo->name, 1871 TCP_CA_NAME_MAX) == 0) { 1872 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1873 /* We can't "see" modules being unloaded */ 1874 continue; 1875 } 1876 break; 1877 } 1878 } 1879 if (algo == NULL) { 1880 CC_LIST_RUNLOCK(); 1881 return(ESRCH); 1882 } 1883 /* 1884 * With a reference the algorithm cannot be removed 1885 * so we hold a reference through the change process. 1886 */ 1887 cc_refer(algo); 1888 CC_LIST_RUNLOCK(); 1889 if (algo->cb_init != NULL) { 1890 /* We can now pre-get the memory for the CC */ 1891 mem_sz = (*algo->cc_data_sz)(); 1892 if (mem_sz == 0) { 1893 goto no_mem_needed; 1894 } 1895 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1896 } else { 1897 no_mem_needed: 1898 mem_sz = 0; 1899 ptr = NULL; 1900 } 1901 /* 1902 * Make sure its all clean and zero and also get 1903 * back the inplock. 1904 */ 1905 memset(&cc_mem, 0, sizeof(cc_mem)); 1906 INP_WLOCK(inp); 1907 if (inp->inp_flags & INP_DROPPED) { 1908 INP_WUNLOCK(inp); 1909 if (ptr) 1910 free(ptr, M_CC_MEM); 1911 /* Release our temp reference */ 1912 CC_LIST_RLOCK(); 1913 cc_release(algo); 1914 CC_LIST_RUNLOCK(); 1915 return (ECONNRESET); 1916 } 1917 tp = intotcpcb(inp); 1918 if (ptr != NULL) 1919 memset(ptr, 0, mem_sz); 1920 cc_mem.ccvc.tcp = tp; 1921 /* 1922 * We once again hold a write lock over the tcb so it's 1923 * safe to do these things without ordering concerns. 1924 * Note here we init into stack memory. 1925 */ 1926 if (algo->cb_init != NULL) 1927 error = algo->cb_init(&cc_mem, ptr); 1928 else 1929 error = 0; 1930 /* 1931 * The CC algorithms, when given their memory 1932 * should not fail we could in theory have a 1933 * KASSERT here. 1934 */ 1935 if (error == 0) { 1936 /* 1937 * Touchdown, lets go ahead and move the 1938 * connection to the new CC module by 1939 * copying in the cc_mem after we call 1940 * the old ones cleanup (if any). 1941 */ 1942 if (CC_ALGO(tp)->cb_destroy != NULL) 1943 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 1944 /* Detach the old CC from the tcpcb */ 1945 cc_detach(tp); 1946 /* Copy in our temp memory that was inited */ 1947 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 1948 /* Now attach the new, which takes a reference */ 1949 cc_attach(tp, algo); 1950 /* Ok now are we where we have gotten past any conn_init? */ 1951 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 1952 /* Yep run the connection init for the new CC */ 1953 CC_ALGO(tp)->conn_init(&tp->t_ccv); 1954 } 1955 } else if (ptr) 1956 free(ptr, M_CC_MEM); 1957 INP_WUNLOCK(inp); 1958 /* Now lets release our temp reference */ 1959 CC_LIST_RLOCK(); 1960 cc_release(algo); 1961 CC_LIST_RUNLOCK(); 1962 return (error); 1963 } 1964 1965 int 1966 tcp_default_ctloutput(struct inpcb *inp, struct sockopt *sopt) 1967 { 1968 struct tcpcb *tp = intotcpcb(inp); 1969 int error, opt, optval; 1970 u_int ui; 1971 struct tcp_info ti; 1972 #ifdef KERN_TLS 1973 struct tls_enable tls; 1974 struct socket *so = inp->inp_socket; 1975 #endif 1976 char *pbuf, buf[TCP_LOG_ID_LEN]; 1977 #ifdef STATS 1978 struct statsblob *sbp; 1979 #endif 1980 size_t len; 1981 1982 INP_WLOCK_ASSERT(inp); 1983 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1984 ("inp_flags == %x", inp->inp_flags)); 1985 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 1986 1987 switch (sopt->sopt_level) { 1988 #ifdef INET6 1989 case IPPROTO_IPV6: 1990 MPASS(inp->inp_vflag & INP_IPV6PROTO); 1991 switch (sopt->sopt_name) { 1992 case IPV6_USE_MIN_MTU: 1993 tcp6_use_min_mtu(tp); 1994 /* FALLTHROUGH */ 1995 } 1996 INP_WUNLOCK(inp); 1997 return (0); 1998 #endif 1999 #ifdef INET 2000 case IPPROTO_IP: 2001 INP_WUNLOCK(inp); 2002 return (0); 2003 #endif 2004 } 2005 2006 /* 2007 * For TCP_CCALGOOPT forward the control to CC module, for both 2008 * SOPT_SET and SOPT_GET. 2009 */ 2010 switch (sopt->sopt_name) { 2011 case TCP_CCALGOOPT: 2012 INP_WUNLOCK(inp); 2013 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2014 return (EINVAL); 2015 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2016 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2017 sopt->sopt_valsize); 2018 if (error) { 2019 free(pbuf, M_TEMP); 2020 return (error); 2021 } 2022 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2023 if (CC_ALGO(tp)->ctl_output != NULL) 2024 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2025 else 2026 error = ENOENT; 2027 INP_WUNLOCK(inp); 2028 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2029 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2030 free(pbuf, M_TEMP); 2031 return (error); 2032 } 2033 2034 switch (sopt->sopt_dir) { 2035 case SOPT_SET: 2036 switch (sopt->sopt_name) { 2037 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2038 case TCP_MD5SIG: 2039 INP_WUNLOCK(inp); 2040 if (!TCPMD5_ENABLED()) 2041 return (ENOPROTOOPT); 2042 error = TCPMD5_PCBCTL(inp, sopt); 2043 if (error) 2044 return (error); 2045 INP_WLOCK_RECHECK(inp); 2046 goto unlock_and_done; 2047 #endif /* IPSEC */ 2048 2049 case TCP_NODELAY: 2050 case TCP_NOOPT: 2051 case TCP_LRD: 2052 INP_WUNLOCK(inp); 2053 error = sooptcopyin(sopt, &optval, sizeof optval, 2054 sizeof optval); 2055 if (error) 2056 return (error); 2057 2058 INP_WLOCK_RECHECK(inp); 2059 switch (sopt->sopt_name) { 2060 case TCP_NODELAY: 2061 opt = TF_NODELAY; 2062 break; 2063 case TCP_NOOPT: 2064 opt = TF_NOOPT; 2065 break; 2066 case TCP_LRD: 2067 opt = TF_LRD; 2068 break; 2069 default: 2070 opt = 0; /* dead code to fool gcc */ 2071 break; 2072 } 2073 2074 if (optval) 2075 tp->t_flags |= opt; 2076 else 2077 tp->t_flags &= ~opt; 2078 unlock_and_done: 2079 #ifdef TCP_OFFLOAD 2080 if (tp->t_flags & TF_TOE) { 2081 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2082 sopt->sopt_name); 2083 } 2084 #endif 2085 INP_WUNLOCK(inp); 2086 break; 2087 2088 case TCP_NOPUSH: 2089 INP_WUNLOCK(inp); 2090 error = sooptcopyin(sopt, &optval, sizeof optval, 2091 sizeof optval); 2092 if (error) 2093 return (error); 2094 2095 INP_WLOCK_RECHECK(inp); 2096 if (optval) 2097 tp->t_flags |= TF_NOPUSH; 2098 else if (tp->t_flags & TF_NOPUSH) { 2099 tp->t_flags &= ~TF_NOPUSH; 2100 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2101 struct epoch_tracker et; 2102 2103 NET_EPOCH_ENTER(et); 2104 error = tcp_output_nodrop(tp); 2105 NET_EPOCH_EXIT(et); 2106 } 2107 } 2108 goto unlock_and_done; 2109 2110 case TCP_REMOTE_UDP_ENCAPS_PORT: 2111 INP_WUNLOCK(inp); 2112 error = sooptcopyin(sopt, &optval, sizeof optval, 2113 sizeof optval); 2114 if (error) 2115 return (error); 2116 if ((optval < TCP_TUNNELING_PORT_MIN) || 2117 (optval > TCP_TUNNELING_PORT_MAX)) { 2118 /* Its got to be in range */ 2119 return (EINVAL); 2120 } 2121 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2122 /* You have to have enabled a UDP tunneling port first */ 2123 return (EINVAL); 2124 } 2125 INP_WLOCK_RECHECK(inp); 2126 if (tp->t_state != TCPS_CLOSED) { 2127 /* You can't change after you are connected */ 2128 error = EINVAL; 2129 } else { 2130 /* Ok we are all good set the port */ 2131 tp->t_port = htons(optval); 2132 } 2133 goto unlock_and_done; 2134 2135 case TCP_MAXSEG: 2136 INP_WUNLOCK(inp); 2137 error = sooptcopyin(sopt, &optval, sizeof optval, 2138 sizeof optval); 2139 if (error) 2140 return (error); 2141 2142 INP_WLOCK_RECHECK(inp); 2143 if (optval > 0 && optval <= tp->t_maxseg && 2144 optval + 40 >= V_tcp_minmss) 2145 tp->t_maxseg = optval; 2146 else 2147 error = EINVAL; 2148 goto unlock_and_done; 2149 2150 case TCP_INFO: 2151 INP_WUNLOCK(inp); 2152 error = EINVAL; 2153 break; 2154 2155 case TCP_STATS: 2156 INP_WUNLOCK(inp); 2157 #ifdef STATS 2158 error = sooptcopyin(sopt, &optval, sizeof optval, 2159 sizeof optval); 2160 if (error) 2161 return (error); 2162 2163 if (optval > 0) 2164 sbp = stats_blob_alloc( 2165 V_tcp_perconn_stats_dflt_tpl, 0); 2166 else 2167 sbp = NULL; 2168 2169 INP_WLOCK_RECHECK(inp); 2170 if ((tp->t_stats != NULL && sbp == NULL) || 2171 (tp->t_stats == NULL && sbp != NULL)) { 2172 struct statsblob *t = tp->t_stats; 2173 tp->t_stats = sbp; 2174 sbp = t; 2175 } 2176 INP_WUNLOCK(inp); 2177 2178 stats_blob_destroy(sbp); 2179 #else 2180 return (EOPNOTSUPP); 2181 #endif /* !STATS */ 2182 break; 2183 2184 case TCP_CONGESTION: 2185 error = tcp_set_cc_mod(inp, sopt); 2186 break; 2187 2188 case TCP_REUSPORT_LB_NUMA: 2189 INP_WUNLOCK(inp); 2190 error = sooptcopyin(sopt, &optval, sizeof(optval), 2191 sizeof(optval)); 2192 INP_WLOCK_RECHECK(inp); 2193 if (!error) 2194 error = in_pcblbgroup_numa(inp, optval); 2195 INP_WUNLOCK(inp); 2196 break; 2197 2198 #ifdef KERN_TLS 2199 case TCP_TXTLS_ENABLE: 2200 INP_WUNLOCK(inp); 2201 error = copyin_tls_enable(sopt, &tls); 2202 if (error) 2203 break; 2204 error = ktls_enable_tx(so, &tls); 2205 break; 2206 case TCP_TXTLS_MODE: 2207 INP_WUNLOCK(inp); 2208 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2209 if (error) 2210 return (error); 2211 2212 INP_WLOCK_RECHECK(inp); 2213 error = ktls_set_tx_mode(so, ui); 2214 INP_WUNLOCK(inp); 2215 break; 2216 case TCP_RXTLS_ENABLE: 2217 INP_WUNLOCK(inp); 2218 error = sooptcopyin(sopt, &tls, sizeof(tls), 2219 sizeof(tls)); 2220 if (error) 2221 break; 2222 error = ktls_enable_rx(so, &tls); 2223 break; 2224 #endif 2225 case TCP_MAXUNACKTIME: 2226 case TCP_KEEPIDLE: 2227 case TCP_KEEPINTVL: 2228 case TCP_KEEPINIT: 2229 INP_WUNLOCK(inp); 2230 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2231 if (error) 2232 return (error); 2233 2234 if (ui > (UINT_MAX / hz)) { 2235 error = EINVAL; 2236 break; 2237 } 2238 ui *= hz; 2239 2240 INP_WLOCK_RECHECK(inp); 2241 switch (sopt->sopt_name) { 2242 case TCP_MAXUNACKTIME: 2243 tp->t_maxunacktime = ui; 2244 break; 2245 2246 case TCP_KEEPIDLE: 2247 tp->t_keepidle = ui; 2248 /* 2249 * XXX: better check current remaining 2250 * timeout and "merge" it with new value. 2251 */ 2252 if ((tp->t_state > TCPS_LISTEN) && 2253 (tp->t_state <= TCPS_CLOSING)) 2254 tcp_timer_activate(tp, TT_KEEP, 2255 TP_KEEPIDLE(tp)); 2256 break; 2257 case TCP_KEEPINTVL: 2258 tp->t_keepintvl = ui; 2259 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2260 (TP_MAXIDLE(tp) > 0)) 2261 tcp_timer_activate(tp, TT_2MSL, 2262 TP_MAXIDLE(tp)); 2263 break; 2264 case TCP_KEEPINIT: 2265 tp->t_keepinit = ui; 2266 if (tp->t_state == TCPS_SYN_RECEIVED || 2267 tp->t_state == TCPS_SYN_SENT) 2268 tcp_timer_activate(tp, TT_KEEP, 2269 TP_KEEPINIT(tp)); 2270 break; 2271 } 2272 goto unlock_and_done; 2273 2274 case TCP_KEEPCNT: 2275 INP_WUNLOCK(inp); 2276 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2277 if (error) 2278 return (error); 2279 2280 INP_WLOCK_RECHECK(inp); 2281 tp->t_keepcnt = ui; 2282 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2283 (TP_MAXIDLE(tp) > 0)) 2284 tcp_timer_activate(tp, TT_2MSL, 2285 TP_MAXIDLE(tp)); 2286 goto unlock_and_done; 2287 2288 #ifdef TCPPCAP 2289 case TCP_PCAP_OUT: 2290 case TCP_PCAP_IN: 2291 INP_WUNLOCK(inp); 2292 error = sooptcopyin(sopt, &optval, sizeof optval, 2293 sizeof optval); 2294 if (error) 2295 return (error); 2296 2297 INP_WLOCK_RECHECK(inp); 2298 if (optval >= 0) 2299 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2300 &(tp->t_outpkts) : &(tp->t_inpkts), 2301 optval); 2302 else 2303 error = EINVAL; 2304 goto unlock_and_done; 2305 #endif 2306 2307 case TCP_FASTOPEN: { 2308 struct tcp_fastopen tfo_optval; 2309 2310 INP_WUNLOCK(inp); 2311 if (!V_tcp_fastopen_client_enable && 2312 !V_tcp_fastopen_server_enable) 2313 return (EPERM); 2314 2315 error = sooptcopyin(sopt, &tfo_optval, 2316 sizeof(tfo_optval), sizeof(int)); 2317 if (error) 2318 return (error); 2319 2320 INP_WLOCK_RECHECK(inp); 2321 if ((tp->t_state != TCPS_CLOSED) && 2322 (tp->t_state != TCPS_LISTEN)) { 2323 error = EINVAL; 2324 goto unlock_and_done; 2325 } 2326 if (tfo_optval.enable) { 2327 if (tp->t_state == TCPS_LISTEN) { 2328 if (!V_tcp_fastopen_server_enable) { 2329 error = EPERM; 2330 goto unlock_and_done; 2331 } 2332 2333 if (tp->t_tfo_pending == NULL) 2334 tp->t_tfo_pending = 2335 tcp_fastopen_alloc_counter(); 2336 } else { 2337 /* 2338 * If a pre-shared key was provided, 2339 * stash it in the client cookie 2340 * field of the tcpcb for use during 2341 * connect. 2342 */ 2343 if (sopt->sopt_valsize == 2344 sizeof(tfo_optval)) { 2345 memcpy(tp->t_tfo_cookie.client, 2346 tfo_optval.psk, 2347 TCP_FASTOPEN_PSK_LEN); 2348 tp->t_tfo_client_cookie_len = 2349 TCP_FASTOPEN_PSK_LEN; 2350 } 2351 } 2352 tp->t_flags |= TF_FASTOPEN; 2353 } else 2354 tp->t_flags &= ~TF_FASTOPEN; 2355 goto unlock_and_done; 2356 } 2357 2358 #ifdef TCP_BLACKBOX 2359 case TCP_LOG: 2360 INP_WUNLOCK(inp); 2361 error = sooptcopyin(sopt, &optval, sizeof optval, 2362 sizeof optval); 2363 if (error) 2364 return (error); 2365 2366 INP_WLOCK_RECHECK(inp); 2367 error = tcp_log_state_change(tp, optval); 2368 goto unlock_and_done; 2369 2370 case TCP_LOGBUF: 2371 INP_WUNLOCK(inp); 2372 error = EINVAL; 2373 break; 2374 2375 case TCP_LOGID: 2376 INP_WUNLOCK(inp); 2377 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2378 if (error) 2379 break; 2380 buf[sopt->sopt_valsize] = '\0'; 2381 INP_WLOCK_RECHECK(inp); 2382 error = tcp_log_set_id(tp, buf); 2383 /* tcp_log_set_id() unlocks the INP. */ 2384 break; 2385 2386 case TCP_LOGDUMP: 2387 case TCP_LOGDUMPID: 2388 INP_WUNLOCK(inp); 2389 error = 2390 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2391 if (error) 2392 break; 2393 buf[sopt->sopt_valsize] = '\0'; 2394 INP_WLOCK_RECHECK(inp); 2395 if (sopt->sopt_name == TCP_LOGDUMP) { 2396 error = tcp_log_dump_tp_logbuf(tp, buf, 2397 M_WAITOK, true); 2398 INP_WUNLOCK(inp); 2399 } else { 2400 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2401 /* 2402 * tcp_log_dump_tp_bucket_logbufs() drops the 2403 * INP lock. 2404 */ 2405 } 2406 break; 2407 #endif 2408 2409 default: 2410 INP_WUNLOCK(inp); 2411 error = ENOPROTOOPT; 2412 break; 2413 } 2414 break; 2415 2416 case SOPT_GET: 2417 tp = intotcpcb(inp); 2418 switch (sopt->sopt_name) { 2419 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2420 case TCP_MD5SIG: 2421 INP_WUNLOCK(inp); 2422 if (!TCPMD5_ENABLED()) 2423 return (ENOPROTOOPT); 2424 error = TCPMD5_PCBCTL(inp, sopt); 2425 break; 2426 #endif 2427 2428 case TCP_NODELAY: 2429 optval = tp->t_flags & TF_NODELAY; 2430 INP_WUNLOCK(inp); 2431 error = sooptcopyout(sopt, &optval, sizeof optval); 2432 break; 2433 case TCP_MAXSEG: 2434 optval = tp->t_maxseg; 2435 INP_WUNLOCK(inp); 2436 error = sooptcopyout(sopt, &optval, sizeof optval); 2437 break; 2438 case TCP_REMOTE_UDP_ENCAPS_PORT: 2439 optval = ntohs(tp->t_port); 2440 INP_WUNLOCK(inp); 2441 error = sooptcopyout(sopt, &optval, sizeof optval); 2442 break; 2443 case TCP_NOOPT: 2444 optval = tp->t_flags & TF_NOOPT; 2445 INP_WUNLOCK(inp); 2446 error = sooptcopyout(sopt, &optval, sizeof optval); 2447 break; 2448 case TCP_NOPUSH: 2449 optval = tp->t_flags & TF_NOPUSH; 2450 INP_WUNLOCK(inp); 2451 error = sooptcopyout(sopt, &optval, sizeof optval); 2452 break; 2453 case TCP_INFO: 2454 tcp_fill_info(tp, &ti); 2455 INP_WUNLOCK(inp); 2456 error = sooptcopyout(sopt, &ti, sizeof ti); 2457 break; 2458 case TCP_STATS: 2459 { 2460 #ifdef STATS 2461 int nheld; 2462 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2463 2464 error = 0; 2465 socklen_t outsbsz = sopt->sopt_valsize; 2466 if (tp->t_stats == NULL) 2467 error = ENOENT; 2468 else if (outsbsz >= tp->t_stats->cursz) 2469 outsbsz = tp->t_stats->cursz; 2470 else if (outsbsz >= sizeof(struct statsblob)) 2471 outsbsz = sizeof(struct statsblob); 2472 else 2473 error = EINVAL; 2474 INP_WUNLOCK(inp); 2475 if (error) 2476 break; 2477 2478 sbp = sopt->sopt_val; 2479 nheld = atop(round_page(((vm_offset_t)sbp) + 2480 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2481 vm_page_t ma[nheld]; 2482 if (vm_fault_quick_hold_pages( 2483 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2484 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2485 nheld) < 0) { 2486 error = EFAULT; 2487 break; 2488 } 2489 2490 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2491 SIZEOF_MEMBER(struct statsblob, flags)))) 2492 goto unhold; 2493 2494 INP_WLOCK_RECHECK(inp); 2495 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2496 sbflags | SB_CLONE_USRDSTNOFAULT); 2497 INP_WUNLOCK(inp); 2498 sopt->sopt_valsize = outsbsz; 2499 unhold: 2500 vm_page_unhold_pages(ma, nheld); 2501 #else 2502 INP_WUNLOCK(inp); 2503 error = EOPNOTSUPP; 2504 #endif /* !STATS */ 2505 break; 2506 } 2507 case TCP_CONGESTION: 2508 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2509 INP_WUNLOCK(inp); 2510 error = sooptcopyout(sopt, buf, len + 1); 2511 break; 2512 case TCP_MAXUNACKTIME: 2513 case TCP_KEEPIDLE: 2514 case TCP_KEEPINTVL: 2515 case TCP_KEEPINIT: 2516 case TCP_KEEPCNT: 2517 switch (sopt->sopt_name) { 2518 case TCP_MAXUNACKTIME: 2519 ui = TP_MAXUNACKTIME(tp) / hz; 2520 break; 2521 case TCP_KEEPIDLE: 2522 ui = TP_KEEPIDLE(tp) / hz; 2523 break; 2524 case TCP_KEEPINTVL: 2525 ui = TP_KEEPINTVL(tp) / hz; 2526 break; 2527 case TCP_KEEPINIT: 2528 ui = TP_KEEPINIT(tp) / hz; 2529 break; 2530 case TCP_KEEPCNT: 2531 ui = TP_KEEPCNT(tp); 2532 break; 2533 } 2534 INP_WUNLOCK(inp); 2535 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2536 break; 2537 #ifdef TCPPCAP 2538 case TCP_PCAP_OUT: 2539 case TCP_PCAP_IN: 2540 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2541 &(tp->t_outpkts) : &(tp->t_inpkts)); 2542 INP_WUNLOCK(inp); 2543 error = sooptcopyout(sopt, &optval, sizeof optval); 2544 break; 2545 #endif 2546 case TCP_FASTOPEN: 2547 optval = tp->t_flags & TF_FASTOPEN; 2548 INP_WUNLOCK(inp); 2549 error = sooptcopyout(sopt, &optval, sizeof optval); 2550 break; 2551 #ifdef TCP_BLACKBOX 2552 case TCP_LOG: 2553 optval = tp->t_logstate; 2554 INP_WUNLOCK(inp); 2555 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2556 break; 2557 case TCP_LOGBUF: 2558 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2559 error = tcp_log_getlogbuf(sopt, tp); 2560 break; 2561 case TCP_LOGID: 2562 len = tcp_log_get_id(tp, buf); 2563 INP_WUNLOCK(inp); 2564 error = sooptcopyout(sopt, buf, len + 1); 2565 break; 2566 case TCP_LOGDUMP: 2567 case TCP_LOGDUMPID: 2568 INP_WUNLOCK(inp); 2569 error = EINVAL; 2570 break; 2571 #endif 2572 #ifdef KERN_TLS 2573 case TCP_TXTLS_MODE: 2574 error = ktls_get_tx_mode(so, &optval); 2575 INP_WUNLOCK(inp); 2576 if (error == 0) 2577 error = sooptcopyout(sopt, &optval, 2578 sizeof(optval)); 2579 break; 2580 case TCP_RXTLS_MODE: 2581 error = ktls_get_rx_mode(so, &optval); 2582 INP_WUNLOCK(inp); 2583 if (error == 0) 2584 error = sooptcopyout(sopt, &optval, 2585 sizeof(optval)); 2586 break; 2587 #endif 2588 case TCP_LRD: 2589 optval = tp->t_flags & TF_LRD; 2590 INP_WUNLOCK(inp); 2591 error = sooptcopyout(sopt, &optval, sizeof optval); 2592 break; 2593 default: 2594 INP_WUNLOCK(inp); 2595 error = ENOPROTOOPT; 2596 break; 2597 } 2598 break; 2599 } 2600 return (error); 2601 } 2602 #undef INP_WLOCK_RECHECK 2603 #undef INP_WLOCK_RECHECK_CLEANUP 2604 2605 /* 2606 * Initiate (or continue) disconnect. 2607 * If embryonic state, just send reset (once). 2608 * If in ``let data drain'' option and linger null, just drop. 2609 * Otherwise (hard), mark socket disconnecting and drop 2610 * current input data; switch states based on user close, and 2611 * send segment to peer (with FIN). 2612 */ 2613 static void 2614 tcp_disconnect(struct tcpcb *tp) 2615 { 2616 struct inpcb *inp = tptoinpcb(tp); 2617 struct socket *so = tptosocket(tp); 2618 2619 NET_EPOCH_ASSERT(); 2620 INP_WLOCK_ASSERT(inp); 2621 2622 /* 2623 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2624 * socket is still open. 2625 */ 2626 if (tp->t_state < TCPS_ESTABLISHED && 2627 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2628 tp = tcp_close(tp); 2629 KASSERT(tp != NULL, 2630 ("tcp_disconnect: tcp_close() returned NULL")); 2631 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2632 tp = tcp_drop(tp, 0); 2633 KASSERT(tp != NULL, 2634 ("tcp_disconnect: tcp_drop() returned NULL")); 2635 } else { 2636 soisdisconnecting(so); 2637 sbflush(&so->so_rcv); 2638 tcp_usrclosed(tp); 2639 if (!(inp->inp_flags & INP_DROPPED)) 2640 /* Ignore stack's drop request, we already at it. */ 2641 (void)tcp_output_nodrop(tp); 2642 } 2643 } 2644 2645 /* 2646 * User issued close, and wish to trail through shutdown states: 2647 * if never received SYN, just forget it. If got a SYN from peer, 2648 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2649 * If already got a FIN from peer, then almost done; go to LAST_ACK 2650 * state. In all other cases, have already sent FIN to peer (e.g. 2651 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2652 * for peer to send FIN or not respond to keep-alives, etc. 2653 * We can let the user exit from the close as soon as the FIN is acked. 2654 */ 2655 static void 2656 tcp_usrclosed(struct tcpcb *tp) 2657 { 2658 2659 NET_EPOCH_ASSERT(); 2660 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2661 2662 switch (tp->t_state) { 2663 case TCPS_LISTEN: 2664 #ifdef TCP_OFFLOAD 2665 tcp_offload_listen_stop(tp); 2666 #endif 2667 tcp_state_change(tp, TCPS_CLOSED); 2668 /* FALLTHROUGH */ 2669 case TCPS_CLOSED: 2670 tp = tcp_close(tp); 2671 /* 2672 * tcp_close() should never return NULL here as the socket is 2673 * still open. 2674 */ 2675 KASSERT(tp != NULL, 2676 ("tcp_usrclosed: tcp_close() returned NULL")); 2677 break; 2678 2679 case TCPS_SYN_SENT: 2680 case TCPS_SYN_RECEIVED: 2681 tp->t_flags |= TF_NEEDFIN; 2682 break; 2683 2684 case TCPS_ESTABLISHED: 2685 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2686 break; 2687 2688 case TCPS_CLOSE_WAIT: 2689 tcp_state_change(tp, TCPS_LAST_ACK); 2690 break; 2691 } 2692 if (tp->t_acktime == 0) 2693 tp->t_acktime = ticks; 2694 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2695 soisdisconnected(tptosocket(tp)); 2696 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2697 if (tp->t_state == TCPS_FIN_WAIT_2) { 2698 int timeout; 2699 2700 timeout = (tcp_fast_finwait2_recycle) ? 2701 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2702 tcp_timer_activate(tp, TT_2MSL, timeout); 2703 } 2704 } 2705 } 2706 2707 #ifdef DDB 2708 static void 2709 db_print_indent(int indent) 2710 { 2711 int i; 2712 2713 for (i = 0; i < indent; i++) 2714 db_printf(" "); 2715 } 2716 2717 static void 2718 db_print_tstate(int t_state) 2719 { 2720 2721 switch (t_state) { 2722 case TCPS_CLOSED: 2723 db_printf("TCPS_CLOSED"); 2724 return; 2725 2726 case TCPS_LISTEN: 2727 db_printf("TCPS_LISTEN"); 2728 return; 2729 2730 case TCPS_SYN_SENT: 2731 db_printf("TCPS_SYN_SENT"); 2732 return; 2733 2734 case TCPS_SYN_RECEIVED: 2735 db_printf("TCPS_SYN_RECEIVED"); 2736 return; 2737 2738 case TCPS_ESTABLISHED: 2739 db_printf("TCPS_ESTABLISHED"); 2740 return; 2741 2742 case TCPS_CLOSE_WAIT: 2743 db_printf("TCPS_CLOSE_WAIT"); 2744 return; 2745 2746 case TCPS_FIN_WAIT_1: 2747 db_printf("TCPS_FIN_WAIT_1"); 2748 return; 2749 2750 case TCPS_CLOSING: 2751 db_printf("TCPS_CLOSING"); 2752 return; 2753 2754 case TCPS_LAST_ACK: 2755 db_printf("TCPS_LAST_ACK"); 2756 return; 2757 2758 case TCPS_FIN_WAIT_2: 2759 db_printf("TCPS_FIN_WAIT_2"); 2760 return; 2761 2762 case TCPS_TIME_WAIT: 2763 db_printf("TCPS_TIME_WAIT"); 2764 return; 2765 2766 default: 2767 db_printf("unknown"); 2768 return; 2769 } 2770 } 2771 2772 static void 2773 db_print_tflags(u_int t_flags) 2774 { 2775 int comma; 2776 2777 comma = 0; 2778 if (t_flags & TF_ACKNOW) { 2779 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2780 comma = 1; 2781 } 2782 if (t_flags & TF_DELACK) { 2783 db_printf("%sTF_DELACK", comma ? ", " : ""); 2784 comma = 1; 2785 } 2786 if (t_flags & TF_NODELAY) { 2787 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2788 comma = 1; 2789 } 2790 if (t_flags & TF_NOOPT) { 2791 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2792 comma = 1; 2793 } 2794 if (t_flags & TF_SENTFIN) { 2795 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2796 comma = 1; 2797 } 2798 if (t_flags & TF_REQ_SCALE) { 2799 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2800 comma = 1; 2801 } 2802 if (t_flags & TF_RCVD_SCALE) { 2803 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2804 comma = 1; 2805 } 2806 if (t_flags & TF_REQ_TSTMP) { 2807 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2808 comma = 1; 2809 } 2810 if (t_flags & TF_RCVD_TSTMP) { 2811 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2812 comma = 1; 2813 } 2814 if (t_flags & TF_SACK_PERMIT) { 2815 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2816 comma = 1; 2817 } 2818 if (t_flags & TF_NEEDSYN) { 2819 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2820 comma = 1; 2821 } 2822 if (t_flags & TF_NEEDFIN) { 2823 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2824 comma = 1; 2825 } 2826 if (t_flags & TF_NOPUSH) { 2827 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2828 comma = 1; 2829 } 2830 if (t_flags & TF_PREVVALID) { 2831 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2832 comma = 1; 2833 } 2834 if (t_flags & TF_MORETOCOME) { 2835 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2836 comma = 1; 2837 } 2838 if (t_flags & TF_SONOTCONN) { 2839 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2840 comma = 1; 2841 } 2842 if (t_flags & TF_LASTIDLE) { 2843 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2844 comma = 1; 2845 } 2846 if (t_flags & TF_RXWIN0SENT) { 2847 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2848 comma = 1; 2849 } 2850 if (t_flags & TF_FASTRECOVERY) { 2851 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2852 comma = 1; 2853 } 2854 if (t_flags & TF_CONGRECOVERY) { 2855 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2856 comma = 1; 2857 } 2858 if (t_flags & TF_WASFRECOVERY) { 2859 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2860 comma = 1; 2861 } 2862 if (t_flags & TF_WASCRECOVERY) { 2863 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2864 comma = 1; 2865 } 2866 if (t_flags & TF_SIGNATURE) { 2867 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2868 comma = 1; 2869 } 2870 if (t_flags & TF_FORCEDATA) { 2871 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2872 comma = 1; 2873 } 2874 if (t_flags & TF_TSO) { 2875 db_printf("%sTF_TSO", comma ? ", " : ""); 2876 comma = 1; 2877 } 2878 if (t_flags & TF_FASTOPEN) { 2879 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2880 comma = 1; 2881 } 2882 } 2883 2884 static void 2885 db_print_tflags2(u_int t_flags2) 2886 { 2887 int comma; 2888 2889 comma = 0; 2890 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2891 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2892 comma = 1; 2893 } 2894 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2895 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2896 comma = 1; 2897 } 2898 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2899 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2900 comma = 1; 2901 } 2902 if (t_flags2 & TF2_LOG_AUTO) { 2903 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2904 comma = 1; 2905 } 2906 if (t_flags2 & TF2_DROP_AF_DATA) { 2907 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2908 comma = 1; 2909 } 2910 if (t_flags2 & TF2_ECN_PERMIT) { 2911 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2912 comma = 1; 2913 } 2914 if (t_flags2 & TF2_ECN_SND_CWR) { 2915 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 2916 comma = 1; 2917 } 2918 if (t_flags2 & TF2_ECN_SND_ECE) { 2919 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 2920 comma = 1; 2921 } 2922 if (t_flags2 & TF2_ACE_PERMIT) { 2923 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 2924 comma = 1; 2925 } 2926 if (t_flags2 & TF2_FBYTES_COMPLETE) { 2927 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 2928 comma = 1; 2929 } 2930 } 2931 2932 static void 2933 db_print_toobflags(char t_oobflags) 2934 { 2935 int comma; 2936 2937 comma = 0; 2938 if (t_oobflags & TCPOOB_HAVEDATA) { 2939 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 2940 comma = 1; 2941 } 2942 if (t_oobflags & TCPOOB_HADDATA) { 2943 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 2944 comma = 1; 2945 } 2946 } 2947 2948 static void 2949 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 2950 { 2951 2952 db_print_indent(indent); 2953 db_printf("%s at %p\n", name, tp); 2954 2955 indent += 2; 2956 2957 db_print_indent(indent); 2958 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 2959 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 2960 2961 db_print_indent(indent); 2962 db_printf("t_callout: %p t_timers: %p\n", 2963 &tp->t_callout, &tp->t_timers); 2964 2965 db_print_indent(indent); 2966 db_printf("t_state: %d (", tp->t_state); 2967 db_print_tstate(tp->t_state); 2968 db_printf(")\n"); 2969 2970 db_print_indent(indent); 2971 db_printf("t_flags: 0x%x (", tp->t_flags); 2972 db_print_tflags(tp->t_flags); 2973 db_printf(")\n"); 2974 2975 db_print_indent(indent); 2976 db_printf("t_flags2: 0x%x (", tp->t_flags2); 2977 db_print_tflags2(tp->t_flags2); 2978 db_printf(")\n"); 2979 2980 db_print_indent(indent); 2981 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 2982 tp->snd_una, tp->snd_max, tp->snd_nxt); 2983 2984 db_print_indent(indent); 2985 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 2986 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 2987 2988 db_print_indent(indent); 2989 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 2990 tp->iss, tp->irs, tp->rcv_nxt); 2991 2992 db_print_indent(indent); 2993 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 2994 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 2995 2996 db_print_indent(indent); 2997 db_printf("snd_wnd: %u snd_cwnd: %u\n", 2998 tp->snd_wnd, tp->snd_cwnd); 2999 3000 db_print_indent(indent); 3001 db_printf("snd_ssthresh: %u snd_recover: " 3002 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3003 3004 db_print_indent(indent); 3005 db_printf("t_rcvtime: %u t_startime: %u\n", 3006 tp->t_rcvtime, tp->t_starttime); 3007 3008 db_print_indent(indent); 3009 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3010 tp->t_rtttime, tp->t_rtseq); 3011 3012 db_print_indent(indent); 3013 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3014 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3015 3016 db_print_indent(indent); 3017 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3018 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3019 3020 db_print_indent(indent); 3021 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3022 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3023 3024 db_print_indent(indent); 3025 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3026 db_print_toobflags(tp->t_oobflags); 3027 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3028 3029 db_print_indent(indent); 3030 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3031 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3032 3033 db_print_indent(indent); 3034 db_printf("ts_recent: %u ts_recent_age: %u\n", 3035 tp->ts_recent, tp->ts_recent_age); 3036 3037 db_print_indent(indent); 3038 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3039 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3040 3041 db_print_indent(indent); 3042 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3043 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3044 tp->snd_recover_prev, tp->t_badrxtwin); 3045 3046 db_print_indent(indent); 3047 db_printf("snd_numholes: %d snd_holes first: %p\n", 3048 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3049 3050 db_print_indent(indent); 3051 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3052 tp->snd_fack, tp->rcv_numsacks); 3053 3054 /* Skip sackblks, sackhint. */ 3055 3056 db_print_indent(indent); 3057 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3058 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3059 } 3060 3061 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3062 { 3063 struct tcpcb *tp; 3064 3065 if (!have_addr) { 3066 db_printf("usage: show tcpcb <addr>\n"); 3067 return; 3068 } 3069 tp = (struct tcpcb *)addr; 3070 3071 db_print_tcpcb(tp, "tcpcb", 0); 3072 } 3073 #endif 3074