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 38 #include <sys/cdefs.h> 39 #include "opt_ddb.h" 40 #include "opt_inet.h" 41 #include "opt_inet6.h" 42 #include "opt_ipsec.h" 43 #include "opt_kern_tls.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/arb.h> 48 #include <sys/limits.h> 49 #include <sys/malloc.h> 50 #include <sys/refcount.h> 51 #include <sys/kernel.h> 52 #include <sys/ktls.h> 53 #include <sys/qmath.h> 54 #include <sys/sysctl.h> 55 #include <sys/mbuf.h> 56 #ifdef INET6 57 #include <sys/domain.h> 58 #endif /* INET6 */ 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/protosw.h> 62 #include <sys/proc.h> 63 #include <sys/jail.h> 64 #include <sys/stats.h> 65 66 #ifdef DDB 67 #include <ddb/ddb.h> 68 #endif 69 70 #include <net/if.h> 71 #include <net/if_var.h> 72 #include <net/route.h> 73 #include <net/vnet.h> 74 75 #include <netinet/in.h> 76 #include <netinet/in_kdtrace.h> 77 #include <netinet/in_pcb.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/in_var.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 #ifdef INET6 83 #include <netinet/ip6.h> 84 #include <netinet6/in6_pcb.h> 85 #include <netinet6/ip6_var.h> 86 #include <netinet6/scope6_var.h> 87 #endif 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/tcp_log_buf.h> 94 #include <netinet/tcpip.h> 95 #include <netinet/cc/cc.h> 96 #include <netinet/tcp_fastopen.h> 97 #include <netinet/tcp_hpts.h> 98 #ifdef TCPPCAP 99 #include <netinet/tcp_pcap.h> 100 #endif 101 #ifdef TCP_OFFLOAD 102 #include <netinet/tcp_offload.h> 103 #endif 104 #include <netipsec/ipsec_support.h> 105 106 #include <vm/vm.h> 107 #include <vm/vm_param.h> 108 #include <vm/pmap.h> 109 #include <vm/vm_extern.h> 110 #include <vm/vm_map.h> 111 #include <vm/vm_page.h> 112 113 /* 114 * TCP protocol interface to socket abstraction. 115 */ 116 #ifdef INET 117 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 118 struct thread *td); 119 #endif /* INET */ 120 #ifdef INET6 121 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 122 struct thread *td); 123 #endif /* INET6 */ 124 static void tcp_disconnect(struct tcpcb *); 125 static void tcp_usrclosed(struct tcpcb *); 126 static void tcp_fill_info(const struct tcpcb *, struct tcp_info *); 127 128 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 129 130 static void 131 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error) 132 { 133 struct tcp_log_buffer *lgb; 134 135 KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL")); 136 INP_WLOCK_ASSERT(tptoinpcb(tp)); 137 if (tcp_bblogging_on(tp)) { 138 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error, 139 0, NULL, false, NULL, NULL, 0, NULL); 140 } else { 141 lgb = NULL; 142 } 143 if (lgb != NULL) { 144 if (error >= 0) { 145 lgb->tlb_errno = (uint32_t)error; 146 } 147 lgb->tlb_flex1 = pru; 148 } 149 } 150 151 /* 152 * TCP attaches to socket via pru_attach(), reserving space, 153 * and an internet control block. 154 */ 155 static int 156 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 157 { 158 struct inpcb *inp; 159 struct tcpcb *tp = NULL; 160 int error; 161 162 inp = sotoinpcb(so); 163 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 164 165 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 166 if (error) 167 goto out; 168 169 so->so_rcv.sb_flags |= SB_AUTOSIZE; 170 so->so_snd.sb_flags |= SB_AUTOSIZE; 171 error = in_pcballoc(so, &V_tcbinfo); 172 if (error) 173 goto out; 174 inp = sotoinpcb(so); 175 tp = tcp_newtcpcb(inp); 176 if (tp == NULL) { 177 error = ENOBUFS; 178 in_pcbfree(inp); 179 goto out; 180 } 181 tp->t_state = TCPS_CLOSED; 182 tcp_bblog_pru(tp, PRU_ATTACH, error); 183 INP_WUNLOCK(inp); 184 TCPSTATES_INC(TCPS_CLOSED); 185 out: 186 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 187 return (error); 188 } 189 190 /* 191 * tcp_usr_detach is called when the socket layer loses its final reference 192 * to the socket, be it a file descriptor reference, a reference from TCP, 193 * etc. At this point, there is only one case in which we will keep around 194 * inpcb state: time wait. 195 */ 196 static void 197 tcp_usr_detach(struct socket *so) 198 { 199 struct inpcb *inp; 200 struct tcpcb *tp; 201 202 inp = sotoinpcb(so); 203 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 204 INP_WLOCK(inp); 205 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 206 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 207 208 tp = intotcpcb(inp); 209 210 KASSERT(inp->inp_flags & INP_DROPPED || 211 tp->t_state < TCPS_SYN_SENT, 212 ("%s: inp %p not dropped or embryonic", __func__, inp)); 213 214 tcp_discardcb(tp); 215 in_pcbfree(inp); 216 } 217 218 #ifdef INET 219 /* 220 * Give the socket an address. 221 */ 222 static int 223 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 224 { 225 int error = 0; 226 struct inpcb *inp; 227 struct tcpcb *tp; 228 struct sockaddr_in *sinp; 229 230 inp = sotoinpcb(so); 231 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 232 INP_WLOCK(inp); 233 if (inp->inp_flags & INP_DROPPED) { 234 INP_WUNLOCK(inp); 235 return (EINVAL); 236 } 237 tp = intotcpcb(inp); 238 239 sinp = (struct sockaddr_in *)nam; 240 if (nam->sa_family != AF_INET) { 241 /* 242 * Preserve compatibility with old programs. 243 */ 244 if (nam->sa_family != AF_UNSPEC || 245 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 246 sinp->sin_addr.s_addr != INADDR_ANY) { 247 error = EAFNOSUPPORT; 248 goto out; 249 } 250 nam->sa_family = AF_INET; 251 } 252 if (nam->sa_len != sizeof(*sinp)) { 253 error = EINVAL; 254 goto out; 255 } 256 /* 257 * Must check for multicast addresses and disallow binding 258 * to them. 259 */ 260 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 261 error = EAFNOSUPPORT; 262 goto out; 263 } 264 INP_HASH_WLOCK(&V_tcbinfo); 265 error = in_pcbbind(inp, sinp, td->td_ucred); 266 INP_HASH_WUNLOCK(&V_tcbinfo); 267 out: 268 tcp_bblog_pru(tp, PRU_BIND, error); 269 TCP_PROBE2(debug__user, tp, PRU_BIND); 270 INP_WUNLOCK(inp); 271 272 return (error); 273 } 274 #endif /* INET */ 275 276 #ifdef INET6 277 static int 278 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 279 { 280 int error = 0; 281 struct inpcb *inp; 282 struct tcpcb *tp; 283 struct sockaddr_in6 *sin6; 284 u_char vflagsav; 285 286 inp = sotoinpcb(so); 287 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 288 INP_WLOCK(inp); 289 if (inp->inp_flags & INP_DROPPED) { 290 INP_WUNLOCK(inp); 291 return (EINVAL); 292 } 293 tp = intotcpcb(inp); 294 295 vflagsav = inp->inp_vflag; 296 297 sin6 = (struct sockaddr_in6 *)nam; 298 if (nam->sa_family != AF_INET6) { 299 error = EAFNOSUPPORT; 300 goto out; 301 } 302 if (nam->sa_len != sizeof(*sin6)) { 303 error = EINVAL; 304 goto out; 305 } 306 /* 307 * Must check for multicast addresses and disallow binding 308 * to them. 309 */ 310 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 311 error = EAFNOSUPPORT; 312 goto out; 313 } 314 315 INP_HASH_WLOCK(&V_tcbinfo); 316 inp->inp_vflag &= ~INP_IPV4; 317 inp->inp_vflag |= INP_IPV6; 318 #ifdef INET 319 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 320 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 321 inp->inp_vflag |= INP_IPV4; 322 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 323 struct sockaddr_in sin; 324 325 in6_sin6_2_sin(&sin, sin6); 326 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 327 error = EAFNOSUPPORT; 328 INP_HASH_WUNLOCK(&V_tcbinfo); 329 goto out; 330 } 331 inp->inp_vflag |= INP_IPV4; 332 inp->inp_vflag &= ~INP_IPV6; 333 error = in_pcbbind(inp, &sin, td->td_ucred); 334 INP_HASH_WUNLOCK(&V_tcbinfo); 335 goto out; 336 } 337 } 338 #endif 339 error = in6_pcbbind(inp, sin6, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 out: 342 if (error != 0) 343 inp->inp_vflag = vflagsav; 344 tcp_bblog_pru(tp, PRU_BIND, error); 345 TCP_PROBE2(debug__user, tp, PRU_BIND); 346 INP_WUNLOCK(inp); 347 return (error); 348 } 349 #endif /* INET6 */ 350 351 #ifdef INET 352 /* 353 * Prepare to accept connections. 354 */ 355 static int 356 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 357 { 358 int error = 0; 359 struct inpcb *inp; 360 struct tcpcb *tp; 361 362 inp = sotoinpcb(so); 363 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 364 INP_WLOCK(inp); 365 if (inp->inp_flags & INP_DROPPED) { 366 INP_WUNLOCK(inp); 367 return (EINVAL); 368 } 369 tp = intotcpcb(inp); 370 371 SOCK_LOCK(so); 372 error = solisten_proto_check(so); 373 if (error != 0) { 374 SOCK_UNLOCK(so); 375 goto out; 376 } 377 if (inp->inp_lport == 0) { 378 INP_HASH_WLOCK(&V_tcbinfo); 379 error = in_pcbbind(inp, NULL, td->td_ucred); 380 INP_HASH_WUNLOCK(&V_tcbinfo); 381 } 382 if (error == 0) { 383 tcp_state_change(tp, TCPS_LISTEN); 384 solisten_proto(so, backlog); 385 #ifdef TCP_OFFLOAD 386 if ((so->so_options & SO_NO_OFFLOAD) == 0) 387 tcp_offload_listen_start(tp); 388 #endif 389 } else { 390 solisten_proto_abort(so); 391 } 392 SOCK_UNLOCK(so); 393 394 if (IS_FASTOPEN(tp->t_flags)) 395 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 396 397 out: 398 tcp_bblog_pru(tp, PRU_LISTEN, error); 399 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 400 INP_WUNLOCK(inp); 401 return (error); 402 } 403 #endif /* INET */ 404 405 #ifdef INET6 406 static int 407 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 408 { 409 int error = 0; 410 struct inpcb *inp; 411 struct tcpcb *tp; 412 u_char vflagsav; 413 414 inp = sotoinpcb(so); 415 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 416 INP_WLOCK(inp); 417 if (inp->inp_flags & INP_DROPPED) { 418 INP_WUNLOCK(inp); 419 return (EINVAL); 420 } 421 tp = intotcpcb(inp); 422 423 vflagsav = inp->inp_vflag; 424 425 SOCK_LOCK(so); 426 error = solisten_proto_check(so); 427 if (error != 0) { 428 SOCK_UNLOCK(so); 429 goto out; 430 } 431 INP_HASH_WLOCK(&V_tcbinfo); 432 if (inp->inp_lport == 0) { 433 inp->inp_vflag &= ~INP_IPV4; 434 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 435 inp->inp_vflag |= INP_IPV4; 436 error = in6_pcbbind(inp, NULL, td->td_ucred); 437 } 438 INP_HASH_WUNLOCK(&V_tcbinfo); 439 if (error == 0) { 440 tcp_state_change(tp, TCPS_LISTEN); 441 solisten_proto(so, backlog); 442 #ifdef TCP_OFFLOAD 443 if ((so->so_options & SO_NO_OFFLOAD) == 0) 444 tcp_offload_listen_start(tp); 445 #endif 446 } else { 447 solisten_proto_abort(so); 448 } 449 SOCK_UNLOCK(so); 450 451 if (IS_FASTOPEN(tp->t_flags)) 452 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 453 454 if (error != 0) 455 inp->inp_vflag = vflagsav; 456 457 out: 458 tcp_bblog_pru(tp, PRU_LISTEN, error); 459 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 460 INP_WUNLOCK(inp); 461 return (error); 462 } 463 #endif /* INET6 */ 464 465 #ifdef INET 466 /* 467 * Initiate connection to peer. 468 * Create a template for use in transmissions on this connection. 469 * Enter SYN_SENT state, and mark socket as connecting. 470 * Start keep-alive timer, and seed output sequence space. 471 * Send initial segment on connection. 472 */ 473 static int 474 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 475 { 476 struct epoch_tracker et; 477 int error = 0; 478 struct inpcb *inp; 479 struct tcpcb *tp; 480 struct sockaddr_in *sinp; 481 482 inp = sotoinpcb(so); 483 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 484 INP_WLOCK(inp); 485 if (inp->inp_flags & INP_DROPPED) { 486 INP_WUNLOCK(inp); 487 return (ECONNREFUSED); 488 } 489 tp = intotcpcb(inp); 490 491 sinp = (struct sockaddr_in *)nam; 492 if (nam->sa_family != AF_INET) { 493 error = EAFNOSUPPORT; 494 goto out; 495 } 496 if (nam->sa_len != sizeof (*sinp)) { 497 error = EINVAL; 498 goto out; 499 } 500 /* 501 * Must disallow TCP ``connections'' to multicast addresses. 502 */ 503 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 504 error = EAFNOSUPPORT; 505 goto out; 506 } 507 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 508 error = EACCES; 509 goto out; 510 } 511 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 512 goto out; 513 if (SOLISTENING(so)) { 514 error = EOPNOTSUPP; 515 goto out; 516 } 517 NET_EPOCH_ENTER(et); 518 if ((error = tcp_connect(tp, sinp, td)) != 0) 519 goto out_in_epoch; 520 #ifdef TCP_OFFLOAD 521 if (registered_toedevs > 0 && 522 (so->so_options & SO_NO_OFFLOAD) == 0 && 523 (error = tcp_offload_connect(so, nam)) == 0) 524 goto out_in_epoch; 525 #endif 526 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 527 error = tcp_output(tp); 528 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 529 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 530 out_in_epoch: 531 NET_EPOCH_EXIT(et); 532 out: 533 tcp_bblog_pru(tp, PRU_CONNECT, error); 534 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 535 INP_WUNLOCK(inp); 536 return (error); 537 } 538 #endif /* INET */ 539 540 #ifdef INET6 541 static int 542 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 543 { 544 struct epoch_tracker et; 545 int error = 0; 546 struct inpcb *inp; 547 struct tcpcb *tp; 548 struct sockaddr_in6 *sin6; 549 u_int8_t incflagsav; 550 u_char vflagsav; 551 552 inp = sotoinpcb(so); 553 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 554 INP_WLOCK(inp); 555 if (inp->inp_flags & INP_DROPPED) { 556 INP_WUNLOCK(inp); 557 return (ECONNREFUSED); 558 } 559 tp = intotcpcb(inp); 560 561 vflagsav = inp->inp_vflag; 562 incflagsav = inp->inp_inc.inc_flags; 563 564 sin6 = (struct sockaddr_in6 *)nam; 565 if (nam->sa_family != AF_INET6) { 566 error = EAFNOSUPPORT; 567 goto out; 568 } 569 if (nam->sa_len != sizeof (*sin6)) { 570 error = EINVAL; 571 goto out; 572 } 573 /* 574 * Must disallow TCP ``connections'' to multicast addresses. 575 */ 576 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 577 error = EAFNOSUPPORT; 578 goto out; 579 } 580 if (SOLISTENING(so)) { 581 error = EINVAL; 582 goto out; 583 } 584 #ifdef INET 585 /* 586 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 587 * therefore probably require the hash lock, which isn't held here. 588 * Is this a significant problem? 589 */ 590 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 591 struct sockaddr_in sin; 592 593 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 594 error = EINVAL; 595 goto out; 596 } 597 if ((inp->inp_vflag & INP_IPV4) == 0) { 598 error = EAFNOSUPPORT; 599 goto out; 600 } 601 602 in6_sin6_2_sin(&sin, sin6); 603 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 604 error = EAFNOSUPPORT; 605 goto out; 606 } 607 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 608 error = EACCES; 609 goto out; 610 } 611 if ((error = prison_remote_ip4(td->td_ucred, 612 &sin.sin_addr)) != 0) 613 goto out; 614 inp->inp_vflag |= INP_IPV4; 615 inp->inp_vflag &= ~INP_IPV6; 616 NET_EPOCH_ENTER(et); 617 if ((error = tcp_connect(tp, &sin, td)) != 0) 618 goto out_in_epoch; 619 #ifdef TCP_OFFLOAD 620 if (registered_toedevs > 0 && 621 (so->so_options & SO_NO_OFFLOAD) == 0 && 622 (error = tcp_offload_connect(so, nam)) == 0) 623 goto out_in_epoch; 624 #endif 625 error = tcp_output(tp); 626 goto out_in_epoch; 627 } else { 628 if ((inp->inp_vflag & INP_IPV6) == 0) { 629 error = EAFNOSUPPORT; 630 goto out; 631 } 632 } 633 #endif 634 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 635 goto out; 636 inp->inp_vflag &= ~INP_IPV4; 637 inp->inp_vflag |= INP_IPV6; 638 inp->inp_inc.inc_flags |= INC_ISIPV6; 639 NET_EPOCH_ENTER(et); 640 if ((error = tcp6_connect(tp, sin6, td)) != 0) 641 goto out_in_epoch; 642 #ifdef TCP_OFFLOAD 643 if (registered_toedevs > 0 && 644 (so->so_options & SO_NO_OFFLOAD) == 0 && 645 (error = tcp_offload_connect(so, nam)) == 0) 646 goto out_in_epoch; 647 #endif 648 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 649 error = tcp_output(tp); 650 out_in_epoch: 651 NET_EPOCH_EXIT(et); 652 out: 653 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 654 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 655 /* 656 * If the implicit bind in the connect call fails, restore 657 * the flags we modified. 658 */ 659 if (error != 0 && inp->inp_lport == 0) { 660 inp->inp_vflag = vflagsav; 661 inp->inp_inc.inc_flags = incflagsav; 662 } 663 664 tcp_bblog_pru(tp, PRU_CONNECT, error); 665 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 666 INP_WUNLOCK(inp); 667 return (error); 668 } 669 #endif /* INET6 */ 670 671 /* 672 * Initiate disconnect from peer. 673 * If connection never passed embryonic stage, just drop; 674 * else if don't need to let data drain, then can just drop anyways, 675 * else have to begin TCP shutdown process: mark socket disconnecting, 676 * drain unread data, state switch to reflect user close, and 677 * send segment (e.g. FIN) to peer. Socket will be really disconnected 678 * when peer sends FIN and acks ours. 679 * 680 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 681 */ 682 static int 683 tcp_usr_disconnect(struct socket *so) 684 { 685 struct inpcb *inp; 686 struct tcpcb *tp = NULL; 687 struct epoch_tracker et; 688 int error = 0; 689 690 NET_EPOCH_ENTER(et); 691 inp = sotoinpcb(so); 692 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 693 INP_WLOCK(inp); 694 if (inp->inp_flags & INP_DROPPED) { 695 INP_WUNLOCK(inp); 696 NET_EPOCH_EXIT(et); 697 return (ECONNRESET); 698 } 699 tp = intotcpcb(inp); 700 701 if (tp->t_state == TCPS_TIME_WAIT) 702 goto out; 703 tcp_disconnect(tp); 704 out: 705 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 706 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 707 INP_WUNLOCK(inp); 708 NET_EPOCH_EXIT(et); 709 return (error); 710 } 711 712 #ifdef INET 713 /* 714 * Accept a connection. Essentially all the work is done at higher levels; 715 * just return the address of the peer, storing through addr. 716 */ 717 static int 718 tcp_usr_accept(struct socket *so, struct sockaddr *sa) 719 { 720 struct inpcb *inp; 721 struct tcpcb *tp; 722 int error = 0; 723 724 inp = sotoinpcb(so); 725 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 726 INP_WLOCK(inp); 727 if (inp->inp_flags & INP_DROPPED) { 728 INP_WUNLOCK(inp); 729 return (ECONNABORTED); 730 } 731 tp = intotcpcb(inp); 732 733 if (so->so_state & SS_ISDISCONNECTED) 734 error = ECONNABORTED; 735 else 736 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 737 .sin_family = AF_INET, 738 .sin_len = sizeof(struct sockaddr_in), 739 .sin_port = inp->inp_fport, 740 .sin_addr = inp->inp_faddr, 741 }; 742 tcp_bblog_pru(tp, PRU_ACCEPT, error); 743 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 744 INP_WUNLOCK(inp); 745 746 return (error); 747 } 748 #endif /* INET */ 749 750 #ifdef INET6 751 static int 752 tcp6_usr_accept(struct socket *so, struct sockaddr *sa) 753 { 754 struct inpcb *inp; 755 struct tcpcb *tp; 756 int error = 0; 757 758 inp = sotoinpcb(so); 759 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 760 INP_WLOCK(inp); 761 if (inp->inp_flags & INP_DROPPED) { 762 INP_WUNLOCK(inp); 763 return (ECONNABORTED); 764 } 765 tp = intotcpcb(inp); 766 767 if (so->so_state & SS_ISDISCONNECTED) { 768 error = ECONNABORTED; 769 } else { 770 if (inp->inp_vflag & INP_IPV4) { 771 struct sockaddr_in sin = { 772 .sin_family = AF_INET, 773 .sin_len = sizeof(struct sockaddr_in), 774 .sin_port = inp->inp_fport, 775 .sin_addr = inp->inp_faddr, 776 }; 777 in6_sin_2_v4mapsin6(&sin, (struct sockaddr_in6 *)sa); 778 } else { 779 *(struct sockaddr_in6 *)sa = (struct sockaddr_in6 ){ 780 .sin6_family = AF_INET6, 781 .sin6_len = sizeof(struct sockaddr_in6), 782 .sin6_port = inp->inp_fport, 783 .sin6_addr = inp->in6p_faddr, 784 }; 785 /* XXX: should catch errors */ 786 (void)sa6_recoverscope((struct sockaddr_in6 *)sa); 787 } 788 } 789 790 tcp_bblog_pru(tp, PRU_ACCEPT, error); 791 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 792 INP_WUNLOCK(inp); 793 794 return (error); 795 } 796 #endif /* INET6 */ 797 798 /* 799 * Mark the connection as being incapable of further output. 800 */ 801 static int 802 tcp_usr_shutdown(struct socket *so) 803 { 804 int error = 0; 805 struct inpcb *inp; 806 struct tcpcb *tp; 807 struct epoch_tracker et; 808 809 inp = sotoinpcb(so); 810 KASSERT(inp != NULL, ("inp == NULL")); 811 INP_WLOCK(inp); 812 if (inp->inp_flags & INP_DROPPED) { 813 INP_WUNLOCK(inp); 814 return (ECONNRESET); 815 } 816 tp = intotcpcb(inp); 817 818 NET_EPOCH_ENTER(et); 819 socantsendmore(so); 820 tcp_usrclosed(tp); 821 if (!(inp->inp_flags & INP_DROPPED)) 822 error = tcp_output_nodrop(tp); 823 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 824 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 825 error = tcp_unlock_or_drop(tp, error); 826 NET_EPOCH_EXIT(et); 827 828 return (error); 829 } 830 831 /* 832 * After a receive, possibly send window update to peer. 833 */ 834 static int 835 tcp_usr_rcvd(struct socket *so, int flags) 836 { 837 struct epoch_tracker et; 838 struct inpcb *inp; 839 struct tcpcb *tp; 840 int outrv = 0, error = 0; 841 842 inp = sotoinpcb(so); 843 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 844 INP_WLOCK(inp); 845 if (inp->inp_flags & INP_DROPPED) { 846 INP_WUNLOCK(inp); 847 return (ECONNRESET); 848 } 849 tp = intotcpcb(inp); 850 851 NET_EPOCH_ENTER(et); 852 /* 853 * For passively-created TFO connections, don't attempt a window 854 * update while still in SYN_RECEIVED as this may trigger an early 855 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 856 * application response data, or failing that, when the DELACK timer 857 * expires. 858 */ 859 if (IS_FASTOPEN(tp->t_flags) && 860 (tp->t_state == TCPS_SYN_RECEIVED)) 861 goto out; 862 #ifdef TCP_OFFLOAD 863 if (tp->t_flags & TF_TOE) 864 tcp_offload_rcvd(tp); 865 else 866 #endif 867 outrv = tcp_output_nodrop(tp); 868 out: 869 tcp_bblog_pru(tp, PRU_RCVD, error); 870 TCP_PROBE2(debug__user, tp, PRU_RCVD); 871 (void) tcp_unlock_or_drop(tp, outrv); 872 NET_EPOCH_EXIT(et); 873 return (error); 874 } 875 876 /* 877 * Do a send by putting data in output queue and updating urgent 878 * marker if URG set. Possibly send more data. Unlike the other 879 * pru_*() routines, the mbuf chains are our responsibility. We 880 * must either enqueue them or free them. The other pru_* routines 881 * generally are caller-frees. 882 */ 883 static int 884 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 885 struct sockaddr *nam, struct mbuf *control, struct thread *td) 886 { 887 struct epoch_tracker et; 888 int error = 0; 889 struct inpcb *inp; 890 struct tcpcb *tp; 891 #ifdef INET 892 #ifdef INET6 893 struct sockaddr_in sin; 894 #endif 895 struct sockaddr_in *sinp; 896 #endif 897 #ifdef INET6 898 struct sockaddr_in6 *sin6; 899 int isipv6; 900 #endif 901 u_int8_t incflagsav; 902 u_char vflagsav; 903 bool restoreflags; 904 905 inp = sotoinpcb(so); 906 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 907 INP_WLOCK(inp); 908 if (inp->inp_flags & INP_DROPPED) { 909 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 910 m_freem(m); 911 INP_WUNLOCK(inp); 912 return (ECONNRESET); 913 } 914 tp = intotcpcb(inp); 915 916 vflagsav = inp->inp_vflag; 917 incflagsav = inp->inp_inc.inc_flags; 918 restoreflags = false; 919 920 NET_EPOCH_ENTER(et); 921 if (control != NULL) { 922 /* TCP doesn't do control messages (rights, creds, etc) */ 923 if (control->m_len > 0) { 924 m_freem(control); 925 error = EINVAL; 926 goto out; 927 } 928 m_freem(control); /* empty control, just free it */ 929 } 930 931 if ((flags & PRUS_OOB) != 0 && 932 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 933 goto out; 934 935 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 936 if (tp->t_state == TCPS_LISTEN) { 937 error = EINVAL; 938 goto out; 939 } 940 switch (nam->sa_family) { 941 #ifdef INET 942 case AF_INET: 943 sinp = (struct sockaddr_in *)nam; 944 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 945 error = EINVAL; 946 goto out; 947 } 948 if ((inp->inp_vflag & INP_IPV6) != 0) { 949 error = EAFNOSUPPORT; 950 goto out; 951 } 952 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 953 error = EAFNOSUPPORT; 954 goto out; 955 } 956 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 957 error = EACCES; 958 goto out; 959 } 960 if ((error = prison_remote_ip4(td->td_ucred, 961 &sinp->sin_addr))) 962 goto out; 963 #ifdef INET6 964 isipv6 = 0; 965 #endif 966 break; 967 #endif /* INET */ 968 #ifdef INET6 969 case AF_INET6: 970 sin6 = (struct sockaddr_in6 *)nam; 971 if (sin6->sin6_len != sizeof(*sin6)) { 972 error = EINVAL; 973 goto out; 974 } 975 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 976 error = EAFNOSUPPORT; 977 goto out; 978 } 979 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 980 error = EAFNOSUPPORT; 981 goto out; 982 } 983 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 984 #ifdef INET 985 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 986 error = EINVAL; 987 goto out; 988 } 989 if ((inp->inp_vflag & INP_IPV4) == 0) { 990 error = EAFNOSUPPORT; 991 goto out; 992 } 993 restoreflags = true; 994 inp->inp_vflag &= ~INP_IPV6; 995 sinp = &sin; 996 in6_sin6_2_sin(sinp, sin6); 997 if (IN_MULTICAST( 998 ntohl(sinp->sin_addr.s_addr))) { 999 error = EAFNOSUPPORT; 1000 goto out; 1001 } 1002 if ((error = prison_remote_ip4(td->td_ucred, 1003 &sinp->sin_addr))) 1004 goto out; 1005 isipv6 = 0; 1006 #else /* !INET */ 1007 error = EAFNOSUPPORT; 1008 goto out; 1009 #endif /* INET */ 1010 } else { 1011 if ((inp->inp_vflag & INP_IPV6) == 0) { 1012 error = EAFNOSUPPORT; 1013 goto out; 1014 } 1015 restoreflags = true; 1016 inp->inp_vflag &= ~INP_IPV4; 1017 inp->inp_inc.inc_flags |= INC_ISIPV6; 1018 if ((error = prison_remote_ip6(td->td_ucred, 1019 &sin6->sin6_addr))) 1020 goto out; 1021 isipv6 = 1; 1022 } 1023 break; 1024 #endif /* INET6 */ 1025 default: 1026 error = EAFNOSUPPORT; 1027 goto out; 1028 } 1029 } 1030 if (!(flags & PRUS_OOB)) { 1031 if (tp->t_acktime == 0) 1032 tp->t_acktime = ticks; 1033 sbappendstream(&so->so_snd, m, flags); 1034 m = NULL; 1035 if (nam && tp->t_state < TCPS_SYN_SENT) { 1036 KASSERT(tp->t_state == TCPS_CLOSED, 1037 ("%s: tp %p is listening", __func__, tp)); 1038 1039 /* 1040 * Do implied connect if not yet connected, 1041 * initialize window to default value, and 1042 * initialize maxseg using peer's cached MSS. 1043 */ 1044 #ifdef INET6 1045 if (isipv6) 1046 error = tcp6_connect(tp, sin6, td); 1047 #endif /* INET6 */ 1048 #if defined(INET6) && defined(INET) 1049 else 1050 #endif 1051 #ifdef INET 1052 error = tcp_connect(tp, sinp, td); 1053 #endif 1054 /* 1055 * The bind operation in tcp_connect succeeded. We 1056 * no longer want to restore the flags if later 1057 * operations fail. 1058 */ 1059 if (error == 0 || inp->inp_lport != 0) 1060 restoreflags = false; 1061 1062 if (error) { 1063 /* m is freed if PRUS_NOTREADY is unset. */ 1064 sbflush(&so->so_snd); 1065 goto out; 1066 } 1067 if (IS_FASTOPEN(tp->t_flags)) 1068 tcp_fastopen_connect(tp); 1069 else { 1070 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1071 tcp_mss(tp, -1); 1072 } 1073 } 1074 if (flags & PRUS_EOF) { 1075 /* 1076 * Close the send side of the connection after 1077 * the data is sent. 1078 */ 1079 socantsendmore(so); 1080 tcp_usrclosed(tp); 1081 } 1082 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1083 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1084 (tp->t_fbyte_out == 0) && 1085 (so->so_snd.sb_ccc > 0)) { 1086 tp->t_fbyte_out = ticks; 1087 if (tp->t_fbyte_out == 0) 1088 tp->t_fbyte_out = 1; 1089 if (tp->t_fbyte_out && tp->t_fbyte_in) 1090 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1091 } 1092 if (!(inp->inp_flags & INP_DROPPED) && 1093 !(flags & PRUS_NOTREADY)) { 1094 if (flags & PRUS_MORETOCOME) 1095 tp->t_flags |= TF_MORETOCOME; 1096 error = tcp_output_nodrop(tp); 1097 if (flags & PRUS_MORETOCOME) 1098 tp->t_flags &= ~TF_MORETOCOME; 1099 } 1100 } else { 1101 /* 1102 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1103 */ 1104 SOCKBUF_LOCK(&so->so_snd); 1105 if (sbspace(&so->so_snd) < -512) { 1106 SOCKBUF_UNLOCK(&so->so_snd); 1107 error = ENOBUFS; 1108 goto out; 1109 } 1110 /* 1111 * According to RFC961 (Assigned Protocols), 1112 * the urgent pointer points to the last octet 1113 * of urgent data. We continue, however, 1114 * to consider it to indicate the first octet 1115 * of data past the urgent section. 1116 * Otherwise, snd_up should be one lower. 1117 */ 1118 if (tp->t_acktime == 0) 1119 tp->t_acktime = ticks; 1120 sbappendstream_locked(&so->so_snd, m, flags); 1121 SOCKBUF_UNLOCK(&so->so_snd); 1122 m = NULL; 1123 if (nam && tp->t_state < TCPS_SYN_SENT) { 1124 /* 1125 * Do implied connect if not yet connected, 1126 * initialize window to default value, and 1127 * initialize maxseg using peer's cached MSS. 1128 */ 1129 1130 /* 1131 * Not going to contemplate SYN|URG 1132 */ 1133 if (IS_FASTOPEN(tp->t_flags)) 1134 tp->t_flags &= ~TF_FASTOPEN; 1135 #ifdef INET6 1136 if (isipv6) 1137 error = tcp6_connect(tp, sin6, td); 1138 #endif /* INET6 */ 1139 #if defined(INET6) && defined(INET) 1140 else 1141 #endif 1142 #ifdef INET 1143 error = tcp_connect(tp, sinp, td); 1144 #endif 1145 /* 1146 * The bind operation in tcp_connect succeeded. We 1147 * no longer want to restore the flags if later 1148 * operations fail. 1149 */ 1150 if (error == 0 || inp->inp_lport != 0) 1151 restoreflags = false; 1152 1153 if (error != 0) { 1154 /* m is freed if PRUS_NOTREADY is unset. */ 1155 sbflush(&so->so_snd); 1156 goto out; 1157 } 1158 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1159 tcp_mss(tp, -1); 1160 } 1161 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1162 if ((flags & PRUS_NOTREADY) == 0) { 1163 tp->t_flags |= TF_FORCEDATA; 1164 error = tcp_output_nodrop(tp); 1165 tp->t_flags &= ~TF_FORCEDATA; 1166 } 1167 } 1168 TCP_LOG_EVENT(tp, NULL, 1169 &inp->inp_socket->so_rcv, 1170 &inp->inp_socket->so_snd, 1171 TCP_LOG_USERSEND, error, 1172 0, NULL, false); 1173 1174 out: 1175 /* 1176 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1177 * responsible for freeing memory. 1178 */ 1179 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1180 m_freem(m); 1181 1182 /* 1183 * If the request was unsuccessful and we changed flags, 1184 * restore the original flags. 1185 */ 1186 if (error != 0 && restoreflags) { 1187 inp->inp_vflag = vflagsav; 1188 inp->inp_inc.inc_flags = incflagsav; 1189 } 1190 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1191 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1192 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1193 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1194 error = tcp_unlock_or_drop(tp, error); 1195 NET_EPOCH_EXIT(et); 1196 return (error); 1197 } 1198 1199 static int 1200 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1201 { 1202 struct epoch_tracker et; 1203 struct inpcb *inp; 1204 struct tcpcb *tp; 1205 int error; 1206 1207 inp = sotoinpcb(so); 1208 INP_WLOCK(inp); 1209 if (inp->inp_flags & INP_DROPPED) { 1210 INP_WUNLOCK(inp); 1211 mb_free_notready(m, count); 1212 return (ECONNRESET); 1213 } 1214 tp = intotcpcb(inp); 1215 1216 SOCKBUF_LOCK(&so->so_snd); 1217 error = sbready(&so->so_snd, m, count); 1218 SOCKBUF_UNLOCK(&so->so_snd); 1219 if (error) { 1220 INP_WUNLOCK(inp); 1221 return (error); 1222 } 1223 NET_EPOCH_ENTER(et); 1224 error = tcp_output_unlock(tp); 1225 NET_EPOCH_EXIT(et); 1226 1227 return (error); 1228 } 1229 1230 /* 1231 * Abort the TCP. Drop the connection abruptly. 1232 */ 1233 static void 1234 tcp_usr_abort(struct socket *so) 1235 { 1236 struct inpcb *inp; 1237 struct tcpcb *tp; 1238 struct epoch_tracker et; 1239 1240 inp = sotoinpcb(so); 1241 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1242 1243 NET_EPOCH_ENTER(et); 1244 INP_WLOCK(inp); 1245 KASSERT(inp->inp_socket != NULL, 1246 ("tcp_usr_abort: inp_socket == NULL")); 1247 1248 /* 1249 * If we still have full TCP state, and we're not dropped, drop. 1250 */ 1251 if (!(inp->inp_flags & INP_DROPPED)) { 1252 tp = intotcpcb(inp); 1253 tp = tcp_drop(tp, ECONNABORTED); 1254 if (tp == NULL) 1255 goto dropped; 1256 tcp_bblog_pru(tp, PRU_ABORT, 0); 1257 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1258 } 1259 if (!(inp->inp_flags & INP_DROPPED)) { 1260 soref(so); 1261 inp->inp_flags |= INP_SOCKREF; 1262 } 1263 INP_WUNLOCK(inp); 1264 dropped: 1265 NET_EPOCH_EXIT(et); 1266 } 1267 1268 /* 1269 * TCP socket is closed. Start friendly disconnect. 1270 */ 1271 static void 1272 tcp_usr_close(struct socket *so) 1273 { 1274 struct inpcb *inp; 1275 struct tcpcb *tp; 1276 struct epoch_tracker et; 1277 1278 inp = sotoinpcb(so); 1279 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1280 1281 NET_EPOCH_ENTER(et); 1282 INP_WLOCK(inp); 1283 KASSERT(inp->inp_socket != NULL, 1284 ("tcp_usr_close: inp_socket == NULL")); 1285 1286 /* 1287 * If we are still connected and we're not dropped, initiate 1288 * a disconnect. 1289 */ 1290 if (!(inp->inp_flags & INP_DROPPED)) { 1291 tp = intotcpcb(inp); 1292 if (tp->t_state != TCPS_TIME_WAIT) { 1293 tp->t_flags |= TF_CLOSED; 1294 tcp_disconnect(tp); 1295 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1296 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1297 } 1298 } 1299 if (!(inp->inp_flags & INP_DROPPED)) { 1300 soref(so); 1301 inp->inp_flags |= INP_SOCKREF; 1302 } 1303 INP_WUNLOCK(inp); 1304 NET_EPOCH_EXIT(et); 1305 } 1306 1307 static int 1308 tcp_pru_options_support(struct tcpcb *tp, int flags) 1309 { 1310 /* 1311 * If the specific TCP stack has a pru_options 1312 * specified then it does not always support 1313 * all the PRU_XX options and we must ask it. 1314 * If the function is not specified then all 1315 * of the PRU_XX options are supported. 1316 */ 1317 int ret = 0; 1318 1319 if (tp->t_fb->tfb_pru_options) { 1320 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1321 } 1322 return (ret); 1323 } 1324 1325 /* 1326 * Receive out-of-band data. 1327 */ 1328 static int 1329 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1330 { 1331 int error = 0; 1332 struct inpcb *inp; 1333 struct tcpcb *tp; 1334 1335 inp = sotoinpcb(so); 1336 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1337 INP_WLOCK(inp); 1338 if (inp->inp_flags & INP_DROPPED) { 1339 INP_WUNLOCK(inp); 1340 return (ECONNRESET); 1341 } 1342 tp = intotcpcb(inp); 1343 1344 error = tcp_pru_options_support(tp, PRUS_OOB); 1345 if (error) { 1346 goto out; 1347 } 1348 if ((so->so_oobmark == 0 && 1349 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1350 so->so_options & SO_OOBINLINE || 1351 tp->t_oobflags & TCPOOB_HADDATA) { 1352 error = EINVAL; 1353 goto out; 1354 } 1355 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1356 error = EWOULDBLOCK; 1357 goto out; 1358 } 1359 m->m_len = 1; 1360 *mtod(m, caddr_t) = tp->t_iobc; 1361 if ((flags & MSG_PEEK) == 0) 1362 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1363 1364 out: 1365 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1366 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1367 INP_WUNLOCK(inp); 1368 return (error); 1369 } 1370 1371 #ifdef INET 1372 struct protosw tcp_protosw = { 1373 .pr_type = SOCK_STREAM, 1374 .pr_protocol = IPPROTO_TCP, 1375 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1376 PR_CAPATTACH, 1377 .pr_ctloutput = tcp_ctloutput, 1378 .pr_abort = tcp_usr_abort, 1379 .pr_accept = tcp_usr_accept, 1380 .pr_attach = tcp_usr_attach, 1381 .pr_bind = tcp_usr_bind, 1382 .pr_connect = tcp_usr_connect, 1383 .pr_control = in_control, 1384 .pr_detach = tcp_usr_detach, 1385 .pr_disconnect = tcp_usr_disconnect, 1386 .pr_listen = tcp_usr_listen, 1387 .pr_peeraddr = in_getpeeraddr, 1388 .pr_rcvd = tcp_usr_rcvd, 1389 .pr_rcvoob = tcp_usr_rcvoob, 1390 .pr_send = tcp_usr_send, 1391 .pr_ready = tcp_usr_ready, 1392 .pr_shutdown = tcp_usr_shutdown, 1393 .pr_sockaddr = in_getsockaddr, 1394 .pr_sosetlabel = in_pcbsosetlabel, 1395 .pr_close = tcp_usr_close, 1396 }; 1397 #endif /* INET */ 1398 1399 #ifdef INET6 1400 struct protosw tcp6_protosw = { 1401 .pr_type = SOCK_STREAM, 1402 .pr_protocol = IPPROTO_TCP, 1403 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1404 PR_CAPATTACH, 1405 .pr_ctloutput = tcp_ctloutput, 1406 .pr_abort = tcp_usr_abort, 1407 .pr_accept = tcp6_usr_accept, 1408 .pr_attach = tcp_usr_attach, 1409 .pr_bind = tcp6_usr_bind, 1410 .pr_connect = tcp6_usr_connect, 1411 .pr_control = in6_control, 1412 .pr_detach = tcp_usr_detach, 1413 .pr_disconnect = tcp_usr_disconnect, 1414 .pr_listen = tcp6_usr_listen, 1415 .pr_peeraddr = in6_mapped_peeraddr, 1416 .pr_rcvd = tcp_usr_rcvd, 1417 .pr_rcvoob = tcp_usr_rcvoob, 1418 .pr_send = tcp_usr_send, 1419 .pr_ready = tcp_usr_ready, 1420 .pr_shutdown = tcp_usr_shutdown, 1421 .pr_sockaddr = in6_mapped_sockaddr, 1422 .pr_sosetlabel = in_pcbsosetlabel, 1423 .pr_close = tcp_usr_close, 1424 }; 1425 #endif /* INET6 */ 1426 1427 #ifdef INET 1428 /* 1429 * Common subroutine to open a TCP connection to remote host specified 1430 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1431 * and assign a local port number and install the inpcb into the hash. 1432 * Initialize connection parameters and enter SYN-SENT state. 1433 */ 1434 static int 1435 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1436 { 1437 struct inpcb *inp = tptoinpcb(tp); 1438 struct socket *so = tptosocket(tp); 1439 int error; 1440 1441 NET_EPOCH_ASSERT(); 1442 INP_WLOCK_ASSERT(inp); 1443 1444 if (__predict_false((so->so_state & 1445 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1446 SS_ISDISCONNECTED)) != 0)) 1447 return (EISCONN); 1448 1449 INP_HASH_WLOCK(&V_tcbinfo); 1450 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1451 INP_HASH_WUNLOCK(&V_tcbinfo); 1452 if (error != 0) 1453 return (error); 1454 1455 /* 1456 * Compute window scaling to request: 1457 * Scale to fit into sweet spot. See tcp_syncache.c. 1458 * XXX: This should move to tcp_output(). 1459 */ 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 /* INET */ 1475 1476 #ifdef INET6 1477 static int 1478 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1479 { 1480 struct inpcb *inp = tptoinpcb(tp); 1481 struct socket *so = tptosocket(tp); 1482 int error; 1483 1484 NET_EPOCH_ASSERT(); 1485 INP_WLOCK_ASSERT(inp); 1486 1487 if (__predict_false((so->so_state & 1488 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1489 return (EISCONN); 1490 1491 INP_HASH_WLOCK(&V_tcbinfo); 1492 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1493 INP_HASH_WUNLOCK(&V_tcbinfo); 1494 if (error != 0) 1495 return (error); 1496 1497 /* Compute window scaling to request. */ 1498 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1499 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1500 tp->request_r_scale++; 1501 1502 soisconnecting(so); 1503 TCPSTAT_INC(tcps_connattempt); 1504 tcp_state_change(tp, TCPS_SYN_SENT); 1505 tp->iss = tcp_new_isn(&inp->inp_inc); 1506 if (tp->t_flags & TF_REQ_TSTMP) 1507 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1508 tcp_sendseqinit(tp); 1509 1510 return (0); 1511 } 1512 #endif /* INET6 */ 1513 1514 /* 1515 * Export TCP internal state information via a struct tcp_info, based on the 1516 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1517 * (TCP state machine, etc). We export all information using FreeBSD-native 1518 * constants -- for example, the numeric values for tcpi_state will differ 1519 * from Linux. 1520 */ 1521 void 1522 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1523 { 1524 1525 INP_LOCK_ASSERT(tptoinpcb(tp)); 1526 bzero(ti, sizeof(*ti)); 1527 1528 ti->tcpi_state = tp->t_state; 1529 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1530 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1531 if (tp->t_flags & TF_SACK_PERMIT) 1532 ti->tcpi_options |= TCPI_OPT_SACK; 1533 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1534 ti->tcpi_options |= TCPI_OPT_WSCALE; 1535 ti->tcpi_snd_wscale = tp->snd_scale; 1536 ti->tcpi_rcv_wscale = tp->rcv_scale; 1537 } 1538 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1539 case TF2_ECN_PERMIT: 1540 ti->tcpi_options |= TCPI_OPT_ECN; 1541 break; 1542 case TF2_ACE_PERMIT: 1543 /* FALLTHROUGH */ 1544 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1545 ti->tcpi_options |= TCPI_OPT_ACE; 1546 break; 1547 default: 1548 break; 1549 } 1550 if (IS_FASTOPEN(tp->t_flags)) 1551 ti->tcpi_options |= TCPI_OPT_TFO; 1552 1553 ti->tcpi_rto = tp->t_rxtcur * tick; 1554 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1555 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1556 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1557 1558 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1559 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1560 1561 /* 1562 * FreeBSD-specific extension fields for tcp_info. 1563 */ 1564 ti->tcpi_rcv_space = tp->rcv_wnd; 1565 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1566 ti->tcpi_snd_wnd = tp->snd_wnd; 1567 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1568 ti->tcpi_snd_nxt = tp->snd_nxt; 1569 ti->tcpi_snd_mss = tp->t_maxseg; 1570 ti->tcpi_rcv_mss = tp->t_maxseg; 1571 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1572 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1573 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1574 ti->tcpi_snd_una = tp->snd_una; 1575 ti->tcpi_snd_max = tp->snd_max; 1576 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1577 ti->tcpi_rcv_adv = tp->rcv_adv; 1578 ti->tcpi_dupacks = tp->t_dupacks; 1579 #ifdef TCP_OFFLOAD 1580 if (tp->t_flags & TF_TOE) { 1581 ti->tcpi_options |= TCPI_OPT_TOE; 1582 tcp_offload_tcp_info(tp, ti); 1583 } 1584 #endif 1585 /* 1586 * AccECN related counters. 1587 */ 1588 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1589 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1590 /* 1591 * Internal counter starts at 5 for AccECN 1592 * but 0 for RFC3168 ECN. 1593 */ 1594 ti->tcpi_delivered_ce = tp->t_scep - 5; 1595 else 1596 ti->tcpi_delivered_ce = tp->t_scep; 1597 ti->tcpi_received_ce = tp->t_rcep; 1598 } 1599 1600 /* 1601 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1602 * socket option arguments. When it re-acquires the lock after the copy, it 1603 * has to revalidate that the connection is still valid for the socket 1604 * option. 1605 */ 1606 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1607 INP_WLOCK(inp); \ 1608 if (inp->inp_flags & INP_DROPPED) { \ 1609 INP_WUNLOCK(inp); \ 1610 cleanup; \ 1611 return (ECONNRESET); \ 1612 } \ 1613 tp = intotcpcb(inp); \ 1614 } while(0) 1615 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1616 1617 int 1618 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1619 { 1620 struct socket *so = inp->inp_socket; 1621 struct tcpcb *tp = intotcpcb(inp); 1622 int error = 0; 1623 1624 MPASS(sopt->sopt_dir == SOPT_SET); 1625 INP_WLOCK_ASSERT(inp); 1626 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1627 ("inp_flags == %x", inp->inp_flags)); 1628 KASSERT(so != NULL, ("inp_socket == NULL")); 1629 1630 if (sopt->sopt_level != IPPROTO_TCP) { 1631 INP_WUNLOCK(inp); 1632 #ifdef INET6 1633 if (inp->inp_vflag & INP_IPV6PROTO) 1634 error = ip6_ctloutput(so, sopt); 1635 #endif 1636 #if defined(INET6) && defined(INET) 1637 else 1638 #endif 1639 #ifdef INET 1640 error = ip_ctloutput(so, sopt); 1641 #endif 1642 /* 1643 * When an IP-level socket option affects TCP, pass control 1644 * down to stack tfb_tcp_ctloutput, otherwise return what 1645 * IP level returned. 1646 */ 1647 switch (sopt->sopt_level) { 1648 #ifdef INET6 1649 case IPPROTO_IPV6: 1650 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1651 return (error); 1652 switch (sopt->sopt_name) { 1653 case IPV6_TCLASS: 1654 /* Notify tcp stacks that care (e.g. RACK). */ 1655 break; 1656 case IPV6_USE_MIN_MTU: 1657 /* Update t_maxseg accordingly. */ 1658 break; 1659 default: 1660 return (error); 1661 } 1662 break; 1663 #endif 1664 #ifdef INET 1665 case IPPROTO_IP: 1666 switch (sopt->sopt_name) { 1667 case IP_TOS: 1668 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1669 break; 1670 case IP_TTL: 1671 /* Notify tcp stacks that care (e.g. RACK). */ 1672 break; 1673 default: 1674 return (error); 1675 } 1676 break; 1677 #endif 1678 default: 1679 return (error); 1680 } 1681 INP_WLOCK(inp); 1682 if (inp->inp_flags & INP_DROPPED) { 1683 INP_WUNLOCK(inp); 1684 return (ECONNRESET); 1685 } 1686 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1687 /* 1688 * Protect the TCP option TCP_FUNCTION_BLK so 1689 * that a sub-function can *never* overwrite this. 1690 */ 1691 struct tcp_function_set fsn; 1692 struct tcp_function_block *blk; 1693 void *ptr = NULL; 1694 1695 INP_WUNLOCK(inp); 1696 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1697 if (error) 1698 return (error); 1699 1700 INP_WLOCK(inp); 1701 tp = intotcpcb(inp); 1702 1703 blk = find_and_ref_tcp_functions(&fsn); 1704 if (blk == NULL) { 1705 INP_WUNLOCK(inp); 1706 return (ENOENT); 1707 } 1708 if (tp->t_fb == blk) { 1709 /* You already have this */ 1710 refcount_release(&blk->tfb_refcnt); 1711 INP_WUNLOCK(inp); 1712 return (0); 1713 } 1714 if (tp->t_state != TCPS_CLOSED) { 1715 /* 1716 * The user has advanced the state 1717 * past the initial point, we may not 1718 * be able to switch. 1719 */ 1720 if (blk->tfb_tcp_handoff_ok != NULL) { 1721 /* 1722 * Does the stack provide a 1723 * query mechanism, if so it may 1724 * still be possible? 1725 */ 1726 error = (*blk->tfb_tcp_handoff_ok)(tp); 1727 } else 1728 error = EINVAL; 1729 if (error) { 1730 refcount_release(&blk->tfb_refcnt); 1731 INP_WUNLOCK(inp); 1732 return(error); 1733 } 1734 } 1735 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1736 refcount_release(&blk->tfb_refcnt); 1737 INP_WUNLOCK(inp); 1738 return (ENOENT); 1739 } 1740 /* 1741 * Ensure the new stack takes ownership with a 1742 * clean slate on peak rate threshold. 1743 */ 1744 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 1745 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1746 if (blk->tfb_tcp_fb_init) { 1747 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1748 if (error) { 1749 /* 1750 * Release the ref count the lookup 1751 * acquired. 1752 */ 1753 refcount_release(&blk->tfb_refcnt); 1754 /* 1755 * Now there is a chance that the 1756 * init() function mucked with some 1757 * things before it failed, such as 1758 * hpts or inp_flags2 or timer granularity. 1759 * It should not of, but lets give the old 1760 * stack a chance to reset to a known good state. 1761 */ 1762 if (tp->t_fb->tfb_switch_failed) { 1763 (*tp->t_fb->tfb_switch_failed)(tp); 1764 } 1765 goto err_out; 1766 } 1767 } 1768 if (tp->t_fb->tfb_tcp_fb_fini) { 1769 struct epoch_tracker et; 1770 /* 1771 * Tell the stack to cleanup with 0 i.e. 1772 * the tcb is not going away. 1773 */ 1774 NET_EPOCH_ENTER(et); 1775 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1776 NET_EPOCH_EXIT(et); 1777 } 1778 /* 1779 * Release the old refcnt, the 1780 * lookup acquired a ref on the 1781 * new one already. 1782 */ 1783 refcount_release(&tp->t_fb->tfb_refcnt); 1784 /* 1785 * Set in the new stack. 1786 */ 1787 tp->t_fb = blk; 1788 tp->t_fb_ptr = ptr; 1789 #ifdef TCP_OFFLOAD 1790 if (tp->t_flags & TF_TOE) { 1791 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1792 sopt->sopt_name); 1793 } 1794 #endif 1795 err_out: 1796 INP_WUNLOCK(inp); 1797 return (error); 1798 1799 } 1800 1801 /* Pass in the INP locked, callee must unlock it. */ 1802 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1803 } 1804 1805 static int 1806 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1807 { 1808 struct socket *so = inp->inp_socket; 1809 struct tcpcb *tp = intotcpcb(inp); 1810 int error = 0; 1811 1812 MPASS(sopt->sopt_dir == SOPT_GET); 1813 INP_WLOCK_ASSERT(inp); 1814 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1815 ("inp_flags == %x", inp->inp_flags)); 1816 KASSERT(so != NULL, ("inp_socket == NULL")); 1817 1818 if (sopt->sopt_level != IPPROTO_TCP) { 1819 INP_WUNLOCK(inp); 1820 #ifdef INET6 1821 if (inp->inp_vflag & INP_IPV6PROTO) 1822 error = ip6_ctloutput(so, sopt); 1823 #endif /* INET6 */ 1824 #if defined(INET6) && defined(INET) 1825 else 1826 #endif 1827 #ifdef INET 1828 error = ip_ctloutput(so, sopt); 1829 #endif 1830 return (error); 1831 } 1832 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1833 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1834 struct tcp_function_set fsn; 1835 1836 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1837 memset(&fsn, 0, sizeof(fsn)); 1838 find_tcp_function_alias(tp->t_fb, &fsn); 1839 } else { 1840 strncpy(fsn.function_set_name, 1841 tp->t_fb->tfb_tcp_block_name, 1842 TCP_FUNCTION_NAME_LEN_MAX); 1843 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1844 } 1845 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1846 INP_WUNLOCK(inp); 1847 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1848 return (error); 1849 } 1850 1851 /* Pass in the INP locked, callee must unlock it. */ 1852 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1853 } 1854 1855 int 1856 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1857 { 1858 struct inpcb *inp; 1859 1860 inp = sotoinpcb(so); 1861 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1862 1863 INP_WLOCK(inp); 1864 if (inp->inp_flags & INP_DROPPED) { 1865 INP_WUNLOCK(inp); 1866 return (ECONNRESET); 1867 } 1868 if (sopt->sopt_dir == SOPT_SET) 1869 return (tcp_ctloutput_set(inp, sopt)); 1870 else if (sopt->sopt_dir == SOPT_GET) 1871 return (tcp_ctloutput_get(inp, sopt)); 1872 else 1873 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1874 } 1875 1876 /* 1877 * If this assert becomes untrue, we need to change the size of the buf 1878 * variable in tcp_default_ctloutput(). 1879 */ 1880 #ifdef CTASSERT 1881 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1882 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1883 #endif 1884 1885 #ifdef KERN_TLS 1886 static int 1887 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1888 { 1889 struct tls_enable_v0 tls_v0; 1890 int error; 1891 1892 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1893 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1894 sizeof(tls_v0)); 1895 if (error) 1896 return (error); 1897 memset(tls, 0, sizeof(*tls)); 1898 tls->cipher_key = tls_v0.cipher_key; 1899 tls->iv = tls_v0.iv; 1900 tls->auth_key = tls_v0.auth_key; 1901 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1902 tls->cipher_key_len = tls_v0.cipher_key_len; 1903 tls->iv_len = tls_v0.iv_len; 1904 tls->auth_algorithm = tls_v0.auth_algorithm; 1905 tls->auth_key_len = tls_v0.auth_key_len; 1906 tls->flags = tls_v0.flags; 1907 tls->tls_vmajor = tls_v0.tls_vmajor; 1908 tls->tls_vminor = tls_v0.tls_vminor; 1909 return (0); 1910 } 1911 1912 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1913 } 1914 #endif 1915 1916 extern struct cc_algo newreno_cc_algo; 1917 1918 static int 1919 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1920 { 1921 struct cc_algo *algo; 1922 void *ptr = NULL; 1923 struct tcpcb *tp; 1924 struct cc_var cc_mem; 1925 char buf[TCP_CA_NAME_MAX]; 1926 size_t mem_sz; 1927 int error; 1928 1929 INP_WUNLOCK(inp); 1930 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1931 if (error) 1932 return(error); 1933 buf[sopt->sopt_valsize] = '\0'; 1934 CC_LIST_RLOCK(); 1935 STAILQ_FOREACH(algo, &cc_list, entries) { 1936 if (strncmp(buf, algo->name, 1937 TCP_CA_NAME_MAX) == 0) { 1938 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1939 /* We can't "see" modules being unloaded */ 1940 continue; 1941 } 1942 break; 1943 } 1944 } 1945 if (algo == NULL) { 1946 CC_LIST_RUNLOCK(); 1947 return(ESRCH); 1948 } 1949 /* 1950 * With a reference the algorithm cannot be removed 1951 * so we hold a reference through the change process. 1952 */ 1953 cc_refer(algo); 1954 CC_LIST_RUNLOCK(); 1955 if (algo->cb_init != NULL) { 1956 /* We can now pre-get the memory for the CC */ 1957 mem_sz = (*algo->cc_data_sz)(); 1958 if (mem_sz == 0) { 1959 goto no_mem_needed; 1960 } 1961 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1962 } else { 1963 no_mem_needed: 1964 mem_sz = 0; 1965 ptr = NULL; 1966 } 1967 /* 1968 * Make sure its all clean and zero and also get 1969 * back the inplock. 1970 */ 1971 memset(&cc_mem, 0, sizeof(cc_mem)); 1972 INP_WLOCK(inp); 1973 if (inp->inp_flags & INP_DROPPED) { 1974 INP_WUNLOCK(inp); 1975 if (ptr) 1976 free(ptr, M_CC_MEM); 1977 /* Release our temp reference */ 1978 CC_LIST_RLOCK(); 1979 cc_release(algo); 1980 CC_LIST_RUNLOCK(); 1981 return (ECONNRESET); 1982 } 1983 tp = intotcpcb(inp); 1984 if (ptr != NULL) 1985 memset(ptr, 0, mem_sz); 1986 cc_mem.ccvc.tcp = tp; 1987 /* 1988 * We once again hold a write lock over the tcb so it's 1989 * safe to do these things without ordering concerns. 1990 * Note here we init into stack memory. 1991 */ 1992 if (algo->cb_init != NULL) 1993 error = algo->cb_init(&cc_mem, ptr); 1994 else 1995 error = 0; 1996 /* 1997 * The CC algorithms, when given their memory 1998 * should not fail we could in theory have a 1999 * KASSERT here. 2000 */ 2001 if (error == 0) { 2002 /* 2003 * Touchdown, lets go ahead and move the 2004 * connection to the new CC module by 2005 * copying in the cc_mem after we call 2006 * the old ones cleanup (if any). 2007 */ 2008 if (CC_ALGO(tp)->cb_destroy != NULL) 2009 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2010 /* Detach the old CC from the tcpcb */ 2011 cc_detach(tp); 2012 /* Copy in our temp memory that was inited */ 2013 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2014 /* Now attach the new, which takes a reference */ 2015 cc_attach(tp, algo); 2016 /* Ok now are we where we have gotten past any conn_init? */ 2017 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2018 /* Yep run the connection init for the new CC */ 2019 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2020 } 2021 } else if (ptr) 2022 free(ptr, M_CC_MEM); 2023 INP_WUNLOCK(inp); 2024 /* Now lets release our temp reference */ 2025 CC_LIST_RLOCK(); 2026 cc_release(algo); 2027 CC_LIST_RUNLOCK(); 2028 return (error); 2029 } 2030 2031 int 2032 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2033 { 2034 struct inpcb *inp = tptoinpcb(tp); 2035 int error, opt, optval; 2036 u_int ui; 2037 struct tcp_info ti; 2038 #ifdef KERN_TLS 2039 struct tls_enable tls; 2040 struct socket *so = inp->inp_socket; 2041 #endif 2042 char *pbuf, buf[TCP_LOG_ID_LEN]; 2043 #ifdef STATS 2044 struct statsblob *sbp; 2045 #endif 2046 size_t len; 2047 2048 INP_WLOCK_ASSERT(inp); 2049 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2050 ("inp_flags == %x", inp->inp_flags)); 2051 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2052 2053 switch (sopt->sopt_level) { 2054 #ifdef INET6 2055 case IPPROTO_IPV6: 2056 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2057 switch (sopt->sopt_name) { 2058 case IPV6_USE_MIN_MTU: 2059 tcp6_use_min_mtu(tp); 2060 /* FALLTHROUGH */ 2061 } 2062 INP_WUNLOCK(inp); 2063 return (0); 2064 #endif 2065 #ifdef INET 2066 case IPPROTO_IP: 2067 INP_WUNLOCK(inp); 2068 return (0); 2069 #endif 2070 } 2071 2072 /* 2073 * For TCP_CCALGOOPT forward the control to CC module, for both 2074 * SOPT_SET and SOPT_GET. 2075 */ 2076 switch (sopt->sopt_name) { 2077 case TCP_CCALGOOPT: 2078 INP_WUNLOCK(inp); 2079 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2080 return (EINVAL); 2081 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2082 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2083 sopt->sopt_valsize); 2084 if (error) { 2085 free(pbuf, M_TEMP); 2086 return (error); 2087 } 2088 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2089 if (CC_ALGO(tp)->ctl_output != NULL) 2090 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2091 else 2092 error = ENOENT; 2093 INP_WUNLOCK(inp); 2094 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2095 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2096 free(pbuf, M_TEMP); 2097 return (error); 2098 } 2099 2100 switch (sopt->sopt_dir) { 2101 case SOPT_SET: 2102 switch (sopt->sopt_name) { 2103 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2104 case TCP_MD5SIG: 2105 INP_WUNLOCK(inp); 2106 if (!TCPMD5_ENABLED()) 2107 return (ENOPROTOOPT); 2108 error = TCPMD5_PCBCTL(inp, sopt); 2109 if (error) 2110 return (error); 2111 INP_WLOCK_RECHECK(inp); 2112 goto unlock_and_done; 2113 #endif /* IPSEC */ 2114 2115 case TCP_NODELAY: 2116 case TCP_NOOPT: 2117 INP_WUNLOCK(inp); 2118 error = sooptcopyin(sopt, &optval, sizeof optval, 2119 sizeof optval); 2120 if (error) 2121 return (error); 2122 2123 INP_WLOCK_RECHECK(inp); 2124 switch (sopt->sopt_name) { 2125 case TCP_NODELAY: 2126 opt = TF_NODELAY; 2127 break; 2128 case TCP_NOOPT: 2129 opt = TF_NOOPT; 2130 break; 2131 default: 2132 opt = 0; /* dead code to fool gcc */ 2133 break; 2134 } 2135 2136 if (optval) 2137 tp->t_flags |= opt; 2138 else 2139 tp->t_flags &= ~opt; 2140 unlock_and_done: 2141 #ifdef TCP_OFFLOAD 2142 if (tp->t_flags & TF_TOE) { 2143 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2144 sopt->sopt_name); 2145 } 2146 #endif 2147 INP_WUNLOCK(inp); 2148 break; 2149 2150 case TCP_NOPUSH: 2151 INP_WUNLOCK(inp); 2152 error = sooptcopyin(sopt, &optval, sizeof optval, 2153 sizeof optval); 2154 if (error) 2155 return (error); 2156 2157 INP_WLOCK_RECHECK(inp); 2158 if (optval) 2159 tp->t_flags |= TF_NOPUSH; 2160 else if (tp->t_flags & TF_NOPUSH) { 2161 tp->t_flags &= ~TF_NOPUSH; 2162 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2163 struct epoch_tracker et; 2164 2165 NET_EPOCH_ENTER(et); 2166 error = tcp_output_nodrop(tp); 2167 NET_EPOCH_EXIT(et); 2168 } 2169 } 2170 goto unlock_and_done; 2171 2172 case TCP_REMOTE_UDP_ENCAPS_PORT: 2173 INP_WUNLOCK(inp); 2174 error = sooptcopyin(sopt, &optval, sizeof optval, 2175 sizeof optval); 2176 if (error) 2177 return (error); 2178 if ((optval < TCP_TUNNELING_PORT_MIN) || 2179 (optval > TCP_TUNNELING_PORT_MAX)) { 2180 /* Its got to be in range */ 2181 return (EINVAL); 2182 } 2183 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2184 /* You have to have enabled a UDP tunneling port first */ 2185 return (EINVAL); 2186 } 2187 INP_WLOCK_RECHECK(inp); 2188 if (tp->t_state != TCPS_CLOSED) { 2189 /* You can't change after you are connected */ 2190 error = EINVAL; 2191 } else { 2192 /* Ok we are all good set the port */ 2193 tp->t_port = htons(optval); 2194 } 2195 goto unlock_and_done; 2196 2197 case TCP_MAXSEG: 2198 INP_WUNLOCK(inp); 2199 error = sooptcopyin(sopt, &optval, sizeof optval, 2200 sizeof optval); 2201 if (error) 2202 return (error); 2203 2204 INP_WLOCK_RECHECK(inp); 2205 if (optval > 0 && optval <= tp->t_maxseg && 2206 optval + 40 >= V_tcp_minmss) 2207 tp->t_maxseg = optval; 2208 else 2209 error = EINVAL; 2210 goto unlock_and_done; 2211 2212 case TCP_INFO: 2213 INP_WUNLOCK(inp); 2214 error = EINVAL; 2215 break; 2216 2217 case TCP_STATS: 2218 INP_WUNLOCK(inp); 2219 #ifdef STATS 2220 error = sooptcopyin(sopt, &optval, sizeof optval, 2221 sizeof optval); 2222 if (error) 2223 return (error); 2224 2225 if (optval > 0) 2226 sbp = stats_blob_alloc( 2227 V_tcp_perconn_stats_dflt_tpl, 0); 2228 else 2229 sbp = NULL; 2230 2231 INP_WLOCK_RECHECK(inp); 2232 if ((tp->t_stats != NULL && sbp == NULL) || 2233 (tp->t_stats == NULL && sbp != NULL)) { 2234 struct statsblob *t = tp->t_stats; 2235 tp->t_stats = sbp; 2236 sbp = t; 2237 } 2238 INP_WUNLOCK(inp); 2239 2240 stats_blob_destroy(sbp); 2241 #else 2242 return (EOPNOTSUPP); 2243 #endif /* !STATS */ 2244 break; 2245 2246 case TCP_CONGESTION: 2247 error = tcp_set_cc_mod(inp, sopt); 2248 break; 2249 2250 case TCP_REUSPORT_LB_NUMA: 2251 INP_WUNLOCK(inp); 2252 error = sooptcopyin(sopt, &optval, sizeof(optval), 2253 sizeof(optval)); 2254 INP_WLOCK_RECHECK(inp); 2255 if (!error) 2256 error = in_pcblbgroup_numa(inp, optval); 2257 INP_WUNLOCK(inp); 2258 break; 2259 2260 #ifdef KERN_TLS 2261 case TCP_TXTLS_ENABLE: 2262 INP_WUNLOCK(inp); 2263 error = copyin_tls_enable(sopt, &tls); 2264 if (error) 2265 break; 2266 error = ktls_enable_tx(so, &tls); 2267 break; 2268 case TCP_TXTLS_MODE: 2269 INP_WUNLOCK(inp); 2270 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2271 if (error) 2272 return (error); 2273 2274 INP_WLOCK_RECHECK(inp); 2275 error = ktls_set_tx_mode(so, ui); 2276 INP_WUNLOCK(inp); 2277 break; 2278 case TCP_RXTLS_ENABLE: 2279 INP_WUNLOCK(inp); 2280 error = sooptcopyin(sopt, &tls, sizeof(tls), 2281 sizeof(tls)); 2282 if (error) 2283 break; 2284 error = ktls_enable_rx(so, &tls); 2285 break; 2286 #endif 2287 case TCP_MAXUNACKTIME: 2288 case TCP_KEEPIDLE: 2289 case TCP_KEEPINTVL: 2290 case TCP_KEEPINIT: 2291 INP_WUNLOCK(inp); 2292 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2293 if (error) 2294 return (error); 2295 2296 if (ui > (UINT_MAX / hz)) { 2297 error = EINVAL; 2298 break; 2299 } 2300 ui *= hz; 2301 2302 INP_WLOCK_RECHECK(inp); 2303 switch (sopt->sopt_name) { 2304 case TCP_MAXUNACKTIME: 2305 tp->t_maxunacktime = ui; 2306 break; 2307 2308 case TCP_KEEPIDLE: 2309 tp->t_keepidle = ui; 2310 /* 2311 * XXX: better check current remaining 2312 * timeout and "merge" it with new value. 2313 */ 2314 if ((tp->t_state > TCPS_LISTEN) && 2315 (tp->t_state <= TCPS_CLOSING)) 2316 tcp_timer_activate(tp, TT_KEEP, 2317 TP_KEEPIDLE(tp)); 2318 break; 2319 case TCP_KEEPINTVL: 2320 tp->t_keepintvl = ui; 2321 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2322 (TP_MAXIDLE(tp) > 0)) 2323 tcp_timer_activate(tp, TT_2MSL, 2324 TP_MAXIDLE(tp)); 2325 break; 2326 case TCP_KEEPINIT: 2327 tp->t_keepinit = ui; 2328 if (tp->t_state == TCPS_SYN_RECEIVED || 2329 tp->t_state == TCPS_SYN_SENT) 2330 tcp_timer_activate(tp, TT_KEEP, 2331 TP_KEEPINIT(tp)); 2332 break; 2333 } 2334 goto unlock_and_done; 2335 2336 case TCP_KEEPCNT: 2337 INP_WUNLOCK(inp); 2338 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2339 if (error) 2340 return (error); 2341 2342 INP_WLOCK_RECHECK(inp); 2343 tp->t_keepcnt = ui; 2344 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2345 (TP_MAXIDLE(tp) > 0)) 2346 tcp_timer_activate(tp, TT_2MSL, 2347 TP_MAXIDLE(tp)); 2348 goto unlock_and_done; 2349 2350 #ifdef TCPPCAP 2351 case TCP_PCAP_OUT: 2352 case TCP_PCAP_IN: 2353 INP_WUNLOCK(inp); 2354 error = sooptcopyin(sopt, &optval, sizeof optval, 2355 sizeof optval); 2356 if (error) 2357 return (error); 2358 2359 INP_WLOCK_RECHECK(inp); 2360 if (optval >= 0) 2361 tcp_pcap_set_sock_max( 2362 (sopt->sopt_name == TCP_PCAP_OUT) ? 2363 &(tp->t_outpkts) : &(tp->t_inpkts), 2364 optval); 2365 else 2366 error = EINVAL; 2367 goto unlock_and_done; 2368 #endif 2369 2370 case TCP_FASTOPEN: { 2371 struct tcp_fastopen tfo_optval; 2372 2373 INP_WUNLOCK(inp); 2374 if (!V_tcp_fastopen_client_enable && 2375 !V_tcp_fastopen_server_enable) 2376 return (EPERM); 2377 2378 error = sooptcopyin(sopt, &tfo_optval, 2379 sizeof(tfo_optval), sizeof(int)); 2380 if (error) 2381 return (error); 2382 2383 INP_WLOCK_RECHECK(inp); 2384 if ((tp->t_state != TCPS_CLOSED) && 2385 (tp->t_state != TCPS_LISTEN)) { 2386 error = EINVAL; 2387 goto unlock_and_done; 2388 } 2389 if (tfo_optval.enable) { 2390 if (tp->t_state == TCPS_LISTEN) { 2391 if (!V_tcp_fastopen_server_enable) { 2392 error = EPERM; 2393 goto unlock_and_done; 2394 } 2395 2396 if (tp->t_tfo_pending == NULL) 2397 tp->t_tfo_pending = 2398 tcp_fastopen_alloc_counter(); 2399 } else { 2400 /* 2401 * If a pre-shared key was provided, 2402 * stash it in the client cookie 2403 * field of the tcpcb for use during 2404 * connect. 2405 */ 2406 if (sopt->sopt_valsize == 2407 sizeof(tfo_optval)) { 2408 memcpy(tp->t_tfo_cookie.client, 2409 tfo_optval.psk, 2410 TCP_FASTOPEN_PSK_LEN); 2411 tp->t_tfo_client_cookie_len = 2412 TCP_FASTOPEN_PSK_LEN; 2413 } 2414 } 2415 tp->t_flags |= TF_FASTOPEN; 2416 } else 2417 tp->t_flags &= ~TF_FASTOPEN; 2418 goto unlock_and_done; 2419 } 2420 2421 #ifdef TCP_BLACKBOX 2422 case TCP_LOG: 2423 INP_WUNLOCK(inp); 2424 error = sooptcopyin(sopt, &optval, sizeof optval, 2425 sizeof optval); 2426 if (error) 2427 return (error); 2428 2429 INP_WLOCK_RECHECK(inp); 2430 error = tcp_log_state_change(tp, optval); 2431 goto unlock_and_done; 2432 2433 case TCP_LOGBUF: 2434 INP_WUNLOCK(inp); 2435 error = EINVAL; 2436 break; 2437 2438 case TCP_LOGID: 2439 INP_WUNLOCK(inp); 2440 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2441 if (error) 2442 break; 2443 buf[sopt->sopt_valsize] = '\0'; 2444 INP_WLOCK_RECHECK(inp); 2445 error = tcp_log_set_id(tp, buf); 2446 /* tcp_log_set_id() unlocks the INP. */ 2447 break; 2448 2449 case TCP_LOGDUMP: 2450 case TCP_LOGDUMPID: 2451 INP_WUNLOCK(inp); 2452 error = 2453 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2454 if (error) 2455 break; 2456 buf[sopt->sopt_valsize] = '\0'; 2457 INP_WLOCK_RECHECK(inp); 2458 if (sopt->sopt_name == TCP_LOGDUMP) { 2459 error = tcp_log_dump_tp_logbuf(tp, buf, 2460 M_WAITOK, true); 2461 INP_WUNLOCK(inp); 2462 } else { 2463 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2464 /* 2465 * tcp_log_dump_tp_bucket_logbufs() drops the 2466 * INP lock. 2467 */ 2468 } 2469 break; 2470 #endif 2471 2472 default: 2473 INP_WUNLOCK(inp); 2474 error = ENOPROTOOPT; 2475 break; 2476 } 2477 break; 2478 2479 case SOPT_GET: 2480 tp = intotcpcb(inp); 2481 switch (sopt->sopt_name) { 2482 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2483 case TCP_MD5SIG: 2484 INP_WUNLOCK(inp); 2485 if (!TCPMD5_ENABLED()) 2486 return (ENOPROTOOPT); 2487 error = TCPMD5_PCBCTL(inp, sopt); 2488 break; 2489 #endif 2490 2491 case TCP_NODELAY: 2492 optval = tp->t_flags & TF_NODELAY; 2493 INP_WUNLOCK(inp); 2494 error = sooptcopyout(sopt, &optval, sizeof optval); 2495 break; 2496 case TCP_MAXSEG: 2497 optval = tp->t_maxseg; 2498 INP_WUNLOCK(inp); 2499 error = sooptcopyout(sopt, &optval, sizeof optval); 2500 break; 2501 case TCP_REMOTE_UDP_ENCAPS_PORT: 2502 optval = ntohs(tp->t_port); 2503 INP_WUNLOCK(inp); 2504 error = sooptcopyout(sopt, &optval, sizeof optval); 2505 break; 2506 case TCP_NOOPT: 2507 optval = tp->t_flags & TF_NOOPT; 2508 INP_WUNLOCK(inp); 2509 error = sooptcopyout(sopt, &optval, sizeof optval); 2510 break; 2511 case TCP_NOPUSH: 2512 optval = tp->t_flags & TF_NOPUSH; 2513 INP_WUNLOCK(inp); 2514 error = sooptcopyout(sopt, &optval, sizeof optval); 2515 break; 2516 case TCP_INFO: 2517 tcp_fill_info(tp, &ti); 2518 INP_WUNLOCK(inp); 2519 error = sooptcopyout(sopt, &ti, sizeof ti); 2520 break; 2521 case TCP_STATS: 2522 { 2523 #ifdef STATS 2524 int nheld; 2525 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2526 2527 error = 0; 2528 socklen_t outsbsz = sopt->sopt_valsize; 2529 if (tp->t_stats == NULL) 2530 error = ENOENT; 2531 else if (outsbsz >= tp->t_stats->cursz) 2532 outsbsz = tp->t_stats->cursz; 2533 else if (outsbsz >= sizeof(struct statsblob)) 2534 outsbsz = sizeof(struct statsblob); 2535 else 2536 error = EINVAL; 2537 INP_WUNLOCK(inp); 2538 if (error) 2539 break; 2540 2541 sbp = sopt->sopt_val; 2542 nheld = atop(round_page(((vm_offset_t)sbp) + 2543 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2544 vm_page_t ma[nheld]; 2545 if (vm_fault_quick_hold_pages( 2546 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2547 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2548 nheld) < 0) { 2549 error = EFAULT; 2550 break; 2551 } 2552 2553 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2554 SIZEOF_MEMBER(struct statsblob, flags)))) 2555 goto unhold; 2556 2557 INP_WLOCK_RECHECK(inp); 2558 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2559 sbflags | SB_CLONE_USRDSTNOFAULT); 2560 INP_WUNLOCK(inp); 2561 sopt->sopt_valsize = outsbsz; 2562 unhold: 2563 vm_page_unhold_pages(ma, nheld); 2564 #else 2565 INP_WUNLOCK(inp); 2566 error = EOPNOTSUPP; 2567 #endif /* !STATS */ 2568 break; 2569 } 2570 case TCP_CONGESTION: 2571 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2572 INP_WUNLOCK(inp); 2573 error = sooptcopyout(sopt, buf, len + 1); 2574 break; 2575 case TCP_MAXUNACKTIME: 2576 case TCP_KEEPIDLE: 2577 case TCP_KEEPINTVL: 2578 case TCP_KEEPINIT: 2579 case TCP_KEEPCNT: 2580 switch (sopt->sopt_name) { 2581 case TCP_MAXUNACKTIME: 2582 ui = TP_MAXUNACKTIME(tp) / hz; 2583 break; 2584 case TCP_KEEPIDLE: 2585 ui = TP_KEEPIDLE(tp) / hz; 2586 break; 2587 case TCP_KEEPINTVL: 2588 ui = TP_KEEPINTVL(tp) / hz; 2589 break; 2590 case TCP_KEEPINIT: 2591 ui = TP_KEEPINIT(tp) / hz; 2592 break; 2593 case TCP_KEEPCNT: 2594 ui = TP_KEEPCNT(tp); 2595 break; 2596 } 2597 INP_WUNLOCK(inp); 2598 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2599 break; 2600 #ifdef TCPPCAP 2601 case TCP_PCAP_OUT: 2602 case TCP_PCAP_IN: 2603 optval = tcp_pcap_get_sock_max( 2604 (sopt->sopt_name == TCP_PCAP_OUT) ? 2605 &(tp->t_outpkts) : &(tp->t_inpkts)); 2606 INP_WUNLOCK(inp); 2607 error = sooptcopyout(sopt, &optval, sizeof optval); 2608 break; 2609 #endif 2610 case TCP_FASTOPEN: 2611 optval = tp->t_flags & TF_FASTOPEN; 2612 INP_WUNLOCK(inp); 2613 error = sooptcopyout(sopt, &optval, sizeof optval); 2614 break; 2615 #ifdef TCP_BLACKBOX 2616 case TCP_LOG: 2617 optval = tcp_get_bblog_state(tp); 2618 INP_WUNLOCK(inp); 2619 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2620 break; 2621 case TCP_LOGBUF: 2622 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2623 error = tcp_log_getlogbuf(sopt, tp); 2624 break; 2625 case TCP_LOGID: 2626 len = tcp_log_get_id(tp, buf); 2627 INP_WUNLOCK(inp); 2628 error = sooptcopyout(sopt, buf, len + 1); 2629 break; 2630 case TCP_LOGDUMP: 2631 case TCP_LOGDUMPID: 2632 INP_WUNLOCK(inp); 2633 error = EINVAL; 2634 break; 2635 #endif 2636 #ifdef KERN_TLS 2637 case TCP_TXTLS_MODE: 2638 error = ktls_get_tx_mode(so, &optval); 2639 INP_WUNLOCK(inp); 2640 if (error == 0) 2641 error = sooptcopyout(sopt, &optval, 2642 sizeof(optval)); 2643 break; 2644 case TCP_RXTLS_MODE: 2645 error = ktls_get_rx_mode(so, &optval); 2646 INP_WUNLOCK(inp); 2647 if (error == 0) 2648 error = sooptcopyout(sopt, &optval, 2649 sizeof(optval)); 2650 break; 2651 #endif 2652 default: 2653 INP_WUNLOCK(inp); 2654 error = ENOPROTOOPT; 2655 break; 2656 } 2657 break; 2658 } 2659 return (error); 2660 } 2661 #undef INP_WLOCK_RECHECK 2662 #undef INP_WLOCK_RECHECK_CLEANUP 2663 2664 /* 2665 * Initiate (or continue) disconnect. 2666 * If embryonic state, just send reset (once). 2667 * If in ``let data drain'' option and linger null, just drop. 2668 * Otherwise (hard), mark socket disconnecting and drop 2669 * current input data; switch states based on user close, and 2670 * send segment to peer (with FIN). 2671 */ 2672 static void 2673 tcp_disconnect(struct tcpcb *tp) 2674 { 2675 struct inpcb *inp = tptoinpcb(tp); 2676 struct socket *so = tptosocket(tp); 2677 2678 NET_EPOCH_ASSERT(); 2679 INP_WLOCK_ASSERT(inp); 2680 2681 /* 2682 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2683 * socket is still open. 2684 */ 2685 if (tp->t_state < TCPS_ESTABLISHED && 2686 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2687 tp = tcp_close(tp); 2688 KASSERT(tp != NULL, 2689 ("tcp_disconnect: tcp_close() returned NULL")); 2690 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2691 tp = tcp_drop(tp, 0); 2692 KASSERT(tp != NULL, 2693 ("tcp_disconnect: tcp_drop() returned NULL")); 2694 } else { 2695 soisdisconnecting(so); 2696 sbflush(&so->so_rcv); 2697 tcp_usrclosed(tp); 2698 if (!(inp->inp_flags & INP_DROPPED)) 2699 /* Ignore stack's drop request, we already at it. */ 2700 (void)tcp_output_nodrop(tp); 2701 } 2702 } 2703 2704 /* 2705 * User issued close, and wish to trail through shutdown states: 2706 * if never received SYN, just forget it. If got a SYN from peer, 2707 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2708 * If already got a FIN from peer, then almost done; go to LAST_ACK 2709 * state. In all other cases, have already sent FIN to peer (e.g. 2710 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2711 * for peer to send FIN or not respond to keep-alives, etc. 2712 * We can let the user exit from the close as soon as the FIN is acked. 2713 */ 2714 static void 2715 tcp_usrclosed(struct tcpcb *tp) 2716 { 2717 2718 NET_EPOCH_ASSERT(); 2719 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2720 2721 switch (tp->t_state) { 2722 case TCPS_LISTEN: 2723 #ifdef TCP_OFFLOAD 2724 tcp_offload_listen_stop(tp); 2725 #endif 2726 tcp_state_change(tp, TCPS_CLOSED); 2727 /* FALLTHROUGH */ 2728 case TCPS_CLOSED: 2729 tp = tcp_close(tp); 2730 /* 2731 * tcp_close() should never return NULL here as the socket is 2732 * still open. 2733 */ 2734 KASSERT(tp != NULL, 2735 ("tcp_usrclosed: tcp_close() returned NULL")); 2736 break; 2737 2738 case TCPS_SYN_SENT: 2739 case TCPS_SYN_RECEIVED: 2740 tp->t_flags |= TF_NEEDFIN; 2741 break; 2742 2743 case TCPS_ESTABLISHED: 2744 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2745 break; 2746 2747 case TCPS_CLOSE_WAIT: 2748 tcp_state_change(tp, TCPS_LAST_ACK); 2749 break; 2750 } 2751 if (tp->t_acktime == 0) 2752 tp->t_acktime = ticks; 2753 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2754 soisdisconnected(tptosocket(tp)); 2755 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2756 if (tp->t_state == TCPS_FIN_WAIT_2) { 2757 int timeout; 2758 2759 timeout = (tcp_fast_finwait2_recycle) ? 2760 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2761 tcp_timer_activate(tp, TT_2MSL, timeout); 2762 } 2763 } 2764 } 2765 2766 #ifdef DDB 2767 static void 2768 db_print_indent(int indent) 2769 { 2770 int i; 2771 2772 for (i = 0; i < indent; i++) 2773 db_printf(" "); 2774 } 2775 2776 static void 2777 db_print_tstate(int t_state) 2778 { 2779 2780 switch (t_state) { 2781 case TCPS_CLOSED: 2782 db_printf("TCPS_CLOSED"); 2783 return; 2784 2785 case TCPS_LISTEN: 2786 db_printf("TCPS_LISTEN"); 2787 return; 2788 2789 case TCPS_SYN_SENT: 2790 db_printf("TCPS_SYN_SENT"); 2791 return; 2792 2793 case TCPS_SYN_RECEIVED: 2794 db_printf("TCPS_SYN_RECEIVED"); 2795 return; 2796 2797 case TCPS_ESTABLISHED: 2798 db_printf("TCPS_ESTABLISHED"); 2799 return; 2800 2801 case TCPS_CLOSE_WAIT: 2802 db_printf("TCPS_CLOSE_WAIT"); 2803 return; 2804 2805 case TCPS_FIN_WAIT_1: 2806 db_printf("TCPS_FIN_WAIT_1"); 2807 return; 2808 2809 case TCPS_CLOSING: 2810 db_printf("TCPS_CLOSING"); 2811 return; 2812 2813 case TCPS_LAST_ACK: 2814 db_printf("TCPS_LAST_ACK"); 2815 return; 2816 2817 case TCPS_FIN_WAIT_2: 2818 db_printf("TCPS_FIN_WAIT_2"); 2819 return; 2820 2821 case TCPS_TIME_WAIT: 2822 db_printf("TCPS_TIME_WAIT"); 2823 return; 2824 2825 default: 2826 db_printf("unknown"); 2827 return; 2828 } 2829 } 2830 2831 static void 2832 db_print_tflags(u_int t_flags) 2833 { 2834 int comma; 2835 2836 comma = 0; 2837 if (t_flags & TF_ACKNOW) { 2838 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2839 comma = 1; 2840 } 2841 if (t_flags & TF_DELACK) { 2842 db_printf("%sTF_DELACK", comma ? ", " : ""); 2843 comma = 1; 2844 } 2845 if (t_flags & TF_NODELAY) { 2846 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2847 comma = 1; 2848 } 2849 if (t_flags & TF_NOOPT) { 2850 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2851 comma = 1; 2852 } 2853 if (t_flags & TF_SENTFIN) { 2854 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2855 comma = 1; 2856 } 2857 if (t_flags & TF_REQ_SCALE) { 2858 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2859 comma = 1; 2860 } 2861 if (t_flags & TF_RCVD_SCALE) { 2862 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2863 comma = 1; 2864 } 2865 if (t_flags & TF_REQ_TSTMP) { 2866 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2867 comma = 1; 2868 } 2869 if (t_flags & TF_RCVD_TSTMP) { 2870 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2871 comma = 1; 2872 } 2873 if (t_flags & TF_SACK_PERMIT) { 2874 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2875 comma = 1; 2876 } 2877 if (t_flags & TF_NEEDSYN) { 2878 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2879 comma = 1; 2880 } 2881 if (t_flags & TF_NEEDFIN) { 2882 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2883 comma = 1; 2884 } 2885 if (t_flags & TF_NOPUSH) { 2886 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2887 comma = 1; 2888 } 2889 if (t_flags & TF_PREVVALID) { 2890 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2891 comma = 1; 2892 } 2893 if (t_flags & TF_MORETOCOME) { 2894 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2895 comma = 1; 2896 } 2897 if (t_flags & TF_SONOTCONN) { 2898 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2899 comma = 1; 2900 } 2901 if (t_flags & TF_LASTIDLE) { 2902 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2903 comma = 1; 2904 } 2905 if (t_flags & TF_RXWIN0SENT) { 2906 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2907 comma = 1; 2908 } 2909 if (t_flags & TF_FASTRECOVERY) { 2910 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2911 comma = 1; 2912 } 2913 if (t_flags & TF_CONGRECOVERY) { 2914 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2915 comma = 1; 2916 } 2917 if (t_flags & TF_WASFRECOVERY) { 2918 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2919 comma = 1; 2920 } 2921 if (t_flags & TF_WASCRECOVERY) { 2922 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2923 comma = 1; 2924 } 2925 if (t_flags & TF_SIGNATURE) { 2926 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2927 comma = 1; 2928 } 2929 if (t_flags & TF_FORCEDATA) { 2930 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2931 comma = 1; 2932 } 2933 if (t_flags & TF_TSO) { 2934 db_printf("%sTF_TSO", comma ? ", " : ""); 2935 comma = 1; 2936 } 2937 if (t_flags & TF_FASTOPEN) { 2938 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2939 comma = 1; 2940 } 2941 } 2942 2943 static void 2944 db_print_tflags2(u_int t_flags2) 2945 { 2946 int comma; 2947 2948 comma = 0; 2949 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2950 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2951 comma = 1; 2952 } 2953 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2954 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2955 comma = 1; 2956 } 2957 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2958 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2959 comma = 1; 2960 } 2961 if (t_flags2 & TF2_LOG_AUTO) { 2962 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2963 comma = 1; 2964 } 2965 if (t_flags2 & TF2_DROP_AF_DATA) { 2966 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2967 comma = 1; 2968 } 2969 if (t_flags2 & TF2_ECN_PERMIT) { 2970 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2971 comma = 1; 2972 } 2973 if (t_flags2 & TF2_ECN_SND_CWR) { 2974 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 2975 comma = 1; 2976 } 2977 if (t_flags2 & TF2_ECN_SND_ECE) { 2978 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 2979 comma = 1; 2980 } 2981 if (t_flags2 & TF2_ACE_PERMIT) { 2982 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 2983 comma = 1; 2984 } 2985 if (t_flags2 & TF2_FBYTES_COMPLETE) { 2986 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 2987 comma = 1; 2988 } 2989 } 2990 2991 static void 2992 db_print_toobflags(char t_oobflags) 2993 { 2994 int comma; 2995 2996 comma = 0; 2997 if (t_oobflags & TCPOOB_HAVEDATA) { 2998 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 2999 comma = 1; 3000 } 3001 if (t_oobflags & TCPOOB_HADDATA) { 3002 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3003 comma = 1; 3004 } 3005 } 3006 3007 static void 3008 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3009 { 3010 3011 db_print_indent(indent); 3012 db_printf("%s at %p\n", name, tp); 3013 3014 indent += 2; 3015 3016 db_print_indent(indent); 3017 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3018 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3019 3020 db_print_indent(indent); 3021 db_printf("t_callout: %p t_timers: %p\n", 3022 &tp->t_callout, &tp->t_timers); 3023 3024 db_print_indent(indent); 3025 db_printf("t_state: %d (", tp->t_state); 3026 db_print_tstate(tp->t_state); 3027 db_printf(")\n"); 3028 3029 db_print_indent(indent); 3030 db_printf("t_flags: 0x%x (", tp->t_flags); 3031 db_print_tflags(tp->t_flags); 3032 db_printf(")\n"); 3033 3034 db_print_indent(indent); 3035 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3036 db_print_tflags2(tp->t_flags2); 3037 db_printf(")\n"); 3038 3039 db_print_indent(indent); 3040 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 3041 tp->snd_una, tp->snd_max, tp->snd_nxt); 3042 3043 db_print_indent(indent); 3044 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3045 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3046 3047 db_print_indent(indent); 3048 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3049 tp->iss, tp->irs, tp->rcv_nxt); 3050 3051 db_print_indent(indent); 3052 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3053 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3054 3055 db_print_indent(indent); 3056 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3057 tp->snd_wnd, tp->snd_cwnd); 3058 3059 db_print_indent(indent); 3060 db_printf("snd_ssthresh: %u snd_recover: " 3061 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3062 3063 db_print_indent(indent); 3064 db_printf("t_rcvtime: %u t_startime: %u\n", 3065 tp->t_rcvtime, tp->t_starttime); 3066 3067 db_print_indent(indent); 3068 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3069 tp->t_rtttime, tp->t_rtseq); 3070 3071 db_print_indent(indent); 3072 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3073 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3074 3075 db_print_indent(indent); 3076 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3077 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3078 3079 db_print_indent(indent); 3080 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3081 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3082 3083 db_print_indent(indent); 3084 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3085 db_print_toobflags(tp->t_oobflags); 3086 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3087 3088 db_print_indent(indent); 3089 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3090 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3091 3092 db_print_indent(indent); 3093 db_printf("ts_recent: %u ts_recent_age: %u\n", 3094 tp->ts_recent, tp->ts_recent_age); 3095 3096 db_print_indent(indent); 3097 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3098 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3099 3100 db_print_indent(indent); 3101 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3102 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3103 tp->snd_recover_prev, tp->t_badrxtwin); 3104 3105 db_print_indent(indent); 3106 db_printf("snd_numholes: %d snd_holes first: %p\n", 3107 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3108 3109 db_print_indent(indent); 3110 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3111 tp->snd_fack, tp->rcv_numsacks); 3112 3113 /* Skip sackblks, sackhint. */ 3114 3115 db_print_indent(indent); 3116 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3117 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3118 } 3119 3120 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3121 { 3122 struct tcpcb *tp; 3123 3124 if (!have_addr) { 3125 db_printf("usage: show tcpcb <addr>\n"); 3126 return; 3127 } 3128 tp = (struct tcpcb *)addr; 3129 3130 db_print_tcpcb(tp, "tcpcb", 0); 3131 } 3132 #endif 3133