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, enum shutdown_how how) 803 { 804 struct epoch_tracker et; 805 struct inpcb *inp = sotoinpcb(so); 806 struct tcpcb *tp = intotcpcb(inp); 807 int error = 0; 808 809 SOCK_LOCK(so); 810 if ((so->so_state & 811 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 812 SOCK_UNLOCK(so); 813 return (ENOTCONN); 814 } 815 if (SOLISTENING(so)) { 816 if (how != SHUT_WR) { 817 so->so_error = ECONNABORTED; 818 solisten_wakeup(so); /* unlocks so */ 819 } else 820 SOCK_UNLOCK(so); 821 return (0); 822 } 823 SOCK_UNLOCK(so); 824 825 switch (how) { 826 case SHUT_RD: 827 sorflush(so); 828 break; 829 case SHUT_RDWR: 830 sorflush(so); 831 /* FALLTHROUGH */ 832 case SHUT_WR: 833 /* 834 * XXXGL: mimicing old soshutdown() here. But shouldn't we 835 * return ECONNRESEST for SHUT_RD as well? 836 */ 837 INP_WLOCK(inp); 838 if (inp->inp_flags & INP_DROPPED) { 839 INP_WUNLOCK(inp); 840 return (ECONNRESET); 841 } 842 843 socantsendmore(so); 844 NET_EPOCH_ENTER(et); 845 tcp_usrclosed(tp); 846 error = tcp_output_nodrop(tp); 847 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 848 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 849 error = tcp_unlock_or_drop(tp, error); 850 NET_EPOCH_EXIT(et); 851 } 852 wakeup(&so->so_timeo); 853 854 return (error); 855 } 856 857 /* 858 * After a receive, possibly send window update to peer. 859 */ 860 static int 861 tcp_usr_rcvd(struct socket *so, int flags) 862 { 863 struct epoch_tracker et; 864 struct inpcb *inp; 865 struct tcpcb *tp; 866 int outrv = 0, error = 0; 867 868 inp = sotoinpcb(so); 869 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 870 INP_WLOCK(inp); 871 if (inp->inp_flags & INP_DROPPED) { 872 INP_WUNLOCK(inp); 873 return (ECONNRESET); 874 } 875 tp = intotcpcb(inp); 876 877 NET_EPOCH_ENTER(et); 878 /* 879 * For passively-created TFO connections, don't attempt a window 880 * update while still in SYN_RECEIVED as this may trigger an early 881 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 882 * application response data, or failing that, when the DELACK timer 883 * expires. 884 */ 885 if (IS_FASTOPEN(tp->t_flags) && 886 (tp->t_state == TCPS_SYN_RECEIVED)) 887 goto out; 888 #ifdef TCP_OFFLOAD 889 if (tp->t_flags & TF_TOE) 890 tcp_offload_rcvd(tp); 891 else 892 #endif 893 outrv = tcp_output_nodrop(tp); 894 out: 895 tcp_bblog_pru(tp, PRU_RCVD, error); 896 TCP_PROBE2(debug__user, tp, PRU_RCVD); 897 (void) tcp_unlock_or_drop(tp, outrv); 898 NET_EPOCH_EXIT(et); 899 return (error); 900 } 901 902 /* 903 * Do a send by putting data in output queue and updating urgent 904 * marker if URG set. Possibly send more data. Unlike the other 905 * pru_*() routines, the mbuf chains are our responsibility. We 906 * must either enqueue them or free them. The other pru_* routines 907 * generally are caller-frees. 908 */ 909 static int 910 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 911 struct sockaddr *nam, struct mbuf *control, struct thread *td) 912 { 913 struct epoch_tracker et; 914 int error = 0; 915 struct inpcb *inp; 916 struct tcpcb *tp; 917 #ifdef INET 918 #ifdef INET6 919 struct sockaddr_in sin; 920 #endif 921 struct sockaddr_in *sinp; 922 #endif 923 #ifdef INET6 924 struct sockaddr_in6 *sin6; 925 int isipv6; 926 #endif 927 u_int8_t incflagsav; 928 u_char vflagsav; 929 bool restoreflags; 930 931 inp = sotoinpcb(so); 932 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 933 INP_WLOCK(inp); 934 if (inp->inp_flags & INP_DROPPED) { 935 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 936 m_freem(m); 937 INP_WUNLOCK(inp); 938 return (ECONNRESET); 939 } 940 tp = intotcpcb(inp); 941 942 vflagsav = inp->inp_vflag; 943 incflagsav = inp->inp_inc.inc_flags; 944 restoreflags = false; 945 946 NET_EPOCH_ENTER(et); 947 if (control != NULL) { 948 /* TCP doesn't do control messages (rights, creds, etc) */ 949 if (control->m_len > 0) { 950 m_freem(control); 951 error = EINVAL; 952 goto out; 953 } 954 m_freem(control); /* empty control, just free it */ 955 } 956 957 if ((flags & PRUS_OOB) != 0 && 958 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 959 goto out; 960 961 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 962 if (tp->t_state == TCPS_LISTEN) { 963 error = EINVAL; 964 goto out; 965 } 966 switch (nam->sa_family) { 967 #ifdef INET 968 case AF_INET: 969 sinp = (struct sockaddr_in *)nam; 970 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 971 error = EINVAL; 972 goto out; 973 } 974 if ((inp->inp_vflag & INP_IPV6) != 0) { 975 error = EAFNOSUPPORT; 976 goto out; 977 } 978 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 979 error = EAFNOSUPPORT; 980 goto out; 981 } 982 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 983 error = EACCES; 984 goto out; 985 } 986 if ((error = prison_remote_ip4(td->td_ucred, 987 &sinp->sin_addr))) 988 goto out; 989 #ifdef INET6 990 isipv6 = 0; 991 #endif 992 break; 993 #endif /* INET */ 994 #ifdef INET6 995 case AF_INET6: 996 sin6 = (struct sockaddr_in6 *)nam; 997 if (sin6->sin6_len != sizeof(*sin6)) { 998 error = EINVAL; 999 goto out; 1000 } 1001 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1002 error = EAFNOSUPPORT; 1003 goto out; 1004 } 1005 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1006 error = EAFNOSUPPORT; 1007 goto out; 1008 } 1009 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1010 #ifdef INET 1011 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1012 error = EINVAL; 1013 goto out; 1014 } 1015 if ((inp->inp_vflag & INP_IPV4) == 0) { 1016 error = EAFNOSUPPORT; 1017 goto out; 1018 } 1019 restoreflags = true; 1020 inp->inp_vflag &= ~INP_IPV6; 1021 sinp = &sin; 1022 in6_sin6_2_sin(sinp, sin6); 1023 if (IN_MULTICAST( 1024 ntohl(sinp->sin_addr.s_addr))) { 1025 error = EAFNOSUPPORT; 1026 goto out; 1027 } 1028 if ((error = prison_remote_ip4(td->td_ucred, 1029 &sinp->sin_addr))) 1030 goto out; 1031 isipv6 = 0; 1032 #else /* !INET */ 1033 error = EAFNOSUPPORT; 1034 goto out; 1035 #endif /* INET */ 1036 } else { 1037 if ((inp->inp_vflag & INP_IPV6) == 0) { 1038 error = EAFNOSUPPORT; 1039 goto out; 1040 } 1041 restoreflags = true; 1042 inp->inp_vflag &= ~INP_IPV4; 1043 inp->inp_inc.inc_flags |= INC_ISIPV6; 1044 if ((error = prison_remote_ip6(td->td_ucred, 1045 &sin6->sin6_addr))) 1046 goto out; 1047 isipv6 = 1; 1048 } 1049 break; 1050 #endif /* INET6 */ 1051 default: 1052 error = EAFNOSUPPORT; 1053 goto out; 1054 } 1055 } 1056 if (!(flags & PRUS_OOB)) { 1057 if (tp->t_acktime == 0) 1058 tp->t_acktime = ticks; 1059 sbappendstream(&so->so_snd, m, flags); 1060 m = NULL; 1061 if (nam && tp->t_state < TCPS_SYN_SENT) { 1062 KASSERT(tp->t_state == TCPS_CLOSED, 1063 ("%s: tp %p is listening", __func__, tp)); 1064 1065 /* 1066 * Do implied connect if not yet connected, 1067 * initialize window to default value, and 1068 * initialize maxseg using peer's cached MSS. 1069 */ 1070 #ifdef INET6 1071 if (isipv6) 1072 error = tcp6_connect(tp, sin6, td); 1073 #endif /* INET6 */ 1074 #if defined(INET6) && defined(INET) 1075 else 1076 #endif 1077 #ifdef INET 1078 error = tcp_connect(tp, sinp, td); 1079 #endif 1080 /* 1081 * The bind operation in tcp_connect succeeded. We 1082 * no longer want to restore the flags if later 1083 * operations fail. 1084 */ 1085 if (error == 0 || inp->inp_lport != 0) 1086 restoreflags = false; 1087 1088 if (error) { 1089 /* m is freed if PRUS_NOTREADY is unset. */ 1090 sbflush(&so->so_snd); 1091 goto out; 1092 } 1093 if (IS_FASTOPEN(tp->t_flags)) 1094 tcp_fastopen_connect(tp); 1095 else { 1096 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1097 tcp_mss(tp, -1); 1098 } 1099 } 1100 if (flags & PRUS_EOF) { 1101 /* 1102 * Close the send side of the connection after 1103 * the data is sent. 1104 */ 1105 socantsendmore(so); 1106 tcp_usrclosed(tp); 1107 } 1108 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1109 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1110 (tp->t_fbyte_out == 0) && 1111 (so->so_snd.sb_ccc > 0)) { 1112 tp->t_fbyte_out = ticks; 1113 if (tp->t_fbyte_out == 0) 1114 tp->t_fbyte_out = 1; 1115 if (tp->t_fbyte_out && tp->t_fbyte_in) 1116 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1117 } 1118 if (!(inp->inp_flags & INP_DROPPED) && 1119 !(flags & PRUS_NOTREADY)) { 1120 if (flags & PRUS_MORETOCOME) 1121 tp->t_flags |= TF_MORETOCOME; 1122 error = tcp_output_nodrop(tp); 1123 if (flags & PRUS_MORETOCOME) 1124 tp->t_flags &= ~TF_MORETOCOME; 1125 } 1126 } else { 1127 /* 1128 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1129 */ 1130 SOCKBUF_LOCK(&so->so_snd); 1131 if (sbspace(&so->so_snd) < -512) { 1132 SOCKBUF_UNLOCK(&so->so_snd); 1133 error = ENOBUFS; 1134 goto out; 1135 } 1136 /* 1137 * According to RFC961 (Assigned Protocols), 1138 * the urgent pointer points to the last octet 1139 * of urgent data. We continue, however, 1140 * to consider it to indicate the first octet 1141 * of data past the urgent section. 1142 * Otherwise, snd_up should be one lower. 1143 */ 1144 if (tp->t_acktime == 0) 1145 tp->t_acktime = ticks; 1146 sbappendstream_locked(&so->so_snd, m, flags); 1147 SOCKBUF_UNLOCK(&so->so_snd); 1148 m = NULL; 1149 if (nam && tp->t_state < TCPS_SYN_SENT) { 1150 /* 1151 * Do implied connect if not yet connected, 1152 * initialize window to default value, and 1153 * initialize maxseg using peer's cached MSS. 1154 */ 1155 1156 /* 1157 * Not going to contemplate SYN|URG 1158 */ 1159 if (IS_FASTOPEN(tp->t_flags)) 1160 tp->t_flags &= ~TF_FASTOPEN; 1161 #ifdef INET6 1162 if (isipv6) 1163 error = tcp6_connect(tp, sin6, td); 1164 #endif /* INET6 */ 1165 #if defined(INET6) && defined(INET) 1166 else 1167 #endif 1168 #ifdef INET 1169 error = tcp_connect(tp, sinp, td); 1170 #endif 1171 /* 1172 * The bind operation in tcp_connect succeeded. We 1173 * no longer want to restore the flags if later 1174 * operations fail. 1175 */ 1176 if (error == 0 || inp->inp_lport != 0) 1177 restoreflags = false; 1178 1179 if (error != 0) { 1180 /* m is freed if PRUS_NOTREADY is unset. */ 1181 sbflush(&so->so_snd); 1182 goto out; 1183 } 1184 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1185 tcp_mss(tp, -1); 1186 } 1187 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1188 if ((flags & PRUS_NOTREADY) == 0) { 1189 tp->t_flags |= TF_FORCEDATA; 1190 error = tcp_output_nodrop(tp); 1191 tp->t_flags &= ~TF_FORCEDATA; 1192 } 1193 } 1194 TCP_LOG_EVENT(tp, NULL, 1195 &inp->inp_socket->so_rcv, 1196 &inp->inp_socket->so_snd, 1197 TCP_LOG_USERSEND, error, 1198 0, NULL, false); 1199 1200 out: 1201 /* 1202 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1203 * responsible for freeing memory. 1204 */ 1205 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1206 m_freem(m); 1207 1208 /* 1209 * If the request was unsuccessful and we changed flags, 1210 * restore the original flags. 1211 */ 1212 if (error != 0 && restoreflags) { 1213 inp->inp_vflag = vflagsav; 1214 inp->inp_inc.inc_flags = incflagsav; 1215 } 1216 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1217 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1218 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1219 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1220 error = tcp_unlock_or_drop(tp, error); 1221 NET_EPOCH_EXIT(et); 1222 return (error); 1223 } 1224 1225 static int 1226 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1227 { 1228 struct epoch_tracker et; 1229 struct inpcb *inp; 1230 struct tcpcb *tp; 1231 int error; 1232 1233 inp = sotoinpcb(so); 1234 INP_WLOCK(inp); 1235 if (inp->inp_flags & INP_DROPPED) { 1236 INP_WUNLOCK(inp); 1237 mb_free_notready(m, count); 1238 return (ECONNRESET); 1239 } 1240 tp = intotcpcb(inp); 1241 1242 SOCKBUF_LOCK(&so->so_snd); 1243 error = sbready(&so->so_snd, m, count); 1244 SOCKBUF_UNLOCK(&so->so_snd); 1245 if (error) { 1246 INP_WUNLOCK(inp); 1247 return (error); 1248 } 1249 NET_EPOCH_ENTER(et); 1250 error = tcp_output_unlock(tp); 1251 NET_EPOCH_EXIT(et); 1252 1253 return (error); 1254 } 1255 1256 /* 1257 * Abort the TCP. Drop the connection abruptly. 1258 */ 1259 static void 1260 tcp_usr_abort(struct socket *so) 1261 { 1262 struct inpcb *inp; 1263 struct tcpcb *tp; 1264 struct epoch_tracker et; 1265 1266 inp = sotoinpcb(so); 1267 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1268 1269 NET_EPOCH_ENTER(et); 1270 INP_WLOCK(inp); 1271 KASSERT(inp->inp_socket != NULL, 1272 ("tcp_usr_abort: inp_socket == NULL")); 1273 1274 /* 1275 * If we still have full TCP state, and we're not dropped, drop. 1276 */ 1277 if (!(inp->inp_flags & INP_DROPPED)) { 1278 tp = intotcpcb(inp); 1279 tp = tcp_drop(tp, ECONNABORTED); 1280 if (tp == NULL) 1281 goto dropped; 1282 tcp_bblog_pru(tp, PRU_ABORT, 0); 1283 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1284 } 1285 if (!(inp->inp_flags & INP_DROPPED)) { 1286 soref(so); 1287 inp->inp_flags |= INP_SOCKREF; 1288 } 1289 INP_WUNLOCK(inp); 1290 dropped: 1291 NET_EPOCH_EXIT(et); 1292 } 1293 1294 /* 1295 * TCP socket is closed. Start friendly disconnect. 1296 */ 1297 static void 1298 tcp_usr_close(struct socket *so) 1299 { 1300 struct inpcb *inp; 1301 struct tcpcb *tp; 1302 struct epoch_tracker et; 1303 1304 inp = sotoinpcb(so); 1305 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1306 1307 NET_EPOCH_ENTER(et); 1308 INP_WLOCK(inp); 1309 KASSERT(inp->inp_socket != NULL, 1310 ("tcp_usr_close: inp_socket == NULL")); 1311 1312 /* 1313 * If we are still connected and we're not dropped, initiate 1314 * a disconnect. 1315 */ 1316 if (!(inp->inp_flags & INP_DROPPED)) { 1317 tp = intotcpcb(inp); 1318 if (tp->t_state != TCPS_TIME_WAIT) { 1319 tp->t_flags |= TF_CLOSED; 1320 tcp_disconnect(tp); 1321 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1322 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1323 } 1324 } 1325 if (!(inp->inp_flags & INP_DROPPED)) { 1326 soref(so); 1327 inp->inp_flags |= INP_SOCKREF; 1328 } 1329 INP_WUNLOCK(inp); 1330 NET_EPOCH_EXIT(et); 1331 } 1332 1333 static int 1334 tcp_pru_options_support(struct tcpcb *tp, int flags) 1335 { 1336 /* 1337 * If the specific TCP stack has a pru_options 1338 * specified then it does not always support 1339 * all the PRU_XX options and we must ask it. 1340 * If the function is not specified then all 1341 * of the PRU_XX options are supported. 1342 */ 1343 int ret = 0; 1344 1345 if (tp->t_fb->tfb_pru_options) { 1346 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1347 } 1348 return (ret); 1349 } 1350 1351 /* 1352 * Receive out-of-band data. 1353 */ 1354 static int 1355 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1356 { 1357 int error = 0; 1358 struct inpcb *inp; 1359 struct tcpcb *tp; 1360 1361 inp = sotoinpcb(so); 1362 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1363 INP_WLOCK(inp); 1364 if (inp->inp_flags & INP_DROPPED) { 1365 INP_WUNLOCK(inp); 1366 return (ECONNRESET); 1367 } 1368 tp = intotcpcb(inp); 1369 1370 error = tcp_pru_options_support(tp, PRUS_OOB); 1371 if (error) { 1372 goto out; 1373 } 1374 if ((so->so_oobmark == 0 && 1375 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1376 so->so_options & SO_OOBINLINE || 1377 tp->t_oobflags & TCPOOB_HADDATA) { 1378 error = EINVAL; 1379 goto out; 1380 } 1381 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1382 error = EWOULDBLOCK; 1383 goto out; 1384 } 1385 m->m_len = 1; 1386 *mtod(m, caddr_t) = tp->t_iobc; 1387 if ((flags & MSG_PEEK) == 0) 1388 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1389 1390 out: 1391 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1392 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1393 INP_WUNLOCK(inp); 1394 return (error); 1395 } 1396 1397 #ifdef INET 1398 struct protosw tcp_protosw = { 1399 .pr_type = SOCK_STREAM, 1400 .pr_protocol = IPPROTO_TCP, 1401 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1402 PR_CAPATTACH, 1403 .pr_ctloutput = tcp_ctloutput, 1404 .pr_abort = tcp_usr_abort, 1405 .pr_accept = tcp_usr_accept, 1406 .pr_attach = tcp_usr_attach, 1407 .pr_bind = tcp_usr_bind, 1408 .pr_connect = tcp_usr_connect, 1409 .pr_control = in_control, 1410 .pr_detach = tcp_usr_detach, 1411 .pr_disconnect = tcp_usr_disconnect, 1412 .pr_listen = tcp_usr_listen, 1413 .pr_peeraddr = in_getpeeraddr, 1414 .pr_rcvd = tcp_usr_rcvd, 1415 .pr_rcvoob = tcp_usr_rcvoob, 1416 .pr_send = tcp_usr_send, 1417 .pr_ready = tcp_usr_ready, 1418 .pr_shutdown = tcp_usr_shutdown, 1419 .pr_sockaddr = in_getsockaddr, 1420 .pr_sosetlabel = in_pcbsosetlabel, 1421 .pr_close = tcp_usr_close, 1422 }; 1423 #endif /* INET */ 1424 1425 #ifdef INET6 1426 struct protosw tcp6_protosw = { 1427 .pr_type = SOCK_STREAM, 1428 .pr_protocol = IPPROTO_TCP, 1429 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1430 PR_CAPATTACH, 1431 .pr_ctloutput = tcp_ctloutput, 1432 .pr_abort = tcp_usr_abort, 1433 .pr_accept = tcp6_usr_accept, 1434 .pr_attach = tcp_usr_attach, 1435 .pr_bind = tcp6_usr_bind, 1436 .pr_connect = tcp6_usr_connect, 1437 .pr_control = in6_control, 1438 .pr_detach = tcp_usr_detach, 1439 .pr_disconnect = tcp_usr_disconnect, 1440 .pr_listen = tcp6_usr_listen, 1441 .pr_peeraddr = in6_mapped_peeraddr, 1442 .pr_rcvd = tcp_usr_rcvd, 1443 .pr_rcvoob = tcp_usr_rcvoob, 1444 .pr_send = tcp_usr_send, 1445 .pr_ready = tcp_usr_ready, 1446 .pr_shutdown = tcp_usr_shutdown, 1447 .pr_sockaddr = in6_mapped_sockaddr, 1448 .pr_sosetlabel = in_pcbsosetlabel, 1449 .pr_close = tcp_usr_close, 1450 }; 1451 #endif /* INET6 */ 1452 1453 #ifdef INET 1454 /* 1455 * Common subroutine to open a TCP connection to remote host specified 1456 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1457 * and assign a local port number and install the inpcb into the hash. 1458 * Initialize connection parameters and enter SYN-SENT state. 1459 */ 1460 static int 1461 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1462 { 1463 struct inpcb *inp = tptoinpcb(tp); 1464 struct socket *so = tptosocket(tp); 1465 int error; 1466 1467 NET_EPOCH_ASSERT(); 1468 INP_WLOCK_ASSERT(inp); 1469 1470 if (__predict_false((so->so_state & 1471 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1472 SS_ISDISCONNECTED)) != 0)) 1473 return (EISCONN); 1474 1475 INP_HASH_WLOCK(&V_tcbinfo); 1476 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1477 INP_HASH_WUNLOCK(&V_tcbinfo); 1478 if (error != 0) 1479 return (error); 1480 1481 /* 1482 * Compute window scaling to request: 1483 * Scale to fit into sweet spot. See tcp_syncache.c. 1484 * XXX: This should move to tcp_output(). 1485 */ 1486 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1487 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1488 tp->request_r_scale++; 1489 1490 soisconnecting(so); 1491 TCPSTAT_INC(tcps_connattempt); 1492 tcp_state_change(tp, TCPS_SYN_SENT); 1493 tp->iss = tcp_new_isn(&inp->inp_inc); 1494 if (tp->t_flags & TF_REQ_TSTMP) 1495 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1496 tcp_sendseqinit(tp); 1497 1498 return (0); 1499 } 1500 #endif /* INET */ 1501 1502 #ifdef INET6 1503 static int 1504 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1505 { 1506 struct inpcb *inp = tptoinpcb(tp); 1507 struct socket *so = tptosocket(tp); 1508 int error; 1509 1510 NET_EPOCH_ASSERT(); 1511 INP_WLOCK_ASSERT(inp); 1512 1513 if (__predict_false((so->so_state & 1514 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1515 return (EISCONN); 1516 1517 INP_HASH_WLOCK(&V_tcbinfo); 1518 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1519 INP_HASH_WUNLOCK(&V_tcbinfo); 1520 if (error != 0) 1521 return (error); 1522 1523 /* Compute window scaling to request. */ 1524 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1525 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1526 tp->request_r_scale++; 1527 1528 soisconnecting(so); 1529 TCPSTAT_INC(tcps_connattempt); 1530 tcp_state_change(tp, TCPS_SYN_SENT); 1531 tp->iss = tcp_new_isn(&inp->inp_inc); 1532 if (tp->t_flags & TF_REQ_TSTMP) 1533 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1534 tcp_sendseqinit(tp); 1535 1536 return (0); 1537 } 1538 #endif /* INET6 */ 1539 1540 /* 1541 * Export TCP internal state information via a struct tcp_info, based on the 1542 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1543 * (TCP state machine, etc). We export all information using FreeBSD-native 1544 * constants -- for example, the numeric values for tcpi_state will differ 1545 * from Linux. 1546 */ 1547 void 1548 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1549 { 1550 1551 INP_LOCK_ASSERT(tptoinpcb(tp)); 1552 bzero(ti, sizeof(*ti)); 1553 1554 ti->tcpi_state = tp->t_state; 1555 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1556 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1557 if (tp->t_flags & TF_SACK_PERMIT) 1558 ti->tcpi_options |= TCPI_OPT_SACK; 1559 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1560 ti->tcpi_options |= TCPI_OPT_WSCALE; 1561 ti->tcpi_snd_wscale = tp->snd_scale; 1562 ti->tcpi_rcv_wscale = tp->rcv_scale; 1563 } 1564 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1565 case TF2_ECN_PERMIT: 1566 ti->tcpi_options |= TCPI_OPT_ECN; 1567 break; 1568 case TF2_ACE_PERMIT: 1569 /* FALLTHROUGH */ 1570 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1571 ti->tcpi_options |= TCPI_OPT_ACE; 1572 break; 1573 default: 1574 break; 1575 } 1576 if (IS_FASTOPEN(tp->t_flags)) 1577 ti->tcpi_options |= TCPI_OPT_TFO; 1578 1579 ti->tcpi_rto = tp->t_rxtcur * tick; 1580 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1581 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1582 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1583 1584 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1585 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1586 1587 /* 1588 * FreeBSD-specific extension fields for tcp_info. 1589 */ 1590 ti->tcpi_rcv_space = tp->rcv_wnd; 1591 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1592 ti->tcpi_snd_wnd = tp->snd_wnd; 1593 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1594 ti->tcpi_snd_nxt = tp->snd_nxt; 1595 ti->tcpi_snd_mss = tp->t_maxseg; 1596 ti->tcpi_rcv_mss = tp->t_maxseg; 1597 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1598 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1599 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1600 ti->tcpi_snd_una = tp->snd_una; 1601 ti->tcpi_snd_max = tp->snd_max; 1602 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1603 ti->tcpi_rcv_adv = tp->rcv_adv; 1604 ti->tcpi_dupacks = tp->t_dupacks; 1605 #ifdef TCP_OFFLOAD 1606 if (tp->t_flags & TF_TOE) { 1607 ti->tcpi_options |= TCPI_OPT_TOE; 1608 tcp_offload_tcp_info(tp, ti); 1609 } 1610 #endif 1611 /* 1612 * AccECN related counters. 1613 */ 1614 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1615 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1616 /* 1617 * Internal counter starts at 5 for AccECN 1618 * but 0 for RFC3168 ECN. 1619 */ 1620 ti->tcpi_delivered_ce = tp->t_scep - 5; 1621 else 1622 ti->tcpi_delivered_ce = tp->t_scep; 1623 ti->tcpi_received_ce = tp->t_rcep; 1624 } 1625 1626 /* 1627 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1628 * socket option arguments. When it re-acquires the lock after the copy, it 1629 * has to revalidate that the connection is still valid for the socket 1630 * option. 1631 */ 1632 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1633 INP_WLOCK(inp); \ 1634 if (inp->inp_flags & INP_DROPPED) { \ 1635 INP_WUNLOCK(inp); \ 1636 cleanup; \ 1637 return (ECONNRESET); \ 1638 } \ 1639 tp = intotcpcb(inp); \ 1640 } while(0) 1641 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1642 1643 int 1644 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1645 { 1646 struct socket *so = inp->inp_socket; 1647 struct tcpcb *tp = intotcpcb(inp); 1648 int error = 0; 1649 1650 MPASS(sopt->sopt_dir == SOPT_SET); 1651 INP_WLOCK_ASSERT(inp); 1652 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1653 ("inp_flags == %x", inp->inp_flags)); 1654 KASSERT(so != NULL, ("inp_socket == NULL")); 1655 1656 if (sopt->sopt_level != IPPROTO_TCP) { 1657 INP_WUNLOCK(inp); 1658 #ifdef INET6 1659 if (inp->inp_vflag & INP_IPV6PROTO) 1660 error = ip6_ctloutput(so, sopt); 1661 #endif 1662 #if defined(INET6) && defined(INET) 1663 else 1664 #endif 1665 #ifdef INET 1666 error = ip_ctloutput(so, sopt); 1667 #endif 1668 /* 1669 * When an IP-level socket option affects TCP, pass control 1670 * down to stack tfb_tcp_ctloutput, otherwise return what 1671 * IP level returned. 1672 */ 1673 switch (sopt->sopt_level) { 1674 #ifdef INET6 1675 case IPPROTO_IPV6: 1676 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1677 return (error); 1678 switch (sopt->sopt_name) { 1679 case IPV6_TCLASS: 1680 /* Notify tcp stacks that care (e.g. RACK). */ 1681 break; 1682 case IPV6_USE_MIN_MTU: 1683 /* Update t_maxseg accordingly. */ 1684 break; 1685 default: 1686 return (error); 1687 } 1688 break; 1689 #endif 1690 #ifdef INET 1691 case IPPROTO_IP: 1692 switch (sopt->sopt_name) { 1693 case IP_TOS: 1694 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1695 break; 1696 case IP_TTL: 1697 /* Notify tcp stacks that care (e.g. RACK). */ 1698 break; 1699 default: 1700 return (error); 1701 } 1702 break; 1703 #endif 1704 default: 1705 return (error); 1706 } 1707 INP_WLOCK(inp); 1708 if (inp->inp_flags & INP_DROPPED) { 1709 INP_WUNLOCK(inp); 1710 return (ECONNRESET); 1711 } 1712 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1713 /* 1714 * Protect the TCP option TCP_FUNCTION_BLK so 1715 * that a sub-function can *never* overwrite this. 1716 */ 1717 struct tcp_function_set fsn; 1718 struct tcp_function_block *blk; 1719 void *ptr = NULL; 1720 1721 INP_WUNLOCK(inp); 1722 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1723 if (error) 1724 return (error); 1725 1726 INP_WLOCK(inp); 1727 tp = intotcpcb(inp); 1728 1729 blk = find_and_ref_tcp_functions(&fsn); 1730 if (blk == NULL) { 1731 INP_WUNLOCK(inp); 1732 return (ENOENT); 1733 } 1734 if (tp->t_fb == blk) { 1735 /* You already have this */ 1736 refcount_release(&blk->tfb_refcnt); 1737 INP_WUNLOCK(inp); 1738 return (0); 1739 } 1740 if (tp->t_state != TCPS_CLOSED) { 1741 /* 1742 * The user has advanced the state 1743 * past the initial point, we may not 1744 * be able to switch. 1745 */ 1746 if (blk->tfb_tcp_handoff_ok != NULL) { 1747 /* 1748 * Does the stack provide a 1749 * query mechanism, if so it may 1750 * still be possible? 1751 */ 1752 error = (*blk->tfb_tcp_handoff_ok)(tp); 1753 } else 1754 error = EINVAL; 1755 if (error) { 1756 refcount_release(&blk->tfb_refcnt); 1757 INP_WUNLOCK(inp); 1758 return(error); 1759 } 1760 } 1761 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1762 refcount_release(&blk->tfb_refcnt); 1763 INP_WUNLOCK(inp); 1764 return (ENOENT); 1765 } 1766 /* 1767 * Ensure the new stack takes ownership with a 1768 * clean slate on peak rate threshold. 1769 */ 1770 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 1771 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1772 if (blk->tfb_tcp_fb_init) { 1773 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1774 if (error) { 1775 /* 1776 * Release the ref count the lookup 1777 * acquired. 1778 */ 1779 refcount_release(&blk->tfb_refcnt); 1780 /* 1781 * Now there is a chance that the 1782 * init() function mucked with some 1783 * things before it failed, such as 1784 * hpts or inp_flags2 or timer granularity. 1785 * It should not of, but lets give the old 1786 * stack a chance to reset to a known good state. 1787 */ 1788 if (tp->t_fb->tfb_switch_failed) { 1789 (*tp->t_fb->tfb_switch_failed)(tp); 1790 } 1791 goto err_out; 1792 } 1793 } 1794 if (tp->t_fb->tfb_tcp_fb_fini) { 1795 struct epoch_tracker et; 1796 /* 1797 * Tell the stack to cleanup with 0 i.e. 1798 * the tcb is not going away. 1799 */ 1800 NET_EPOCH_ENTER(et); 1801 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1802 NET_EPOCH_EXIT(et); 1803 } 1804 /* 1805 * Release the old refcnt, the 1806 * lookup acquired a ref on the 1807 * new one already. 1808 */ 1809 refcount_release(&tp->t_fb->tfb_refcnt); 1810 /* 1811 * Set in the new stack. 1812 */ 1813 tp->t_fb = blk; 1814 tp->t_fb_ptr = ptr; 1815 #ifdef TCP_OFFLOAD 1816 if (tp->t_flags & TF_TOE) { 1817 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1818 sopt->sopt_name); 1819 } 1820 #endif 1821 err_out: 1822 INP_WUNLOCK(inp); 1823 return (error); 1824 1825 } 1826 1827 /* Pass in the INP locked, callee must unlock it. */ 1828 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1829 } 1830 1831 static int 1832 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1833 { 1834 struct socket *so = inp->inp_socket; 1835 struct tcpcb *tp = intotcpcb(inp); 1836 int error = 0; 1837 1838 MPASS(sopt->sopt_dir == SOPT_GET); 1839 INP_WLOCK_ASSERT(inp); 1840 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1841 ("inp_flags == %x", inp->inp_flags)); 1842 KASSERT(so != NULL, ("inp_socket == NULL")); 1843 1844 if (sopt->sopt_level != IPPROTO_TCP) { 1845 INP_WUNLOCK(inp); 1846 #ifdef INET6 1847 if (inp->inp_vflag & INP_IPV6PROTO) 1848 error = ip6_ctloutput(so, sopt); 1849 #endif /* INET6 */ 1850 #if defined(INET6) && defined(INET) 1851 else 1852 #endif 1853 #ifdef INET 1854 error = ip_ctloutput(so, sopt); 1855 #endif 1856 return (error); 1857 } 1858 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1859 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1860 struct tcp_function_set fsn; 1861 1862 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1863 memset(&fsn, 0, sizeof(fsn)); 1864 find_tcp_function_alias(tp->t_fb, &fsn); 1865 } else { 1866 strncpy(fsn.function_set_name, 1867 tp->t_fb->tfb_tcp_block_name, 1868 TCP_FUNCTION_NAME_LEN_MAX); 1869 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1870 } 1871 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1872 INP_WUNLOCK(inp); 1873 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1874 return (error); 1875 } 1876 1877 /* Pass in the INP locked, callee must unlock it. */ 1878 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1879 } 1880 1881 int 1882 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1883 { 1884 struct inpcb *inp; 1885 1886 inp = sotoinpcb(so); 1887 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1888 1889 INP_WLOCK(inp); 1890 if (inp->inp_flags & INP_DROPPED) { 1891 INP_WUNLOCK(inp); 1892 return (ECONNRESET); 1893 } 1894 if (sopt->sopt_dir == SOPT_SET) 1895 return (tcp_ctloutput_set(inp, sopt)); 1896 else if (sopt->sopt_dir == SOPT_GET) 1897 return (tcp_ctloutput_get(inp, sopt)); 1898 else 1899 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1900 } 1901 1902 /* 1903 * If this assert becomes untrue, we need to change the size of the buf 1904 * variable in tcp_default_ctloutput(). 1905 */ 1906 #ifdef CTASSERT 1907 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1908 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1909 #endif 1910 1911 #ifdef KERN_TLS 1912 static int 1913 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1914 { 1915 struct tls_enable_v0 tls_v0; 1916 int error; 1917 1918 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1919 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1920 sizeof(tls_v0)); 1921 if (error) 1922 return (error); 1923 memset(tls, 0, sizeof(*tls)); 1924 tls->cipher_key = tls_v0.cipher_key; 1925 tls->iv = tls_v0.iv; 1926 tls->auth_key = tls_v0.auth_key; 1927 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1928 tls->cipher_key_len = tls_v0.cipher_key_len; 1929 tls->iv_len = tls_v0.iv_len; 1930 tls->auth_algorithm = tls_v0.auth_algorithm; 1931 tls->auth_key_len = tls_v0.auth_key_len; 1932 tls->flags = tls_v0.flags; 1933 tls->tls_vmajor = tls_v0.tls_vmajor; 1934 tls->tls_vminor = tls_v0.tls_vminor; 1935 return (0); 1936 } 1937 1938 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1939 } 1940 #endif 1941 1942 extern struct cc_algo newreno_cc_algo; 1943 1944 static int 1945 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1946 { 1947 struct cc_algo *algo; 1948 void *ptr = NULL; 1949 struct tcpcb *tp; 1950 struct cc_var cc_mem; 1951 char buf[TCP_CA_NAME_MAX]; 1952 size_t mem_sz; 1953 int error; 1954 1955 INP_WUNLOCK(inp); 1956 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1957 if (error) 1958 return(error); 1959 buf[sopt->sopt_valsize] = '\0'; 1960 CC_LIST_RLOCK(); 1961 STAILQ_FOREACH(algo, &cc_list, entries) { 1962 if (strncmp(buf, algo->name, 1963 TCP_CA_NAME_MAX) == 0) { 1964 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1965 /* We can't "see" modules being unloaded */ 1966 continue; 1967 } 1968 break; 1969 } 1970 } 1971 if (algo == NULL) { 1972 CC_LIST_RUNLOCK(); 1973 return(ESRCH); 1974 } 1975 /* 1976 * With a reference the algorithm cannot be removed 1977 * so we hold a reference through the change process. 1978 */ 1979 cc_refer(algo); 1980 CC_LIST_RUNLOCK(); 1981 if (algo->cb_init != NULL) { 1982 /* We can now pre-get the memory for the CC */ 1983 mem_sz = (*algo->cc_data_sz)(); 1984 if (mem_sz == 0) { 1985 goto no_mem_needed; 1986 } 1987 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1988 } else { 1989 no_mem_needed: 1990 mem_sz = 0; 1991 ptr = NULL; 1992 } 1993 /* 1994 * Make sure its all clean and zero and also get 1995 * back the inplock. 1996 */ 1997 memset(&cc_mem, 0, sizeof(cc_mem)); 1998 INP_WLOCK(inp); 1999 if (inp->inp_flags & INP_DROPPED) { 2000 INP_WUNLOCK(inp); 2001 if (ptr) 2002 free(ptr, M_CC_MEM); 2003 /* Release our temp reference */ 2004 CC_LIST_RLOCK(); 2005 cc_release(algo); 2006 CC_LIST_RUNLOCK(); 2007 return (ECONNRESET); 2008 } 2009 tp = intotcpcb(inp); 2010 if (ptr != NULL) 2011 memset(ptr, 0, mem_sz); 2012 cc_mem.ccvc.tcp = tp; 2013 /* 2014 * We once again hold a write lock over the tcb so it's 2015 * safe to do these things without ordering concerns. 2016 * Note here we init into stack memory. 2017 */ 2018 if (algo->cb_init != NULL) 2019 error = algo->cb_init(&cc_mem, ptr); 2020 else 2021 error = 0; 2022 /* 2023 * The CC algorithms, when given their memory 2024 * should not fail we could in theory have a 2025 * KASSERT here. 2026 */ 2027 if (error == 0) { 2028 /* 2029 * Touchdown, lets go ahead and move the 2030 * connection to the new CC module by 2031 * copying in the cc_mem after we call 2032 * the old ones cleanup (if any). 2033 */ 2034 if (CC_ALGO(tp)->cb_destroy != NULL) 2035 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2036 /* Detach the old CC from the tcpcb */ 2037 cc_detach(tp); 2038 /* Copy in our temp memory that was inited */ 2039 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2040 /* Now attach the new, which takes a reference */ 2041 cc_attach(tp, algo); 2042 /* Ok now are we where we have gotten past any conn_init? */ 2043 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2044 /* Yep run the connection init for the new CC */ 2045 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2046 } 2047 } else if (ptr) 2048 free(ptr, M_CC_MEM); 2049 INP_WUNLOCK(inp); 2050 /* Now lets release our temp reference */ 2051 CC_LIST_RLOCK(); 2052 cc_release(algo); 2053 CC_LIST_RUNLOCK(); 2054 return (error); 2055 } 2056 2057 int 2058 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2059 { 2060 struct inpcb *inp = tptoinpcb(tp); 2061 int error, opt, optval; 2062 u_int ui; 2063 struct tcp_info ti; 2064 #ifdef KERN_TLS 2065 struct tls_enable tls; 2066 struct socket *so = inp->inp_socket; 2067 #endif 2068 char *pbuf, buf[TCP_LOG_ID_LEN]; 2069 #ifdef STATS 2070 struct statsblob *sbp; 2071 #endif 2072 size_t len; 2073 2074 INP_WLOCK_ASSERT(inp); 2075 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2076 ("inp_flags == %x", inp->inp_flags)); 2077 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2078 2079 switch (sopt->sopt_level) { 2080 #ifdef INET6 2081 case IPPROTO_IPV6: 2082 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2083 switch (sopt->sopt_name) { 2084 case IPV6_USE_MIN_MTU: 2085 tcp6_use_min_mtu(tp); 2086 /* FALLTHROUGH */ 2087 } 2088 INP_WUNLOCK(inp); 2089 return (0); 2090 #endif 2091 #ifdef INET 2092 case IPPROTO_IP: 2093 INP_WUNLOCK(inp); 2094 return (0); 2095 #endif 2096 } 2097 2098 /* 2099 * For TCP_CCALGOOPT forward the control to CC module, for both 2100 * SOPT_SET and SOPT_GET. 2101 */ 2102 switch (sopt->sopt_name) { 2103 case TCP_CCALGOOPT: 2104 INP_WUNLOCK(inp); 2105 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2106 return (EINVAL); 2107 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2108 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2109 sopt->sopt_valsize); 2110 if (error) { 2111 free(pbuf, M_TEMP); 2112 return (error); 2113 } 2114 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2115 if (CC_ALGO(tp)->ctl_output != NULL) 2116 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2117 else 2118 error = ENOENT; 2119 INP_WUNLOCK(inp); 2120 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2121 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2122 free(pbuf, M_TEMP); 2123 return (error); 2124 } 2125 2126 switch (sopt->sopt_dir) { 2127 case SOPT_SET: 2128 switch (sopt->sopt_name) { 2129 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2130 case TCP_MD5SIG: 2131 INP_WUNLOCK(inp); 2132 if (!TCPMD5_ENABLED()) 2133 return (ENOPROTOOPT); 2134 error = TCPMD5_PCBCTL(inp, sopt); 2135 if (error) 2136 return (error); 2137 INP_WLOCK_RECHECK(inp); 2138 goto unlock_and_done; 2139 #endif /* IPSEC */ 2140 2141 case TCP_NODELAY: 2142 case TCP_NOOPT: 2143 INP_WUNLOCK(inp); 2144 error = sooptcopyin(sopt, &optval, sizeof optval, 2145 sizeof optval); 2146 if (error) 2147 return (error); 2148 2149 INP_WLOCK_RECHECK(inp); 2150 switch (sopt->sopt_name) { 2151 case TCP_NODELAY: 2152 opt = TF_NODELAY; 2153 break; 2154 case TCP_NOOPT: 2155 opt = TF_NOOPT; 2156 break; 2157 default: 2158 opt = 0; /* dead code to fool gcc */ 2159 break; 2160 } 2161 2162 if (optval) 2163 tp->t_flags |= opt; 2164 else 2165 tp->t_flags &= ~opt; 2166 unlock_and_done: 2167 #ifdef TCP_OFFLOAD 2168 if (tp->t_flags & TF_TOE) { 2169 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2170 sopt->sopt_name); 2171 } 2172 #endif 2173 INP_WUNLOCK(inp); 2174 break; 2175 2176 case TCP_NOPUSH: 2177 INP_WUNLOCK(inp); 2178 error = sooptcopyin(sopt, &optval, sizeof optval, 2179 sizeof optval); 2180 if (error) 2181 return (error); 2182 2183 INP_WLOCK_RECHECK(inp); 2184 if (optval) 2185 tp->t_flags |= TF_NOPUSH; 2186 else if (tp->t_flags & TF_NOPUSH) { 2187 tp->t_flags &= ~TF_NOPUSH; 2188 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2189 struct epoch_tracker et; 2190 2191 NET_EPOCH_ENTER(et); 2192 error = tcp_output_nodrop(tp); 2193 NET_EPOCH_EXIT(et); 2194 } 2195 } 2196 goto unlock_and_done; 2197 2198 case TCP_REMOTE_UDP_ENCAPS_PORT: 2199 INP_WUNLOCK(inp); 2200 error = sooptcopyin(sopt, &optval, sizeof optval, 2201 sizeof optval); 2202 if (error) 2203 return (error); 2204 if ((optval < TCP_TUNNELING_PORT_MIN) || 2205 (optval > TCP_TUNNELING_PORT_MAX)) { 2206 /* Its got to be in range */ 2207 return (EINVAL); 2208 } 2209 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2210 /* You have to have enabled a UDP tunneling port first */ 2211 return (EINVAL); 2212 } 2213 INP_WLOCK_RECHECK(inp); 2214 if (tp->t_state != TCPS_CLOSED) { 2215 /* You can't change after you are connected */ 2216 error = EINVAL; 2217 } else { 2218 /* Ok we are all good set the port */ 2219 tp->t_port = htons(optval); 2220 } 2221 goto unlock_and_done; 2222 2223 case TCP_MAXSEG: 2224 INP_WUNLOCK(inp); 2225 error = sooptcopyin(sopt, &optval, sizeof optval, 2226 sizeof optval); 2227 if (error) 2228 return (error); 2229 2230 INP_WLOCK_RECHECK(inp); 2231 if (optval > 0 && optval <= tp->t_maxseg && 2232 optval + 40 >= V_tcp_minmss) 2233 tp->t_maxseg = optval; 2234 else 2235 error = EINVAL; 2236 goto unlock_and_done; 2237 2238 case TCP_INFO: 2239 INP_WUNLOCK(inp); 2240 error = EINVAL; 2241 break; 2242 2243 case TCP_STATS: 2244 INP_WUNLOCK(inp); 2245 #ifdef STATS 2246 error = sooptcopyin(sopt, &optval, sizeof optval, 2247 sizeof optval); 2248 if (error) 2249 return (error); 2250 2251 if (optval > 0) 2252 sbp = stats_blob_alloc( 2253 V_tcp_perconn_stats_dflt_tpl, 0); 2254 else 2255 sbp = NULL; 2256 2257 INP_WLOCK_RECHECK(inp); 2258 if ((tp->t_stats != NULL && sbp == NULL) || 2259 (tp->t_stats == NULL && sbp != NULL)) { 2260 struct statsblob *t = tp->t_stats; 2261 tp->t_stats = sbp; 2262 sbp = t; 2263 } 2264 INP_WUNLOCK(inp); 2265 2266 stats_blob_destroy(sbp); 2267 #else 2268 return (EOPNOTSUPP); 2269 #endif /* !STATS */ 2270 break; 2271 2272 case TCP_CONGESTION: 2273 error = tcp_set_cc_mod(inp, sopt); 2274 break; 2275 2276 case TCP_REUSPORT_LB_NUMA: 2277 INP_WUNLOCK(inp); 2278 error = sooptcopyin(sopt, &optval, sizeof(optval), 2279 sizeof(optval)); 2280 INP_WLOCK_RECHECK(inp); 2281 if (!error) 2282 error = in_pcblbgroup_numa(inp, optval); 2283 INP_WUNLOCK(inp); 2284 break; 2285 2286 #ifdef KERN_TLS 2287 case TCP_TXTLS_ENABLE: 2288 INP_WUNLOCK(inp); 2289 error = copyin_tls_enable(sopt, &tls); 2290 if (error) 2291 break; 2292 error = ktls_enable_tx(so, &tls); 2293 break; 2294 case TCP_TXTLS_MODE: 2295 INP_WUNLOCK(inp); 2296 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2297 if (error) 2298 return (error); 2299 2300 INP_WLOCK_RECHECK(inp); 2301 error = ktls_set_tx_mode(so, ui); 2302 INP_WUNLOCK(inp); 2303 break; 2304 case TCP_RXTLS_ENABLE: 2305 INP_WUNLOCK(inp); 2306 error = sooptcopyin(sopt, &tls, sizeof(tls), 2307 sizeof(tls)); 2308 if (error) 2309 break; 2310 error = ktls_enable_rx(so, &tls); 2311 break; 2312 #endif 2313 case TCP_MAXUNACKTIME: 2314 case TCP_KEEPIDLE: 2315 case TCP_KEEPINTVL: 2316 case TCP_KEEPINIT: 2317 INP_WUNLOCK(inp); 2318 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2319 if (error) 2320 return (error); 2321 2322 if (ui > (UINT_MAX / hz)) { 2323 error = EINVAL; 2324 break; 2325 } 2326 ui *= hz; 2327 2328 INP_WLOCK_RECHECK(inp); 2329 switch (sopt->sopt_name) { 2330 case TCP_MAXUNACKTIME: 2331 tp->t_maxunacktime = ui; 2332 break; 2333 2334 case TCP_KEEPIDLE: 2335 tp->t_keepidle = ui; 2336 /* 2337 * XXX: better check current remaining 2338 * timeout and "merge" it with new value. 2339 */ 2340 if ((tp->t_state > TCPS_LISTEN) && 2341 (tp->t_state <= TCPS_CLOSING)) 2342 tcp_timer_activate(tp, TT_KEEP, 2343 TP_KEEPIDLE(tp)); 2344 break; 2345 case TCP_KEEPINTVL: 2346 tp->t_keepintvl = ui; 2347 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2348 (TP_MAXIDLE(tp) > 0)) 2349 tcp_timer_activate(tp, TT_2MSL, 2350 TP_MAXIDLE(tp)); 2351 break; 2352 case TCP_KEEPINIT: 2353 tp->t_keepinit = ui; 2354 if (tp->t_state == TCPS_SYN_RECEIVED || 2355 tp->t_state == TCPS_SYN_SENT) 2356 tcp_timer_activate(tp, TT_KEEP, 2357 TP_KEEPINIT(tp)); 2358 break; 2359 } 2360 goto unlock_and_done; 2361 2362 case TCP_KEEPCNT: 2363 INP_WUNLOCK(inp); 2364 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2365 if (error) 2366 return (error); 2367 2368 INP_WLOCK_RECHECK(inp); 2369 tp->t_keepcnt = ui; 2370 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2371 (TP_MAXIDLE(tp) > 0)) 2372 tcp_timer_activate(tp, TT_2MSL, 2373 TP_MAXIDLE(tp)); 2374 goto unlock_and_done; 2375 2376 #ifdef TCPPCAP 2377 case TCP_PCAP_OUT: 2378 case TCP_PCAP_IN: 2379 INP_WUNLOCK(inp); 2380 error = sooptcopyin(sopt, &optval, sizeof optval, 2381 sizeof optval); 2382 if (error) 2383 return (error); 2384 2385 INP_WLOCK_RECHECK(inp); 2386 if (optval >= 0) 2387 tcp_pcap_set_sock_max( 2388 (sopt->sopt_name == TCP_PCAP_OUT) ? 2389 &(tp->t_outpkts) : &(tp->t_inpkts), 2390 optval); 2391 else 2392 error = EINVAL; 2393 goto unlock_and_done; 2394 #endif 2395 2396 case TCP_FASTOPEN: { 2397 struct tcp_fastopen tfo_optval; 2398 2399 INP_WUNLOCK(inp); 2400 if (!V_tcp_fastopen_client_enable && 2401 !V_tcp_fastopen_server_enable) 2402 return (EPERM); 2403 2404 error = sooptcopyin(sopt, &tfo_optval, 2405 sizeof(tfo_optval), sizeof(int)); 2406 if (error) 2407 return (error); 2408 2409 INP_WLOCK_RECHECK(inp); 2410 if ((tp->t_state != TCPS_CLOSED) && 2411 (tp->t_state != TCPS_LISTEN)) { 2412 error = EINVAL; 2413 goto unlock_and_done; 2414 } 2415 if (tfo_optval.enable) { 2416 if (tp->t_state == TCPS_LISTEN) { 2417 if (!V_tcp_fastopen_server_enable) { 2418 error = EPERM; 2419 goto unlock_and_done; 2420 } 2421 2422 if (tp->t_tfo_pending == NULL) 2423 tp->t_tfo_pending = 2424 tcp_fastopen_alloc_counter(); 2425 } else { 2426 /* 2427 * If a pre-shared key was provided, 2428 * stash it in the client cookie 2429 * field of the tcpcb for use during 2430 * connect. 2431 */ 2432 if (sopt->sopt_valsize == 2433 sizeof(tfo_optval)) { 2434 memcpy(tp->t_tfo_cookie.client, 2435 tfo_optval.psk, 2436 TCP_FASTOPEN_PSK_LEN); 2437 tp->t_tfo_client_cookie_len = 2438 TCP_FASTOPEN_PSK_LEN; 2439 } 2440 } 2441 tp->t_flags |= TF_FASTOPEN; 2442 } else 2443 tp->t_flags &= ~TF_FASTOPEN; 2444 goto unlock_and_done; 2445 } 2446 2447 #ifdef TCP_BLACKBOX 2448 case TCP_LOG: 2449 INP_WUNLOCK(inp); 2450 error = sooptcopyin(sopt, &optval, sizeof optval, 2451 sizeof optval); 2452 if (error) 2453 return (error); 2454 2455 INP_WLOCK_RECHECK(inp); 2456 error = tcp_log_state_change(tp, optval); 2457 goto unlock_and_done; 2458 2459 case TCP_LOGBUF: 2460 INP_WUNLOCK(inp); 2461 error = EINVAL; 2462 break; 2463 2464 case TCP_LOGID: 2465 INP_WUNLOCK(inp); 2466 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2467 if (error) 2468 break; 2469 buf[sopt->sopt_valsize] = '\0'; 2470 INP_WLOCK_RECHECK(inp); 2471 error = tcp_log_set_id(tp, buf); 2472 /* tcp_log_set_id() unlocks the INP. */ 2473 break; 2474 2475 case TCP_LOGDUMP: 2476 case TCP_LOGDUMPID: 2477 INP_WUNLOCK(inp); 2478 error = 2479 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2480 if (error) 2481 break; 2482 buf[sopt->sopt_valsize] = '\0'; 2483 INP_WLOCK_RECHECK(inp); 2484 if (sopt->sopt_name == TCP_LOGDUMP) { 2485 error = tcp_log_dump_tp_logbuf(tp, buf, 2486 M_WAITOK, true); 2487 INP_WUNLOCK(inp); 2488 } else { 2489 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2490 /* 2491 * tcp_log_dump_tp_bucket_logbufs() drops the 2492 * INP lock. 2493 */ 2494 } 2495 break; 2496 #endif 2497 2498 default: 2499 INP_WUNLOCK(inp); 2500 error = ENOPROTOOPT; 2501 break; 2502 } 2503 break; 2504 2505 case SOPT_GET: 2506 tp = intotcpcb(inp); 2507 switch (sopt->sopt_name) { 2508 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2509 case TCP_MD5SIG: 2510 INP_WUNLOCK(inp); 2511 if (!TCPMD5_ENABLED()) 2512 return (ENOPROTOOPT); 2513 error = TCPMD5_PCBCTL(inp, sopt); 2514 break; 2515 #endif 2516 2517 case TCP_NODELAY: 2518 optval = tp->t_flags & TF_NODELAY; 2519 INP_WUNLOCK(inp); 2520 error = sooptcopyout(sopt, &optval, sizeof optval); 2521 break; 2522 case TCP_MAXSEG: 2523 optval = tp->t_maxseg; 2524 INP_WUNLOCK(inp); 2525 error = sooptcopyout(sopt, &optval, sizeof optval); 2526 break; 2527 case TCP_REMOTE_UDP_ENCAPS_PORT: 2528 optval = ntohs(tp->t_port); 2529 INP_WUNLOCK(inp); 2530 error = sooptcopyout(sopt, &optval, sizeof optval); 2531 break; 2532 case TCP_NOOPT: 2533 optval = tp->t_flags & TF_NOOPT; 2534 INP_WUNLOCK(inp); 2535 error = sooptcopyout(sopt, &optval, sizeof optval); 2536 break; 2537 case TCP_NOPUSH: 2538 optval = tp->t_flags & TF_NOPUSH; 2539 INP_WUNLOCK(inp); 2540 error = sooptcopyout(sopt, &optval, sizeof optval); 2541 break; 2542 case TCP_INFO: 2543 tcp_fill_info(tp, &ti); 2544 INP_WUNLOCK(inp); 2545 error = sooptcopyout(sopt, &ti, sizeof ti); 2546 break; 2547 case TCP_STATS: 2548 { 2549 #ifdef STATS 2550 int nheld; 2551 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2552 2553 error = 0; 2554 socklen_t outsbsz = sopt->sopt_valsize; 2555 if (tp->t_stats == NULL) 2556 error = ENOENT; 2557 else if (outsbsz >= tp->t_stats->cursz) 2558 outsbsz = tp->t_stats->cursz; 2559 else if (outsbsz >= sizeof(struct statsblob)) 2560 outsbsz = sizeof(struct statsblob); 2561 else 2562 error = EINVAL; 2563 INP_WUNLOCK(inp); 2564 if (error) 2565 break; 2566 2567 sbp = sopt->sopt_val; 2568 nheld = atop(round_page(((vm_offset_t)sbp) + 2569 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2570 vm_page_t ma[nheld]; 2571 if (vm_fault_quick_hold_pages( 2572 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2573 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2574 nheld) < 0) { 2575 error = EFAULT; 2576 break; 2577 } 2578 2579 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2580 SIZEOF_MEMBER(struct statsblob, flags)))) 2581 goto unhold; 2582 2583 INP_WLOCK_RECHECK(inp); 2584 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2585 sbflags | SB_CLONE_USRDSTNOFAULT); 2586 INP_WUNLOCK(inp); 2587 sopt->sopt_valsize = outsbsz; 2588 unhold: 2589 vm_page_unhold_pages(ma, nheld); 2590 #else 2591 INP_WUNLOCK(inp); 2592 error = EOPNOTSUPP; 2593 #endif /* !STATS */ 2594 break; 2595 } 2596 case TCP_CONGESTION: 2597 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2598 INP_WUNLOCK(inp); 2599 error = sooptcopyout(sopt, buf, len + 1); 2600 break; 2601 case TCP_MAXUNACKTIME: 2602 case TCP_KEEPIDLE: 2603 case TCP_KEEPINTVL: 2604 case TCP_KEEPINIT: 2605 case TCP_KEEPCNT: 2606 switch (sopt->sopt_name) { 2607 case TCP_MAXUNACKTIME: 2608 ui = TP_MAXUNACKTIME(tp) / hz; 2609 break; 2610 case TCP_KEEPIDLE: 2611 ui = TP_KEEPIDLE(tp) / hz; 2612 break; 2613 case TCP_KEEPINTVL: 2614 ui = TP_KEEPINTVL(tp) / hz; 2615 break; 2616 case TCP_KEEPINIT: 2617 ui = TP_KEEPINIT(tp) / hz; 2618 break; 2619 case TCP_KEEPCNT: 2620 ui = TP_KEEPCNT(tp); 2621 break; 2622 } 2623 INP_WUNLOCK(inp); 2624 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2625 break; 2626 #ifdef TCPPCAP 2627 case TCP_PCAP_OUT: 2628 case TCP_PCAP_IN: 2629 optval = tcp_pcap_get_sock_max( 2630 (sopt->sopt_name == TCP_PCAP_OUT) ? 2631 &(tp->t_outpkts) : &(tp->t_inpkts)); 2632 INP_WUNLOCK(inp); 2633 error = sooptcopyout(sopt, &optval, sizeof optval); 2634 break; 2635 #endif 2636 case TCP_FASTOPEN: 2637 optval = tp->t_flags & TF_FASTOPEN; 2638 INP_WUNLOCK(inp); 2639 error = sooptcopyout(sopt, &optval, sizeof optval); 2640 break; 2641 #ifdef TCP_BLACKBOX 2642 case TCP_LOG: 2643 optval = tcp_get_bblog_state(tp); 2644 INP_WUNLOCK(inp); 2645 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2646 break; 2647 case TCP_LOGBUF: 2648 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2649 error = tcp_log_getlogbuf(sopt, tp); 2650 break; 2651 case TCP_LOGID: 2652 len = tcp_log_get_id(tp, buf); 2653 INP_WUNLOCK(inp); 2654 error = sooptcopyout(sopt, buf, len + 1); 2655 break; 2656 case TCP_LOGDUMP: 2657 case TCP_LOGDUMPID: 2658 INP_WUNLOCK(inp); 2659 error = EINVAL; 2660 break; 2661 #endif 2662 #ifdef KERN_TLS 2663 case TCP_TXTLS_MODE: 2664 error = ktls_get_tx_mode(so, &optval); 2665 INP_WUNLOCK(inp); 2666 if (error == 0) 2667 error = sooptcopyout(sopt, &optval, 2668 sizeof(optval)); 2669 break; 2670 case TCP_RXTLS_MODE: 2671 error = ktls_get_rx_mode(so, &optval); 2672 INP_WUNLOCK(inp); 2673 if (error == 0) 2674 error = sooptcopyout(sopt, &optval, 2675 sizeof(optval)); 2676 break; 2677 #endif 2678 default: 2679 INP_WUNLOCK(inp); 2680 error = ENOPROTOOPT; 2681 break; 2682 } 2683 break; 2684 } 2685 return (error); 2686 } 2687 #undef INP_WLOCK_RECHECK 2688 #undef INP_WLOCK_RECHECK_CLEANUP 2689 2690 /* 2691 * Initiate (or continue) disconnect. 2692 * If embryonic state, just send reset (once). 2693 * If in ``let data drain'' option and linger null, just drop. 2694 * Otherwise (hard), mark socket disconnecting and drop 2695 * current input data; switch states based on user close, and 2696 * send segment to peer (with FIN). 2697 */ 2698 static void 2699 tcp_disconnect(struct tcpcb *tp) 2700 { 2701 struct inpcb *inp = tptoinpcb(tp); 2702 struct socket *so = tptosocket(tp); 2703 2704 NET_EPOCH_ASSERT(); 2705 INP_WLOCK_ASSERT(inp); 2706 2707 /* 2708 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2709 * socket is still open. 2710 */ 2711 if (tp->t_state < TCPS_ESTABLISHED && 2712 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2713 tp = tcp_close(tp); 2714 KASSERT(tp != NULL, 2715 ("tcp_disconnect: tcp_close() returned NULL")); 2716 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2717 tp = tcp_drop(tp, 0); 2718 KASSERT(tp != NULL, 2719 ("tcp_disconnect: tcp_drop() returned NULL")); 2720 } else { 2721 soisdisconnecting(so); 2722 sbflush(&so->so_rcv); 2723 tcp_usrclosed(tp); 2724 if (!(inp->inp_flags & INP_DROPPED)) 2725 /* Ignore stack's drop request, we already at it. */ 2726 (void)tcp_output_nodrop(tp); 2727 } 2728 } 2729 2730 /* 2731 * User issued close, and wish to trail through shutdown states: 2732 * if never received SYN, just forget it. If got a SYN from peer, 2733 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2734 * If already got a FIN from peer, then almost done; go to LAST_ACK 2735 * state. In all other cases, have already sent FIN to peer (e.g. 2736 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2737 * for peer to send FIN or not respond to keep-alives, etc. 2738 * We can let the user exit from the close as soon as the FIN is acked. 2739 */ 2740 static void 2741 tcp_usrclosed(struct tcpcb *tp) 2742 { 2743 2744 NET_EPOCH_ASSERT(); 2745 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2746 2747 switch (tp->t_state) { 2748 case TCPS_LISTEN: 2749 #ifdef TCP_OFFLOAD 2750 tcp_offload_listen_stop(tp); 2751 #endif 2752 tcp_state_change(tp, TCPS_CLOSED); 2753 /* FALLTHROUGH */ 2754 case TCPS_CLOSED: 2755 tp = tcp_close(tp); 2756 /* 2757 * tcp_close() should never return NULL here as the socket is 2758 * still open. 2759 */ 2760 KASSERT(tp != NULL, 2761 ("tcp_usrclosed: tcp_close() returned NULL")); 2762 break; 2763 2764 case TCPS_SYN_SENT: 2765 case TCPS_SYN_RECEIVED: 2766 tp->t_flags |= TF_NEEDFIN; 2767 break; 2768 2769 case TCPS_ESTABLISHED: 2770 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2771 break; 2772 2773 case TCPS_CLOSE_WAIT: 2774 tcp_state_change(tp, TCPS_LAST_ACK); 2775 break; 2776 } 2777 if (tp->t_acktime == 0) 2778 tp->t_acktime = ticks; 2779 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2780 tcp_free_sackholes(tp); 2781 soisdisconnected(tptosocket(tp)); 2782 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2783 if (tp->t_state == TCPS_FIN_WAIT_2) { 2784 int timeout; 2785 2786 timeout = (tcp_fast_finwait2_recycle) ? 2787 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2788 tcp_timer_activate(tp, TT_2MSL, timeout); 2789 } 2790 } 2791 } 2792 2793 #ifdef DDB 2794 static void 2795 db_print_indent(int indent) 2796 { 2797 int i; 2798 2799 for (i = 0; i < indent; i++) 2800 db_printf(" "); 2801 } 2802 2803 static void 2804 db_print_tstate(int t_state) 2805 { 2806 2807 switch (t_state) { 2808 case TCPS_CLOSED: 2809 db_printf("TCPS_CLOSED"); 2810 return; 2811 2812 case TCPS_LISTEN: 2813 db_printf("TCPS_LISTEN"); 2814 return; 2815 2816 case TCPS_SYN_SENT: 2817 db_printf("TCPS_SYN_SENT"); 2818 return; 2819 2820 case TCPS_SYN_RECEIVED: 2821 db_printf("TCPS_SYN_RECEIVED"); 2822 return; 2823 2824 case TCPS_ESTABLISHED: 2825 db_printf("TCPS_ESTABLISHED"); 2826 return; 2827 2828 case TCPS_CLOSE_WAIT: 2829 db_printf("TCPS_CLOSE_WAIT"); 2830 return; 2831 2832 case TCPS_FIN_WAIT_1: 2833 db_printf("TCPS_FIN_WAIT_1"); 2834 return; 2835 2836 case TCPS_CLOSING: 2837 db_printf("TCPS_CLOSING"); 2838 return; 2839 2840 case TCPS_LAST_ACK: 2841 db_printf("TCPS_LAST_ACK"); 2842 return; 2843 2844 case TCPS_FIN_WAIT_2: 2845 db_printf("TCPS_FIN_WAIT_2"); 2846 return; 2847 2848 case TCPS_TIME_WAIT: 2849 db_printf("TCPS_TIME_WAIT"); 2850 return; 2851 2852 default: 2853 db_printf("unknown"); 2854 return; 2855 } 2856 } 2857 2858 static void 2859 db_print_tflags(u_int t_flags) 2860 { 2861 int comma; 2862 2863 comma = 0; 2864 if (t_flags & TF_ACKNOW) { 2865 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2866 comma = 1; 2867 } 2868 if (t_flags & TF_DELACK) { 2869 db_printf("%sTF_DELACK", comma ? ", " : ""); 2870 comma = 1; 2871 } 2872 if (t_flags & TF_NODELAY) { 2873 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2874 comma = 1; 2875 } 2876 if (t_flags & TF_NOOPT) { 2877 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2878 comma = 1; 2879 } 2880 if (t_flags & TF_SENTFIN) { 2881 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2882 comma = 1; 2883 } 2884 if (t_flags & TF_REQ_SCALE) { 2885 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2886 comma = 1; 2887 } 2888 if (t_flags & TF_RCVD_SCALE) { 2889 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2890 comma = 1; 2891 } 2892 if (t_flags & TF_REQ_TSTMP) { 2893 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2894 comma = 1; 2895 } 2896 if (t_flags & TF_RCVD_TSTMP) { 2897 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2898 comma = 1; 2899 } 2900 if (t_flags & TF_SACK_PERMIT) { 2901 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2902 comma = 1; 2903 } 2904 if (t_flags & TF_NEEDSYN) { 2905 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2906 comma = 1; 2907 } 2908 if (t_flags & TF_NEEDFIN) { 2909 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2910 comma = 1; 2911 } 2912 if (t_flags & TF_NOPUSH) { 2913 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2914 comma = 1; 2915 } 2916 if (t_flags & TF_PREVVALID) { 2917 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2918 comma = 1; 2919 } 2920 if (t_flags & TF_MORETOCOME) { 2921 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2922 comma = 1; 2923 } 2924 if (t_flags & TF_SONOTCONN) { 2925 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2926 comma = 1; 2927 } 2928 if (t_flags & TF_LASTIDLE) { 2929 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2930 comma = 1; 2931 } 2932 if (t_flags & TF_RXWIN0SENT) { 2933 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2934 comma = 1; 2935 } 2936 if (t_flags & TF_FASTRECOVERY) { 2937 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2938 comma = 1; 2939 } 2940 if (t_flags & TF_CONGRECOVERY) { 2941 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2942 comma = 1; 2943 } 2944 if (t_flags & TF_WASFRECOVERY) { 2945 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2946 comma = 1; 2947 } 2948 if (t_flags & TF_WASCRECOVERY) { 2949 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2950 comma = 1; 2951 } 2952 if (t_flags & TF_SIGNATURE) { 2953 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2954 comma = 1; 2955 } 2956 if (t_flags & TF_FORCEDATA) { 2957 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2958 comma = 1; 2959 } 2960 if (t_flags & TF_TSO) { 2961 db_printf("%sTF_TSO", comma ? ", " : ""); 2962 comma = 1; 2963 } 2964 if (t_flags & TF_FASTOPEN) { 2965 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2966 comma = 1; 2967 } 2968 } 2969 2970 static void 2971 db_print_tflags2(u_int t_flags2) 2972 { 2973 int comma; 2974 2975 comma = 0; 2976 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2977 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2978 comma = 1; 2979 } 2980 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2981 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2982 comma = 1; 2983 } 2984 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2985 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2986 comma = 1; 2987 } 2988 if (t_flags2 & TF2_LOG_AUTO) { 2989 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2990 comma = 1; 2991 } 2992 if (t_flags2 & TF2_DROP_AF_DATA) { 2993 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2994 comma = 1; 2995 } 2996 if (t_flags2 & TF2_ECN_PERMIT) { 2997 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2998 comma = 1; 2999 } 3000 if (t_flags2 & TF2_ECN_SND_CWR) { 3001 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 3002 comma = 1; 3003 } 3004 if (t_flags2 & TF2_ECN_SND_ECE) { 3005 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 3006 comma = 1; 3007 } 3008 if (t_flags2 & TF2_ACE_PERMIT) { 3009 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 3010 comma = 1; 3011 } 3012 if (t_flags2 & TF2_FBYTES_COMPLETE) { 3013 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 3014 comma = 1; 3015 } 3016 } 3017 3018 static void 3019 db_print_toobflags(char t_oobflags) 3020 { 3021 int comma; 3022 3023 comma = 0; 3024 if (t_oobflags & TCPOOB_HAVEDATA) { 3025 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3026 comma = 1; 3027 } 3028 if (t_oobflags & TCPOOB_HADDATA) { 3029 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3030 comma = 1; 3031 } 3032 } 3033 3034 static void 3035 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3036 { 3037 3038 db_print_indent(indent); 3039 db_printf("%s at %p\n", name, tp); 3040 3041 indent += 2; 3042 3043 db_print_indent(indent); 3044 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3045 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3046 3047 db_print_indent(indent); 3048 db_printf("t_callout: %p t_timers: %p\n", 3049 &tp->t_callout, &tp->t_timers); 3050 3051 db_print_indent(indent); 3052 db_printf("t_state: %d (", tp->t_state); 3053 db_print_tstate(tp->t_state); 3054 db_printf(")\n"); 3055 3056 db_print_indent(indent); 3057 db_printf("t_flags: 0x%x (", tp->t_flags); 3058 db_print_tflags(tp->t_flags); 3059 db_printf(")\n"); 3060 3061 db_print_indent(indent); 3062 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3063 db_print_tflags2(tp->t_flags2); 3064 db_printf(")\n"); 3065 3066 db_print_indent(indent); 3067 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 3068 tp->snd_una, tp->snd_max, tp->snd_nxt); 3069 3070 db_print_indent(indent); 3071 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3072 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3073 3074 db_print_indent(indent); 3075 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3076 tp->iss, tp->irs, tp->rcv_nxt); 3077 3078 db_print_indent(indent); 3079 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3080 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3081 3082 db_print_indent(indent); 3083 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3084 tp->snd_wnd, tp->snd_cwnd); 3085 3086 db_print_indent(indent); 3087 db_printf("snd_ssthresh: %u snd_recover: " 3088 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3089 3090 db_print_indent(indent); 3091 db_printf("t_rcvtime: %u t_startime: %u\n", 3092 tp->t_rcvtime, tp->t_starttime); 3093 3094 db_print_indent(indent); 3095 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3096 tp->t_rtttime, tp->t_rtseq); 3097 3098 db_print_indent(indent); 3099 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3100 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3101 3102 db_print_indent(indent); 3103 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3104 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3105 3106 db_print_indent(indent); 3107 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3108 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3109 3110 db_print_indent(indent); 3111 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3112 db_print_toobflags(tp->t_oobflags); 3113 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3114 3115 db_print_indent(indent); 3116 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3117 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3118 3119 db_print_indent(indent); 3120 db_printf("ts_recent: %u ts_recent_age: %u\n", 3121 tp->ts_recent, tp->ts_recent_age); 3122 3123 db_print_indent(indent); 3124 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3125 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3126 3127 db_print_indent(indent); 3128 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3129 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3130 tp->snd_recover_prev, tp->t_badrxtwin); 3131 3132 db_print_indent(indent); 3133 db_printf("snd_numholes: %d snd_holes first: %p\n", 3134 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3135 3136 db_print_indent(indent); 3137 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3138 tp->snd_fack, tp->rcv_numsacks); 3139 3140 /* Skip sackblks, sackhint. */ 3141 3142 db_print_indent(indent); 3143 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3144 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3145 } 3146 3147 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3148 { 3149 struct tcpcb *tp; 3150 3151 if (!have_addr) { 3152 db_printf("usage: show tcpcb <addr>\n"); 3153 return; 3154 } 3155 tp = (struct tcpcb *)addr; 3156 3157 db_print_tcpcb(tp, "tcpcb", 0); 3158 } 3159 #endif 3160