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