1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_ddb.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_ipsec.h" 45 #include "opt_kern_tls.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/arb.h> 50 #include <sys/limits.h> 51 #include <sys/malloc.h> 52 #include <sys/refcount.h> 53 #include <sys/kernel.h> 54 #include <sys/ktls.h> 55 #include <sys/qmath.h> 56 #include <sys/sysctl.h> 57 #include <sys/mbuf.h> 58 #ifdef INET6 59 #include <sys/domain.h> 60 #endif /* INET6 */ 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/protosw.h> 64 #include <sys/proc.h> 65 #include <sys/jail.h> 66 #include <sys/stats.h> 67 68 #ifdef DDB 69 #include <ddb/ddb.h> 70 #endif 71 72 #include <net/if.h> 73 #include <net/if_var.h> 74 #include <net/route.h> 75 #include <net/vnet.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_kdtrace.h> 79 #include <netinet/in_pcb.h> 80 #include <netinet/in_systm.h> 81 #include <netinet/in_var.h> 82 #include <netinet/ip.h> 83 #include <netinet/ip_var.h> 84 #ifdef INET6 85 #include <netinet/ip6.h> 86 #include <netinet6/in6_pcb.h> 87 #include <netinet6/ip6_var.h> 88 #include <netinet6/scope6_var.h> 89 #endif 90 #include <netinet/tcp.h> 91 #include <netinet/tcp_fsm.h> 92 #include <netinet/tcp_seq.h> 93 #include <netinet/tcp_timer.h> 94 #include <netinet/tcp_var.h> 95 #include <netinet/tcp_log_buf.h> 96 #include <netinet/tcpip.h> 97 #include <netinet/cc/cc.h> 98 #include <netinet/tcp_fastopen.h> 99 #include <netinet/tcp_hpts.h> 100 #ifdef TCPPCAP 101 #include <netinet/tcp_pcap.h> 102 #endif 103 #ifdef TCP_OFFLOAD 104 #include <netinet/tcp_offload.h> 105 #endif 106 #include <netipsec/ipsec_support.h> 107 108 #include <vm/vm.h> 109 #include <vm/vm_param.h> 110 #include <vm/pmap.h> 111 #include <vm/vm_extern.h> 112 #include <vm/vm_map.h> 113 #include <vm/vm_page.h> 114 115 /* 116 * TCP protocol interface to socket abstraction. 117 */ 118 #ifdef INET 119 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 120 struct thread *td); 121 #endif /* INET */ 122 #ifdef INET6 123 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 124 struct thread *td); 125 #endif /* INET6 */ 126 static void tcp_disconnect(struct tcpcb *); 127 static void tcp_usrclosed(struct tcpcb *); 128 static void tcp_fill_info(const struct tcpcb *, struct tcp_info *); 129 130 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 131 132 static void 133 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error) 134 { 135 struct tcp_log_buffer *lgb; 136 137 KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL")); 138 INP_WLOCK_ASSERT(tptoinpcb(tp)); 139 if (tcp_bblogging_on(tp)) { 140 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error, 141 0, NULL, false, NULL, NULL, 0, NULL); 142 } else { 143 lgb = NULL; 144 } 145 if (lgb != NULL) { 146 if (error >= 0) { 147 lgb->tlb_errno = (uint32_t)error; 148 } 149 lgb->tlb_flex1 = pru; 150 } 151 } 152 153 /* 154 * TCP attaches to socket via pru_attach(), reserving space, 155 * and an internet control block. 156 */ 157 static int 158 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 159 { 160 struct inpcb *inp; 161 struct tcpcb *tp = NULL; 162 int error; 163 164 inp = sotoinpcb(so); 165 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 166 167 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 168 if (error) 169 goto out; 170 171 so->so_rcv.sb_flags |= SB_AUTOSIZE; 172 so->so_snd.sb_flags |= SB_AUTOSIZE; 173 error = in_pcballoc(so, &V_tcbinfo); 174 if (error) 175 goto out; 176 inp = sotoinpcb(so); 177 tp = tcp_newtcpcb(inp); 178 if (tp == NULL) { 179 error = ENOBUFS; 180 in_pcbdetach(inp); 181 in_pcbfree(inp); 182 goto out; 183 } 184 tp->t_state = TCPS_CLOSED; 185 tcp_bblog_pru(tp, PRU_ATTACH, error); 186 INP_WUNLOCK(inp); 187 TCPSTATES_INC(TCPS_CLOSED); 188 out: 189 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 190 return (error); 191 } 192 193 /* 194 * tcp_usr_detach is called when the socket layer loses its final reference 195 * to the socket, be it a file descriptor reference, a reference from TCP, 196 * etc. At this point, there is only one case in which we will keep around 197 * inpcb state: time wait. 198 */ 199 static void 200 tcp_usr_detach(struct socket *so) 201 { 202 struct inpcb *inp; 203 struct tcpcb *tp; 204 205 inp = sotoinpcb(so); 206 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 207 INP_WLOCK(inp); 208 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 209 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 210 211 tp = intotcpcb(inp); 212 213 KASSERT(inp->inp_flags & INP_DROPPED || 214 tp->t_state < TCPS_SYN_SENT, 215 ("%s: inp %p not dropped or embryonic", __func__, inp)); 216 217 tcp_discardcb(tp); 218 in_pcbdetach(inp); 219 in_pcbfree(inp); 220 } 221 222 #ifdef INET 223 /* 224 * Give the socket an address. 225 */ 226 static int 227 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 228 { 229 int error = 0; 230 struct inpcb *inp; 231 struct tcpcb *tp; 232 struct sockaddr_in *sinp; 233 234 inp = sotoinpcb(so); 235 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 236 INP_WLOCK(inp); 237 if (inp->inp_flags & INP_DROPPED) { 238 INP_WUNLOCK(inp); 239 return (EINVAL); 240 } 241 tp = intotcpcb(inp); 242 243 sinp = (struct sockaddr_in *)nam; 244 if (nam->sa_family != AF_INET) { 245 /* 246 * Preserve compatibility with old programs. 247 */ 248 if (nam->sa_family != AF_UNSPEC || 249 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 250 sinp->sin_addr.s_addr != INADDR_ANY) { 251 error = EAFNOSUPPORT; 252 goto out; 253 } 254 nam->sa_family = AF_INET; 255 } 256 if (nam->sa_len != sizeof(*sinp)) { 257 error = EINVAL; 258 goto out; 259 } 260 /* 261 * Must check for multicast addresses and disallow binding 262 * to them. 263 */ 264 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 265 error = EAFNOSUPPORT; 266 goto out; 267 } 268 INP_HASH_WLOCK(&V_tcbinfo); 269 error = in_pcbbind(inp, sinp, td->td_ucred); 270 INP_HASH_WUNLOCK(&V_tcbinfo); 271 out: 272 tcp_bblog_pru(tp, PRU_BIND, error); 273 TCP_PROBE2(debug__user, tp, PRU_BIND); 274 INP_WUNLOCK(inp); 275 276 return (error); 277 } 278 #endif /* INET */ 279 280 #ifdef INET6 281 static int 282 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 283 { 284 int error = 0; 285 struct inpcb *inp; 286 struct tcpcb *tp; 287 struct sockaddr_in6 *sin6; 288 u_char vflagsav; 289 290 inp = sotoinpcb(so); 291 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 292 INP_WLOCK(inp); 293 if (inp->inp_flags & INP_DROPPED) { 294 INP_WUNLOCK(inp); 295 return (EINVAL); 296 } 297 tp = intotcpcb(inp); 298 299 vflagsav = inp->inp_vflag; 300 301 sin6 = (struct sockaddr_in6 *)nam; 302 if (nam->sa_family != AF_INET6) { 303 error = EAFNOSUPPORT; 304 goto out; 305 } 306 if (nam->sa_len != sizeof(*sin6)) { 307 error = EINVAL; 308 goto out; 309 } 310 /* 311 * Must check for multicast addresses and disallow binding 312 * to them. 313 */ 314 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 315 error = EAFNOSUPPORT; 316 goto out; 317 } 318 319 INP_HASH_WLOCK(&V_tcbinfo); 320 inp->inp_vflag &= ~INP_IPV4; 321 inp->inp_vflag |= INP_IPV6; 322 #ifdef INET 323 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 324 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 325 inp->inp_vflag |= INP_IPV4; 326 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 327 struct sockaddr_in sin; 328 329 in6_sin6_2_sin(&sin, sin6); 330 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 331 error = EAFNOSUPPORT; 332 INP_HASH_WUNLOCK(&V_tcbinfo); 333 goto out; 334 } 335 inp->inp_vflag |= INP_IPV4; 336 inp->inp_vflag &= ~INP_IPV6; 337 error = in_pcbbind(inp, &sin, td->td_ucred); 338 INP_HASH_WUNLOCK(&V_tcbinfo); 339 goto out; 340 } 341 } 342 #endif 343 error = in6_pcbbind(inp, sin6, td->td_ucred); 344 INP_HASH_WUNLOCK(&V_tcbinfo); 345 out: 346 if (error != 0) 347 inp->inp_vflag = vflagsav; 348 tcp_bblog_pru(tp, PRU_BIND, error); 349 TCP_PROBE2(debug__user, tp, PRU_BIND); 350 INP_WUNLOCK(inp); 351 return (error); 352 } 353 #endif /* INET6 */ 354 355 #ifdef INET 356 /* 357 * Prepare to accept connections. 358 */ 359 static int 360 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 361 { 362 int error = 0; 363 struct inpcb *inp; 364 struct tcpcb *tp; 365 366 inp = sotoinpcb(so); 367 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 368 INP_WLOCK(inp); 369 if (inp->inp_flags & INP_DROPPED) { 370 INP_WUNLOCK(inp); 371 return (EINVAL); 372 } 373 tp = intotcpcb(inp); 374 375 SOCK_LOCK(so); 376 error = solisten_proto_check(so); 377 if (error != 0) { 378 SOCK_UNLOCK(so); 379 goto out; 380 } 381 if (inp->inp_lport == 0) { 382 INP_HASH_WLOCK(&V_tcbinfo); 383 error = in_pcbbind(inp, NULL, td->td_ucred); 384 INP_HASH_WUNLOCK(&V_tcbinfo); 385 } 386 if (error == 0) { 387 tcp_state_change(tp, TCPS_LISTEN); 388 solisten_proto(so, backlog); 389 #ifdef TCP_OFFLOAD 390 if ((so->so_options & SO_NO_OFFLOAD) == 0) 391 tcp_offload_listen_start(tp); 392 #endif 393 } else { 394 solisten_proto_abort(so); 395 } 396 SOCK_UNLOCK(so); 397 398 if (IS_FASTOPEN(tp->t_flags)) 399 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 400 401 out: 402 tcp_bblog_pru(tp, PRU_LISTEN, error); 403 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 404 INP_WUNLOCK(inp); 405 return (error); 406 } 407 #endif /* INET */ 408 409 #ifdef INET6 410 static int 411 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 412 { 413 int error = 0; 414 struct inpcb *inp; 415 struct tcpcb *tp; 416 u_char vflagsav; 417 418 inp = sotoinpcb(so); 419 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 420 INP_WLOCK(inp); 421 if (inp->inp_flags & INP_DROPPED) { 422 INP_WUNLOCK(inp); 423 return (EINVAL); 424 } 425 tp = intotcpcb(inp); 426 427 vflagsav = inp->inp_vflag; 428 429 SOCK_LOCK(so); 430 error = solisten_proto_check(so); 431 if (error != 0) { 432 SOCK_UNLOCK(so); 433 goto out; 434 } 435 INP_HASH_WLOCK(&V_tcbinfo); 436 if (inp->inp_lport == 0) { 437 inp->inp_vflag &= ~INP_IPV4; 438 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 439 inp->inp_vflag |= INP_IPV4; 440 error = in6_pcbbind(inp, NULL, td->td_ucred); 441 } 442 INP_HASH_WUNLOCK(&V_tcbinfo); 443 if (error == 0) { 444 tcp_state_change(tp, TCPS_LISTEN); 445 solisten_proto(so, backlog); 446 #ifdef TCP_OFFLOAD 447 if ((so->so_options & SO_NO_OFFLOAD) == 0) 448 tcp_offload_listen_start(tp); 449 #endif 450 } else { 451 solisten_proto_abort(so); 452 } 453 SOCK_UNLOCK(so); 454 455 if (IS_FASTOPEN(tp->t_flags)) 456 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 457 458 if (error != 0) 459 inp->inp_vflag = vflagsav; 460 461 out: 462 tcp_bblog_pru(tp, PRU_LISTEN, error); 463 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 464 INP_WUNLOCK(inp); 465 return (error); 466 } 467 #endif /* INET6 */ 468 469 #ifdef INET 470 /* 471 * Initiate connection to peer. 472 * Create a template for use in transmissions on this connection. 473 * Enter SYN_SENT state, and mark socket as connecting. 474 * Start keep-alive timer, and seed output sequence space. 475 * Send initial segment on connection. 476 */ 477 static int 478 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 479 { 480 struct epoch_tracker et; 481 int error = 0; 482 struct inpcb *inp; 483 struct tcpcb *tp; 484 struct sockaddr_in *sinp; 485 486 inp = sotoinpcb(so); 487 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 488 INP_WLOCK(inp); 489 if (inp->inp_flags & INP_DROPPED) { 490 INP_WUNLOCK(inp); 491 return (ECONNREFUSED); 492 } 493 tp = intotcpcb(inp); 494 495 sinp = (struct sockaddr_in *)nam; 496 if (nam->sa_family != AF_INET) { 497 error = EAFNOSUPPORT; 498 goto out; 499 } 500 if (nam->sa_len != sizeof (*sinp)) { 501 error = EINVAL; 502 goto out; 503 } 504 /* 505 * Must disallow TCP ``connections'' to multicast addresses. 506 */ 507 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 508 error = EAFNOSUPPORT; 509 goto out; 510 } 511 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 512 error = EACCES; 513 goto out; 514 } 515 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 516 goto out; 517 if (SOLISTENING(so)) { 518 error = EOPNOTSUPP; 519 goto out; 520 } 521 NET_EPOCH_ENTER(et); 522 if ((error = tcp_connect(tp, sinp, td)) != 0) 523 goto out_in_epoch; 524 #ifdef TCP_OFFLOAD 525 if (registered_toedevs > 0 && 526 (so->so_options & SO_NO_OFFLOAD) == 0 && 527 (error = tcp_offload_connect(so, nam)) == 0) 528 goto out_in_epoch; 529 #endif 530 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 531 error = tcp_output(tp); 532 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 533 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 534 out_in_epoch: 535 NET_EPOCH_EXIT(et); 536 out: 537 tcp_bblog_pru(tp, PRU_CONNECT, error); 538 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 539 INP_WUNLOCK(inp); 540 return (error); 541 } 542 #endif /* INET */ 543 544 #ifdef INET6 545 static int 546 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 547 { 548 struct epoch_tracker et; 549 int error = 0; 550 struct inpcb *inp; 551 struct tcpcb *tp; 552 struct sockaddr_in6 *sin6; 553 u_int8_t incflagsav; 554 u_char vflagsav; 555 556 inp = sotoinpcb(so); 557 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 558 INP_WLOCK(inp); 559 if (inp->inp_flags & INP_DROPPED) { 560 INP_WUNLOCK(inp); 561 return (ECONNREFUSED); 562 } 563 tp = intotcpcb(inp); 564 565 vflagsav = inp->inp_vflag; 566 incflagsav = inp->inp_inc.inc_flags; 567 568 sin6 = (struct sockaddr_in6 *)nam; 569 if (nam->sa_family != AF_INET6) { 570 error = EAFNOSUPPORT; 571 goto out; 572 } 573 if (nam->sa_len != sizeof (*sin6)) { 574 error = EINVAL; 575 goto out; 576 } 577 /* 578 * Must disallow TCP ``connections'' to multicast addresses. 579 */ 580 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 581 error = EAFNOSUPPORT; 582 goto out; 583 } 584 if (SOLISTENING(so)) { 585 error = EINVAL; 586 goto out; 587 } 588 #ifdef INET 589 /* 590 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 591 * therefore probably require the hash lock, which isn't held here. 592 * Is this a significant problem? 593 */ 594 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 595 struct sockaddr_in sin; 596 597 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 598 error = EINVAL; 599 goto out; 600 } 601 if ((inp->inp_vflag & INP_IPV4) == 0) { 602 error = EAFNOSUPPORT; 603 goto out; 604 } 605 606 in6_sin6_2_sin(&sin, sin6); 607 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 608 error = EAFNOSUPPORT; 609 goto out; 610 } 611 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 612 error = EACCES; 613 goto out; 614 } 615 if ((error = prison_remote_ip4(td->td_ucred, 616 &sin.sin_addr)) != 0) 617 goto out; 618 inp->inp_vflag |= INP_IPV4; 619 inp->inp_vflag &= ~INP_IPV6; 620 NET_EPOCH_ENTER(et); 621 if ((error = tcp_connect(tp, &sin, td)) != 0) 622 goto out_in_epoch; 623 #ifdef TCP_OFFLOAD 624 if (registered_toedevs > 0 && 625 (so->so_options & SO_NO_OFFLOAD) == 0 && 626 (error = tcp_offload_connect(so, nam)) == 0) 627 goto out_in_epoch; 628 #endif 629 error = tcp_output(tp); 630 goto out_in_epoch; 631 } else { 632 if ((inp->inp_vflag & INP_IPV6) == 0) { 633 error = EAFNOSUPPORT; 634 goto out; 635 } 636 } 637 #endif 638 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 639 goto out; 640 inp->inp_vflag &= ~INP_IPV4; 641 inp->inp_vflag |= INP_IPV6; 642 inp->inp_inc.inc_flags |= INC_ISIPV6; 643 NET_EPOCH_ENTER(et); 644 if ((error = tcp6_connect(tp, sin6, td)) != 0) 645 goto out_in_epoch; 646 #ifdef TCP_OFFLOAD 647 if (registered_toedevs > 0 && 648 (so->so_options & SO_NO_OFFLOAD) == 0 && 649 (error = tcp_offload_connect(so, nam)) == 0) 650 goto out_in_epoch; 651 #endif 652 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 653 error = tcp_output(tp); 654 out_in_epoch: 655 NET_EPOCH_EXIT(et); 656 out: 657 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 658 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 659 /* 660 * If the implicit bind in the connect call fails, restore 661 * the flags we modified. 662 */ 663 if (error != 0 && inp->inp_lport == 0) { 664 inp->inp_vflag = vflagsav; 665 inp->inp_inc.inc_flags = incflagsav; 666 } 667 668 tcp_bblog_pru(tp, PRU_CONNECT, error); 669 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 670 INP_WUNLOCK(inp); 671 return (error); 672 } 673 #endif /* INET6 */ 674 675 /* 676 * Initiate disconnect from peer. 677 * If connection never passed embryonic stage, just drop; 678 * else if don't need to let data drain, then can just drop anyways, 679 * else have to begin TCP shutdown process: mark socket disconnecting, 680 * drain unread data, state switch to reflect user close, and 681 * send segment (e.g. FIN) to peer. Socket will be really disconnected 682 * when peer sends FIN and acks ours. 683 * 684 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 685 */ 686 static int 687 tcp_usr_disconnect(struct socket *so) 688 { 689 struct inpcb *inp; 690 struct tcpcb *tp = NULL; 691 struct epoch_tracker et; 692 int error = 0; 693 694 NET_EPOCH_ENTER(et); 695 inp = sotoinpcb(so); 696 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 697 INP_WLOCK(inp); 698 if (inp->inp_flags & INP_DROPPED) { 699 INP_WUNLOCK(inp); 700 NET_EPOCH_EXIT(et); 701 return (ECONNRESET); 702 } 703 tp = intotcpcb(inp); 704 705 if (tp->t_state == TCPS_TIME_WAIT) 706 goto out; 707 tcp_disconnect(tp); 708 out: 709 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 710 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 711 INP_WUNLOCK(inp); 712 NET_EPOCH_EXIT(et); 713 return (error); 714 } 715 716 #ifdef INET 717 /* 718 * Accept a connection. Essentially all the work is done at higher levels; 719 * just return the address of the peer, storing through addr. 720 */ 721 static int 722 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 723 { 724 int error = 0; 725 struct inpcb *inp; 726 struct tcpcb *tp; 727 struct in_addr addr; 728 in_port_t port = 0; 729 730 inp = sotoinpcb(so); 731 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 732 INP_WLOCK(inp); 733 if (inp->inp_flags & INP_DROPPED) { 734 INP_WUNLOCK(inp); 735 return (ECONNABORTED); 736 } 737 tp = intotcpcb(inp); 738 739 if (so->so_state & SS_ISDISCONNECTED) { 740 error = ECONNABORTED; 741 goto out; 742 } 743 /* 744 * We inline in_getpeeraddr and COMMON_END here, so that we can 745 * copy the data of interest and defer the malloc until after we 746 * release the lock. 747 */ 748 port = inp->inp_fport; 749 addr = inp->inp_faddr; 750 751 out: 752 tcp_bblog_pru(tp, PRU_ACCEPT, error); 753 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 754 INP_WUNLOCK(inp); 755 if (error == 0) 756 *nam = in_sockaddr(port, &addr); 757 return error; 758 } 759 #endif /* INET */ 760 761 #ifdef INET6 762 static int 763 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 764 { 765 struct inpcb *inp; 766 int error = 0; 767 struct tcpcb *tp; 768 struct in_addr addr; 769 struct in6_addr addr6; 770 in_port_t port = 0; 771 int v4 = 0; 772 773 inp = sotoinpcb(so); 774 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 775 INP_WLOCK(inp); 776 if (inp->inp_flags & INP_DROPPED) { 777 INP_WUNLOCK(inp); 778 return (ECONNABORTED); 779 } 780 tp = intotcpcb(inp); 781 782 if (so->so_state & SS_ISDISCONNECTED) { 783 error = ECONNABORTED; 784 goto out; 785 } 786 /* 787 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 788 * copy the data of interest and defer the malloc until after we 789 * release the lock. 790 */ 791 if (inp->inp_vflag & INP_IPV4) { 792 v4 = 1; 793 port = inp->inp_fport; 794 addr = inp->inp_faddr; 795 } else { 796 port = inp->inp_fport; 797 addr6 = inp->in6p_faddr; 798 } 799 800 out: 801 tcp_bblog_pru(tp, PRU_ACCEPT, error); 802 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 803 INP_WUNLOCK(inp); 804 if (error == 0) { 805 if (v4) 806 *nam = in6_v4mapsin6_sockaddr(port, &addr); 807 else 808 *nam = in6_sockaddr(port, &addr6); 809 } 810 return error; 811 } 812 #endif /* INET6 */ 813 814 /* 815 * Mark the connection as being incapable of further output. 816 */ 817 static int 818 tcp_usr_shutdown(struct socket *so) 819 { 820 int error = 0; 821 struct inpcb *inp; 822 struct tcpcb *tp; 823 struct epoch_tracker et; 824 825 inp = sotoinpcb(so); 826 KASSERT(inp != NULL, ("inp == NULL")); 827 INP_WLOCK(inp); 828 if (inp->inp_flags & INP_DROPPED) { 829 INP_WUNLOCK(inp); 830 return (ECONNRESET); 831 } 832 tp = intotcpcb(inp); 833 834 NET_EPOCH_ENTER(et); 835 socantsendmore(so); 836 tcp_usrclosed(tp); 837 if (!(inp->inp_flags & INP_DROPPED)) 838 error = tcp_output_nodrop(tp); 839 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 840 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 841 error = tcp_unlock_or_drop(tp, error); 842 NET_EPOCH_EXIT(et); 843 844 return (error); 845 } 846 847 /* 848 * After a receive, possibly send window update to peer. 849 */ 850 static int 851 tcp_usr_rcvd(struct socket *so, int flags) 852 { 853 struct epoch_tracker et; 854 struct inpcb *inp; 855 struct tcpcb *tp; 856 int outrv = 0, error = 0; 857 858 inp = sotoinpcb(so); 859 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 860 INP_WLOCK(inp); 861 if (inp->inp_flags & INP_DROPPED) { 862 INP_WUNLOCK(inp); 863 return (ECONNRESET); 864 } 865 tp = intotcpcb(inp); 866 867 NET_EPOCH_ENTER(et); 868 /* 869 * For passively-created TFO connections, don't attempt a window 870 * update while still in SYN_RECEIVED as this may trigger an early 871 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 872 * application response data, or failing that, when the DELACK timer 873 * expires. 874 */ 875 if (IS_FASTOPEN(tp->t_flags) && 876 (tp->t_state == TCPS_SYN_RECEIVED)) 877 goto out; 878 #ifdef TCP_OFFLOAD 879 if (tp->t_flags & TF_TOE) 880 tcp_offload_rcvd(tp); 881 else 882 #endif 883 outrv = tcp_output_nodrop(tp); 884 out: 885 tcp_bblog_pru(tp, PRU_RCVD, error); 886 TCP_PROBE2(debug__user, tp, PRU_RCVD); 887 (void) tcp_unlock_or_drop(tp, outrv); 888 NET_EPOCH_EXIT(et); 889 return (error); 890 } 891 892 /* 893 * Do a send by putting data in output queue and updating urgent 894 * marker if URG set. Possibly send more data. Unlike the other 895 * pru_*() routines, the mbuf chains are our responsibility. We 896 * must either enqueue them or free them. The other pru_* routines 897 * generally are caller-frees. 898 */ 899 static int 900 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 901 struct sockaddr *nam, struct mbuf *control, struct thread *td) 902 { 903 struct epoch_tracker et; 904 int error = 0; 905 struct inpcb *inp; 906 struct tcpcb *tp; 907 #ifdef INET 908 #ifdef INET6 909 struct sockaddr_in sin; 910 #endif 911 struct sockaddr_in *sinp; 912 #endif 913 #ifdef INET6 914 struct sockaddr_in6 *sin6; 915 int isipv6; 916 #endif 917 u_int8_t incflagsav; 918 u_char vflagsav; 919 bool restoreflags; 920 921 inp = sotoinpcb(so); 922 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 923 INP_WLOCK(inp); 924 if (inp->inp_flags & INP_DROPPED) { 925 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 926 m_freem(m); 927 INP_WUNLOCK(inp); 928 return (ECONNRESET); 929 } 930 tp = intotcpcb(inp); 931 932 vflagsav = inp->inp_vflag; 933 incflagsav = inp->inp_inc.inc_flags; 934 restoreflags = false; 935 936 NET_EPOCH_ENTER(et); 937 if (control != NULL) { 938 /* TCP doesn't do control messages (rights, creds, etc) */ 939 if (control->m_len > 0) { 940 m_freem(control); 941 error = EINVAL; 942 goto out; 943 } 944 m_freem(control); /* empty control, just free it */ 945 } 946 947 if ((flags & PRUS_OOB) != 0 && 948 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 949 goto out; 950 951 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 952 if (tp->t_state == TCPS_LISTEN) { 953 error = EINVAL; 954 goto out; 955 } 956 switch (nam->sa_family) { 957 #ifdef INET 958 case AF_INET: 959 sinp = (struct sockaddr_in *)nam; 960 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 961 error = EINVAL; 962 goto out; 963 } 964 if ((inp->inp_vflag & INP_IPV6) != 0) { 965 error = EAFNOSUPPORT; 966 goto out; 967 } 968 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 969 error = EAFNOSUPPORT; 970 goto out; 971 } 972 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 973 error = EACCES; 974 goto out; 975 } 976 if ((error = prison_remote_ip4(td->td_ucred, 977 &sinp->sin_addr))) 978 goto out; 979 #ifdef INET6 980 isipv6 = 0; 981 #endif 982 break; 983 #endif /* INET */ 984 #ifdef INET6 985 case AF_INET6: 986 sin6 = (struct sockaddr_in6 *)nam; 987 if (sin6->sin6_len != sizeof(*sin6)) { 988 error = EINVAL; 989 goto out; 990 } 991 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 992 error = EAFNOSUPPORT; 993 goto out; 994 } 995 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 996 error = EAFNOSUPPORT; 997 goto out; 998 } 999 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1000 #ifdef INET 1001 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1002 error = EINVAL; 1003 goto out; 1004 } 1005 if ((inp->inp_vflag & INP_IPV4) == 0) { 1006 error = EAFNOSUPPORT; 1007 goto out; 1008 } 1009 restoreflags = true; 1010 inp->inp_vflag &= ~INP_IPV6; 1011 sinp = &sin; 1012 in6_sin6_2_sin(sinp, sin6); 1013 if (IN_MULTICAST( 1014 ntohl(sinp->sin_addr.s_addr))) { 1015 error = EAFNOSUPPORT; 1016 goto out; 1017 } 1018 if ((error = prison_remote_ip4(td->td_ucred, 1019 &sinp->sin_addr))) 1020 goto out; 1021 isipv6 = 0; 1022 #else /* !INET */ 1023 error = EAFNOSUPPORT; 1024 goto out; 1025 #endif /* INET */ 1026 } else { 1027 if ((inp->inp_vflag & INP_IPV6) == 0) { 1028 error = EAFNOSUPPORT; 1029 goto out; 1030 } 1031 restoreflags = true; 1032 inp->inp_vflag &= ~INP_IPV4; 1033 inp->inp_inc.inc_flags |= INC_ISIPV6; 1034 if ((error = prison_remote_ip6(td->td_ucred, 1035 &sin6->sin6_addr))) 1036 goto out; 1037 isipv6 = 1; 1038 } 1039 break; 1040 #endif /* INET6 */ 1041 default: 1042 error = EAFNOSUPPORT; 1043 goto out; 1044 } 1045 } 1046 if (!(flags & PRUS_OOB)) { 1047 if (tp->t_acktime == 0) 1048 tp->t_acktime = ticks; 1049 sbappendstream(&so->so_snd, m, flags); 1050 m = NULL; 1051 if (nam && tp->t_state < TCPS_SYN_SENT) { 1052 KASSERT(tp->t_state == TCPS_CLOSED, 1053 ("%s: tp %p is listening", __func__, tp)); 1054 1055 /* 1056 * Do implied connect if not yet connected, 1057 * initialize window to default value, and 1058 * initialize maxseg using peer's cached MSS. 1059 */ 1060 #ifdef INET6 1061 if (isipv6) 1062 error = tcp6_connect(tp, sin6, td); 1063 #endif /* INET6 */ 1064 #if defined(INET6) && defined(INET) 1065 else 1066 #endif 1067 #ifdef INET 1068 error = tcp_connect(tp, sinp, td); 1069 #endif 1070 /* 1071 * The bind operation in tcp_connect succeeded. We 1072 * no longer want to restore the flags if later 1073 * operations fail. 1074 */ 1075 if (error == 0 || inp->inp_lport != 0) 1076 restoreflags = false; 1077 1078 if (error) { 1079 /* m is freed if PRUS_NOTREADY is unset. */ 1080 sbflush(&so->so_snd); 1081 goto out; 1082 } 1083 if (IS_FASTOPEN(tp->t_flags)) 1084 tcp_fastopen_connect(tp); 1085 else { 1086 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1087 tcp_mss(tp, -1); 1088 } 1089 } 1090 if (flags & PRUS_EOF) { 1091 /* 1092 * Close the send side of the connection after 1093 * the data is sent. 1094 */ 1095 socantsendmore(so); 1096 tcp_usrclosed(tp); 1097 } 1098 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1099 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1100 (tp->t_fbyte_out == 0) && 1101 (so->so_snd.sb_ccc > 0)) { 1102 tp->t_fbyte_out = ticks; 1103 if (tp->t_fbyte_out == 0) 1104 tp->t_fbyte_out = 1; 1105 if (tp->t_fbyte_out && tp->t_fbyte_in) 1106 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1107 } 1108 if (!(inp->inp_flags & INP_DROPPED) && 1109 !(flags & PRUS_NOTREADY)) { 1110 if (flags & PRUS_MORETOCOME) 1111 tp->t_flags |= TF_MORETOCOME; 1112 error = tcp_output_nodrop(tp); 1113 if (flags & PRUS_MORETOCOME) 1114 tp->t_flags &= ~TF_MORETOCOME; 1115 } 1116 } else { 1117 /* 1118 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1119 */ 1120 SOCKBUF_LOCK(&so->so_snd); 1121 if (sbspace(&so->so_snd) < -512) { 1122 SOCKBUF_UNLOCK(&so->so_snd); 1123 error = ENOBUFS; 1124 goto out; 1125 } 1126 /* 1127 * According to RFC961 (Assigned Protocols), 1128 * the urgent pointer points to the last octet 1129 * of urgent data. We continue, however, 1130 * to consider it to indicate the first octet 1131 * of data past the urgent section. 1132 * Otherwise, snd_up should be one lower. 1133 */ 1134 if (tp->t_acktime == 0) 1135 tp->t_acktime = ticks; 1136 sbappendstream_locked(&so->so_snd, m, flags); 1137 SOCKBUF_UNLOCK(&so->so_snd); 1138 m = NULL; 1139 if (nam && tp->t_state < TCPS_SYN_SENT) { 1140 /* 1141 * Do implied connect if not yet connected, 1142 * initialize window to default value, and 1143 * initialize maxseg using peer's cached MSS. 1144 */ 1145 1146 /* 1147 * Not going to contemplate SYN|URG 1148 */ 1149 if (IS_FASTOPEN(tp->t_flags)) 1150 tp->t_flags &= ~TF_FASTOPEN; 1151 #ifdef INET6 1152 if (isipv6) 1153 error = tcp6_connect(tp, sin6, td); 1154 #endif /* INET6 */ 1155 #if defined(INET6) && defined(INET) 1156 else 1157 #endif 1158 #ifdef INET 1159 error = tcp_connect(tp, sinp, td); 1160 #endif 1161 /* 1162 * The bind operation in tcp_connect succeeded. We 1163 * no longer want to restore the flags if later 1164 * operations fail. 1165 */ 1166 if (error == 0 || inp->inp_lport != 0) 1167 restoreflags = false; 1168 1169 if (error != 0) { 1170 /* m is freed if PRUS_NOTREADY is unset. */ 1171 sbflush(&so->so_snd); 1172 goto out; 1173 } 1174 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1175 tcp_mss(tp, -1); 1176 } 1177 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1178 if ((flags & PRUS_NOTREADY) == 0) { 1179 tp->t_flags |= TF_FORCEDATA; 1180 error = tcp_output_nodrop(tp); 1181 tp->t_flags &= ~TF_FORCEDATA; 1182 } 1183 } 1184 TCP_LOG_EVENT(tp, NULL, 1185 &inp->inp_socket->so_rcv, 1186 &inp->inp_socket->so_snd, 1187 TCP_LOG_USERSEND, error, 1188 0, NULL, false); 1189 1190 out: 1191 /* 1192 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1193 * responsible for freeing memory. 1194 */ 1195 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1196 m_freem(m); 1197 1198 /* 1199 * If the request was unsuccessful and we changed flags, 1200 * restore the original flags. 1201 */ 1202 if (error != 0 && restoreflags) { 1203 inp->inp_vflag = vflagsav; 1204 inp->inp_inc.inc_flags = incflagsav; 1205 } 1206 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1207 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1208 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1209 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1210 error = tcp_unlock_or_drop(tp, error); 1211 NET_EPOCH_EXIT(et); 1212 return (error); 1213 } 1214 1215 static int 1216 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1217 { 1218 struct epoch_tracker et; 1219 struct inpcb *inp; 1220 struct tcpcb *tp; 1221 int error; 1222 1223 inp = sotoinpcb(so); 1224 INP_WLOCK(inp); 1225 if (inp->inp_flags & INP_DROPPED) { 1226 INP_WUNLOCK(inp); 1227 mb_free_notready(m, count); 1228 return (ECONNRESET); 1229 } 1230 tp = intotcpcb(inp); 1231 1232 SOCKBUF_LOCK(&so->so_snd); 1233 error = sbready(&so->so_snd, m, count); 1234 SOCKBUF_UNLOCK(&so->so_snd); 1235 if (error) { 1236 INP_WUNLOCK(inp); 1237 return (error); 1238 } 1239 NET_EPOCH_ENTER(et); 1240 error = tcp_output_unlock(tp); 1241 NET_EPOCH_EXIT(et); 1242 1243 return (error); 1244 } 1245 1246 /* 1247 * Abort the TCP. Drop the connection abruptly. 1248 */ 1249 static void 1250 tcp_usr_abort(struct socket *so) 1251 { 1252 struct inpcb *inp; 1253 struct tcpcb *tp; 1254 struct epoch_tracker et; 1255 1256 inp = sotoinpcb(so); 1257 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1258 1259 NET_EPOCH_ENTER(et); 1260 INP_WLOCK(inp); 1261 KASSERT(inp->inp_socket != NULL, 1262 ("tcp_usr_abort: inp_socket == NULL")); 1263 1264 /* 1265 * If we still have full TCP state, and we're not dropped, drop. 1266 */ 1267 if (!(inp->inp_flags & INP_DROPPED)) { 1268 tp = intotcpcb(inp); 1269 tp = tcp_drop(tp, ECONNABORTED); 1270 if (tp == NULL) 1271 goto dropped; 1272 tcp_bblog_pru(tp, PRU_ABORT, 0); 1273 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1274 } 1275 if (!(inp->inp_flags & INP_DROPPED)) { 1276 soref(so); 1277 inp->inp_flags |= INP_SOCKREF; 1278 } 1279 INP_WUNLOCK(inp); 1280 dropped: 1281 NET_EPOCH_EXIT(et); 1282 } 1283 1284 /* 1285 * TCP socket is closed. Start friendly disconnect. 1286 */ 1287 static void 1288 tcp_usr_close(struct socket *so) 1289 { 1290 struct inpcb *inp; 1291 struct tcpcb *tp; 1292 struct epoch_tracker et; 1293 1294 inp = sotoinpcb(so); 1295 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1296 1297 NET_EPOCH_ENTER(et); 1298 INP_WLOCK(inp); 1299 KASSERT(inp->inp_socket != NULL, 1300 ("tcp_usr_close: inp_socket == NULL")); 1301 1302 /* 1303 * If we are still connected and we're not dropped, initiate 1304 * a disconnect. 1305 */ 1306 if (!(inp->inp_flags & INP_DROPPED)) { 1307 tp = intotcpcb(inp); 1308 if (tp->t_state != TCPS_TIME_WAIT) { 1309 tp->t_flags |= TF_CLOSED; 1310 tcp_disconnect(tp); 1311 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1312 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1313 } 1314 } 1315 if (!(inp->inp_flags & INP_DROPPED)) { 1316 soref(so); 1317 inp->inp_flags |= INP_SOCKREF; 1318 } 1319 INP_WUNLOCK(inp); 1320 NET_EPOCH_EXIT(et); 1321 } 1322 1323 static int 1324 tcp_pru_options_support(struct tcpcb *tp, int flags) 1325 { 1326 /* 1327 * If the specific TCP stack has a pru_options 1328 * specified then it does not always support 1329 * all the PRU_XX options and we must ask it. 1330 * If the function is not specified then all 1331 * of the PRU_XX options are supported. 1332 */ 1333 int ret = 0; 1334 1335 if (tp->t_fb->tfb_pru_options) { 1336 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1337 } 1338 return (ret); 1339 } 1340 1341 /* 1342 * Receive out-of-band data. 1343 */ 1344 static int 1345 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1346 { 1347 int error = 0; 1348 struct inpcb *inp; 1349 struct tcpcb *tp; 1350 1351 inp = sotoinpcb(so); 1352 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1353 INP_WLOCK(inp); 1354 if (inp->inp_flags & INP_DROPPED) { 1355 INP_WUNLOCK(inp); 1356 return (ECONNRESET); 1357 } 1358 tp = intotcpcb(inp); 1359 1360 error = tcp_pru_options_support(tp, PRUS_OOB); 1361 if (error) { 1362 goto out; 1363 } 1364 if ((so->so_oobmark == 0 && 1365 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1366 so->so_options & SO_OOBINLINE || 1367 tp->t_oobflags & TCPOOB_HADDATA) { 1368 error = EINVAL; 1369 goto out; 1370 } 1371 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1372 error = EWOULDBLOCK; 1373 goto out; 1374 } 1375 m->m_len = 1; 1376 *mtod(m, caddr_t) = tp->t_iobc; 1377 if ((flags & MSG_PEEK) == 0) 1378 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1379 1380 out: 1381 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1382 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1383 INP_WUNLOCK(inp); 1384 return (error); 1385 } 1386 1387 #ifdef INET 1388 struct protosw tcp_protosw = { 1389 .pr_type = SOCK_STREAM, 1390 .pr_protocol = IPPROTO_TCP, 1391 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1392 PR_CAPATTACH, 1393 .pr_ctloutput = tcp_ctloutput, 1394 .pr_abort = tcp_usr_abort, 1395 .pr_accept = tcp_usr_accept, 1396 .pr_attach = tcp_usr_attach, 1397 .pr_bind = tcp_usr_bind, 1398 .pr_connect = tcp_usr_connect, 1399 .pr_control = in_control, 1400 .pr_detach = tcp_usr_detach, 1401 .pr_disconnect = tcp_usr_disconnect, 1402 .pr_listen = tcp_usr_listen, 1403 .pr_peeraddr = in_getpeeraddr, 1404 .pr_rcvd = tcp_usr_rcvd, 1405 .pr_rcvoob = tcp_usr_rcvoob, 1406 .pr_send = tcp_usr_send, 1407 .pr_ready = tcp_usr_ready, 1408 .pr_shutdown = tcp_usr_shutdown, 1409 .pr_sockaddr = in_getsockaddr, 1410 .pr_sosetlabel = in_pcbsosetlabel, 1411 .pr_close = tcp_usr_close, 1412 }; 1413 #endif /* INET */ 1414 1415 #ifdef INET6 1416 struct protosw tcp6_protosw = { 1417 .pr_type = SOCK_STREAM, 1418 .pr_protocol = IPPROTO_TCP, 1419 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1420 PR_CAPATTACH, 1421 .pr_ctloutput = tcp_ctloutput, 1422 .pr_abort = tcp_usr_abort, 1423 .pr_accept = tcp6_usr_accept, 1424 .pr_attach = tcp_usr_attach, 1425 .pr_bind = tcp6_usr_bind, 1426 .pr_connect = tcp6_usr_connect, 1427 .pr_control = in6_control, 1428 .pr_detach = tcp_usr_detach, 1429 .pr_disconnect = tcp_usr_disconnect, 1430 .pr_listen = tcp6_usr_listen, 1431 .pr_peeraddr = in6_mapped_peeraddr, 1432 .pr_rcvd = tcp_usr_rcvd, 1433 .pr_rcvoob = tcp_usr_rcvoob, 1434 .pr_send = tcp_usr_send, 1435 .pr_ready = tcp_usr_ready, 1436 .pr_shutdown = tcp_usr_shutdown, 1437 .pr_sockaddr = in6_mapped_sockaddr, 1438 .pr_sosetlabel = in_pcbsosetlabel, 1439 .pr_close = tcp_usr_close, 1440 }; 1441 #endif /* INET6 */ 1442 1443 #ifdef INET 1444 /* 1445 * Common subroutine to open a TCP connection to remote host specified 1446 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1447 * and assign a local port number and install the inpcb into the hash. 1448 * Initialize connection parameters and enter SYN-SENT state. 1449 */ 1450 static int 1451 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1452 { 1453 struct inpcb *inp = tptoinpcb(tp); 1454 struct socket *so = tptosocket(tp); 1455 int error; 1456 1457 NET_EPOCH_ASSERT(); 1458 INP_WLOCK_ASSERT(inp); 1459 1460 if (__predict_false((so->so_state & 1461 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1462 SS_ISDISCONNECTED)) != 0)) 1463 return (EISCONN); 1464 1465 INP_HASH_WLOCK(&V_tcbinfo); 1466 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1467 INP_HASH_WUNLOCK(&V_tcbinfo); 1468 if (error != 0) 1469 return (error); 1470 1471 /* 1472 * Compute window scaling to request: 1473 * Scale to fit into sweet spot. See tcp_syncache.c. 1474 * XXX: This should move to tcp_output(). 1475 */ 1476 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1477 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1478 tp->request_r_scale++; 1479 1480 soisconnecting(so); 1481 TCPSTAT_INC(tcps_connattempt); 1482 tcp_state_change(tp, TCPS_SYN_SENT); 1483 tp->iss = tcp_new_isn(&inp->inp_inc); 1484 if (tp->t_flags & TF_REQ_TSTMP) 1485 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1486 tcp_sendseqinit(tp); 1487 1488 return (0); 1489 } 1490 #endif /* INET */ 1491 1492 #ifdef INET6 1493 static int 1494 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1495 { 1496 struct inpcb *inp = tptoinpcb(tp); 1497 struct socket *so = tptosocket(tp); 1498 int error; 1499 1500 NET_EPOCH_ASSERT(); 1501 INP_WLOCK_ASSERT(inp); 1502 1503 if (__predict_false((so->so_state & 1504 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1505 return (EISCONN); 1506 1507 INP_HASH_WLOCK(&V_tcbinfo); 1508 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1509 INP_HASH_WUNLOCK(&V_tcbinfo); 1510 if (error != 0) 1511 return (error); 1512 1513 /* Compute window scaling to request. */ 1514 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1515 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1516 tp->request_r_scale++; 1517 1518 soisconnecting(so); 1519 TCPSTAT_INC(tcps_connattempt); 1520 tcp_state_change(tp, TCPS_SYN_SENT); 1521 tp->iss = tcp_new_isn(&inp->inp_inc); 1522 if (tp->t_flags & TF_REQ_TSTMP) 1523 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1524 tcp_sendseqinit(tp); 1525 1526 return (0); 1527 } 1528 #endif /* INET6 */ 1529 1530 /* 1531 * Export TCP internal state information via a struct tcp_info, based on the 1532 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1533 * (TCP state machine, etc). We export all information using FreeBSD-native 1534 * constants -- for example, the numeric values for tcpi_state will differ 1535 * from Linux. 1536 */ 1537 void 1538 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1539 { 1540 1541 INP_LOCK_ASSERT(tptoinpcb(tp)); 1542 bzero(ti, sizeof(*ti)); 1543 1544 ti->tcpi_state = tp->t_state; 1545 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1546 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1547 if (tp->t_flags & TF_SACK_PERMIT) 1548 ti->tcpi_options |= TCPI_OPT_SACK; 1549 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1550 ti->tcpi_options |= TCPI_OPT_WSCALE; 1551 ti->tcpi_snd_wscale = tp->snd_scale; 1552 ti->tcpi_rcv_wscale = tp->rcv_scale; 1553 } 1554 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1555 case TF2_ECN_PERMIT: 1556 ti->tcpi_options |= TCPI_OPT_ECN; 1557 break; 1558 case TF2_ACE_PERMIT: 1559 /* FALLTHROUGH */ 1560 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1561 ti->tcpi_options |= TCPI_OPT_ACE; 1562 break; 1563 default: 1564 break; 1565 } 1566 if (IS_FASTOPEN(tp->t_flags)) 1567 ti->tcpi_options |= TCPI_OPT_TFO; 1568 1569 ti->tcpi_rto = tp->t_rxtcur * tick; 1570 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1571 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1572 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1573 1574 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1575 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1576 1577 /* 1578 * FreeBSD-specific extension fields for tcp_info. 1579 */ 1580 ti->tcpi_rcv_space = tp->rcv_wnd; 1581 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1582 ti->tcpi_snd_wnd = tp->snd_wnd; 1583 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1584 ti->tcpi_snd_nxt = tp->snd_nxt; 1585 ti->tcpi_snd_mss = tp->t_maxseg; 1586 ti->tcpi_rcv_mss = tp->t_maxseg; 1587 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1588 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1589 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1590 ti->tcpi_snd_una = tp->snd_una; 1591 ti->tcpi_snd_max = tp->snd_max; 1592 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1593 ti->tcpi_rcv_adv = tp->rcv_adv; 1594 ti->tcpi_dupacks = tp->t_dupacks; 1595 #ifdef TCP_OFFLOAD 1596 if (tp->t_flags & TF_TOE) { 1597 ti->tcpi_options |= TCPI_OPT_TOE; 1598 tcp_offload_tcp_info(tp, ti); 1599 } 1600 #endif 1601 /* 1602 * AccECN related counters. 1603 */ 1604 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1605 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1606 /* 1607 * Internal counter starts at 5 for AccECN 1608 * but 0 for RFC3168 ECN. 1609 */ 1610 ti->tcpi_delivered_ce = tp->t_scep - 5; 1611 else 1612 ti->tcpi_delivered_ce = tp->t_scep; 1613 ti->tcpi_received_ce = tp->t_rcep; 1614 } 1615 1616 /* 1617 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1618 * socket option arguments. When it re-acquires the lock after the copy, it 1619 * has to revalidate that the connection is still valid for the socket 1620 * option. 1621 */ 1622 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1623 INP_WLOCK(inp); \ 1624 if (inp->inp_flags & INP_DROPPED) { \ 1625 INP_WUNLOCK(inp); \ 1626 cleanup; \ 1627 return (ECONNRESET); \ 1628 } \ 1629 tp = intotcpcb(inp); \ 1630 } while(0) 1631 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1632 1633 int 1634 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1635 { 1636 struct socket *so = inp->inp_socket; 1637 struct tcpcb *tp = intotcpcb(inp); 1638 int error = 0; 1639 1640 MPASS(sopt->sopt_dir == SOPT_SET); 1641 INP_WLOCK_ASSERT(inp); 1642 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1643 ("inp_flags == %x", inp->inp_flags)); 1644 KASSERT(so != NULL, ("inp_socket == NULL")); 1645 1646 if (sopt->sopt_level != IPPROTO_TCP) { 1647 INP_WUNLOCK(inp); 1648 #ifdef INET6 1649 if (inp->inp_vflag & INP_IPV6PROTO) 1650 error = ip6_ctloutput(so, sopt); 1651 #endif 1652 #if defined(INET6) && defined(INET) 1653 else 1654 #endif 1655 #ifdef INET 1656 error = ip_ctloutput(so, sopt); 1657 #endif 1658 /* 1659 * When an IP-level socket option affects TCP, pass control 1660 * down to stack tfb_tcp_ctloutput, otherwise return what 1661 * IP level returned. 1662 */ 1663 switch (sopt->sopt_level) { 1664 #ifdef INET6 1665 case IPPROTO_IPV6: 1666 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1667 return (error); 1668 switch (sopt->sopt_name) { 1669 case IPV6_TCLASS: 1670 /* Notify tcp stacks that care (e.g. RACK). */ 1671 break; 1672 case IPV6_USE_MIN_MTU: 1673 /* Update t_maxseg accordingly. */ 1674 break; 1675 default: 1676 return (error); 1677 } 1678 break; 1679 #endif 1680 #ifdef INET 1681 case IPPROTO_IP: 1682 switch (sopt->sopt_name) { 1683 case IP_TOS: 1684 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1685 break; 1686 case IP_TTL: 1687 /* Notify tcp stacks that care (e.g. RACK). */ 1688 break; 1689 default: 1690 return (error); 1691 } 1692 break; 1693 #endif 1694 default: 1695 return (error); 1696 } 1697 INP_WLOCK(inp); 1698 if (inp->inp_flags & INP_DROPPED) { 1699 INP_WUNLOCK(inp); 1700 return (ECONNRESET); 1701 } 1702 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1703 /* 1704 * Protect the TCP option TCP_FUNCTION_BLK so 1705 * that a sub-function can *never* overwrite this. 1706 */ 1707 struct tcp_function_set fsn; 1708 struct tcp_function_block *blk; 1709 void *ptr = NULL; 1710 1711 INP_WUNLOCK(inp); 1712 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1713 if (error) 1714 return (error); 1715 1716 INP_WLOCK(inp); 1717 tp = intotcpcb(inp); 1718 1719 blk = find_and_ref_tcp_functions(&fsn); 1720 if (blk == NULL) { 1721 INP_WUNLOCK(inp); 1722 return (ENOENT); 1723 } 1724 if (tp->t_fb == blk) { 1725 /* You already have this */ 1726 refcount_release(&blk->tfb_refcnt); 1727 INP_WUNLOCK(inp); 1728 return (0); 1729 } 1730 if (tp->t_state != TCPS_CLOSED) { 1731 /* 1732 * The user has advanced the state 1733 * past the initial point, we may not 1734 * be able to switch. 1735 */ 1736 if (blk->tfb_tcp_handoff_ok != NULL) { 1737 /* 1738 * Does the stack provide a 1739 * query mechanism, if so it may 1740 * still be possible? 1741 */ 1742 error = (*blk->tfb_tcp_handoff_ok)(tp); 1743 } else 1744 error = EINVAL; 1745 if (error) { 1746 refcount_release(&blk->tfb_refcnt); 1747 INP_WUNLOCK(inp); 1748 return(error); 1749 } 1750 } 1751 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1752 refcount_release(&blk->tfb_refcnt); 1753 INP_WUNLOCK(inp); 1754 return (ENOENT); 1755 } 1756 /* 1757 * Ensure the new stack takes ownership with a 1758 * clean slate on peak rate threshold. 1759 */ 1760 #ifdef TCPHPTS 1761 /* Assure that we are not on any hpts */ 1762 tcp_hpts_remove(tp); 1763 #endif 1764 if (blk->tfb_tcp_fb_init) { 1765 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1766 if (error) { 1767 /* 1768 * Release the ref count the lookup 1769 * acquired. 1770 */ 1771 refcount_release(&blk->tfb_refcnt); 1772 /* 1773 * Now there is a chance that the 1774 * init() function mucked with some 1775 * things before it failed, such as 1776 * hpts or inp_flags2 or timer granularity. 1777 * It should not of, but lets give the old 1778 * stack a chance to reset to a known good state. 1779 */ 1780 if (tp->t_fb->tfb_switch_failed) { 1781 (*tp->t_fb->tfb_switch_failed)(tp); 1782 } 1783 goto err_out; 1784 } 1785 } 1786 if (tp->t_fb->tfb_tcp_fb_fini) { 1787 struct epoch_tracker et; 1788 /* 1789 * Tell the stack to cleanup with 0 i.e. 1790 * the tcb is not going away. 1791 */ 1792 NET_EPOCH_ENTER(et); 1793 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1794 NET_EPOCH_EXIT(et); 1795 } 1796 /* 1797 * Release the old refcnt, the 1798 * lookup acquired a ref on the 1799 * new one already. 1800 */ 1801 refcount_release(&tp->t_fb->tfb_refcnt); 1802 /* 1803 * Set in the new stack. 1804 */ 1805 tp->t_fb = blk; 1806 tp->t_fb_ptr = ptr; 1807 #ifdef TCP_OFFLOAD 1808 if (tp->t_flags & TF_TOE) { 1809 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1810 sopt->sopt_name); 1811 } 1812 #endif 1813 err_out: 1814 INP_WUNLOCK(inp); 1815 return (error); 1816 1817 } 1818 1819 /* Pass in the INP locked, callee must unlock it. */ 1820 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1821 } 1822 1823 static int 1824 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1825 { 1826 struct socket *so = inp->inp_socket; 1827 struct tcpcb *tp = intotcpcb(inp); 1828 int error = 0; 1829 1830 MPASS(sopt->sopt_dir == SOPT_GET); 1831 INP_WLOCK_ASSERT(inp); 1832 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1833 ("inp_flags == %x", inp->inp_flags)); 1834 KASSERT(so != NULL, ("inp_socket == NULL")); 1835 1836 if (sopt->sopt_level != IPPROTO_TCP) { 1837 INP_WUNLOCK(inp); 1838 #ifdef INET6 1839 if (inp->inp_vflag & INP_IPV6PROTO) 1840 error = ip6_ctloutput(so, sopt); 1841 #endif /* INET6 */ 1842 #if defined(INET6) && defined(INET) 1843 else 1844 #endif 1845 #ifdef INET 1846 error = ip_ctloutput(so, sopt); 1847 #endif 1848 return (error); 1849 } 1850 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1851 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1852 struct tcp_function_set fsn; 1853 1854 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1855 memset(&fsn, 0, sizeof(fsn)); 1856 find_tcp_function_alias(tp->t_fb, &fsn); 1857 } else { 1858 strncpy(fsn.function_set_name, 1859 tp->t_fb->tfb_tcp_block_name, 1860 TCP_FUNCTION_NAME_LEN_MAX); 1861 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1862 } 1863 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1864 INP_WUNLOCK(inp); 1865 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1866 return (error); 1867 } 1868 1869 /* Pass in the INP locked, callee must unlock it. */ 1870 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1871 } 1872 1873 int 1874 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1875 { 1876 struct inpcb *inp; 1877 1878 inp = sotoinpcb(so); 1879 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1880 1881 INP_WLOCK(inp); 1882 if (inp->inp_flags & INP_DROPPED) { 1883 INP_WUNLOCK(inp); 1884 return (ECONNRESET); 1885 } 1886 if (sopt->sopt_dir == SOPT_SET) 1887 return (tcp_ctloutput_set(inp, sopt)); 1888 else if (sopt->sopt_dir == SOPT_GET) 1889 return (tcp_ctloutput_get(inp, sopt)); 1890 else 1891 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1892 } 1893 1894 /* 1895 * If this assert becomes untrue, we need to change the size of the buf 1896 * variable in tcp_default_ctloutput(). 1897 */ 1898 #ifdef CTASSERT 1899 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1900 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1901 #endif 1902 1903 #ifdef KERN_TLS 1904 static int 1905 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1906 { 1907 struct tls_enable_v0 tls_v0; 1908 int error; 1909 1910 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1911 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1912 sizeof(tls_v0)); 1913 if (error) 1914 return (error); 1915 memset(tls, 0, sizeof(*tls)); 1916 tls->cipher_key = tls_v0.cipher_key; 1917 tls->iv = tls_v0.iv; 1918 tls->auth_key = tls_v0.auth_key; 1919 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1920 tls->cipher_key_len = tls_v0.cipher_key_len; 1921 tls->iv_len = tls_v0.iv_len; 1922 tls->auth_algorithm = tls_v0.auth_algorithm; 1923 tls->auth_key_len = tls_v0.auth_key_len; 1924 tls->flags = tls_v0.flags; 1925 tls->tls_vmajor = tls_v0.tls_vmajor; 1926 tls->tls_vminor = tls_v0.tls_vminor; 1927 return (0); 1928 } 1929 1930 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1931 } 1932 #endif 1933 1934 extern struct cc_algo newreno_cc_algo; 1935 1936 static int 1937 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1938 { 1939 struct cc_algo *algo; 1940 void *ptr = NULL; 1941 struct tcpcb *tp; 1942 struct cc_var cc_mem; 1943 char buf[TCP_CA_NAME_MAX]; 1944 size_t mem_sz; 1945 int error; 1946 1947 INP_WUNLOCK(inp); 1948 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1949 if (error) 1950 return(error); 1951 buf[sopt->sopt_valsize] = '\0'; 1952 CC_LIST_RLOCK(); 1953 STAILQ_FOREACH(algo, &cc_list, entries) { 1954 if (strncmp(buf, algo->name, 1955 TCP_CA_NAME_MAX) == 0) { 1956 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1957 /* We can't "see" modules being unloaded */ 1958 continue; 1959 } 1960 break; 1961 } 1962 } 1963 if (algo == NULL) { 1964 CC_LIST_RUNLOCK(); 1965 return(ESRCH); 1966 } 1967 /* 1968 * With a reference the algorithm cannot be removed 1969 * so we hold a reference through the change process. 1970 */ 1971 cc_refer(algo); 1972 CC_LIST_RUNLOCK(); 1973 if (algo->cb_init != NULL) { 1974 /* We can now pre-get the memory for the CC */ 1975 mem_sz = (*algo->cc_data_sz)(); 1976 if (mem_sz == 0) { 1977 goto no_mem_needed; 1978 } 1979 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1980 } else { 1981 no_mem_needed: 1982 mem_sz = 0; 1983 ptr = NULL; 1984 } 1985 /* 1986 * Make sure its all clean and zero and also get 1987 * back the inplock. 1988 */ 1989 memset(&cc_mem, 0, sizeof(cc_mem)); 1990 INP_WLOCK(inp); 1991 if (inp->inp_flags & INP_DROPPED) { 1992 INP_WUNLOCK(inp); 1993 if (ptr) 1994 free(ptr, M_CC_MEM); 1995 /* Release our temp reference */ 1996 CC_LIST_RLOCK(); 1997 cc_release(algo); 1998 CC_LIST_RUNLOCK(); 1999 return (ECONNRESET); 2000 } 2001 tp = intotcpcb(inp); 2002 if (ptr != NULL) 2003 memset(ptr, 0, mem_sz); 2004 cc_mem.ccvc.tcp = tp; 2005 /* 2006 * We once again hold a write lock over the tcb so it's 2007 * safe to do these things without ordering concerns. 2008 * Note here we init into stack memory. 2009 */ 2010 if (algo->cb_init != NULL) 2011 error = algo->cb_init(&cc_mem, ptr); 2012 else 2013 error = 0; 2014 /* 2015 * The CC algorithms, when given their memory 2016 * should not fail we could in theory have a 2017 * KASSERT here. 2018 */ 2019 if (error == 0) { 2020 /* 2021 * Touchdown, lets go ahead and move the 2022 * connection to the new CC module by 2023 * copying in the cc_mem after we call 2024 * the old ones cleanup (if any). 2025 */ 2026 if (CC_ALGO(tp)->cb_destroy != NULL) 2027 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2028 /* Detach the old CC from the tcpcb */ 2029 cc_detach(tp); 2030 /* Copy in our temp memory that was inited */ 2031 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2032 /* Now attach the new, which takes a reference */ 2033 cc_attach(tp, algo); 2034 /* Ok now are we where we have gotten past any conn_init? */ 2035 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2036 /* Yep run the connection init for the new CC */ 2037 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2038 } 2039 } else if (ptr) 2040 free(ptr, M_CC_MEM); 2041 INP_WUNLOCK(inp); 2042 /* Now lets release our temp reference */ 2043 CC_LIST_RLOCK(); 2044 cc_release(algo); 2045 CC_LIST_RUNLOCK(); 2046 return (error); 2047 } 2048 2049 int 2050 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2051 { 2052 struct inpcb *inp = tptoinpcb(tp); 2053 int error, opt, optval; 2054 u_int ui; 2055 struct tcp_info ti; 2056 #ifdef KERN_TLS 2057 struct tls_enable tls; 2058 struct socket *so = inp->inp_socket; 2059 #endif 2060 char *pbuf, buf[TCP_LOG_ID_LEN]; 2061 #ifdef STATS 2062 struct statsblob *sbp; 2063 #endif 2064 size_t len; 2065 2066 INP_WLOCK_ASSERT(inp); 2067 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2068 ("inp_flags == %x", inp->inp_flags)); 2069 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2070 2071 switch (sopt->sopt_level) { 2072 #ifdef INET6 2073 case IPPROTO_IPV6: 2074 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2075 switch (sopt->sopt_name) { 2076 case IPV6_USE_MIN_MTU: 2077 tcp6_use_min_mtu(tp); 2078 /* FALLTHROUGH */ 2079 } 2080 INP_WUNLOCK(inp); 2081 return (0); 2082 #endif 2083 #ifdef INET 2084 case IPPROTO_IP: 2085 INP_WUNLOCK(inp); 2086 return (0); 2087 #endif 2088 } 2089 2090 /* 2091 * For TCP_CCALGOOPT forward the control to CC module, for both 2092 * SOPT_SET and SOPT_GET. 2093 */ 2094 switch (sopt->sopt_name) { 2095 case TCP_CCALGOOPT: 2096 INP_WUNLOCK(inp); 2097 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2098 return (EINVAL); 2099 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2100 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2101 sopt->sopt_valsize); 2102 if (error) { 2103 free(pbuf, M_TEMP); 2104 return (error); 2105 } 2106 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2107 if (CC_ALGO(tp)->ctl_output != NULL) 2108 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2109 else 2110 error = ENOENT; 2111 INP_WUNLOCK(inp); 2112 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2113 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2114 free(pbuf, M_TEMP); 2115 return (error); 2116 } 2117 2118 switch (sopt->sopt_dir) { 2119 case SOPT_SET: 2120 switch (sopt->sopt_name) { 2121 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2122 case TCP_MD5SIG: 2123 INP_WUNLOCK(inp); 2124 if (!TCPMD5_ENABLED()) 2125 return (ENOPROTOOPT); 2126 error = TCPMD5_PCBCTL(inp, sopt); 2127 if (error) 2128 return (error); 2129 INP_WLOCK_RECHECK(inp); 2130 goto unlock_and_done; 2131 #endif /* IPSEC */ 2132 2133 case TCP_NODELAY: 2134 case TCP_NOOPT: 2135 case TCP_LRD: 2136 INP_WUNLOCK(inp); 2137 error = sooptcopyin(sopt, &optval, sizeof optval, 2138 sizeof optval); 2139 if (error) 2140 return (error); 2141 2142 INP_WLOCK_RECHECK(inp); 2143 switch (sopt->sopt_name) { 2144 case TCP_NODELAY: 2145 opt = TF_NODELAY; 2146 break; 2147 case TCP_NOOPT: 2148 opt = TF_NOOPT; 2149 break; 2150 case TCP_LRD: 2151 opt = TF_LRD; 2152 break; 2153 default: 2154 opt = 0; /* dead code to fool gcc */ 2155 break; 2156 } 2157 2158 if (optval) 2159 tp->t_flags |= opt; 2160 else 2161 tp->t_flags &= ~opt; 2162 unlock_and_done: 2163 #ifdef TCP_OFFLOAD 2164 if (tp->t_flags & TF_TOE) { 2165 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2166 sopt->sopt_name); 2167 } 2168 #endif 2169 INP_WUNLOCK(inp); 2170 break; 2171 2172 case TCP_NOPUSH: 2173 INP_WUNLOCK(inp); 2174 error = sooptcopyin(sopt, &optval, sizeof optval, 2175 sizeof optval); 2176 if (error) 2177 return (error); 2178 2179 INP_WLOCK_RECHECK(inp); 2180 if (optval) 2181 tp->t_flags |= TF_NOPUSH; 2182 else if (tp->t_flags & TF_NOPUSH) { 2183 tp->t_flags &= ~TF_NOPUSH; 2184 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2185 struct epoch_tracker et; 2186 2187 NET_EPOCH_ENTER(et); 2188 error = tcp_output_nodrop(tp); 2189 NET_EPOCH_EXIT(et); 2190 } 2191 } 2192 goto unlock_and_done; 2193 2194 case TCP_REMOTE_UDP_ENCAPS_PORT: 2195 INP_WUNLOCK(inp); 2196 error = sooptcopyin(sopt, &optval, sizeof optval, 2197 sizeof optval); 2198 if (error) 2199 return (error); 2200 if ((optval < TCP_TUNNELING_PORT_MIN) || 2201 (optval > TCP_TUNNELING_PORT_MAX)) { 2202 /* Its got to be in range */ 2203 return (EINVAL); 2204 } 2205 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2206 /* You have to have enabled a UDP tunneling port first */ 2207 return (EINVAL); 2208 } 2209 INP_WLOCK_RECHECK(inp); 2210 if (tp->t_state != TCPS_CLOSED) { 2211 /* You can't change after you are connected */ 2212 error = EINVAL; 2213 } else { 2214 /* Ok we are all good set the port */ 2215 tp->t_port = htons(optval); 2216 } 2217 goto unlock_and_done; 2218 2219 case TCP_MAXSEG: 2220 INP_WUNLOCK(inp); 2221 error = sooptcopyin(sopt, &optval, sizeof optval, 2222 sizeof optval); 2223 if (error) 2224 return (error); 2225 2226 INP_WLOCK_RECHECK(inp); 2227 if (optval > 0 && optval <= tp->t_maxseg && 2228 optval + 40 >= V_tcp_minmss) 2229 tp->t_maxseg = optval; 2230 else 2231 error = EINVAL; 2232 goto unlock_and_done; 2233 2234 case TCP_INFO: 2235 INP_WUNLOCK(inp); 2236 error = EINVAL; 2237 break; 2238 2239 case TCP_STATS: 2240 INP_WUNLOCK(inp); 2241 #ifdef STATS 2242 error = sooptcopyin(sopt, &optval, sizeof optval, 2243 sizeof optval); 2244 if (error) 2245 return (error); 2246 2247 if (optval > 0) 2248 sbp = stats_blob_alloc( 2249 V_tcp_perconn_stats_dflt_tpl, 0); 2250 else 2251 sbp = NULL; 2252 2253 INP_WLOCK_RECHECK(inp); 2254 if ((tp->t_stats != NULL && sbp == NULL) || 2255 (tp->t_stats == NULL && sbp != NULL)) { 2256 struct statsblob *t = tp->t_stats; 2257 tp->t_stats = sbp; 2258 sbp = t; 2259 } 2260 INP_WUNLOCK(inp); 2261 2262 stats_blob_destroy(sbp); 2263 #else 2264 return (EOPNOTSUPP); 2265 #endif /* !STATS */ 2266 break; 2267 2268 case TCP_CONGESTION: 2269 error = tcp_set_cc_mod(inp, sopt); 2270 break; 2271 2272 case TCP_REUSPORT_LB_NUMA: 2273 INP_WUNLOCK(inp); 2274 error = sooptcopyin(sopt, &optval, sizeof(optval), 2275 sizeof(optval)); 2276 INP_WLOCK_RECHECK(inp); 2277 if (!error) 2278 error = in_pcblbgroup_numa(inp, optval); 2279 INP_WUNLOCK(inp); 2280 break; 2281 2282 #ifdef KERN_TLS 2283 case TCP_TXTLS_ENABLE: 2284 INP_WUNLOCK(inp); 2285 error = copyin_tls_enable(sopt, &tls); 2286 if (error) 2287 break; 2288 error = ktls_enable_tx(so, &tls); 2289 break; 2290 case TCP_TXTLS_MODE: 2291 INP_WUNLOCK(inp); 2292 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2293 if (error) 2294 return (error); 2295 2296 INP_WLOCK_RECHECK(inp); 2297 error = ktls_set_tx_mode(so, ui); 2298 INP_WUNLOCK(inp); 2299 break; 2300 case TCP_RXTLS_ENABLE: 2301 INP_WUNLOCK(inp); 2302 error = sooptcopyin(sopt, &tls, sizeof(tls), 2303 sizeof(tls)); 2304 if (error) 2305 break; 2306 error = ktls_enable_rx(so, &tls); 2307 break; 2308 #endif 2309 case TCP_MAXUNACKTIME: 2310 case TCP_KEEPIDLE: 2311 case TCP_KEEPINTVL: 2312 case TCP_KEEPINIT: 2313 INP_WUNLOCK(inp); 2314 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2315 if (error) 2316 return (error); 2317 2318 if (ui > (UINT_MAX / hz)) { 2319 error = EINVAL; 2320 break; 2321 } 2322 ui *= hz; 2323 2324 INP_WLOCK_RECHECK(inp); 2325 switch (sopt->sopt_name) { 2326 case TCP_MAXUNACKTIME: 2327 tp->t_maxunacktime = ui; 2328 break; 2329 2330 case TCP_KEEPIDLE: 2331 tp->t_keepidle = ui; 2332 /* 2333 * XXX: better check current remaining 2334 * timeout and "merge" it with new value. 2335 */ 2336 if ((tp->t_state > TCPS_LISTEN) && 2337 (tp->t_state <= TCPS_CLOSING)) 2338 tcp_timer_activate(tp, TT_KEEP, 2339 TP_KEEPIDLE(tp)); 2340 break; 2341 case TCP_KEEPINTVL: 2342 tp->t_keepintvl = ui; 2343 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2344 (TP_MAXIDLE(tp) > 0)) 2345 tcp_timer_activate(tp, TT_2MSL, 2346 TP_MAXIDLE(tp)); 2347 break; 2348 case TCP_KEEPINIT: 2349 tp->t_keepinit = ui; 2350 if (tp->t_state == TCPS_SYN_RECEIVED || 2351 tp->t_state == TCPS_SYN_SENT) 2352 tcp_timer_activate(tp, TT_KEEP, 2353 TP_KEEPINIT(tp)); 2354 break; 2355 } 2356 goto unlock_and_done; 2357 2358 case TCP_KEEPCNT: 2359 INP_WUNLOCK(inp); 2360 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2361 if (error) 2362 return (error); 2363 2364 INP_WLOCK_RECHECK(inp); 2365 tp->t_keepcnt = ui; 2366 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2367 (TP_MAXIDLE(tp) > 0)) 2368 tcp_timer_activate(tp, TT_2MSL, 2369 TP_MAXIDLE(tp)); 2370 goto unlock_and_done; 2371 2372 #ifdef TCPPCAP 2373 case TCP_PCAP_OUT: 2374 case TCP_PCAP_IN: 2375 INP_WUNLOCK(inp); 2376 error = sooptcopyin(sopt, &optval, sizeof optval, 2377 sizeof optval); 2378 if (error) 2379 return (error); 2380 2381 INP_WLOCK_RECHECK(inp); 2382 if (optval >= 0) 2383 tcp_pcap_set_sock_max( 2384 (sopt->sopt_name == TCP_PCAP_OUT) ? 2385 &(tp->t_outpkts) : &(tp->t_inpkts), 2386 optval); 2387 else 2388 error = EINVAL; 2389 goto unlock_and_done; 2390 #endif 2391 2392 case TCP_FASTOPEN: { 2393 struct tcp_fastopen tfo_optval; 2394 2395 INP_WUNLOCK(inp); 2396 if (!V_tcp_fastopen_client_enable && 2397 !V_tcp_fastopen_server_enable) 2398 return (EPERM); 2399 2400 error = sooptcopyin(sopt, &tfo_optval, 2401 sizeof(tfo_optval), sizeof(int)); 2402 if (error) 2403 return (error); 2404 2405 INP_WLOCK_RECHECK(inp); 2406 if ((tp->t_state != TCPS_CLOSED) && 2407 (tp->t_state != TCPS_LISTEN)) { 2408 error = EINVAL; 2409 goto unlock_and_done; 2410 } 2411 if (tfo_optval.enable) { 2412 if (tp->t_state == TCPS_LISTEN) { 2413 if (!V_tcp_fastopen_server_enable) { 2414 error = EPERM; 2415 goto unlock_and_done; 2416 } 2417 2418 if (tp->t_tfo_pending == NULL) 2419 tp->t_tfo_pending = 2420 tcp_fastopen_alloc_counter(); 2421 } else { 2422 /* 2423 * If a pre-shared key was provided, 2424 * stash it in the client cookie 2425 * field of the tcpcb for use during 2426 * connect. 2427 */ 2428 if (sopt->sopt_valsize == 2429 sizeof(tfo_optval)) { 2430 memcpy(tp->t_tfo_cookie.client, 2431 tfo_optval.psk, 2432 TCP_FASTOPEN_PSK_LEN); 2433 tp->t_tfo_client_cookie_len = 2434 TCP_FASTOPEN_PSK_LEN; 2435 } 2436 } 2437 tp->t_flags |= TF_FASTOPEN; 2438 } else 2439 tp->t_flags &= ~TF_FASTOPEN; 2440 goto unlock_and_done; 2441 } 2442 2443 #ifdef TCP_BLACKBOX 2444 case TCP_LOG: 2445 INP_WUNLOCK(inp); 2446 error = sooptcopyin(sopt, &optval, sizeof optval, 2447 sizeof optval); 2448 if (error) 2449 return (error); 2450 2451 INP_WLOCK_RECHECK(inp); 2452 error = tcp_log_state_change(tp, optval); 2453 goto unlock_and_done; 2454 2455 case TCP_LOGBUF: 2456 INP_WUNLOCK(inp); 2457 error = EINVAL; 2458 break; 2459 2460 case TCP_LOGID: 2461 INP_WUNLOCK(inp); 2462 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2463 if (error) 2464 break; 2465 buf[sopt->sopt_valsize] = '\0'; 2466 INP_WLOCK_RECHECK(inp); 2467 error = tcp_log_set_id(tp, buf); 2468 /* tcp_log_set_id() unlocks the INP. */ 2469 break; 2470 2471 case TCP_LOGDUMP: 2472 case TCP_LOGDUMPID: 2473 INP_WUNLOCK(inp); 2474 error = 2475 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2476 if (error) 2477 break; 2478 buf[sopt->sopt_valsize] = '\0'; 2479 INP_WLOCK_RECHECK(inp); 2480 if (sopt->sopt_name == TCP_LOGDUMP) { 2481 error = tcp_log_dump_tp_logbuf(tp, buf, 2482 M_WAITOK, true); 2483 INP_WUNLOCK(inp); 2484 } else { 2485 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2486 /* 2487 * tcp_log_dump_tp_bucket_logbufs() drops the 2488 * INP lock. 2489 */ 2490 } 2491 break; 2492 #endif 2493 2494 default: 2495 INP_WUNLOCK(inp); 2496 error = ENOPROTOOPT; 2497 break; 2498 } 2499 break; 2500 2501 case SOPT_GET: 2502 tp = intotcpcb(inp); 2503 switch (sopt->sopt_name) { 2504 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2505 case TCP_MD5SIG: 2506 INP_WUNLOCK(inp); 2507 if (!TCPMD5_ENABLED()) 2508 return (ENOPROTOOPT); 2509 error = TCPMD5_PCBCTL(inp, sopt); 2510 break; 2511 #endif 2512 2513 case TCP_NODELAY: 2514 optval = tp->t_flags & TF_NODELAY; 2515 INP_WUNLOCK(inp); 2516 error = sooptcopyout(sopt, &optval, sizeof optval); 2517 break; 2518 case TCP_MAXSEG: 2519 optval = tp->t_maxseg; 2520 INP_WUNLOCK(inp); 2521 error = sooptcopyout(sopt, &optval, sizeof optval); 2522 break; 2523 case TCP_REMOTE_UDP_ENCAPS_PORT: 2524 optval = ntohs(tp->t_port); 2525 INP_WUNLOCK(inp); 2526 error = sooptcopyout(sopt, &optval, sizeof optval); 2527 break; 2528 case TCP_NOOPT: 2529 optval = tp->t_flags & TF_NOOPT; 2530 INP_WUNLOCK(inp); 2531 error = sooptcopyout(sopt, &optval, sizeof optval); 2532 break; 2533 case TCP_NOPUSH: 2534 optval = tp->t_flags & TF_NOPUSH; 2535 INP_WUNLOCK(inp); 2536 error = sooptcopyout(sopt, &optval, sizeof optval); 2537 break; 2538 case TCP_INFO: 2539 tcp_fill_info(tp, &ti); 2540 INP_WUNLOCK(inp); 2541 error = sooptcopyout(sopt, &ti, sizeof ti); 2542 break; 2543 case TCP_STATS: 2544 { 2545 #ifdef STATS 2546 int nheld; 2547 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2548 2549 error = 0; 2550 socklen_t outsbsz = sopt->sopt_valsize; 2551 if (tp->t_stats == NULL) 2552 error = ENOENT; 2553 else if (outsbsz >= tp->t_stats->cursz) 2554 outsbsz = tp->t_stats->cursz; 2555 else if (outsbsz >= sizeof(struct statsblob)) 2556 outsbsz = sizeof(struct statsblob); 2557 else 2558 error = EINVAL; 2559 INP_WUNLOCK(inp); 2560 if (error) 2561 break; 2562 2563 sbp = sopt->sopt_val; 2564 nheld = atop(round_page(((vm_offset_t)sbp) + 2565 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2566 vm_page_t ma[nheld]; 2567 if (vm_fault_quick_hold_pages( 2568 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2569 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2570 nheld) < 0) { 2571 error = EFAULT; 2572 break; 2573 } 2574 2575 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2576 SIZEOF_MEMBER(struct statsblob, flags)))) 2577 goto unhold; 2578 2579 INP_WLOCK_RECHECK(inp); 2580 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2581 sbflags | SB_CLONE_USRDSTNOFAULT); 2582 INP_WUNLOCK(inp); 2583 sopt->sopt_valsize = outsbsz; 2584 unhold: 2585 vm_page_unhold_pages(ma, nheld); 2586 #else 2587 INP_WUNLOCK(inp); 2588 error = EOPNOTSUPP; 2589 #endif /* !STATS */ 2590 break; 2591 } 2592 case TCP_CONGESTION: 2593 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2594 INP_WUNLOCK(inp); 2595 error = sooptcopyout(sopt, buf, len + 1); 2596 break; 2597 case TCP_MAXUNACKTIME: 2598 case TCP_KEEPIDLE: 2599 case TCP_KEEPINTVL: 2600 case TCP_KEEPINIT: 2601 case TCP_KEEPCNT: 2602 switch (sopt->sopt_name) { 2603 case TCP_MAXUNACKTIME: 2604 ui = TP_MAXUNACKTIME(tp) / hz; 2605 break; 2606 case TCP_KEEPIDLE: 2607 ui = TP_KEEPIDLE(tp) / hz; 2608 break; 2609 case TCP_KEEPINTVL: 2610 ui = TP_KEEPINTVL(tp) / hz; 2611 break; 2612 case TCP_KEEPINIT: 2613 ui = TP_KEEPINIT(tp) / hz; 2614 break; 2615 case TCP_KEEPCNT: 2616 ui = TP_KEEPCNT(tp); 2617 break; 2618 } 2619 INP_WUNLOCK(inp); 2620 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2621 break; 2622 #ifdef TCPPCAP 2623 case TCP_PCAP_OUT: 2624 case TCP_PCAP_IN: 2625 optval = tcp_pcap_get_sock_max( 2626 (sopt->sopt_name == TCP_PCAP_OUT) ? 2627 &(tp->t_outpkts) : &(tp->t_inpkts)); 2628 INP_WUNLOCK(inp); 2629 error = sooptcopyout(sopt, &optval, sizeof optval); 2630 break; 2631 #endif 2632 case TCP_FASTOPEN: 2633 optval = tp->t_flags & TF_FASTOPEN; 2634 INP_WUNLOCK(inp); 2635 error = sooptcopyout(sopt, &optval, sizeof optval); 2636 break; 2637 #ifdef TCP_BLACKBOX 2638 case TCP_LOG: 2639 optval = tcp_get_bblog_state(tp); 2640 INP_WUNLOCK(inp); 2641 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2642 break; 2643 case TCP_LOGBUF: 2644 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2645 error = tcp_log_getlogbuf(sopt, tp); 2646 break; 2647 case TCP_LOGID: 2648 len = tcp_log_get_id(tp, buf); 2649 INP_WUNLOCK(inp); 2650 error = sooptcopyout(sopt, buf, len + 1); 2651 break; 2652 case TCP_LOGDUMP: 2653 case TCP_LOGDUMPID: 2654 INP_WUNLOCK(inp); 2655 error = EINVAL; 2656 break; 2657 #endif 2658 #ifdef KERN_TLS 2659 case TCP_TXTLS_MODE: 2660 error = ktls_get_tx_mode(so, &optval); 2661 INP_WUNLOCK(inp); 2662 if (error == 0) 2663 error = sooptcopyout(sopt, &optval, 2664 sizeof(optval)); 2665 break; 2666 case TCP_RXTLS_MODE: 2667 error = ktls_get_rx_mode(so, &optval); 2668 INP_WUNLOCK(inp); 2669 if (error == 0) 2670 error = sooptcopyout(sopt, &optval, 2671 sizeof(optval)); 2672 break; 2673 #endif 2674 case TCP_LRD: 2675 optval = tp->t_flags & TF_LRD; 2676 INP_WUNLOCK(inp); 2677 error = sooptcopyout(sopt, &optval, sizeof optval); 2678 break; 2679 default: 2680 INP_WUNLOCK(inp); 2681 error = ENOPROTOOPT; 2682 break; 2683 } 2684 break; 2685 } 2686 return (error); 2687 } 2688 #undef INP_WLOCK_RECHECK 2689 #undef INP_WLOCK_RECHECK_CLEANUP 2690 2691 /* 2692 * Initiate (or continue) disconnect. 2693 * If embryonic state, just send reset (once). 2694 * If in ``let data drain'' option and linger null, just drop. 2695 * Otherwise (hard), mark socket disconnecting and drop 2696 * current input data; switch states based on user close, and 2697 * send segment to peer (with FIN). 2698 */ 2699 static void 2700 tcp_disconnect(struct tcpcb *tp) 2701 { 2702 struct inpcb *inp = tptoinpcb(tp); 2703 struct socket *so = tptosocket(tp); 2704 2705 NET_EPOCH_ASSERT(); 2706 INP_WLOCK_ASSERT(inp); 2707 2708 /* 2709 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2710 * socket is still open. 2711 */ 2712 if (tp->t_state < TCPS_ESTABLISHED && 2713 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2714 tp = tcp_close(tp); 2715 KASSERT(tp != NULL, 2716 ("tcp_disconnect: tcp_close() returned NULL")); 2717 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2718 tp = tcp_drop(tp, 0); 2719 KASSERT(tp != NULL, 2720 ("tcp_disconnect: tcp_drop() returned NULL")); 2721 } else { 2722 soisdisconnecting(so); 2723 sbflush(&so->so_rcv); 2724 tcp_usrclosed(tp); 2725 if (!(inp->inp_flags & INP_DROPPED)) 2726 /* Ignore stack's drop request, we already at it. */ 2727 (void)tcp_output_nodrop(tp); 2728 } 2729 } 2730 2731 /* 2732 * User issued close, and wish to trail through shutdown states: 2733 * if never received SYN, just forget it. If got a SYN from peer, 2734 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2735 * If already got a FIN from peer, then almost done; go to LAST_ACK 2736 * state. In all other cases, have already sent FIN to peer (e.g. 2737 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2738 * for peer to send FIN or not respond to keep-alives, etc. 2739 * We can let the user exit from the close as soon as the FIN is acked. 2740 */ 2741 static void 2742 tcp_usrclosed(struct tcpcb *tp) 2743 { 2744 2745 NET_EPOCH_ASSERT(); 2746 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2747 2748 switch (tp->t_state) { 2749 case TCPS_LISTEN: 2750 #ifdef TCP_OFFLOAD 2751 tcp_offload_listen_stop(tp); 2752 #endif 2753 tcp_state_change(tp, TCPS_CLOSED); 2754 /* FALLTHROUGH */ 2755 case TCPS_CLOSED: 2756 tp = tcp_close(tp); 2757 /* 2758 * tcp_close() should never return NULL here as the socket is 2759 * still open. 2760 */ 2761 KASSERT(tp != NULL, 2762 ("tcp_usrclosed: tcp_close() returned NULL")); 2763 break; 2764 2765 case TCPS_SYN_SENT: 2766 case TCPS_SYN_RECEIVED: 2767 tp->t_flags |= TF_NEEDFIN; 2768 break; 2769 2770 case TCPS_ESTABLISHED: 2771 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2772 break; 2773 2774 case TCPS_CLOSE_WAIT: 2775 tcp_state_change(tp, TCPS_LAST_ACK); 2776 break; 2777 } 2778 if (tp->t_acktime == 0) 2779 tp->t_acktime = ticks; 2780 if (tp->t_state >= TCPS_FIN_WAIT_2) { 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