1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_ipsec.h" 47 #include "opt_kern_tls.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/arb.h> 52 #include <sys/limits.h> 53 #include <sys/malloc.h> 54 #include <sys/refcount.h> 55 #include <sys/kernel.h> 56 #include <sys/ktls.h> 57 #include <sys/qmath.h> 58 #include <sys/sysctl.h> 59 #include <sys/mbuf.h> 60 #ifdef INET6 61 #include <sys/domain.h> 62 #endif /* INET6 */ 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/protosw.h> 66 #include <sys/proc.h> 67 #include <sys/jail.h> 68 #include <sys/stats.h> 69 70 #ifdef DDB 71 #include <ddb/ddb.h> 72 #endif 73 74 #include <net/if.h> 75 #include <net/if_var.h> 76 #include <net/route.h> 77 #include <net/vnet.h> 78 79 #include <netinet/in.h> 80 #include <netinet/in_kdtrace.h> 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_systm.h> 83 #include <netinet/in_var.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_var.h> 86 #ifdef INET6 87 #include <netinet/ip6.h> 88 #include <netinet6/in6_pcb.h> 89 #include <netinet6/ip6_var.h> 90 #include <netinet6/scope6_var.h> 91 #endif 92 #include <netinet/tcp.h> 93 #include <netinet/tcp_fsm.h> 94 #include <netinet/tcp_seq.h> 95 #include <netinet/tcp_timer.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/tcp_log_buf.h> 98 #include <netinet/tcpip.h> 99 #include <netinet/cc/cc.h> 100 #include <netinet/tcp_fastopen.h> 101 #include <netinet/tcp_hpts.h> 102 #ifdef TCPPCAP 103 #include <netinet/tcp_pcap.h> 104 #endif 105 #ifdef TCP_OFFLOAD 106 #include <netinet/tcp_offload.h> 107 #endif 108 #include <netipsec/ipsec_support.h> 109 110 #include <vm/vm.h> 111 #include <vm/vm_param.h> 112 #include <vm/pmap.h> 113 #include <vm/vm_extern.h> 114 #include <vm/vm_map.h> 115 #include <vm/vm_page.h> 116 117 /* 118 * TCP protocol interface to socket abstraction. 119 */ 120 #ifdef INET 121 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 122 struct thread *td); 123 #endif /* INET */ 124 #ifdef INET6 125 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 126 struct thread *td); 127 #endif /* INET6 */ 128 static void tcp_disconnect(struct tcpcb *); 129 static void tcp_usrclosed(struct tcpcb *); 130 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 131 132 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 133 134 static void 135 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error) 136 { 137 struct tcp_log_buffer *lgb; 138 139 KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL")); 140 INP_WLOCK_ASSERT(tptoinpcb(tp)); 141 if (tcp_bblogging_on(tp)) { 142 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error, 143 0, NULL, false, NULL, NULL, 0, NULL); 144 } else { 145 lgb = NULL; 146 } 147 if (lgb != NULL) { 148 if (error >= 0) { 149 lgb->tlb_errno = (uint32_t)error; 150 } 151 lgb->tlb_flex1 = pru; 152 } 153 } 154 155 /* 156 * TCP attaches to socket via pru_attach(), reserving space, 157 * and an internet control block. 158 */ 159 static int 160 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 161 { 162 struct inpcb *inp; 163 struct tcpcb *tp = NULL; 164 int error; 165 166 inp = sotoinpcb(so); 167 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 168 169 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 170 if (error) 171 goto out; 172 173 so->so_rcv.sb_flags |= SB_AUTOSIZE; 174 so->so_snd.sb_flags |= SB_AUTOSIZE; 175 error = in_pcballoc(so, &V_tcbinfo); 176 if (error) 177 goto out; 178 inp = sotoinpcb(so); 179 tp = tcp_newtcpcb(inp); 180 if (tp == NULL) { 181 error = ENOBUFS; 182 in_pcbdetach(inp); 183 in_pcbfree(inp); 184 goto out; 185 } 186 tp->t_state = TCPS_CLOSED; 187 tcp_bblog_pru(tp, PRU_ATTACH, error); 188 INP_WUNLOCK(inp); 189 TCPSTATES_INC(TCPS_CLOSED); 190 out: 191 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 192 return (error); 193 } 194 195 /* 196 * tcp_usr_detach is called when the socket layer loses its final reference 197 * to the socket, be it a file descriptor reference, a reference from TCP, 198 * etc. At this point, there is only one case in which we will keep around 199 * inpcb state: time wait. 200 */ 201 static void 202 tcp_usr_detach(struct socket *so) 203 { 204 struct inpcb *inp; 205 struct tcpcb *tp; 206 207 inp = sotoinpcb(so); 208 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 209 INP_WLOCK(inp); 210 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 211 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 212 213 tp = intotcpcb(inp); 214 215 KASSERT(inp->inp_flags & INP_DROPPED || 216 tp->t_state < TCPS_SYN_SENT, 217 ("%s: inp %p not dropped or embryonic", __func__, inp)); 218 219 tcp_discardcb(tp); 220 in_pcbdetach(inp); 221 in_pcbfree(inp); 222 } 223 224 #ifdef INET 225 /* 226 * Give the socket an address. 227 */ 228 static int 229 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 230 { 231 int error = 0; 232 struct inpcb *inp; 233 struct tcpcb *tp; 234 struct sockaddr_in *sinp; 235 236 inp = sotoinpcb(so); 237 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 238 INP_WLOCK(inp); 239 if (inp->inp_flags & INP_DROPPED) { 240 INP_WUNLOCK(inp); 241 return (EINVAL); 242 } 243 tp = intotcpcb(inp); 244 245 sinp = (struct sockaddr_in *)nam; 246 if (nam->sa_family != AF_INET) { 247 /* 248 * Preserve compatibility with old programs. 249 */ 250 if (nam->sa_family != AF_UNSPEC || 251 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 252 sinp->sin_addr.s_addr != INADDR_ANY) { 253 error = EAFNOSUPPORT; 254 goto out; 255 } 256 nam->sa_family = AF_INET; 257 } 258 if (nam->sa_len != sizeof(*sinp)) { 259 error = EINVAL; 260 goto out; 261 } 262 /* 263 * Must check for multicast addresses and disallow binding 264 * to them. 265 */ 266 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 267 error = EAFNOSUPPORT; 268 goto out; 269 } 270 INP_HASH_WLOCK(&V_tcbinfo); 271 error = in_pcbbind(inp, sinp, td->td_ucred); 272 INP_HASH_WUNLOCK(&V_tcbinfo); 273 out: 274 tcp_bblog_pru(tp, PRU_BIND, error); 275 TCP_PROBE2(debug__user, tp, PRU_BIND); 276 INP_WUNLOCK(inp); 277 278 return (error); 279 } 280 #endif /* INET */ 281 282 #ifdef INET6 283 static int 284 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 285 { 286 int error = 0; 287 struct inpcb *inp; 288 struct tcpcb *tp; 289 struct sockaddr_in6 *sin6; 290 u_char vflagsav; 291 292 inp = sotoinpcb(so); 293 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 294 INP_WLOCK(inp); 295 if (inp->inp_flags & INP_DROPPED) { 296 INP_WUNLOCK(inp); 297 return (EINVAL); 298 } 299 tp = intotcpcb(inp); 300 301 vflagsav = inp->inp_vflag; 302 303 sin6 = (struct sockaddr_in6 *)nam; 304 if (nam->sa_family != AF_INET6) { 305 error = EAFNOSUPPORT; 306 goto out; 307 } 308 if (nam->sa_len != sizeof(*sin6)) { 309 error = EINVAL; 310 goto out; 311 } 312 /* 313 * Must check for multicast addresses and disallow binding 314 * to them. 315 */ 316 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 317 error = EAFNOSUPPORT; 318 goto out; 319 } 320 321 INP_HASH_WLOCK(&V_tcbinfo); 322 inp->inp_vflag &= ~INP_IPV4; 323 inp->inp_vflag |= INP_IPV6; 324 #ifdef INET 325 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 326 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 327 inp->inp_vflag |= INP_IPV4; 328 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 329 struct sockaddr_in sin; 330 331 in6_sin6_2_sin(&sin, sin6); 332 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 333 error = EAFNOSUPPORT; 334 INP_HASH_WUNLOCK(&V_tcbinfo); 335 goto out; 336 } 337 inp->inp_vflag |= INP_IPV4; 338 inp->inp_vflag &= ~INP_IPV6; 339 error = in_pcbbind(inp, &sin, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 goto out; 342 } 343 } 344 #endif 345 error = in6_pcbbind(inp, sin6, td->td_ucred); 346 INP_HASH_WUNLOCK(&V_tcbinfo); 347 out: 348 if (error != 0) 349 inp->inp_vflag = vflagsav; 350 tcp_bblog_pru(tp, PRU_BIND, error); 351 TCP_PROBE2(debug__user, tp, PRU_BIND); 352 INP_WUNLOCK(inp); 353 return (error); 354 } 355 #endif /* INET6 */ 356 357 #ifdef INET 358 /* 359 * Prepare to accept connections. 360 */ 361 static int 362 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 363 { 364 int error = 0; 365 struct inpcb *inp; 366 struct tcpcb *tp; 367 368 inp = sotoinpcb(so); 369 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 370 INP_WLOCK(inp); 371 if (inp->inp_flags & INP_DROPPED) { 372 INP_WUNLOCK(inp); 373 return (EINVAL); 374 } 375 tp = intotcpcb(inp); 376 377 SOCK_LOCK(so); 378 error = solisten_proto_check(so); 379 if (error != 0) { 380 SOCK_UNLOCK(so); 381 goto out; 382 } 383 if (inp->inp_lport == 0) { 384 INP_HASH_WLOCK(&V_tcbinfo); 385 error = in_pcbbind(inp, NULL, td->td_ucred); 386 INP_HASH_WUNLOCK(&V_tcbinfo); 387 } 388 if (error == 0) { 389 tcp_state_change(tp, TCPS_LISTEN); 390 solisten_proto(so, backlog); 391 #ifdef TCP_OFFLOAD 392 if ((so->so_options & SO_NO_OFFLOAD) == 0) 393 tcp_offload_listen_start(tp); 394 #endif 395 } else { 396 solisten_proto_abort(so); 397 } 398 SOCK_UNLOCK(so); 399 400 if (IS_FASTOPEN(tp->t_flags)) 401 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 402 403 out: 404 tcp_bblog_pru(tp, PRU_LISTEN, error); 405 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 406 INP_WUNLOCK(inp); 407 return (error); 408 } 409 #endif /* INET */ 410 411 #ifdef INET6 412 static int 413 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 414 { 415 int error = 0; 416 struct inpcb *inp; 417 struct tcpcb *tp; 418 u_char vflagsav; 419 420 inp = sotoinpcb(so); 421 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 422 INP_WLOCK(inp); 423 if (inp->inp_flags & INP_DROPPED) { 424 INP_WUNLOCK(inp); 425 return (EINVAL); 426 } 427 tp = intotcpcb(inp); 428 429 vflagsav = inp->inp_vflag; 430 431 SOCK_LOCK(so); 432 error = solisten_proto_check(so); 433 if (error != 0) { 434 SOCK_UNLOCK(so); 435 goto out; 436 } 437 INP_HASH_WLOCK(&V_tcbinfo); 438 if (inp->inp_lport == 0) { 439 inp->inp_vflag &= ~INP_IPV4; 440 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 441 inp->inp_vflag |= INP_IPV4; 442 error = in6_pcbbind(inp, NULL, td->td_ucred); 443 } 444 INP_HASH_WUNLOCK(&V_tcbinfo); 445 if (error == 0) { 446 tcp_state_change(tp, TCPS_LISTEN); 447 solisten_proto(so, backlog); 448 #ifdef TCP_OFFLOAD 449 if ((so->so_options & SO_NO_OFFLOAD) == 0) 450 tcp_offload_listen_start(tp); 451 #endif 452 } else { 453 solisten_proto_abort(so); 454 } 455 SOCK_UNLOCK(so); 456 457 if (IS_FASTOPEN(tp->t_flags)) 458 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 459 460 if (error != 0) 461 inp->inp_vflag = vflagsav; 462 463 out: 464 tcp_bblog_pru(tp, PRU_LISTEN, error); 465 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 466 INP_WUNLOCK(inp); 467 return (error); 468 } 469 #endif /* INET6 */ 470 471 #ifdef INET 472 /* 473 * Initiate connection to peer. 474 * Create a template for use in transmissions on this connection. 475 * Enter SYN_SENT state, and mark socket as connecting. 476 * Start keep-alive timer, and seed output sequence space. 477 * Send initial segment on connection. 478 */ 479 static int 480 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 481 { 482 struct epoch_tracker et; 483 int error = 0; 484 struct inpcb *inp; 485 struct tcpcb *tp; 486 struct sockaddr_in *sinp; 487 488 inp = sotoinpcb(so); 489 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 490 INP_WLOCK(inp); 491 if (inp->inp_flags & INP_DROPPED) { 492 INP_WUNLOCK(inp); 493 return (ECONNREFUSED); 494 } 495 tp = intotcpcb(inp); 496 497 sinp = (struct sockaddr_in *)nam; 498 if (nam->sa_family != AF_INET) { 499 error = EAFNOSUPPORT; 500 goto out; 501 } 502 if (nam->sa_len != sizeof (*sinp)) { 503 error = EINVAL; 504 goto out; 505 } 506 /* 507 * Must disallow TCP ``connections'' to multicast addresses. 508 */ 509 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 510 error = EAFNOSUPPORT; 511 goto out; 512 } 513 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 514 error = EACCES; 515 goto out; 516 } 517 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 518 goto out; 519 if (SOLISTENING(so)) { 520 error = EOPNOTSUPP; 521 goto out; 522 } 523 NET_EPOCH_ENTER(et); 524 if ((error = tcp_connect(tp, sinp, td)) != 0) 525 goto out_in_epoch; 526 #ifdef TCP_OFFLOAD 527 if (registered_toedevs > 0 && 528 (so->so_options & SO_NO_OFFLOAD) == 0 && 529 (error = tcp_offload_connect(so, nam)) == 0) 530 goto out_in_epoch; 531 #endif 532 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 533 error = tcp_output(tp); 534 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 535 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 536 out_in_epoch: 537 NET_EPOCH_EXIT(et); 538 out: 539 tcp_bblog_pru(tp, PRU_CONNECT, error); 540 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 541 INP_WUNLOCK(inp); 542 return (error); 543 } 544 #endif /* INET */ 545 546 #ifdef INET6 547 static int 548 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 549 { 550 struct epoch_tracker et; 551 int error = 0; 552 struct inpcb *inp; 553 struct tcpcb *tp; 554 struct sockaddr_in6 *sin6; 555 u_int8_t incflagsav; 556 u_char vflagsav; 557 558 inp = sotoinpcb(so); 559 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 560 INP_WLOCK(inp); 561 if (inp->inp_flags & INP_DROPPED) { 562 INP_WUNLOCK(inp); 563 return (ECONNREFUSED); 564 } 565 tp = intotcpcb(inp); 566 567 vflagsav = inp->inp_vflag; 568 incflagsav = inp->inp_inc.inc_flags; 569 570 sin6 = (struct sockaddr_in6 *)nam; 571 if (nam->sa_family != AF_INET6) { 572 error = EAFNOSUPPORT; 573 goto out; 574 } 575 if (nam->sa_len != sizeof (*sin6)) { 576 error = EINVAL; 577 goto out; 578 } 579 /* 580 * Must disallow TCP ``connections'' to multicast addresses. 581 */ 582 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 583 error = EAFNOSUPPORT; 584 goto out; 585 } 586 if (SOLISTENING(so)) { 587 error = EINVAL; 588 goto out; 589 } 590 #ifdef INET 591 /* 592 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 593 * therefore probably require the hash lock, which isn't held here. 594 * Is this a significant problem? 595 */ 596 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 597 struct sockaddr_in sin; 598 599 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 600 error = EINVAL; 601 goto out; 602 } 603 if ((inp->inp_vflag & INP_IPV4) == 0) { 604 error = EAFNOSUPPORT; 605 goto out; 606 } 607 608 in6_sin6_2_sin(&sin, sin6); 609 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 610 error = EAFNOSUPPORT; 611 goto out; 612 } 613 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 614 error = EACCES; 615 goto out; 616 } 617 if ((error = prison_remote_ip4(td->td_ucred, 618 &sin.sin_addr)) != 0) 619 goto out; 620 inp->inp_vflag |= INP_IPV4; 621 inp->inp_vflag &= ~INP_IPV6; 622 NET_EPOCH_ENTER(et); 623 if ((error = tcp_connect(tp, &sin, td)) != 0) 624 goto out_in_epoch; 625 #ifdef TCP_OFFLOAD 626 if (registered_toedevs > 0 && 627 (so->so_options & SO_NO_OFFLOAD) == 0 && 628 (error = tcp_offload_connect(so, nam)) == 0) 629 goto out_in_epoch; 630 #endif 631 error = tcp_output(tp); 632 goto out_in_epoch; 633 } else { 634 if ((inp->inp_vflag & INP_IPV6) == 0) { 635 error = EAFNOSUPPORT; 636 goto out; 637 } 638 } 639 #endif 640 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 641 goto out; 642 inp->inp_vflag &= ~INP_IPV4; 643 inp->inp_vflag |= INP_IPV6; 644 inp->inp_inc.inc_flags |= INC_ISIPV6; 645 NET_EPOCH_ENTER(et); 646 if ((error = tcp6_connect(tp, sin6, td)) != 0) 647 goto out_in_epoch; 648 #ifdef TCP_OFFLOAD 649 if (registered_toedevs > 0 && 650 (so->so_options & SO_NO_OFFLOAD) == 0 && 651 (error = tcp_offload_connect(so, nam)) == 0) 652 goto out_in_epoch; 653 #endif 654 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 655 error = tcp_output(tp); 656 out_in_epoch: 657 NET_EPOCH_EXIT(et); 658 out: 659 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 660 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 661 /* 662 * If the implicit bind in the connect call fails, restore 663 * the flags we modified. 664 */ 665 if (error != 0 && inp->inp_lport == 0) { 666 inp->inp_vflag = vflagsav; 667 inp->inp_inc.inc_flags = incflagsav; 668 } 669 670 tcp_bblog_pru(tp, PRU_CONNECT, error); 671 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 672 INP_WUNLOCK(inp); 673 return (error); 674 } 675 #endif /* INET6 */ 676 677 /* 678 * Initiate disconnect from peer. 679 * If connection never passed embryonic stage, just drop; 680 * else if don't need to let data drain, then can just drop anyways, 681 * else have to begin TCP shutdown process: mark socket disconnecting, 682 * drain unread data, state switch to reflect user close, and 683 * send segment (e.g. FIN) to peer. Socket will be really disconnected 684 * when peer sends FIN and acks ours. 685 * 686 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 687 */ 688 static int 689 tcp_usr_disconnect(struct socket *so) 690 { 691 struct inpcb *inp; 692 struct tcpcb *tp = NULL; 693 struct epoch_tracker et; 694 int error = 0; 695 696 NET_EPOCH_ENTER(et); 697 inp = sotoinpcb(so); 698 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 699 INP_WLOCK(inp); 700 if (inp->inp_flags & INP_DROPPED) { 701 INP_WUNLOCK(inp); 702 NET_EPOCH_EXIT(et); 703 return (ECONNRESET); 704 } 705 tp = intotcpcb(inp); 706 707 if (tp->t_state == TCPS_TIME_WAIT) 708 goto out; 709 tcp_disconnect(tp); 710 out: 711 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 712 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 713 INP_WUNLOCK(inp); 714 NET_EPOCH_EXIT(et); 715 return (error); 716 } 717 718 #ifdef INET 719 /* 720 * Accept a connection. Essentially all the work is done at higher levels; 721 * just return the address of the peer, storing through addr. 722 */ 723 static int 724 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 725 { 726 int error = 0; 727 struct inpcb *inp; 728 struct tcpcb *tp; 729 struct in_addr addr; 730 in_port_t port = 0; 731 732 inp = sotoinpcb(so); 733 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 734 INP_WLOCK(inp); 735 if (inp->inp_flags & INP_DROPPED) { 736 INP_WUNLOCK(inp); 737 return (ECONNABORTED); 738 } 739 tp = intotcpcb(inp); 740 741 if (so->so_state & SS_ISDISCONNECTED) { 742 error = ECONNABORTED; 743 goto out; 744 } 745 /* 746 * We inline in_getpeeraddr and COMMON_END here, so that we can 747 * copy the data of interest and defer the malloc until after we 748 * release the lock. 749 */ 750 port = inp->inp_fport; 751 addr = inp->inp_faddr; 752 753 out: 754 tcp_bblog_pru(tp, PRU_ACCEPT, error); 755 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 756 INP_WUNLOCK(inp); 757 if (error == 0) 758 *nam = in_sockaddr(port, &addr); 759 return error; 760 } 761 #endif /* INET */ 762 763 #ifdef INET6 764 static int 765 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 766 { 767 struct inpcb *inp; 768 int error = 0; 769 struct tcpcb *tp; 770 struct in_addr addr; 771 struct in6_addr addr6; 772 struct epoch_tracker et; 773 in_port_t port = 0; 774 int v4 = 0; 775 776 inp = sotoinpcb(so); 777 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 778 NET_EPOCH_ENTER(et); /* XXXMT Why is this needed? */ 779 INP_WLOCK(inp); 780 if (inp->inp_flags & INP_DROPPED) { 781 INP_WUNLOCK(inp); 782 NET_EPOCH_EXIT(et); 783 return (ECONNABORTED); 784 } 785 tp = intotcpcb(inp); 786 787 if (so->so_state & SS_ISDISCONNECTED) { 788 error = ECONNABORTED; 789 goto out; 790 } 791 /* 792 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 793 * copy the data of interest and defer the malloc until after we 794 * release the lock. 795 */ 796 if (inp->inp_vflag & INP_IPV4) { 797 v4 = 1; 798 port = inp->inp_fport; 799 addr = inp->inp_faddr; 800 } else { 801 port = inp->inp_fport; 802 addr6 = inp->in6p_faddr; 803 } 804 805 out: 806 tcp_bblog_pru(tp, PRU_ACCEPT, error); 807 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 808 INP_WUNLOCK(inp); 809 NET_EPOCH_EXIT(et); 810 if (error == 0) { 811 if (v4) 812 *nam = in6_v4mapsin6_sockaddr(port, &addr); 813 else 814 *nam = in6_sockaddr(port, &addr6); 815 } 816 return error; 817 } 818 #endif /* INET6 */ 819 820 /* 821 * Mark the connection as being incapable of further output. 822 */ 823 static int 824 tcp_usr_shutdown(struct socket *so) 825 { 826 int error = 0; 827 struct inpcb *inp; 828 struct tcpcb *tp; 829 struct epoch_tracker et; 830 831 inp = sotoinpcb(so); 832 KASSERT(inp != NULL, ("inp == NULL")); 833 INP_WLOCK(inp); 834 if (inp->inp_flags & INP_DROPPED) { 835 INP_WUNLOCK(inp); 836 return (ECONNRESET); 837 } 838 tp = intotcpcb(inp); 839 840 NET_EPOCH_ENTER(et); 841 socantsendmore(so); 842 tcp_usrclosed(tp); 843 if (!(inp->inp_flags & INP_DROPPED)) 844 error = tcp_output_nodrop(tp); 845 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 846 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 847 error = tcp_unlock_or_drop(tp, error); 848 NET_EPOCH_EXIT(et); 849 850 return (error); 851 } 852 853 /* 854 * After a receive, possibly send window update to peer. 855 */ 856 static int 857 tcp_usr_rcvd(struct socket *so, int flags) 858 { 859 struct epoch_tracker et; 860 struct inpcb *inp; 861 struct tcpcb *tp; 862 int outrv = 0, error = 0; 863 864 inp = sotoinpcb(so); 865 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 866 INP_WLOCK(inp); 867 if (inp->inp_flags & INP_DROPPED) { 868 INP_WUNLOCK(inp); 869 return (ECONNRESET); 870 } 871 tp = intotcpcb(inp); 872 873 NET_EPOCH_ENTER(et); 874 /* 875 * For passively-created TFO connections, don't attempt a window 876 * update while still in SYN_RECEIVED as this may trigger an early 877 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 878 * application response data, or failing that, when the DELACK timer 879 * expires. 880 */ 881 if (IS_FASTOPEN(tp->t_flags) && 882 (tp->t_state == TCPS_SYN_RECEIVED)) 883 goto out; 884 #ifdef TCP_OFFLOAD 885 if (tp->t_flags & TF_TOE) 886 tcp_offload_rcvd(tp); 887 else 888 #endif 889 outrv = tcp_output_nodrop(tp); 890 out: 891 tcp_bblog_pru(tp, PRU_RCVD, error); 892 TCP_PROBE2(debug__user, tp, PRU_RCVD); 893 (void) tcp_unlock_or_drop(tp, outrv); 894 NET_EPOCH_EXIT(et); 895 return (error); 896 } 897 898 /* 899 * Do a send by putting data in output queue and updating urgent 900 * marker if URG set. Possibly send more data. Unlike the other 901 * pru_*() routines, the mbuf chains are our responsibility. We 902 * must either enqueue them or free them. The other pru_* routines 903 * generally are caller-frees. 904 */ 905 static int 906 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 907 struct sockaddr *nam, struct mbuf *control, struct thread *td) 908 { 909 struct epoch_tracker et; 910 int error = 0; 911 struct inpcb *inp; 912 struct tcpcb *tp; 913 #ifdef INET 914 #ifdef INET6 915 struct sockaddr_in sin; 916 #endif 917 struct sockaddr_in *sinp; 918 #endif 919 #ifdef INET6 920 struct sockaddr_in6 *sin6; 921 int isipv6; 922 #endif 923 u_int8_t incflagsav; 924 u_char vflagsav; 925 bool restoreflags; 926 927 inp = sotoinpcb(so); 928 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 929 INP_WLOCK(inp); 930 if (inp->inp_flags & INP_DROPPED) { 931 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 932 m_freem(m); 933 INP_WUNLOCK(inp); 934 return (ECONNRESET); 935 } 936 tp = intotcpcb(inp); 937 938 vflagsav = inp->inp_vflag; 939 incflagsav = inp->inp_inc.inc_flags; 940 restoreflags = false; 941 942 NET_EPOCH_ENTER(et); 943 if (control != NULL) { 944 /* TCP doesn't do control messages (rights, creds, etc) */ 945 if (control->m_len > 0) { 946 m_freem(control); 947 error = EINVAL; 948 goto out; 949 } 950 m_freem(control); /* empty control, just free it */ 951 } 952 953 if ((flags & PRUS_OOB) != 0 && 954 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 955 goto out; 956 957 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 958 if (tp->t_state == TCPS_LISTEN) { 959 error = EINVAL; 960 goto out; 961 } 962 switch (nam->sa_family) { 963 #ifdef INET 964 case AF_INET: 965 sinp = (struct sockaddr_in *)nam; 966 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 967 error = EINVAL; 968 goto out; 969 } 970 if ((inp->inp_vflag & INP_IPV6) != 0) { 971 error = EAFNOSUPPORT; 972 goto out; 973 } 974 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 975 error = EAFNOSUPPORT; 976 goto out; 977 } 978 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 979 error = EACCES; 980 goto out; 981 } 982 if ((error = prison_remote_ip4(td->td_ucred, 983 &sinp->sin_addr))) 984 goto out; 985 #ifdef INET6 986 isipv6 = 0; 987 #endif 988 break; 989 #endif /* INET */ 990 #ifdef INET6 991 case AF_INET6: 992 sin6 = (struct sockaddr_in6 *)nam; 993 if (sin6->sin6_len != sizeof(*sin6)) { 994 error = EINVAL; 995 goto out; 996 } 997 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 998 error = EAFNOSUPPORT; 999 goto out; 1000 } 1001 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1002 error = EAFNOSUPPORT; 1003 goto out; 1004 } 1005 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1006 #ifdef INET 1007 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1008 error = EINVAL; 1009 goto out; 1010 } 1011 if ((inp->inp_vflag & INP_IPV4) == 0) { 1012 error = EAFNOSUPPORT; 1013 goto out; 1014 } 1015 restoreflags = true; 1016 inp->inp_vflag &= ~INP_IPV6; 1017 sinp = &sin; 1018 in6_sin6_2_sin(sinp, sin6); 1019 if (IN_MULTICAST( 1020 ntohl(sinp->sin_addr.s_addr))) { 1021 error = EAFNOSUPPORT; 1022 goto out; 1023 } 1024 if ((error = prison_remote_ip4(td->td_ucred, 1025 &sinp->sin_addr))) 1026 goto out; 1027 isipv6 = 0; 1028 #else /* !INET */ 1029 error = EAFNOSUPPORT; 1030 goto out; 1031 #endif /* INET */ 1032 } else { 1033 if ((inp->inp_vflag & INP_IPV6) == 0) { 1034 error = EAFNOSUPPORT; 1035 goto out; 1036 } 1037 restoreflags = true; 1038 inp->inp_vflag &= ~INP_IPV4; 1039 inp->inp_inc.inc_flags |= INC_ISIPV6; 1040 if ((error = prison_remote_ip6(td->td_ucred, 1041 &sin6->sin6_addr))) 1042 goto out; 1043 isipv6 = 1; 1044 } 1045 break; 1046 #endif /* INET6 */ 1047 default: 1048 error = EAFNOSUPPORT; 1049 goto out; 1050 } 1051 } 1052 if (!(flags & PRUS_OOB)) { 1053 if (tp->t_acktime == 0) 1054 tp->t_acktime = ticks; 1055 sbappendstream(&so->so_snd, m, flags); 1056 m = NULL; 1057 if (nam && tp->t_state < TCPS_SYN_SENT) { 1058 KASSERT(tp->t_state == TCPS_CLOSED, 1059 ("%s: tp %p is listening", __func__, tp)); 1060 1061 /* 1062 * Do implied connect if not yet connected, 1063 * initialize window to default value, and 1064 * initialize maxseg using peer's cached MSS. 1065 */ 1066 #ifdef INET6 1067 if (isipv6) 1068 error = tcp6_connect(tp, sin6, td); 1069 #endif /* INET6 */ 1070 #if defined(INET6) && defined(INET) 1071 else 1072 #endif 1073 #ifdef INET 1074 error = tcp_connect(tp, sinp, td); 1075 #endif 1076 /* 1077 * The bind operation in tcp_connect succeeded. We 1078 * no longer want to restore the flags if later 1079 * operations fail. 1080 */ 1081 if (error == 0 || inp->inp_lport != 0) 1082 restoreflags = false; 1083 1084 if (error) { 1085 /* m is freed if PRUS_NOTREADY is unset. */ 1086 sbflush(&so->so_snd); 1087 goto out; 1088 } 1089 if (IS_FASTOPEN(tp->t_flags)) 1090 tcp_fastopen_connect(tp); 1091 else { 1092 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1093 tcp_mss(tp, -1); 1094 } 1095 } 1096 if (flags & PRUS_EOF) { 1097 /* 1098 * Close the send side of the connection after 1099 * the data is sent. 1100 */ 1101 socantsendmore(so); 1102 tcp_usrclosed(tp); 1103 } 1104 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1105 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1106 (tp->t_fbyte_out == 0) && 1107 (so->so_snd.sb_ccc > 0)) { 1108 tp->t_fbyte_out = ticks; 1109 if (tp->t_fbyte_out == 0) 1110 tp->t_fbyte_out = 1; 1111 if (tp->t_fbyte_out && tp->t_fbyte_in) 1112 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1113 } 1114 if (!(inp->inp_flags & INP_DROPPED) && 1115 !(flags & PRUS_NOTREADY)) { 1116 if (flags & PRUS_MORETOCOME) 1117 tp->t_flags |= TF_MORETOCOME; 1118 error = tcp_output_nodrop(tp); 1119 if (flags & PRUS_MORETOCOME) 1120 tp->t_flags &= ~TF_MORETOCOME; 1121 } 1122 } else { 1123 /* 1124 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1125 */ 1126 SOCKBUF_LOCK(&so->so_snd); 1127 if (sbspace(&so->so_snd) < -512) { 1128 SOCKBUF_UNLOCK(&so->so_snd); 1129 error = ENOBUFS; 1130 goto out; 1131 } 1132 /* 1133 * According to RFC961 (Assigned Protocols), 1134 * the urgent pointer points to the last octet 1135 * of urgent data. We continue, however, 1136 * to consider it to indicate the first octet 1137 * of data past the urgent section. 1138 * Otherwise, snd_up should be one lower. 1139 */ 1140 if (tp->t_acktime == 0) 1141 tp->t_acktime = ticks; 1142 sbappendstream_locked(&so->so_snd, m, flags); 1143 SOCKBUF_UNLOCK(&so->so_snd); 1144 m = NULL; 1145 if (nam && tp->t_state < TCPS_SYN_SENT) { 1146 /* 1147 * Do implied connect if not yet connected, 1148 * initialize window to default value, and 1149 * initialize maxseg using peer's cached MSS. 1150 */ 1151 1152 /* 1153 * Not going to contemplate SYN|URG 1154 */ 1155 if (IS_FASTOPEN(tp->t_flags)) 1156 tp->t_flags &= ~TF_FASTOPEN; 1157 #ifdef INET6 1158 if (isipv6) 1159 error = tcp6_connect(tp, sin6, td); 1160 #endif /* INET6 */ 1161 #if defined(INET6) && defined(INET) 1162 else 1163 #endif 1164 #ifdef INET 1165 error = tcp_connect(tp, sinp, td); 1166 #endif 1167 /* 1168 * The bind operation in tcp_connect succeeded. We 1169 * no longer want to restore the flags if later 1170 * operations fail. 1171 */ 1172 if (error == 0 || inp->inp_lport != 0) 1173 restoreflags = false; 1174 1175 if (error != 0) { 1176 /* m is freed if PRUS_NOTREADY is unset. */ 1177 sbflush(&so->so_snd); 1178 goto out; 1179 } 1180 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1181 tcp_mss(tp, -1); 1182 } 1183 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1184 if ((flags & PRUS_NOTREADY) == 0) { 1185 tp->t_flags |= TF_FORCEDATA; 1186 error = tcp_output_nodrop(tp); 1187 tp->t_flags &= ~TF_FORCEDATA; 1188 } 1189 } 1190 TCP_LOG_EVENT(tp, NULL, 1191 &inp->inp_socket->so_rcv, 1192 &inp->inp_socket->so_snd, 1193 TCP_LOG_USERSEND, error, 1194 0, NULL, false); 1195 1196 out: 1197 /* 1198 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1199 * responsible for freeing memory. 1200 */ 1201 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1202 m_freem(m); 1203 1204 /* 1205 * If the request was unsuccessful and we changed flags, 1206 * restore the original flags. 1207 */ 1208 if (error != 0 && restoreflags) { 1209 inp->inp_vflag = vflagsav; 1210 inp->inp_inc.inc_flags = incflagsav; 1211 } 1212 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1213 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1214 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1215 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1216 error = tcp_unlock_or_drop(tp, error); 1217 NET_EPOCH_EXIT(et); 1218 return (error); 1219 } 1220 1221 static int 1222 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1223 { 1224 struct epoch_tracker et; 1225 struct inpcb *inp; 1226 struct tcpcb *tp; 1227 int error; 1228 1229 inp = sotoinpcb(so); 1230 INP_WLOCK(inp); 1231 if (inp->inp_flags & INP_DROPPED) { 1232 INP_WUNLOCK(inp); 1233 mb_free_notready(m, count); 1234 return (ECONNRESET); 1235 } 1236 tp = intotcpcb(inp); 1237 1238 SOCKBUF_LOCK(&so->so_snd); 1239 error = sbready(&so->so_snd, m, count); 1240 SOCKBUF_UNLOCK(&so->so_snd); 1241 if (error) { 1242 INP_WUNLOCK(inp); 1243 return (error); 1244 } 1245 NET_EPOCH_ENTER(et); 1246 error = tcp_output_unlock(tp); 1247 NET_EPOCH_EXIT(et); 1248 1249 return (error); 1250 } 1251 1252 /* 1253 * Abort the TCP. Drop the connection abruptly. 1254 */ 1255 static void 1256 tcp_usr_abort(struct socket *so) 1257 { 1258 struct inpcb *inp; 1259 struct tcpcb *tp; 1260 struct epoch_tracker et; 1261 1262 inp = sotoinpcb(so); 1263 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1264 1265 NET_EPOCH_ENTER(et); 1266 INP_WLOCK(inp); 1267 KASSERT(inp->inp_socket != NULL, 1268 ("tcp_usr_abort: inp_socket == NULL")); 1269 1270 /* 1271 * If we still have full TCP state, and we're not dropped, drop. 1272 */ 1273 if (!(inp->inp_flags & INP_DROPPED)) { 1274 tp = intotcpcb(inp); 1275 tp = tcp_drop(tp, ECONNABORTED); 1276 if (tp == NULL) 1277 goto dropped; 1278 tcp_bblog_pru(tp, PRU_ABORT, 0); 1279 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1280 } 1281 if (!(inp->inp_flags & INP_DROPPED)) { 1282 soref(so); 1283 inp->inp_flags |= INP_SOCKREF; 1284 } 1285 INP_WUNLOCK(inp); 1286 dropped: 1287 NET_EPOCH_EXIT(et); 1288 } 1289 1290 /* 1291 * TCP socket is closed. Start friendly disconnect. 1292 */ 1293 static void 1294 tcp_usr_close(struct socket *so) 1295 { 1296 struct inpcb *inp; 1297 struct tcpcb *tp; 1298 struct epoch_tracker et; 1299 1300 inp = sotoinpcb(so); 1301 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1302 1303 NET_EPOCH_ENTER(et); 1304 INP_WLOCK(inp); 1305 KASSERT(inp->inp_socket != NULL, 1306 ("tcp_usr_close: inp_socket == NULL")); 1307 1308 /* 1309 * If we are still connected and we're not dropped, initiate 1310 * a disconnect. 1311 */ 1312 if (!(inp->inp_flags & INP_DROPPED)) { 1313 tp = intotcpcb(inp); 1314 if (tp->t_state != TCPS_TIME_WAIT) { 1315 tp->t_flags |= TF_CLOSED; 1316 tcp_disconnect(tp); 1317 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1318 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1319 } 1320 } 1321 if (!(inp->inp_flags & INP_DROPPED)) { 1322 soref(so); 1323 inp->inp_flags |= INP_SOCKREF; 1324 } 1325 INP_WUNLOCK(inp); 1326 NET_EPOCH_EXIT(et); 1327 } 1328 1329 static int 1330 tcp_pru_options_support(struct tcpcb *tp, int flags) 1331 { 1332 /* 1333 * If the specific TCP stack has a pru_options 1334 * specified then it does not always support 1335 * all the PRU_XX options and we must ask it. 1336 * If the function is not specified then all 1337 * of the PRU_XX options are supported. 1338 */ 1339 int ret = 0; 1340 1341 if (tp->t_fb->tfb_pru_options) { 1342 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1343 } 1344 return (ret); 1345 } 1346 1347 /* 1348 * Receive out-of-band data. 1349 */ 1350 static int 1351 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1352 { 1353 int error = 0; 1354 struct inpcb *inp; 1355 struct tcpcb *tp; 1356 1357 inp = sotoinpcb(so); 1358 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1359 INP_WLOCK(inp); 1360 if (inp->inp_flags & INP_DROPPED) { 1361 INP_WUNLOCK(inp); 1362 return (ECONNRESET); 1363 } 1364 tp = intotcpcb(inp); 1365 1366 error = tcp_pru_options_support(tp, PRUS_OOB); 1367 if (error) { 1368 goto out; 1369 } 1370 if ((so->so_oobmark == 0 && 1371 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1372 so->so_options & SO_OOBINLINE || 1373 tp->t_oobflags & TCPOOB_HADDATA) { 1374 error = EINVAL; 1375 goto out; 1376 } 1377 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1378 error = EWOULDBLOCK; 1379 goto out; 1380 } 1381 m->m_len = 1; 1382 *mtod(m, caddr_t) = tp->t_iobc; 1383 if ((flags & MSG_PEEK) == 0) 1384 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1385 1386 out: 1387 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1388 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1389 INP_WUNLOCK(inp); 1390 return (error); 1391 } 1392 1393 #ifdef INET 1394 struct protosw tcp_protosw = { 1395 .pr_type = SOCK_STREAM, 1396 .pr_protocol = IPPROTO_TCP, 1397 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1398 PR_CAPATTACH, 1399 .pr_ctloutput = tcp_ctloutput, 1400 .pr_abort = tcp_usr_abort, 1401 .pr_accept = tcp_usr_accept, 1402 .pr_attach = tcp_usr_attach, 1403 .pr_bind = tcp_usr_bind, 1404 .pr_connect = tcp_usr_connect, 1405 .pr_control = in_control, 1406 .pr_detach = tcp_usr_detach, 1407 .pr_disconnect = tcp_usr_disconnect, 1408 .pr_listen = tcp_usr_listen, 1409 .pr_peeraddr = in_getpeeraddr, 1410 .pr_rcvd = tcp_usr_rcvd, 1411 .pr_rcvoob = tcp_usr_rcvoob, 1412 .pr_send = tcp_usr_send, 1413 .pr_ready = tcp_usr_ready, 1414 .pr_shutdown = tcp_usr_shutdown, 1415 .pr_sockaddr = in_getsockaddr, 1416 .pr_sosetlabel = in_pcbsosetlabel, 1417 .pr_close = tcp_usr_close, 1418 }; 1419 #endif /* INET */ 1420 1421 #ifdef INET6 1422 struct protosw tcp6_protosw = { 1423 .pr_type = SOCK_STREAM, 1424 .pr_protocol = IPPROTO_TCP, 1425 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1426 PR_CAPATTACH, 1427 .pr_ctloutput = tcp_ctloutput, 1428 .pr_abort = tcp_usr_abort, 1429 .pr_accept = tcp6_usr_accept, 1430 .pr_attach = tcp_usr_attach, 1431 .pr_bind = tcp6_usr_bind, 1432 .pr_connect = tcp6_usr_connect, 1433 .pr_control = in6_control, 1434 .pr_detach = tcp_usr_detach, 1435 .pr_disconnect = tcp_usr_disconnect, 1436 .pr_listen = tcp6_usr_listen, 1437 .pr_peeraddr = in6_mapped_peeraddr, 1438 .pr_rcvd = tcp_usr_rcvd, 1439 .pr_rcvoob = tcp_usr_rcvoob, 1440 .pr_send = tcp_usr_send, 1441 .pr_ready = tcp_usr_ready, 1442 .pr_shutdown = tcp_usr_shutdown, 1443 .pr_sockaddr = in6_mapped_sockaddr, 1444 .pr_sosetlabel = in_pcbsosetlabel, 1445 .pr_close = tcp_usr_close, 1446 }; 1447 #endif /* INET6 */ 1448 1449 #ifdef INET 1450 /* 1451 * Common subroutine to open a TCP connection to remote host specified 1452 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1453 * and assign a local port number and install the inpcb into the hash. 1454 * Initialize connection parameters and enter SYN-SENT state. 1455 */ 1456 static int 1457 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1458 { 1459 struct inpcb *inp = tptoinpcb(tp); 1460 struct socket *so = tptosocket(tp); 1461 int error; 1462 1463 NET_EPOCH_ASSERT(); 1464 INP_WLOCK_ASSERT(inp); 1465 1466 if (__predict_false((so->so_state & 1467 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1468 SS_ISDISCONNECTED)) != 0)) 1469 return (EISCONN); 1470 1471 INP_HASH_WLOCK(&V_tcbinfo); 1472 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1473 INP_HASH_WUNLOCK(&V_tcbinfo); 1474 if (error != 0) 1475 return (error); 1476 1477 /* 1478 * Compute window scaling to request: 1479 * Scale to fit into sweet spot. See tcp_syncache.c. 1480 * XXX: This should move to tcp_output(). 1481 */ 1482 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1483 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1484 tp->request_r_scale++; 1485 1486 soisconnecting(so); 1487 TCPSTAT_INC(tcps_connattempt); 1488 tcp_state_change(tp, TCPS_SYN_SENT); 1489 tp->iss = tcp_new_isn(&inp->inp_inc); 1490 if (tp->t_flags & TF_REQ_TSTMP) 1491 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1492 tcp_sendseqinit(tp); 1493 1494 return (0); 1495 } 1496 #endif /* INET */ 1497 1498 #ifdef INET6 1499 static int 1500 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1501 { 1502 struct inpcb *inp = tptoinpcb(tp); 1503 struct socket *so = tptosocket(tp); 1504 int error; 1505 1506 NET_EPOCH_ASSERT(); 1507 INP_WLOCK_ASSERT(inp); 1508 1509 if (__predict_false((so->so_state & 1510 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1511 return (EISCONN); 1512 1513 INP_HASH_WLOCK(&V_tcbinfo); 1514 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1515 INP_HASH_WUNLOCK(&V_tcbinfo); 1516 if (error != 0) 1517 return (error); 1518 1519 /* Compute window scaling to request. */ 1520 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1521 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1522 tp->request_r_scale++; 1523 1524 soisconnecting(so); 1525 TCPSTAT_INC(tcps_connattempt); 1526 tcp_state_change(tp, TCPS_SYN_SENT); 1527 tp->iss = tcp_new_isn(&inp->inp_inc); 1528 if (tp->t_flags & TF_REQ_TSTMP) 1529 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1530 tcp_sendseqinit(tp); 1531 1532 return (0); 1533 } 1534 #endif /* INET6 */ 1535 1536 /* 1537 * Export TCP internal state information via a struct tcp_info, based on the 1538 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1539 * (TCP state machine, etc). We export all information using FreeBSD-native 1540 * constants -- for example, the numeric values for tcpi_state will differ 1541 * from Linux. 1542 */ 1543 static void 1544 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1545 { 1546 1547 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1548 bzero(ti, sizeof(*ti)); 1549 1550 ti->tcpi_state = tp->t_state; 1551 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1552 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1553 if (tp->t_flags & TF_SACK_PERMIT) 1554 ti->tcpi_options |= TCPI_OPT_SACK; 1555 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1556 ti->tcpi_options |= TCPI_OPT_WSCALE; 1557 ti->tcpi_snd_wscale = tp->snd_scale; 1558 ti->tcpi_rcv_wscale = tp->rcv_scale; 1559 } 1560 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1561 case TF2_ECN_PERMIT: 1562 ti->tcpi_options |= TCPI_OPT_ECN; 1563 break; 1564 case TF2_ACE_PERMIT: 1565 /* FALLTHROUGH */ 1566 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1567 ti->tcpi_options |= TCPI_OPT_ACE; 1568 break; 1569 default: 1570 break; 1571 } 1572 if (IS_FASTOPEN(tp->t_flags)) 1573 ti->tcpi_options |= TCPI_OPT_TFO; 1574 1575 ti->tcpi_rto = tp->t_rxtcur * tick; 1576 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1577 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1578 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1579 1580 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1581 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1582 1583 /* 1584 * FreeBSD-specific extension fields for tcp_info. 1585 */ 1586 ti->tcpi_rcv_space = tp->rcv_wnd; 1587 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1588 ti->tcpi_snd_wnd = tp->snd_wnd; 1589 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1590 ti->tcpi_snd_nxt = tp->snd_nxt; 1591 ti->tcpi_snd_mss = tp->t_maxseg; 1592 ti->tcpi_rcv_mss = tp->t_maxseg; 1593 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1594 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1595 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1596 #ifdef TCP_OFFLOAD 1597 if (tp->t_flags & TF_TOE) { 1598 ti->tcpi_options |= TCPI_OPT_TOE; 1599 tcp_offload_tcp_info(tp, ti); 1600 } 1601 #endif 1602 /* 1603 * AccECN related counters. 1604 */ 1605 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1606 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1607 /* 1608 * Internal counter starts at 5 for AccECN 1609 * but 0 for RFC3168 ECN. 1610 */ 1611 ti->tcpi_delivered_ce = tp->t_scep - 5; 1612 else 1613 ti->tcpi_delivered_ce = tp->t_scep; 1614 ti->tcpi_received_ce = tp->t_rcep; 1615 } 1616 1617 /* 1618 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1619 * socket option arguments. When it re-acquires the lock after the copy, it 1620 * has to revalidate that the connection is still valid for the socket 1621 * option. 1622 */ 1623 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1624 INP_WLOCK(inp); \ 1625 if (inp->inp_flags & INP_DROPPED) { \ 1626 INP_WUNLOCK(inp); \ 1627 cleanup; \ 1628 return (ECONNRESET); \ 1629 } \ 1630 tp = intotcpcb(inp); \ 1631 } while(0) 1632 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1633 1634 int 1635 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1636 { 1637 struct socket *so = inp->inp_socket; 1638 struct tcpcb *tp = intotcpcb(inp); 1639 int error = 0; 1640 1641 MPASS(sopt->sopt_dir == SOPT_SET); 1642 INP_WLOCK_ASSERT(inp); 1643 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1644 ("inp_flags == %x", inp->inp_flags)); 1645 KASSERT(so != NULL, ("inp_socket == NULL")); 1646 1647 if (sopt->sopt_level != IPPROTO_TCP) { 1648 INP_WUNLOCK(inp); 1649 #ifdef INET6 1650 if (inp->inp_vflag & INP_IPV6PROTO) 1651 error = ip6_ctloutput(so, sopt); 1652 #endif 1653 #if defined(INET6) && defined(INET) 1654 else 1655 #endif 1656 #ifdef INET 1657 error = ip_ctloutput(so, sopt); 1658 #endif 1659 /* 1660 * When an IP-level socket option affects TCP, pass control 1661 * down to stack tfb_tcp_ctloutput, otherwise return what 1662 * IP level returned. 1663 */ 1664 switch (sopt->sopt_level) { 1665 #ifdef INET6 1666 case IPPROTO_IPV6: 1667 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1668 return (error); 1669 switch (sopt->sopt_name) { 1670 case IPV6_TCLASS: 1671 /* Notify tcp stacks that care (e.g. RACK). */ 1672 break; 1673 case IPV6_USE_MIN_MTU: 1674 /* Update t_maxseg accordingly. */ 1675 break; 1676 default: 1677 return (error); 1678 } 1679 break; 1680 #endif 1681 #ifdef INET 1682 case IPPROTO_IP: 1683 switch (sopt->sopt_name) { 1684 case IP_TOS: 1685 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1686 break; 1687 case IP_TTL: 1688 /* Notify tcp stacks that care (e.g. RACK). */ 1689 break; 1690 default: 1691 return (error); 1692 } 1693 break; 1694 #endif 1695 default: 1696 return (error); 1697 } 1698 INP_WLOCK(inp); 1699 if (inp->inp_flags & INP_DROPPED) { 1700 INP_WUNLOCK(inp); 1701 return (ECONNRESET); 1702 } 1703 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1704 /* 1705 * Protect the TCP option TCP_FUNCTION_BLK so 1706 * that a sub-function can *never* overwrite this. 1707 */ 1708 struct tcp_function_set fsn; 1709 struct tcp_function_block *blk; 1710 void *ptr = NULL; 1711 1712 INP_WUNLOCK(inp); 1713 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1714 if (error) 1715 return (error); 1716 1717 INP_WLOCK(inp); 1718 tp = intotcpcb(inp); 1719 1720 blk = find_and_ref_tcp_functions(&fsn); 1721 if (blk == NULL) { 1722 INP_WUNLOCK(inp); 1723 return (ENOENT); 1724 } 1725 if (tp->t_fb == blk) { 1726 /* You already have this */ 1727 refcount_release(&blk->tfb_refcnt); 1728 INP_WUNLOCK(inp); 1729 return (0); 1730 } 1731 if (tp->t_state != TCPS_CLOSED) { 1732 /* 1733 * The user has advanced the state 1734 * past the initial point, we may not 1735 * be able to switch. 1736 */ 1737 if (blk->tfb_tcp_handoff_ok != NULL) { 1738 /* 1739 * Does the stack provide a 1740 * query mechanism, if so it may 1741 * still be possible? 1742 */ 1743 error = (*blk->tfb_tcp_handoff_ok)(tp); 1744 } else 1745 error = EINVAL; 1746 if (error) { 1747 refcount_release(&blk->tfb_refcnt); 1748 INP_WUNLOCK(inp); 1749 return(error); 1750 } 1751 } 1752 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1753 refcount_release(&blk->tfb_refcnt); 1754 INP_WUNLOCK(inp); 1755 return (ENOENT); 1756 } 1757 /* 1758 * Ensure the new stack takes ownership with a 1759 * clean slate on peak rate threshold. 1760 */ 1761 #ifdef TCPHPTS 1762 /* Assure that we are not on any hpts */ 1763 tcp_hpts_remove(tp); 1764 #endif 1765 if (blk->tfb_tcp_fb_init) { 1766 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1767 if (error) { 1768 /* 1769 * Release the ref count the lookup 1770 * acquired. 1771 */ 1772 refcount_release(&blk->tfb_refcnt); 1773 /* 1774 * Now there is a chance that the 1775 * init() function mucked with some 1776 * things before it failed, such as 1777 * hpts or inp_flags2 or timer granularity. 1778 * It should not of, but lets give the old 1779 * stack a chance to reset to a known good state. 1780 */ 1781 if (tp->t_fb->tfb_switch_failed) { 1782 (*tp->t_fb->tfb_switch_failed)(tp); 1783 } 1784 goto err_out; 1785 } 1786 } 1787 if (tp->t_fb->tfb_tcp_fb_fini) { 1788 struct epoch_tracker et; 1789 /* 1790 * Tell the stack to cleanup with 0 i.e. 1791 * the tcb is not going away. 1792 */ 1793 NET_EPOCH_ENTER(et); 1794 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1795 NET_EPOCH_EXIT(et); 1796 } 1797 /* 1798 * Release the old refcnt, the 1799 * lookup acquired a ref on the 1800 * new one already. 1801 */ 1802 refcount_release(&tp->t_fb->tfb_refcnt); 1803 /* 1804 * Set in the new stack. 1805 */ 1806 tp->t_fb = blk; 1807 tp->t_fb_ptr = ptr; 1808 #ifdef TCP_OFFLOAD 1809 if (tp->t_flags & TF_TOE) { 1810 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1811 sopt->sopt_name); 1812 } 1813 #endif 1814 err_out: 1815 INP_WUNLOCK(inp); 1816 return (error); 1817 1818 } 1819 1820 /* Pass in the INP locked, callee must unlock it. */ 1821 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1822 } 1823 1824 static int 1825 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1826 { 1827 struct socket *so = inp->inp_socket; 1828 struct tcpcb *tp = intotcpcb(inp); 1829 int error = 0; 1830 1831 MPASS(sopt->sopt_dir == SOPT_GET); 1832 INP_WLOCK_ASSERT(inp); 1833 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1834 ("inp_flags == %x", inp->inp_flags)); 1835 KASSERT(so != NULL, ("inp_socket == NULL")); 1836 1837 if (sopt->sopt_level != IPPROTO_TCP) { 1838 INP_WUNLOCK(inp); 1839 #ifdef INET6 1840 if (inp->inp_vflag & INP_IPV6PROTO) 1841 error = ip6_ctloutput(so, sopt); 1842 #endif /* INET6 */ 1843 #if defined(INET6) && defined(INET) 1844 else 1845 #endif 1846 #ifdef INET 1847 error = ip_ctloutput(so, sopt); 1848 #endif 1849 return (error); 1850 } 1851 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1852 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1853 struct tcp_function_set fsn; 1854 1855 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1856 memset(&fsn, 0, sizeof(fsn)); 1857 find_tcp_function_alias(tp->t_fb, &fsn); 1858 } else { 1859 strncpy(fsn.function_set_name, 1860 tp->t_fb->tfb_tcp_block_name, 1861 TCP_FUNCTION_NAME_LEN_MAX); 1862 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1863 } 1864 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1865 INP_WUNLOCK(inp); 1866 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1867 return (error); 1868 } 1869 1870 /* Pass in the INP locked, callee must unlock it. */ 1871 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1872 } 1873 1874 int 1875 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1876 { 1877 struct inpcb *inp; 1878 1879 inp = sotoinpcb(so); 1880 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1881 1882 INP_WLOCK(inp); 1883 if (inp->inp_flags & INP_DROPPED) { 1884 INP_WUNLOCK(inp); 1885 return (ECONNRESET); 1886 } 1887 if (sopt->sopt_dir == SOPT_SET) 1888 return (tcp_ctloutput_set(inp, sopt)); 1889 else if (sopt->sopt_dir == SOPT_GET) 1890 return (tcp_ctloutput_get(inp, sopt)); 1891 else 1892 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1893 } 1894 1895 /* 1896 * If this assert becomes untrue, we need to change the size of the buf 1897 * variable in tcp_default_ctloutput(). 1898 */ 1899 #ifdef CTASSERT 1900 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1901 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1902 #endif 1903 1904 #ifdef KERN_TLS 1905 static int 1906 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1907 { 1908 struct tls_enable_v0 tls_v0; 1909 int error; 1910 1911 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1912 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1913 sizeof(tls_v0)); 1914 if (error) 1915 return (error); 1916 memset(tls, 0, sizeof(*tls)); 1917 tls->cipher_key = tls_v0.cipher_key; 1918 tls->iv = tls_v0.iv; 1919 tls->auth_key = tls_v0.auth_key; 1920 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1921 tls->cipher_key_len = tls_v0.cipher_key_len; 1922 tls->iv_len = tls_v0.iv_len; 1923 tls->auth_algorithm = tls_v0.auth_algorithm; 1924 tls->auth_key_len = tls_v0.auth_key_len; 1925 tls->flags = tls_v0.flags; 1926 tls->tls_vmajor = tls_v0.tls_vmajor; 1927 tls->tls_vminor = tls_v0.tls_vminor; 1928 return (0); 1929 } 1930 1931 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1932 } 1933 #endif 1934 1935 extern struct cc_algo newreno_cc_algo; 1936 1937 static int 1938 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1939 { 1940 struct cc_algo *algo; 1941 void *ptr = NULL; 1942 struct tcpcb *tp; 1943 struct cc_var cc_mem; 1944 char buf[TCP_CA_NAME_MAX]; 1945 size_t mem_sz; 1946 int error; 1947 1948 INP_WUNLOCK(inp); 1949 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1950 if (error) 1951 return(error); 1952 buf[sopt->sopt_valsize] = '\0'; 1953 CC_LIST_RLOCK(); 1954 STAILQ_FOREACH(algo, &cc_list, entries) { 1955 if (strncmp(buf, algo->name, 1956 TCP_CA_NAME_MAX) == 0) { 1957 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1958 /* We can't "see" modules being unloaded */ 1959 continue; 1960 } 1961 break; 1962 } 1963 } 1964 if (algo == NULL) { 1965 CC_LIST_RUNLOCK(); 1966 return(ESRCH); 1967 } 1968 /* 1969 * With a reference the algorithm cannot be removed 1970 * so we hold a reference through the change process. 1971 */ 1972 cc_refer(algo); 1973 CC_LIST_RUNLOCK(); 1974 if (algo->cb_init != NULL) { 1975 /* We can now pre-get the memory for the CC */ 1976 mem_sz = (*algo->cc_data_sz)(); 1977 if (mem_sz == 0) { 1978 goto no_mem_needed; 1979 } 1980 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1981 } else { 1982 no_mem_needed: 1983 mem_sz = 0; 1984 ptr = NULL; 1985 } 1986 /* 1987 * Make sure its all clean and zero and also get 1988 * back the inplock. 1989 */ 1990 memset(&cc_mem, 0, sizeof(cc_mem)); 1991 INP_WLOCK(inp); 1992 if (inp->inp_flags & INP_DROPPED) { 1993 INP_WUNLOCK(inp); 1994 if (ptr) 1995 free(ptr, M_CC_MEM); 1996 /* Release our temp reference */ 1997 CC_LIST_RLOCK(); 1998 cc_release(algo); 1999 CC_LIST_RUNLOCK(); 2000 return (ECONNRESET); 2001 } 2002 tp = intotcpcb(inp); 2003 if (ptr != NULL) 2004 memset(ptr, 0, mem_sz); 2005 cc_mem.ccvc.tcp = tp; 2006 /* 2007 * We once again hold a write lock over the tcb so it's 2008 * safe to do these things without ordering concerns. 2009 * Note here we init into stack memory. 2010 */ 2011 if (algo->cb_init != NULL) 2012 error = algo->cb_init(&cc_mem, ptr); 2013 else 2014 error = 0; 2015 /* 2016 * The CC algorithms, when given their memory 2017 * should not fail we could in theory have a 2018 * KASSERT here. 2019 */ 2020 if (error == 0) { 2021 /* 2022 * Touchdown, lets go ahead and move the 2023 * connection to the new CC module by 2024 * copying in the cc_mem after we call 2025 * the old ones cleanup (if any). 2026 */ 2027 if (CC_ALGO(tp)->cb_destroy != NULL) 2028 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2029 /* Detach the old CC from the tcpcb */ 2030 cc_detach(tp); 2031 /* Copy in our temp memory that was inited */ 2032 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2033 /* Now attach the new, which takes a reference */ 2034 cc_attach(tp, algo); 2035 /* Ok now are we where we have gotten past any conn_init? */ 2036 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2037 /* Yep run the connection init for the new CC */ 2038 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2039 } 2040 } else if (ptr) 2041 free(ptr, M_CC_MEM); 2042 INP_WUNLOCK(inp); 2043 /* Now lets release our temp reference */ 2044 CC_LIST_RLOCK(); 2045 cc_release(algo); 2046 CC_LIST_RUNLOCK(); 2047 return (error); 2048 } 2049 2050 int 2051 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2052 { 2053 struct inpcb *inp = tptoinpcb(tp); 2054 int error, opt, optval; 2055 u_int ui; 2056 struct tcp_info ti; 2057 #ifdef KERN_TLS 2058 struct tls_enable tls; 2059 struct socket *so = inp->inp_socket; 2060 #endif 2061 char *pbuf, buf[TCP_LOG_ID_LEN]; 2062 #ifdef STATS 2063 struct statsblob *sbp; 2064 #endif 2065 size_t len; 2066 2067 INP_WLOCK_ASSERT(inp); 2068 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2069 ("inp_flags == %x", inp->inp_flags)); 2070 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2071 2072 switch (sopt->sopt_level) { 2073 #ifdef INET6 2074 case IPPROTO_IPV6: 2075 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2076 switch (sopt->sopt_name) { 2077 case IPV6_USE_MIN_MTU: 2078 tcp6_use_min_mtu(tp); 2079 /* FALLTHROUGH */ 2080 } 2081 INP_WUNLOCK(inp); 2082 return (0); 2083 #endif 2084 #ifdef INET 2085 case IPPROTO_IP: 2086 INP_WUNLOCK(inp); 2087 return (0); 2088 #endif 2089 } 2090 2091 /* 2092 * For TCP_CCALGOOPT forward the control to CC module, for both 2093 * SOPT_SET and SOPT_GET. 2094 */ 2095 switch (sopt->sopt_name) { 2096 case TCP_CCALGOOPT: 2097 INP_WUNLOCK(inp); 2098 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2099 return (EINVAL); 2100 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2101 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2102 sopt->sopt_valsize); 2103 if (error) { 2104 free(pbuf, M_TEMP); 2105 return (error); 2106 } 2107 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2108 if (CC_ALGO(tp)->ctl_output != NULL) 2109 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2110 else 2111 error = ENOENT; 2112 INP_WUNLOCK(inp); 2113 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2114 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2115 free(pbuf, M_TEMP); 2116 return (error); 2117 } 2118 2119 switch (sopt->sopt_dir) { 2120 case SOPT_SET: 2121 switch (sopt->sopt_name) { 2122 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2123 case TCP_MD5SIG: 2124 INP_WUNLOCK(inp); 2125 if (!TCPMD5_ENABLED()) 2126 return (ENOPROTOOPT); 2127 error = TCPMD5_PCBCTL(inp, sopt); 2128 if (error) 2129 return (error); 2130 INP_WLOCK_RECHECK(inp); 2131 goto unlock_and_done; 2132 #endif /* IPSEC */ 2133 2134 case TCP_NODELAY: 2135 case TCP_NOOPT: 2136 case TCP_LRD: 2137 INP_WUNLOCK(inp); 2138 error = sooptcopyin(sopt, &optval, sizeof optval, 2139 sizeof optval); 2140 if (error) 2141 return (error); 2142 2143 INP_WLOCK_RECHECK(inp); 2144 switch (sopt->sopt_name) { 2145 case TCP_NODELAY: 2146 opt = TF_NODELAY; 2147 break; 2148 case TCP_NOOPT: 2149 opt = TF_NOOPT; 2150 break; 2151 case TCP_LRD: 2152 opt = TF_LRD; 2153 break; 2154 default: 2155 opt = 0; /* dead code to fool gcc */ 2156 break; 2157 } 2158 2159 if (optval) 2160 tp->t_flags |= opt; 2161 else 2162 tp->t_flags &= ~opt; 2163 unlock_and_done: 2164 #ifdef TCP_OFFLOAD 2165 if (tp->t_flags & TF_TOE) { 2166 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2167 sopt->sopt_name); 2168 } 2169 #endif 2170 INP_WUNLOCK(inp); 2171 break; 2172 2173 case TCP_NOPUSH: 2174 INP_WUNLOCK(inp); 2175 error = sooptcopyin(sopt, &optval, sizeof optval, 2176 sizeof optval); 2177 if (error) 2178 return (error); 2179 2180 INP_WLOCK_RECHECK(inp); 2181 if (optval) 2182 tp->t_flags |= TF_NOPUSH; 2183 else if (tp->t_flags & TF_NOPUSH) { 2184 tp->t_flags &= ~TF_NOPUSH; 2185 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2186 struct epoch_tracker et; 2187 2188 NET_EPOCH_ENTER(et); 2189 error = tcp_output_nodrop(tp); 2190 NET_EPOCH_EXIT(et); 2191 } 2192 } 2193 goto unlock_and_done; 2194 2195 case TCP_REMOTE_UDP_ENCAPS_PORT: 2196 INP_WUNLOCK(inp); 2197 error = sooptcopyin(sopt, &optval, sizeof optval, 2198 sizeof optval); 2199 if (error) 2200 return (error); 2201 if ((optval < TCP_TUNNELING_PORT_MIN) || 2202 (optval > TCP_TUNNELING_PORT_MAX)) { 2203 /* Its got to be in range */ 2204 return (EINVAL); 2205 } 2206 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2207 /* You have to have enabled a UDP tunneling port first */ 2208 return (EINVAL); 2209 } 2210 INP_WLOCK_RECHECK(inp); 2211 if (tp->t_state != TCPS_CLOSED) { 2212 /* You can't change after you are connected */ 2213 error = EINVAL; 2214 } else { 2215 /* Ok we are all good set the port */ 2216 tp->t_port = htons(optval); 2217 } 2218 goto unlock_and_done; 2219 2220 case TCP_MAXSEG: 2221 INP_WUNLOCK(inp); 2222 error = sooptcopyin(sopt, &optval, sizeof optval, 2223 sizeof optval); 2224 if (error) 2225 return (error); 2226 2227 INP_WLOCK_RECHECK(inp); 2228 if (optval > 0 && optval <= tp->t_maxseg && 2229 optval + 40 >= V_tcp_minmss) 2230 tp->t_maxseg = optval; 2231 else 2232 error = EINVAL; 2233 goto unlock_and_done; 2234 2235 case TCP_INFO: 2236 INP_WUNLOCK(inp); 2237 error = EINVAL; 2238 break; 2239 2240 case TCP_STATS: 2241 INP_WUNLOCK(inp); 2242 #ifdef STATS 2243 error = sooptcopyin(sopt, &optval, sizeof optval, 2244 sizeof optval); 2245 if (error) 2246 return (error); 2247 2248 if (optval > 0) 2249 sbp = stats_blob_alloc( 2250 V_tcp_perconn_stats_dflt_tpl, 0); 2251 else 2252 sbp = NULL; 2253 2254 INP_WLOCK_RECHECK(inp); 2255 if ((tp->t_stats != NULL && sbp == NULL) || 2256 (tp->t_stats == NULL && sbp != NULL)) { 2257 struct statsblob *t = tp->t_stats; 2258 tp->t_stats = sbp; 2259 sbp = t; 2260 } 2261 INP_WUNLOCK(inp); 2262 2263 stats_blob_destroy(sbp); 2264 #else 2265 return (EOPNOTSUPP); 2266 #endif /* !STATS */ 2267 break; 2268 2269 case TCP_CONGESTION: 2270 error = tcp_set_cc_mod(inp, sopt); 2271 break; 2272 2273 case TCP_REUSPORT_LB_NUMA: 2274 INP_WUNLOCK(inp); 2275 error = sooptcopyin(sopt, &optval, sizeof(optval), 2276 sizeof(optval)); 2277 INP_WLOCK_RECHECK(inp); 2278 if (!error) 2279 error = in_pcblbgroup_numa(inp, optval); 2280 INP_WUNLOCK(inp); 2281 break; 2282 2283 #ifdef KERN_TLS 2284 case TCP_TXTLS_ENABLE: 2285 INP_WUNLOCK(inp); 2286 error = copyin_tls_enable(sopt, &tls); 2287 if (error) 2288 break; 2289 error = ktls_enable_tx(so, &tls); 2290 break; 2291 case TCP_TXTLS_MODE: 2292 INP_WUNLOCK(inp); 2293 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2294 if (error) 2295 return (error); 2296 2297 INP_WLOCK_RECHECK(inp); 2298 error = ktls_set_tx_mode(so, ui); 2299 INP_WUNLOCK(inp); 2300 break; 2301 case TCP_RXTLS_ENABLE: 2302 INP_WUNLOCK(inp); 2303 error = sooptcopyin(sopt, &tls, sizeof(tls), 2304 sizeof(tls)); 2305 if (error) 2306 break; 2307 error = ktls_enable_rx(so, &tls); 2308 break; 2309 #endif 2310 case TCP_MAXUNACKTIME: 2311 case TCP_KEEPIDLE: 2312 case TCP_KEEPINTVL: 2313 case TCP_KEEPINIT: 2314 INP_WUNLOCK(inp); 2315 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2316 if (error) 2317 return (error); 2318 2319 if (ui > (UINT_MAX / hz)) { 2320 error = EINVAL; 2321 break; 2322 } 2323 ui *= hz; 2324 2325 INP_WLOCK_RECHECK(inp); 2326 switch (sopt->sopt_name) { 2327 case TCP_MAXUNACKTIME: 2328 tp->t_maxunacktime = ui; 2329 break; 2330 2331 case TCP_KEEPIDLE: 2332 tp->t_keepidle = ui; 2333 /* 2334 * XXX: better check current remaining 2335 * timeout and "merge" it with new value. 2336 */ 2337 if ((tp->t_state > TCPS_LISTEN) && 2338 (tp->t_state <= TCPS_CLOSING)) 2339 tcp_timer_activate(tp, TT_KEEP, 2340 TP_KEEPIDLE(tp)); 2341 break; 2342 case TCP_KEEPINTVL: 2343 tp->t_keepintvl = ui; 2344 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2345 (TP_MAXIDLE(tp) > 0)) 2346 tcp_timer_activate(tp, TT_2MSL, 2347 TP_MAXIDLE(tp)); 2348 break; 2349 case TCP_KEEPINIT: 2350 tp->t_keepinit = ui; 2351 if (tp->t_state == TCPS_SYN_RECEIVED || 2352 tp->t_state == TCPS_SYN_SENT) 2353 tcp_timer_activate(tp, TT_KEEP, 2354 TP_KEEPINIT(tp)); 2355 break; 2356 } 2357 goto unlock_and_done; 2358 2359 case TCP_KEEPCNT: 2360 INP_WUNLOCK(inp); 2361 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2362 if (error) 2363 return (error); 2364 2365 INP_WLOCK_RECHECK(inp); 2366 tp->t_keepcnt = ui; 2367 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2368 (TP_MAXIDLE(tp) > 0)) 2369 tcp_timer_activate(tp, TT_2MSL, 2370 TP_MAXIDLE(tp)); 2371 goto unlock_and_done; 2372 2373 #ifdef TCPPCAP 2374 case TCP_PCAP_OUT: 2375 case TCP_PCAP_IN: 2376 INP_WUNLOCK(inp); 2377 error = sooptcopyin(sopt, &optval, sizeof optval, 2378 sizeof optval); 2379 if (error) 2380 return (error); 2381 2382 INP_WLOCK_RECHECK(inp); 2383 if (optval >= 0) 2384 tcp_pcap_set_sock_max( 2385 (sopt->sopt_name == TCP_PCAP_OUT) ? 2386 &(tp->t_outpkts) : &(tp->t_inpkts), 2387 optval); 2388 else 2389 error = EINVAL; 2390 goto unlock_and_done; 2391 #endif 2392 2393 case TCP_FASTOPEN: { 2394 struct tcp_fastopen tfo_optval; 2395 2396 INP_WUNLOCK(inp); 2397 if (!V_tcp_fastopen_client_enable && 2398 !V_tcp_fastopen_server_enable) 2399 return (EPERM); 2400 2401 error = sooptcopyin(sopt, &tfo_optval, 2402 sizeof(tfo_optval), sizeof(int)); 2403 if (error) 2404 return (error); 2405 2406 INP_WLOCK_RECHECK(inp); 2407 if ((tp->t_state != TCPS_CLOSED) && 2408 (tp->t_state != TCPS_LISTEN)) { 2409 error = EINVAL; 2410 goto unlock_and_done; 2411 } 2412 if (tfo_optval.enable) { 2413 if (tp->t_state == TCPS_LISTEN) { 2414 if (!V_tcp_fastopen_server_enable) { 2415 error = EPERM; 2416 goto unlock_and_done; 2417 } 2418 2419 if (tp->t_tfo_pending == NULL) 2420 tp->t_tfo_pending = 2421 tcp_fastopen_alloc_counter(); 2422 } else { 2423 /* 2424 * If a pre-shared key was provided, 2425 * stash it in the client cookie 2426 * field of the tcpcb for use during 2427 * connect. 2428 */ 2429 if (sopt->sopt_valsize == 2430 sizeof(tfo_optval)) { 2431 memcpy(tp->t_tfo_cookie.client, 2432 tfo_optval.psk, 2433 TCP_FASTOPEN_PSK_LEN); 2434 tp->t_tfo_client_cookie_len = 2435 TCP_FASTOPEN_PSK_LEN; 2436 } 2437 } 2438 tp->t_flags |= TF_FASTOPEN; 2439 } else 2440 tp->t_flags &= ~TF_FASTOPEN; 2441 goto unlock_and_done; 2442 } 2443 2444 #ifdef TCP_BLACKBOX 2445 case TCP_LOG: 2446 INP_WUNLOCK(inp); 2447 error = sooptcopyin(sopt, &optval, sizeof optval, 2448 sizeof optval); 2449 if (error) 2450 return (error); 2451 2452 INP_WLOCK_RECHECK(inp); 2453 error = tcp_log_state_change(tp, optval); 2454 goto unlock_and_done; 2455 2456 case TCP_LOGBUF: 2457 INP_WUNLOCK(inp); 2458 error = EINVAL; 2459 break; 2460 2461 case TCP_LOGID: 2462 INP_WUNLOCK(inp); 2463 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2464 if (error) 2465 break; 2466 buf[sopt->sopt_valsize] = '\0'; 2467 INP_WLOCK_RECHECK(inp); 2468 error = tcp_log_set_id(tp, buf); 2469 /* tcp_log_set_id() unlocks the INP. */ 2470 break; 2471 2472 case TCP_LOGDUMP: 2473 case TCP_LOGDUMPID: 2474 INP_WUNLOCK(inp); 2475 error = 2476 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2477 if (error) 2478 break; 2479 buf[sopt->sopt_valsize] = '\0'; 2480 INP_WLOCK_RECHECK(inp); 2481 if (sopt->sopt_name == TCP_LOGDUMP) { 2482 error = tcp_log_dump_tp_logbuf(tp, buf, 2483 M_WAITOK, true); 2484 INP_WUNLOCK(inp); 2485 } else { 2486 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2487 /* 2488 * tcp_log_dump_tp_bucket_logbufs() drops the 2489 * INP lock. 2490 */ 2491 } 2492 break; 2493 #endif 2494 2495 default: 2496 INP_WUNLOCK(inp); 2497 error = ENOPROTOOPT; 2498 break; 2499 } 2500 break; 2501 2502 case SOPT_GET: 2503 tp = intotcpcb(inp); 2504 switch (sopt->sopt_name) { 2505 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2506 case TCP_MD5SIG: 2507 INP_WUNLOCK(inp); 2508 if (!TCPMD5_ENABLED()) 2509 return (ENOPROTOOPT); 2510 error = TCPMD5_PCBCTL(inp, sopt); 2511 break; 2512 #endif 2513 2514 case TCP_NODELAY: 2515 optval = tp->t_flags & TF_NODELAY; 2516 INP_WUNLOCK(inp); 2517 error = sooptcopyout(sopt, &optval, sizeof optval); 2518 break; 2519 case TCP_MAXSEG: 2520 optval = tp->t_maxseg; 2521 INP_WUNLOCK(inp); 2522 error = sooptcopyout(sopt, &optval, sizeof optval); 2523 break; 2524 case TCP_REMOTE_UDP_ENCAPS_PORT: 2525 optval = ntohs(tp->t_port); 2526 INP_WUNLOCK(inp); 2527 error = sooptcopyout(sopt, &optval, sizeof optval); 2528 break; 2529 case TCP_NOOPT: 2530 optval = tp->t_flags & TF_NOOPT; 2531 INP_WUNLOCK(inp); 2532 error = sooptcopyout(sopt, &optval, sizeof optval); 2533 break; 2534 case TCP_NOPUSH: 2535 optval = tp->t_flags & TF_NOPUSH; 2536 INP_WUNLOCK(inp); 2537 error = sooptcopyout(sopt, &optval, sizeof optval); 2538 break; 2539 case TCP_INFO: 2540 tcp_fill_info(tp, &ti); 2541 INP_WUNLOCK(inp); 2542 error = sooptcopyout(sopt, &ti, sizeof ti); 2543 break; 2544 case TCP_STATS: 2545 { 2546 #ifdef STATS 2547 int nheld; 2548 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2549 2550 error = 0; 2551 socklen_t outsbsz = sopt->sopt_valsize; 2552 if (tp->t_stats == NULL) 2553 error = ENOENT; 2554 else if (outsbsz >= tp->t_stats->cursz) 2555 outsbsz = tp->t_stats->cursz; 2556 else if (outsbsz >= sizeof(struct statsblob)) 2557 outsbsz = sizeof(struct statsblob); 2558 else 2559 error = EINVAL; 2560 INP_WUNLOCK(inp); 2561 if (error) 2562 break; 2563 2564 sbp = sopt->sopt_val; 2565 nheld = atop(round_page(((vm_offset_t)sbp) + 2566 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2567 vm_page_t ma[nheld]; 2568 if (vm_fault_quick_hold_pages( 2569 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2570 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2571 nheld) < 0) { 2572 error = EFAULT; 2573 break; 2574 } 2575 2576 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2577 SIZEOF_MEMBER(struct statsblob, flags)))) 2578 goto unhold; 2579 2580 INP_WLOCK_RECHECK(inp); 2581 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2582 sbflags | SB_CLONE_USRDSTNOFAULT); 2583 INP_WUNLOCK(inp); 2584 sopt->sopt_valsize = outsbsz; 2585 unhold: 2586 vm_page_unhold_pages(ma, nheld); 2587 #else 2588 INP_WUNLOCK(inp); 2589 error = EOPNOTSUPP; 2590 #endif /* !STATS */ 2591 break; 2592 } 2593 case TCP_CONGESTION: 2594 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2595 INP_WUNLOCK(inp); 2596 error = sooptcopyout(sopt, buf, len + 1); 2597 break; 2598 case TCP_MAXUNACKTIME: 2599 case TCP_KEEPIDLE: 2600 case TCP_KEEPINTVL: 2601 case TCP_KEEPINIT: 2602 case TCP_KEEPCNT: 2603 switch (sopt->sopt_name) { 2604 case TCP_MAXUNACKTIME: 2605 ui = TP_MAXUNACKTIME(tp) / hz; 2606 break; 2607 case TCP_KEEPIDLE: 2608 ui = TP_KEEPIDLE(tp) / hz; 2609 break; 2610 case TCP_KEEPINTVL: 2611 ui = TP_KEEPINTVL(tp) / hz; 2612 break; 2613 case TCP_KEEPINIT: 2614 ui = TP_KEEPINIT(tp) / hz; 2615 break; 2616 case TCP_KEEPCNT: 2617 ui = TP_KEEPCNT(tp); 2618 break; 2619 } 2620 INP_WUNLOCK(inp); 2621 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2622 break; 2623 #ifdef TCPPCAP 2624 case TCP_PCAP_OUT: 2625 case TCP_PCAP_IN: 2626 optval = tcp_pcap_get_sock_max( 2627 (sopt->sopt_name == TCP_PCAP_OUT) ? 2628 &(tp->t_outpkts) : &(tp->t_inpkts)); 2629 INP_WUNLOCK(inp); 2630 error = sooptcopyout(sopt, &optval, sizeof optval); 2631 break; 2632 #endif 2633 case TCP_FASTOPEN: 2634 optval = tp->t_flags & TF_FASTOPEN; 2635 INP_WUNLOCK(inp); 2636 error = sooptcopyout(sopt, &optval, sizeof optval); 2637 break; 2638 #ifdef TCP_BLACKBOX 2639 case TCP_LOG: 2640 optval = tcp_get_bblog_state(tp); 2641 INP_WUNLOCK(inp); 2642 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2643 break; 2644 case TCP_LOGBUF: 2645 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2646 error = tcp_log_getlogbuf(sopt, tp); 2647 break; 2648 case TCP_LOGID: 2649 len = tcp_log_get_id(tp, buf); 2650 INP_WUNLOCK(inp); 2651 error = sooptcopyout(sopt, buf, len + 1); 2652 break; 2653 case TCP_LOGDUMP: 2654 case TCP_LOGDUMPID: 2655 INP_WUNLOCK(inp); 2656 error = EINVAL; 2657 break; 2658 #endif 2659 #ifdef KERN_TLS 2660 case TCP_TXTLS_MODE: 2661 error = ktls_get_tx_mode(so, &optval); 2662 INP_WUNLOCK(inp); 2663 if (error == 0) 2664 error = sooptcopyout(sopt, &optval, 2665 sizeof(optval)); 2666 break; 2667 case TCP_RXTLS_MODE: 2668 error = ktls_get_rx_mode(so, &optval); 2669 INP_WUNLOCK(inp); 2670 if (error == 0) 2671 error = sooptcopyout(sopt, &optval, 2672 sizeof(optval)); 2673 break; 2674 #endif 2675 case TCP_LRD: 2676 optval = tp->t_flags & TF_LRD; 2677 INP_WUNLOCK(inp); 2678 error = sooptcopyout(sopt, &optval, sizeof optval); 2679 break; 2680 default: 2681 INP_WUNLOCK(inp); 2682 error = ENOPROTOOPT; 2683 break; 2684 } 2685 break; 2686 } 2687 return (error); 2688 } 2689 #undef INP_WLOCK_RECHECK 2690 #undef INP_WLOCK_RECHECK_CLEANUP 2691 2692 /* 2693 * Initiate (or continue) disconnect. 2694 * If embryonic state, just send reset (once). 2695 * If in ``let data drain'' option and linger null, just drop. 2696 * Otherwise (hard), mark socket disconnecting and drop 2697 * current input data; switch states based on user close, and 2698 * send segment to peer (with FIN). 2699 */ 2700 static void 2701 tcp_disconnect(struct tcpcb *tp) 2702 { 2703 struct inpcb *inp = tptoinpcb(tp); 2704 struct socket *so = tptosocket(tp); 2705 2706 NET_EPOCH_ASSERT(); 2707 INP_WLOCK_ASSERT(inp); 2708 2709 /* 2710 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2711 * socket is still open. 2712 */ 2713 if (tp->t_state < TCPS_ESTABLISHED && 2714 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2715 tp = tcp_close(tp); 2716 KASSERT(tp != NULL, 2717 ("tcp_disconnect: tcp_close() returned NULL")); 2718 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2719 tp = tcp_drop(tp, 0); 2720 KASSERT(tp != NULL, 2721 ("tcp_disconnect: tcp_drop() returned NULL")); 2722 } else { 2723 soisdisconnecting(so); 2724 sbflush(&so->so_rcv); 2725 tcp_usrclosed(tp); 2726 if (!(inp->inp_flags & INP_DROPPED)) 2727 /* Ignore stack's drop request, we already at it. */ 2728 (void)tcp_output_nodrop(tp); 2729 } 2730 } 2731 2732 /* 2733 * User issued close, and wish to trail through shutdown states: 2734 * if never received SYN, just forget it. If got a SYN from peer, 2735 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2736 * If already got a FIN from peer, then almost done; go to LAST_ACK 2737 * state. In all other cases, have already sent FIN to peer (e.g. 2738 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2739 * for peer to send FIN or not respond to keep-alives, etc. 2740 * We can let the user exit from the close as soon as the FIN is acked. 2741 */ 2742 static void 2743 tcp_usrclosed(struct tcpcb *tp) 2744 { 2745 2746 NET_EPOCH_ASSERT(); 2747 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2748 2749 switch (tp->t_state) { 2750 case TCPS_LISTEN: 2751 #ifdef TCP_OFFLOAD 2752 tcp_offload_listen_stop(tp); 2753 #endif 2754 tcp_state_change(tp, TCPS_CLOSED); 2755 /* FALLTHROUGH */ 2756 case TCPS_CLOSED: 2757 tp = tcp_close(tp); 2758 /* 2759 * tcp_close() should never return NULL here as the socket is 2760 * still open. 2761 */ 2762 KASSERT(tp != NULL, 2763 ("tcp_usrclosed: tcp_close() returned NULL")); 2764 break; 2765 2766 case TCPS_SYN_SENT: 2767 case TCPS_SYN_RECEIVED: 2768 tp->t_flags |= TF_NEEDFIN; 2769 break; 2770 2771 case TCPS_ESTABLISHED: 2772 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2773 break; 2774 2775 case TCPS_CLOSE_WAIT: 2776 tcp_state_change(tp, TCPS_LAST_ACK); 2777 break; 2778 } 2779 if (tp->t_acktime == 0) 2780 tp->t_acktime = ticks; 2781 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2782 soisdisconnected(tptosocket(tp)); 2783 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2784 if (tp->t_state == TCPS_FIN_WAIT_2) { 2785 int timeout; 2786 2787 timeout = (tcp_fast_finwait2_recycle) ? 2788 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2789 tcp_timer_activate(tp, TT_2MSL, timeout); 2790 } 2791 } 2792 } 2793 2794 #ifdef DDB 2795 static void 2796 db_print_indent(int indent) 2797 { 2798 int i; 2799 2800 for (i = 0; i < indent; i++) 2801 db_printf(" "); 2802 } 2803 2804 static void 2805 db_print_tstate(int t_state) 2806 { 2807 2808 switch (t_state) { 2809 case TCPS_CLOSED: 2810 db_printf("TCPS_CLOSED"); 2811 return; 2812 2813 case TCPS_LISTEN: 2814 db_printf("TCPS_LISTEN"); 2815 return; 2816 2817 case TCPS_SYN_SENT: 2818 db_printf("TCPS_SYN_SENT"); 2819 return; 2820 2821 case TCPS_SYN_RECEIVED: 2822 db_printf("TCPS_SYN_RECEIVED"); 2823 return; 2824 2825 case TCPS_ESTABLISHED: 2826 db_printf("TCPS_ESTABLISHED"); 2827 return; 2828 2829 case TCPS_CLOSE_WAIT: 2830 db_printf("TCPS_CLOSE_WAIT"); 2831 return; 2832 2833 case TCPS_FIN_WAIT_1: 2834 db_printf("TCPS_FIN_WAIT_1"); 2835 return; 2836 2837 case TCPS_CLOSING: 2838 db_printf("TCPS_CLOSING"); 2839 return; 2840 2841 case TCPS_LAST_ACK: 2842 db_printf("TCPS_LAST_ACK"); 2843 return; 2844 2845 case TCPS_FIN_WAIT_2: 2846 db_printf("TCPS_FIN_WAIT_2"); 2847 return; 2848 2849 case TCPS_TIME_WAIT: 2850 db_printf("TCPS_TIME_WAIT"); 2851 return; 2852 2853 default: 2854 db_printf("unknown"); 2855 return; 2856 } 2857 } 2858 2859 static void 2860 db_print_tflags(u_int t_flags) 2861 { 2862 int comma; 2863 2864 comma = 0; 2865 if (t_flags & TF_ACKNOW) { 2866 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2867 comma = 1; 2868 } 2869 if (t_flags & TF_DELACK) { 2870 db_printf("%sTF_DELACK", comma ? ", " : ""); 2871 comma = 1; 2872 } 2873 if (t_flags & TF_NODELAY) { 2874 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2875 comma = 1; 2876 } 2877 if (t_flags & TF_NOOPT) { 2878 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2879 comma = 1; 2880 } 2881 if (t_flags & TF_SENTFIN) { 2882 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2883 comma = 1; 2884 } 2885 if (t_flags & TF_REQ_SCALE) { 2886 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2887 comma = 1; 2888 } 2889 if (t_flags & TF_RCVD_SCALE) { 2890 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2891 comma = 1; 2892 } 2893 if (t_flags & TF_REQ_TSTMP) { 2894 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2895 comma = 1; 2896 } 2897 if (t_flags & TF_RCVD_TSTMP) { 2898 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2899 comma = 1; 2900 } 2901 if (t_flags & TF_SACK_PERMIT) { 2902 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2903 comma = 1; 2904 } 2905 if (t_flags & TF_NEEDSYN) { 2906 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2907 comma = 1; 2908 } 2909 if (t_flags & TF_NEEDFIN) { 2910 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2911 comma = 1; 2912 } 2913 if (t_flags & TF_NOPUSH) { 2914 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2915 comma = 1; 2916 } 2917 if (t_flags & TF_PREVVALID) { 2918 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2919 comma = 1; 2920 } 2921 if (t_flags & TF_MORETOCOME) { 2922 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2923 comma = 1; 2924 } 2925 if (t_flags & TF_SONOTCONN) { 2926 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2927 comma = 1; 2928 } 2929 if (t_flags & TF_LASTIDLE) { 2930 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2931 comma = 1; 2932 } 2933 if (t_flags & TF_RXWIN0SENT) { 2934 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2935 comma = 1; 2936 } 2937 if (t_flags & TF_FASTRECOVERY) { 2938 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2939 comma = 1; 2940 } 2941 if (t_flags & TF_CONGRECOVERY) { 2942 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2943 comma = 1; 2944 } 2945 if (t_flags & TF_WASFRECOVERY) { 2946 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2947 comma = 1; 2948 } 2949 if (t_flags & TF_WASCRECOVERY) { 2950 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2951 comma = 1; 2952 } 2953 if (t_flags & TF_SIGNATURE) { 2954 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2955 comma = 1; 2956 } 2957 if (t_flags & TF_FORCEDATA) { 2958 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2959 comma = 1; 2960 } 2961 if (t_flags & TF_TSO) { 2962 db_printf("%sTF_TSO", comma ? ", " : ""); 2963 comma = 1; 2964 } 2965 if (t_flags & TF_FASTOPEN) { 2966 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2967 comma = 1; 2968 } 2969 } 2970 2971 static void 2972 db_print_tflags2(u_int t_flags2) 2973 { 2974 int comma; 2975 2976 comma = 0; 2977 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2978 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2979 comma = 1; 2980 } 2981 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2982 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2983 comma = 1; 2984 } 2985 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2986 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2987 comma = 1; 2988 } 2989 if (t_flags2 & TF2_LOG_AUTO) { 2990 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2991 comma = 1; 2992 } 2993 if (t_flags2 & TF2_DROP_AF_DATA) { 2994 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2995 comma = 1; 2996 } 2997 if (t_flags2 & TF2_ECN_PERMIT) { 2998 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2999 comma = 1; 3000 } 3001 if (t_flags2 & TF2_ECN_SND_CWR) { 3002 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 3003 comma = 1; 3004 } 3005 if (t_flags2 & TF2_ECN_SND_ECE) { 3006 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 3007 comma = 1; 3008 } 3009 if (t_flags2 & TF2_ACE_PERMIT) { 3010 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 3011 comma = 1; 3012 } 3013 if (t_flags2 & TF2_FBYTES_COMPLETE) { 3014 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 3015 comma = 1; 3016 } 3017 } 3018 3019 static void 3020 db_print_toobflags(char t_oobflags) 3021 { 3022 int comma; 3023 3024 comma = 0; 3025 if (t_oobflags & TCPOOB_HAVEDATA) { 3026 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3027 comma = 1; 3028 } 3029 if (t_oobflags & TCPOOB_HADDATA) { 3030 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3031 comma = 1; 3032 } 3033 } 3034 3035 static void 3036 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3037 { 3038 3039 db_print_indent(indent); 3040 db_printf("%s at %p\n", name, tp); 3041 3042 indent += 2; 3043 3044 db_print_indent(indent); 3045 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3046 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3047 3048 db_print_indent(indent); 3049 db_printf("t_callout: %p t_timers: %p\n", 3050 &tp->t_callout, &tp->t_timers); 3051 3052 db_print_indent(indent); 3053 db_printf("t_state: %d (", tp->t_state); 3054 db_print_tstate(tp->t_state); 3055 db_printf(")\n"); 3056 3057 db_print_indent(indent); 3058 db_printf("t_flags: 0x%x (", tp->t_flags); 3059 db_print_tflags(tp->t_flags); 3060 db_printf(")\n"); 3061 3062 db_print_indent(indent); 3063 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3064 db_print_tflags2(tp->t_flags2); 3065 db_printf(")\n"); 3066 3067 db_print_indent(indent); 3068 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 3069 tp->snd_una, tp->snd_max, tp->snd_nxt); 3070 3071 db_print_indent(indent); 3072 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3073 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3074 3075 db_print_indent(indent); 3076 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3077 tp->iss, tp->irs, tp->rcv_nxt); 3078 3079 db_print_indent(indent); 3080 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3081 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3082 3083 db_print_indent(indent); 3084 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3085 tp->snd_wnd, tp->snd_cwnd); 3086 3087 db_print_indent(indent); 3088 db_printf("snd_ssthresh: %u snd_recover: " 3089 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3090 3091 db_print_indent(indent); 3092 db_printf("t_rcvtime: %u t_startime: %u\n", 3093 tp->t_rcvtime, tp->t_starttime); 3094 3095 db_print_indent(indent); 3096 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3097 tp->t_rtttime, tp->t_rtseq); 3098 3099 db_print_indent(indent); 3100 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3101 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3102 3103 db_print_indent(indent); 3104 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3105 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3106 3107 db_print_indent(indent); 3108 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3109 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3110 3111 db_print_indent(indent); 3112 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3113 db_print_toobflags(tp->t_oobflags); 3114 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3115 3116 db_print_indent(indent); 3117 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3118 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3119 3120 db_print_indent(indent); 3121 db_printf("ts_recent: %u ts_recent_age: %u\n", 3122 tp->ts_recent, tp->ts_recent_age); 3123 3124 db_print_indent(indent); 3125 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3126 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3127 3128 db_print_indent(indent); 3129 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3130 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3131 tp->snd_recover_prev, tp->t_badrxtwin); 3132 3133 db_print_indent(indent); 3134 db_printf("snd_numholes: %d snd_holes first: %p\n", 3135 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3136 3137 db_print_indent(indent); 3138 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3139 tp->snd_fack, tp->rcv_numsacks); 3140 3141 /* Skip sackblks, sackhint. */ 3142 3143 db_print_indent(indent); 3144 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3145 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3146 } 3147 3148 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3149 { 3150 struct tcpcb *tp; 3151 3152 if (!have_addr) { 3153 db_printf("usage: show tcpcb <addr>\n"); 3154 return; 3155 } 3156 tp = (struct tcpcb *)addr; 3157 3158 db_print_tcpcb(tp, "tcpcb", 0); 3159 } 3160 #endif 3161