1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 #include <sys/cdefs.h> 39 #include "opt_ddb.h" 40 #include "opt_inet.h" 41 #include "opt_inet6.h" 42 #include "opt_ipsec.h" 43 #include "opt_kern_tls.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/arb.h> 48 #include <sys/limits.h> 49 #include <sys/malloc.h> 50 #include <sys/refcount.h> 51 #include <sys/kernel.h> 52 #include <sys/ktls.h> 53 #include <sys/qmath.h> 54 #include <sys/sysctl.h> 55 #include <sys/mbuf.h> 56 #ifdef INET6 57 #include <sys/domain.h> 58 #endif /* INET6 */ 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/protosw.h> 62 #include <sys/proc.h> 63 #include <sys/jail.h> 64 #include <sys/stats.h> 65 66 #ifdef DDB 67 #include <ddb/ddb.h> 68 #endif 69 70 #include <net/if.h> 71 #include <net/if_var.h> 72 #include <net/route.h> 73 #include <net/vnet.h> 74 75 #include <netinet/in.h> 76 #include <netinet/in_kdtrace.h> 77 #include <netinet/in_pcb.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/in_var.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 #ifdef INET6 83 #include <netinet/ip6.h> 84 #include <netinet6/in6_pcb.h> 85 #include <netinet6/ip6_var.h> 86 #include <netinet6/scope6_var.h> 87 #endif 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/tcp_log_buf.h> 94 #include <netinet/tcpip.h> 95 #include <netinet/cc/cc.h> 96 #include <netinet/tcp_fastopen.h> 97 #include <netinet/tcp_hpts.h> 98 #ifdef TCPPCAP 99 #include <netinet/tcp_pcap.h> 100 #endif 101 #ifdef TCP_OFFLOAD 102 #include <netinet/tcp_offload.h> 103 #endif 104 #include <netipsec/ipsec_support.h> 105 106 #include <vm/vm.h> 107 #include <vm/vm_param.h> 108 #include <vm/pmap.h> 109 #include <vm/vm_extern.h> 110 #include <vm/vm_map.h> 111 #include <vm/vm_page.h> 112 113 /* 114 * TCP protocol interface to socket abstraction. 115 */ 116 #ifdef INET 117 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 118 struct thread *td); 119 #endif /* INET */ 120 #ifdef INET6 121 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 122 struct thread *td); 123 #endif /* INET6 */ 124 static void tcp_disconnect(struct tcpcb *); 125 static void tcp_usrclosed(struct tcpcb *); 126 static void tcp_fill_info(const struct tcpcb *, struct tcp_info *); 127 128 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 129 130 static void 131 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error) 132 { 133 struct tcp_log_buffer *lgb; 134 135 KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL")); 136 INP_WLOCK_ASSERT(tptoinpcb(tp)); 137 if (tcp_bblogging_on(tp)) { 138 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error, 139 0, NULL, false, NULL, NULL, 0, NULL); 140 } else { 141 lgb = NULL; 142 } 143 if (lgb != NULL) { 144 if (error >= 0) { 145 lgb->tlb_errno = (uint32_t)error; 146 } 147 lgb->tlb_flex1 = pru; 148 } 149 } 150 151 /* 152 * TCP attaches to socket via pru_attach(), reserving space, 153 * and an internet control block. 154 */ 155 static int 156 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 157 { 158 struct inpcb *inp; 159 struct tcpcb *tp = NULL; 160 int error; 161 162 inp = sotoinpcb(so); 163 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 164 165 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 166 if (error) 167 goto out; 168 169 so->so_rcv.sb_flags |= SB_AUTOSIZE; 170 so->so_snd.sb_flags |= SB_AUTOSIZE; 171 error = in_pcballoc(so, &V_tcbinfo); 172 if (error) 173 goto out; 174 inp = sotoinpcb(so); 175 tp = tcp_newtcpcb(inp); 176 if (tp == NULL) { 177 error = ENOBUFS; 178 in_pcbfree(inp); 179 goto out; 180 } 181 tp->t_state = TCPS_CLOSED; 182 tcp_bblog_pru(tp, PRU_ATTACH, error); 183 INP_WUNLOCK(inp); 184 TCPSTATES_INC(TCPS_CLOSED); 185 out: 186 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 187 return (error); 188 } 189 190 /* 191 * tcp_usr_detach is called when the socket layer loses its final reference 192 * to the socket, be it a file descriptor reference, a reference from TCP, 193 * etc. At this point, there is only one case in which we will keep around 194 * inpcb state: time wait. 195 */ 196 static void 197 tcp_usr_detach(struct socket *so) 198 { 199 struct inpcb *inp; 200 struct tcpcb *tp; 201 202 inp = sotoinpcb(so); 203 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 204 INP_WLOCK(inp); 205 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 206 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 207 208 tp = intotcpcb(inp); 209 210 KASSERT(inp->inp_flags & INP_DROPPED || 211 tp->t_state < TCPS_SYN_SENT, 212 ("%s: inp %p not dropped or embryonic", __func__, inp)); 213 214 tcp_discardcb(tp); 215 in_pcbfree(inp); 216 } 217 218 #ifdef INET 219 /* 220 * Give the socket an address. 221 */ 222 static int 223 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 224 { 225 int error = 0; 226 struct inpcb *inp; 227 struct tcpcb *tp; 228 struct sockaddr_in *sinp; 229 230 inp = sotoinpcb(so); 231 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 232 INP_WLOCK(inp); 233 if (inp->inp_flags & INP_DROPPED) { 234 INP_WUNLOCK(inp); 235 return (EINVAL); 236 } 237 tp = intotcpcb(inp); 238 239 sinp = (struct sockaddr_in *)nam; 240 if (nam->sa_family != AF_INET) { 241 /* 242 * Preserve compatibility with old programs. 243 */ 244 if (nam->sa_family != AF_UNSPEC || 245 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 246 sinp->sin_addr.s_addr != INADDR_ANY) { 247 error = EAFNOSUPPORT; 248 goto out; 249 } 250 nam->sa_family = AF_INET; 251 } 252 if (nam->sa_len != sizeof(*sinp)) { 253 error = EINVAL; 254 goto out; 255 } 256 /* 257 * Must check for multicast addresses and disallow binding 258 * to them. 259 */ 260 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 261 error = EAFNOSUPPORT; 262 goto out; 263 } 264 INP_HASH_WLOCK(&V_tcbinfo); 265 error = in_pcbbind(inp, sinp, td->td_ucred); 266 INP_HASH_WUNLOCK(&V_tcbinfo); 267 out: 268 tcp_bblog_pru(tp, PRU_BIND, error); 269 TCP_PROBE2(debug__user, tp, PRU_BIND); 270 INP_WUNLOCK(inp); 271 272 return (error); 273 } 274 #endif /* INET */ 275 276 #ifdef INET6 277 static int 278 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 279 { 280 int error = 0; 281 struct inpcb *inp; 282 struct tcpcb *tp; 283 struct sockaddr_in6 *sin6; 284 u_char vflagsav; 285 286 inp = sotoinpcb(so); 287 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 288 INP_WLOCK(inp); 289 if (inp->inp_flags & INP_DROPPED) { 290 INP_WUNLOCK(inp); 291 return (EINVAL); 292 } 293 tp = intotcpcb(inp); 294 295 vflagsav = inp->inp_vflag; 296 297 sin6 = (struct sockaddr_in6 *)nam; 298 if (nam->sa_family != AF_INET6) { 299 error = EAFNOSUPPORT; 300 goto out; 301 } 302 if (nam->sa_len != sizeof(*sin6)) { 303 error = EINVAL; 304 goto out; 305 } 306 /* 307 * Must check for multicast addresses and disallow binding 308 * to them. 309 */ 310 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 311 error = EAFNOSUPPORT; 312 goto out; 313 } 314 315 INP_HASH_WLOCK(&V_tcbinfo); 316 inp->inp_vflag &= ~INP_IPV4; 317 inp->inp_vflag |= INP_IPV6; 318 #ifdef INET 319 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 320 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 321 inp->inp_vflag |= INP_IPV4; 322 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 323 struct sockaddr_in sin; 324 325 in6_sin6_2_sin(&sin, sin6); 326 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 327 error = EAFNOSUPPORT; 328 INP_HASH_WUNLOCK(&V_tcbinfo); 329 goto out; 330 } 331 inp->inp_vflag |= INP_IPV4; 332 inp->inp_vflag &= ~INP_IPV6; 333 error = in_pcbbind(inp, &sin, td->td_ucred); 334 INP_HASH_WUNLOCK(&V_tcbinfo); 335 goto out; 336 } 337 } 338 #endif 339 error = in6_pcbbind(inp, sin6, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 out: 342 if (error != 0) 343 inp->inp_vflag = vflagsav; 344 tcp_bblog_pru(tp, PRU_BIND, error); 345 TCP_PROBE2(debug__user, tp, PRU_BIND); 346 INP_WUNLOCK(inp); 347 return (error); 348 } 349 #endif /* INET6 */ 350 351 #ifdef INET 352 /* 353 * Prepare to accept connections. 354 */ 355 static int 356 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 357 { 358 int error = 0; 359 struct inpcb *inp; 360 struct tcpcb *tp; 361 362 inp = sotoinpcb(so); 363 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 364 INP_WLOCK(inp); 365 if (inp->inp_flags & INP_DROPPED) { 366 INP_WUNLOCK(inp); 367 return (EINVAL); 368 } 369 tp = intotcpcb(inp); 370 371 SOCK_LOCK(so); 372 error = solisten_proto_check(so); 373 if (error != 0) { 374 SOCK_UNLOCK(so); 375 goto out; 376 } 377 if (inp->inp_lport == 0) { 378 INP_HASH_WLOCK(&V_tcbinfo); 379 error = in_pcbbind(inp, NULL, td->td_ucred); 380 INP_HASH_WUNLOCK(&V_tcbinfo); 381 } 382 if (error == 0) { 383 tcp_state_change(tp, TCPS_LISTEN); 384 solisten_proto(so, backlog); 385 #ifdef TCP_OFFLOAD 386 if ((so->so_options & SO_NO_OFFLOAD) == 0) 387 tcp_offload_listen_start(tp); 388 #endif 389 } else { 390 solisten_proto_abort(so); 391 } 392 SOCK_UNLOCK(so); 393 394 if (IS_FASTOPEN(tp->t_flags)) 395 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 396 397 out: 398 tcp_bblog_pru(tp, PRU_LISTEN, error); 399 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 400 INP_WUNLOCK(inp); 401 return (error); 402 } 403 #endif /* INET */ 404 405 #ifdef INET6 406 static int 407 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 408 { 409 int error = 0; 410 struct inpcb *inp; 411 struct tcpcb *tp; 412 u_char vflagsav; 413 414 inp = sotoinpcb(so); 415 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 416 INP_WLOCK(inp); 417 if (inp->inp_flags & INP_DROPPED) { 418 INP_WUNLOCK(inp); 419 return (EINVAL); 420 } 421 tp = intotcpcb(inp); 422 423 vflagsav = inp->inp_vflag; 424 425 SOCK_LOCK(so); 426 error = solisten_proto_check(so); 427 if (error != 0) { 428 SOCK_UNLOCK(so); 429 goto out; 430 } 431 INP_HASH_WLOCK(&V_tcbinfo); 432 if (inp->inp_lport == 0) { 433 inp->inp_vflag &= ~INP_IPV4; 434 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 435 inp->inp_vflag |= INP_IPV4; 436 error = in6_pcbbind(inp, NULL, td->td_ucred); 437 } 438 INP_HASH_WUNLOCK(&V_tcbinfo); 439 if (error == 0) { 440 tcp_state_change(tp, TCPS_LISTEN); 441 solisten_proto(so, backlog); 442 #ifdef TCP_OFFLOAD 443 if ((so->so_options & SO_NO_OFFLOAD) == 0) 444 tcp_offload_listen_start(tp); 445 #endif 446 } else { 447 solisten_proto_abort(so); 448 } 449 SOCK_UNLOCK(so); 450 451 if (IS_FASTOPEN(tp->t_flags)) 452 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 453 454 if (error != 0) 455 inp->inp_vflag = vflagsav; 456 457 out: 458 tcp_bblog_pru(tp, PRU_LISTEN, error); 459 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 460 INP_WUNLOCK(inp); 461 return (error); 462 } 463 #endif /* INET6 */ 464 465 #ifdef INET 466 /* 467 * Initiate connection to peer. 468 * Create a template for use in transmissions on this connection. 469 * Enter SYN_SENT state, and mark socket as connecting. 470 * Start keep-alive timer, and seed output sequence space. 471 * Send initial segment on connection. 472 */ 473 static int 474 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 475 { 476 struct epoch_tracker et; 477 int error = 0; 478 struct inpcb *inp; 479 struct tcpcb *tp; 480 struct sockaddr_in *sinp; 481 482 inp = sotoinpcb(so); 483 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 484 INP_WLOCK(inp); 485 if (inp->inp_flags & INP_DROPPED) { 486 INP_WUNLOCK(inp); 487 return (ECONNREFUSED); 488 } 489 tp = intotcpcb(inp); 490 491 sinp = (struct sockaddr_in *)nam; 492 if (nam->sa_family != AF_INET) { 493 error = EAFNOSUPPORT; 494 goto out; 495 } 496 if (nam->sa_len != sizeof (*sinp)) { 497 error = EINVAL; 498 goto out; 499 } 500 /* 501 * Must disallow TCP ``connections'' to multicast addresses. 502 */ 503 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 504 error = EAFNOSUPPORT; 505 goto out; 506 } 507 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 508 error = EACCES; 509 goto out; 510 } 511 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 512 goto out; 513 if (SOLISTENING(so)) { 514 error = EOPNOTSUPP; 515 goto out; 516 } 517 NET_EPOCH_ENTER(et); 518 if ((error = tcp_connect(tp, sinp, td)) != 0) 519 goto out_in_epoch; 520 #ifdef TCP_OFFLOAD 521 if (registered_toedevs > 0 && 522 (so->so_options & SO_NO_OFFLOAD) == 0 && 523 (error = tcp_offload_connect(so, nam)) == 0) 524 goto out_in_epoch; 525 #endif 526 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 527 error = tcp_output(tp); 528 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 529 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 530 out_in_epoch: 531 NET_EPOCH_EXIT(et); 532 out: 533 tcp_bblog_pru(tp, PRU_CONNECT, error); 534 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 535 INP_WUNLOCK(inp); 536 return (error); 537 } 538 #endif /* INET */ 539 540 #ifdef INET6 541 static int 542 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 543 { 544 struct epoch_tracker et; 545 int error = 0; 546 struct inpcb *inp; 547 struct tcpcb *tp; 548 struct sockaddr_in6 *sin6; 549 u_int8_t incflagsav; 550 u_char vflagsav; 551 552 inp = sotoinpcb(so); 553 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 554 INP_WLOCK(inp); 555 if (inp->inp_flags & INP_DROPPED) { 556 INP_WUNLOCK(inp); 557 return (ECONNREFUSED); 558 } 559 tp = intotcpcb(inp); 560 561 vflagsav = inp->inp_vflag; 562 incflagsav = inp->inp_inc.inc_flags; 563 564 sin6 = (struct sockaddr_in6 *)nam; 565 if (nam->sa_family != AF_INET6) { 566 error = EAFNOSUPPORT; 567 goto out; 568 } 569 if (nam->sa_len != sizeof (*sin6)) { 570 error = EINVAL; 571 goto out; 572 } 573 /* 574 * Must disallow TCP ``connections'' to multicast addresses. 575 */ 576 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 577 error = EAFNOSUPPORT; 578 goto out; 579 } 580 if (SOLISTENING(so)) { 581 error = EINVAL; 582 goto out; 583 } 584 #ifdef INET 585 /* 586 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 587 * therefore probably require the hash lock, which isn't held here. 588 * Is this a significant problem? 589 */ 590 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 591 struct sockaddr_in sin; 592 593 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 594 error = EINVAL; 595 goto out; 596 } 597 if ((inp->inp_vflag & INP_IPV4) == 0) { 598 error = EAFNOSUPPORT; 599 goto out; 600 } 601 602 in6_sin6_2_sin(&sin, sin6); 603 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 604 error = EAFNOSUPPORT; 605 goto out; 606 } 607 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 608 error = EACCES; 609 goto out; 610 } 611 if ((error = prison_remote_ip4(td->td_ucred, 612 &sin.sin_addr)) != 0) 613 goto out; 614 inp->inp_vflag |= INP_IPV4; 615 inp->inp_vflag &= ~INP_IPV6; 616 NET_EPOCH_ENTER(et); 617 if ((error = tcp_connect(tp, &sin, td)) != 0) 618 goto out_in_epoch; 619 #ifdef TCP_OFFLOAD 620 if (registered_toedevs > 0 && 621 (so->so_options & SO_NO_OFFLOAD) == 0 && 622 (error = tcp_offload_connect(so, nam)) == 0) 623 goto out_in_epoch; 624 #endif 625 error = tcp_output(tp); 626 goto out_in_epoch; 627 } else { 628 if ((inp->inp_vflag & INP_IPV6) == 0) { 629 error = EAFNOSUPPORT; 630 goto out; 631 } 632 } 633 #endif 634 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 635 goto out; 636 inp->inp_vflag &= ~INP_IPV4; 637 inp->inp_vflag |= INP_IPV6; 638 inp->inp_inc.inc_flags |= INC_ISIPV6; 639 NET_EPOCH_ENTER(et); 640 if ((error = tcp6_connect(tp, sin6, td)) != 0) 641 goto out_in_epoch; 642 #ifdef TCP_OFFLOAD 643 if (registered_toedevs > 0 && 644 (so->so_options & SO_NO_OFFLOAD) == 0 && 645 (error = tcp_offload_connect(so, nam)) == 0) 646 goto out_in_epoch; 647 #endif 648 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 649 error = tcp_output(tp); 650 out_in_epoch: 651 NET_EPOCH_EXIT(et); 652 out: 653 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 654 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 655 /* 656 * If the implicit bind in the connect call fails, restore 657 * the flags we modified. 658 */ 659 if (error != 0 && inp->inp_lport == 0) { 660 inp->inp_vflag = vflagsav; 661 inp->inp_inc.inc_flags = incflagsav; 662 } 663 664 tcp_bblog_pru(tp, PRU_CONNECT, error); 665 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 666 INP_WUNLOCK(inp); 667 return (error); 668 } 669 #endif /* INET6 */ 670 671 /* 672 * Initiate disconnect from peer. 673 * If connection never passed embryonic stage, just drop; 674 * else if don't need to let data drain, then can just drop anyways, 675 * else have to begin TCP shutdown process: mark socket disconnecting, 676 * drain unread data, state switch to reflect user close, and 677 * send segment (e.g. FIN) to peer. Socket will be really disconnected 678 * when peer sends FIN and acks ours. 679 * 680 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 681 */ 682 static int 683 tcp_usr_disconnect(struct socket *so) 684 { 685 struct inpcb *inp; 686 struct tcpcb *tp = NULL; 687 struct epoch_tracker et; 688 int error = 0; 689 690 NET_EPOCH_ENTER(et); 691 inp = sotoinpcb(so); 692 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 693 INP_WLOCK(inp); 694 if (inp->inp_flags & INP_DROPPED) { 695 INP_WUNLOCK(inp); 696 NET_EPOCH_EXIT(et); 697 return (ECONNRESET); 698 } 699 tp = intotcpcb(inp); 700 701 if (tp->t_state == TCPS_TIME_WAIT) 702 goto out; 703 tcp_disconnect(tp); 704 out: 705 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 706 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 707 INP_WUNLOCK(inp); 708 NET_EPOCH_EXIT(et); 709 return (error); 710 } 711 712 #ifdef INET 713 /* 714 * Accept a connection. Essentially all the work is done at higher levels; 715 * just return the address of the peer, storing through addr. 716 */ 717 static int 718 tcp_usr_accept(struct socket *so, struct sockaddr *sa) 719 { 720 struct inpcb *inp; 721 struct tcpcb *tp; 722 int error = 0; 723 724 inp = sotoinpcb(so); 725 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 726 INP_WLOCK(inp); 727 if (inp->inp_flags & INP_DROPPED) { 728 INP_WUNLOCK(inp); 729 return (ECONNABORTED); 730 } 731 tp = intotcpcb(inp); 732 733 if (so->so_state & SS_ISDISCONNECTED) 734 error = ECONNABORTED; 735 else 736 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 737 .sin_family = AF_INET, 738 .sin_len = sizeof(struct sockaddr_in), 739 .sin_port = inp->inp_fport, 740 .sin_addr = inp->inp_faddr, 741 }; 742 tcp_bblog_pru(tp, PRU_ACCEPT, error); 743 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 744 INP_WUNLOCK(inp); 745 746 return (error); 747 } 748 #endif /* INET */ 749 750 #ifdef INET6 751 static int 752 tcp6_usr_accept(struct socket *so, struct sockaddr *sa) 753 { 754 struct inpcb *inp; 755 struct tcpcb *tp; 756 int error = 0; 757 758 inp = sotoinpcb(so); 759 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 760 INP_WLOCK(inp); 761 if (inp->inp_flags & INP_DROPPED) { 762 INP_WUNLOCK(inp); 763 return (ECONNABORTED); 764 } 765 tp = intotcpcb(inp); 766 767 if (so->so_state & SS_ISDISCONNECTED) { 768 error = ECONNABORTED; 769 } else { 770 if (inp->inp_vflag & INP_IPV4) { 771 struct sockaddr_in sin = { 772 .sin_family = AF_INET, 773 .sin_len = sizeof(struct sockaddr_in), 774 .sin_port = inp->inp_fport, 775 .sin_addr = inp->inp_faddr, 776 }; 777 in6_sin_2_v4mapsin6(&sin, (struct sockaddr_in6 *)sa); 778 } else { 779 *(struct sockaddr_in6 *)sa = (struct sockaddr_in6 ){ 780 .sin6_family = AF_INET6, 781 .sin6_len = sizeof(struct sockaddr_in6), 782 .sin6_port = inp->inp_fport, 783 .sin6_addr = inp->in6p_faddr, 784 }; 785 /* XXX: should catch errors */ 786 (void)sa6_recoverscope((struct sockaddr_in6 *)sa); 787 } 788 } 789 790 tcp_bblog_pru(tp, PRU_ACCEPT, error); 791 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 792 INP_WUNLOCK(inp); 793 794 return (error); 795 } 796 #endif /* INET6 */ 797 798 /* 799 * Mark the connection as being incapable of further output. 800 */ 801 static int 802 tcp_usr_shutdown(struct socket *so, enum shutdown_how how) 803 { 804 struct epoch_tracker et; 805 struct inpcb *inp = sotoinpcb(so); 806 struct tcpcb *tp = intotcpcb(inp); 807 int error = 0; 808 809 SOCK_LOCK(so); 810 if ((so->so_state & 811 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 812 SOCK_UNLOCK(so); 813 return (ENOTCONN); 814 } 815 if (SOLISTENING(so)) { 816 if (how != SHUT_WR) { 817 so->so_error = ECONNABORTED; 818 solisten_wakeup(so); /* unlocks so */ 819 } else 820 SOCK_UNLOCK(so); 821 return (0); 822 } 823 SOCK_UNLOCK(so); 824 825 switch (how) { 826 case SHUT_RD: 827 socantrcvmore(so); 828 sbrelease(so, SO_RCV); 829 break; 830 case SHUT_RDWR: 831 socantrcvmore(so); 832 sbrelease(so, SO_RCV); 833 /* FALLTHROUGH */ 834 case SHUT_WR: 835 /* 836 * XXXGL: mimicing old soshutdown() here. But shouldn't we 837 * return ECONNRESEST for SHUT_RD as well? 838 */ 839 INP_WLOCK(inp); 840 if (inp->inp_flags & INP_DROPPED) { 841 INP_WUNLOCK(inp); 842 return (ECONNRESET); 843 } 844 845 socantsendmore(so); 846 NET_EPOCH_ENTER(et); 847 tcp_usrclosed(tp); 848 error = tcp_output_nodrop(tp); 849 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 850 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 851 error = tcp_unlock_or_drop(tp, error); 852 NET_EPOCH_EXIT(et); 853 } 854 wakeup(&so->so_timeo); 855 856 return (error); 857 } 858 859 /* 860 * After a receive, possibly send window update to peer. 861 */ 862 static int 863 tcp_usr_rcvd(struct socket *so, int flags) 864 { 865 struct epoch_tracker et; 866 struct inpcb *inp; 867 struct tcpcb *tp; 868 int outrv = 0, error = 0; 869 870 inp = sotoinpcb(so); 871 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 872 INP_WLOCK(inp); 873 if (inp->inp_flags & INP_DROPPED) { 874 INP_WUNLOCK(inp); 875 return (ECONNRESET); 876 } 877 tp = intotcpcb(inp); 878 879 NET_EPOCH_ENTER(et); 880 /* 881 * For passively-created TFO connections, don't attempt a window 882 * update while still in SYN_RECEIVED as this may trigger an early 883 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 884 * application response data, or failing that, when the DELACK timer 885 * expires. 886 */ 887 if (IS_FASTOPEN(tp->t_flags) && 888 (tp->t_state == TCPS_SYN_RECEIVED)) 889 goto out; 890 #ifdef TCP_OFFLOAD 891 if (tp->t_flags & TF_TOE) 892 tcp_offload_rcvd(tp); 893 else 894 #endif 895 outrv = tcp_output_nodrop(tp); 896 out: 897 tcp_bblog_pru(tp, PRU_RCVD, error); 898 TCP_PROBE2(debug__user, tp, PRU_RCVD); 899 (void) tcp_unlock_or_drop(tp, outrv); 900 NET_EPOCH_EXIT(et); 901 return (error); 902 } 903 904 /* 905 * Do a send by putting data in output queue and updating urgent 906 * marker if URG set. Possibly send more data. Unlike the other 907 * pru_*() routines, the mbuf chains are our responsibility. We 908 * must either enqueue them or free them. The other pru_* routines 909 * generally are caller-frees. 910 */ 911 static int 912 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 913 struct sockaddr *nam, struct mbuf *control, struct thread *td) 914 { 915 struct epoch_tracker et; 916 int error = 0; 917 struct inpcb *inp; 918 struct tcpcb *tp; 919 #ifdef INET 920 #ifdef INET6 921 struct sockaddr_in sin; 922 #endif 923 struct sockaddr_in *sinp; 924 #endif 925 #ifdef INET6 926 struct sockaddr_in6 *sin6; 927 int isipv6; 928 #endif 929 u_int8_t incflagsav; 930 u_char vflagsav; 931 bool restoreflags; 932 933 inp = sotoinpcb(so); 934 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 935 INP_WLOCK(inp); 936 if (inp->inp_flags & INP_DROPPED) { 937 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 938 m_freem(m); 939 INP_WUNLOCK(inp); 940 return (ECONNRESET); 941 } 942 tp = intotcpcb(inp); 943 944 vflagsav = inp->inp_vflag; 945 incflagsav = inp->inp_inc.inc_flags; 946 restoreflags = false; 947 948 NET_EPOCH_ENTER(et); 949 if (control != NULL) { 950 /* TCP doesn't do control messages (rights, creds, etc) */ 951 if (control->m_len > 0) { 952 m_freem(control); 953 error = EINVAL; 954 goto out; 955 } 956 m_freem(control); /* empty control, just free it */ 957 } 958 959 if ((flags & PRUS_OOB) != 0 && 960 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 961 goto out; 962 963 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 964 if (tp->t_state == TCPS_LISTEN) { 965 error = EINVAL; 966 goto out; 967 } 968 switch (nam->sa_family) { 969 #ifdef INET 970 case AF_INET: 971 sinp = (struct sockaddr_in *)nam; 972 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 973 error = EINVAL; 974 goto out; 975 } 976 if ((inp->inp_vflag & INP_IPV6) != 0) { 977 error = EAFNOSUPPORT; 978 goto out; 979 } 980 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 981 error = EAFNOSUPPORT; 982 goto out; 983 } 984 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 985 error = EACCES; 986 goto out; 987 } 988 if ((error = prison_remote_ip4(td->td_ucred, 989 &sinp->sin_addr))) 990 goto out; 991 #ifdef INET6 992 isipv6 = 0; 993 #endif 994 break; 995 #endif /* INET */ 996 #ifdef INET6 997 case AF_INET6: 998 sin6 = (struct sockaddr_in6 *)nam; 999 if (sin6->sin6_len != sizeof(*sin6)) { 1000 error = EINVAL; 1001 goto out; 1002 } 1003 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1004 error = EAFNOSUPPORT; 1005 goto out; 1006 } 1007 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1008 error = EAFNOSUPPORT; 1009 goto out; 1010 } 1011 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1012 #ifdef INET 1013 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1014 error = EINVAL; 1015 goto out; 1016 } 1017 if ((inp->inp_vflag & INP_IPV4) == 0) { 1018 error = EAFNOSUPPORT; 1019 goto out; 1020 } 1021 restoreflags = true; 1022 inp->inp_vflag &= ~INP_IPV6; 1023 sinp = &sin; 1024 in6_sin6_2_sin(sinp, sin6); 1025 if (IN_MULTICAST( 1026 ntohl(sinp->sin_addr.s_addr))) { 1027 error = EAFNOSUPPORT; 1028 goto out; 1029 } 1030 if ((error = prison_remote_ip4(td->td_ucred, 1031 &sinp->sin_addr))) 1032 goto out; 1033 isipv6 = 0; 1034 #else /* !INET */ 1035 error = EAFNOSUPPORT; 1036 goto out; 1037 #endif /* INET */ 1038 } else { 1039 if ((inp->inp_vflag & INP_IPV6) == 0) { 1040 error = EAFNOSUPPORT; 1041 goto out; 1042 } 1043 restoreflags = true; 1044 inp->inp_vflag &= ~INP_IPV4; 1045 inp->inp_inc.inc_flags |= INC_ISIPV6; 1046 if ((error = prison_remote_ip6(td->td_ucred, 1047 &sin6->sin6_addr))) 1048 goto out; 1049 isipv6 = 1; 1050 } 1051 break; 1052 #endif /* INET6 */ 1053 default: 1054 error = EAFNOSUPPORT; 1055 goto out; 1056 } 1057 } 1058 if (!(flags & PRUS_OOB)) { 1059 if (tp->t_acktime == 0) 1060 tp->t_acktime = ticks; 1061 sbappendstream(&so->so_snd, m, flags); 1062 m = NULL; 1063 if (nam && tp->t_state < TCPS_SYN_SENT) { 1064 KASSERT(tp->t_state == TCPS_CLOSED, 1065 ("%s: tp %p is listening", __func__, tp)); 1066 1067 /* 1068 * Do implied connect if not yet connected, 1069 * initialize window to default value, and 1070 * initialize maxseg using peer's cached MSS. 1071 */ 1072 #ifdef INET6 1073 if (isipv6) 1074 error = tcp6_connect(tp, sin6, td); 1075 #endif /* INET6 */ 1076 #if defined(INET6) && defined(INET) 1077 else 1078 #endif 1079 #ifdef INET 1080 error = tcp_connect(tp, sinp, td); 1081 #endif 1082 /* 1083 * The bind operation in tcp_connect succeeded. We 1084 * no longer want to restore the flags if later 1085 * operations fail. 1086 */ 1087 if (error == 0 || inp->inp_lport != 0) 1088 restoreflags = false; 1089 1090 if (error) { 1091 /* m is freed if PRUS_NOTREADY is unset. */ 1092 sbflush(&so->so_snd); 1093 goto out; 1094 } 1095 if (IS_FASTOPEN(tp->t_flags)) 1096 tcp_fastopen_connect(tp); 1097 else { 1098 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1099 tcp_mss(tp, -1); 1100 } 1101 } 1102 if (flags & PRUS_EOF) { 1103 /* 1104 * Close the send side of the connection after 1105 * the data is sent. 1106 */ 1107 socantsendmore(so); 1108 tcp_usrclosed(tp); 1109 } 1110 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1111 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1112 (tp->t_fbyte_out == 0) && 1113 (so->so_snd.sb_ccc > 0)) { 1114 tp->t_fbyte_out = ticks; 1115 if (tp->t_fbyte_out == 0) 1116 tp->t_fbyte_out = 1; 1117 if (tp->t_fbyte_out && tp->t_fbyte_in) 1118 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1119 } 1120 if (!(inp->inp_flags & INP_DROPPED) && 1121 !(flags & PRUS_NOTREADY)) { 1122 if (flags & PRUS_MORETOCOME) 1123 tp->t_flags |= TF_MORETOCOME; 1124 error = tcp_output_nodrop(tp); 1125 if (flags & PRUS_MORETOCOME) 1126 tp->t_flags &= ~TF_MORETOCOME; 1127 } 1128 } else { 1129 /* 1130 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1131 */ 1132 SOCKBUF_LOCK(&so->so_snd); 1133 if (sbspace(&so->so_snd) < -512) { 1134 SOCKBUF_UNLOCK(&so->so_snd); 1135 error = ENOBUFS; 1136 goto out; 1137 } 1138 /* 1139 * According to RFC961 (Assigned Protocols), 1140 * the urgent pointer points to the last octet 1141 * of urgent data. We continue, however, 1142 * to consider it to indicate the first octet 1143 * of data past the urgent section. 1144 * Otherwise, snd_up should be one lower. 1145 */ 1146 if (tp->t_acktime == 0) 1147 tp->t_acktime = ticks; 1148 sbappendstream_locked(&so->so_snd, m, flags); 1149 SOCKBUF_UNLOCK(&so->so_snd); 1150 m = NULL; 1151 if (nam && tp->t_state < TCPS_SYN_SENT) { 1152 /* 1153 * Do implied connect if not yet connected, 1154 * initialize window to default value, and 1155 * initialize maxseg using peer's cached MSS. 1156 */ 1157 1158 /* 1159 * Not going to contemplate SYN|URG 1160 */ 1161 if (IS_FASTOPEN(tp->t_flags)) 1162 tp->t_flags &= ~TF_FASTOPEN; 1163 #ifdef INET6 1164 if (isipv6) 1165 error = tcp6_connect(tp, sin6, td); 1166 #endif /* INET6 */ 1167 #if defined(INET6) && defined(INET) 1168 else 1169 #endif 1170 #ifdef INET 1171 error = tcp_connect(tp, sinp, td); 1172 #endif 1173 /* 1174 * The bind operation in tcp_connect succeeded. We 1175 * no longer want to restore the flags if later 1176 * operations fail. 1177 */ 1178 if (error == 0 || inp->inp_lport != 0) 1179 restoreflags = false; 1180 1181 if (error != 0) { 1182 /* m is freed if PRUS_NOTREADY is unset. */ 1183 sbflush(&so->so_snd); 1184 goto out; 1185 } 1186 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1187 tcp_mss(tp, -1); 1188 } 1189 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1190 if ((flags & PRUS_NOTREADY) == 0) { 1191 tp->t_flags |= TF_FORCEDATA; 1192 error = tcp_output_nodrop(tp); 1193 tp->t_flags &= ~TF_FORCEDATA; 1194 } 1195 } 1196 TCP_LOG_EVENT(tp, NULL, 1197 &inp->inp_socket->so_rcv, 1198 &inp->inp_socket->so_snd, 1199 TCP_LOG_USERSEND, error, 1200 0, NULL, false); 1201 1202 out: 1203 /* 1204 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1205 * responsible for freeing memory. 1206 */ 1207 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1208 m_freem(m); 1209 1210 /* 1211 * If the request was unsuccessful and we changed flags, 1212 * restore the original flags. 1213 */ 1214 if (error != 0 && restoreflags) { 1215 inp->inp_vflag = vflagsav; 1216 inp->inp_inc.inc_flags = incflagsav; 1217 } 1218 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1219 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1220 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1221 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1222 error = tcp_unlock_or_drop(tp, error); 1223 NET_EPOCH_EXIT(et); 1224 return (error); 1225 } 1226 1227 static int 1228 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1229 { 1230 struct epoch_tracker et; 1231 struct inpcb *inp; 1232 struct tcpcb *tp; 1233 int error; 1234 1235 inp = sotoinpcb(so); 1236 INP_WLOCK(inp); 1237 if (inp->inp_flags & INP_DROPPED) { 1238 INP_WUNLOCK(inp); 1239 mb_free_notready(m, count); 1240 return (ECONNRESET); 1241 } 1242 tp = intotcpcb(inp); 1243 1244 SOCKBUF_LOCK(&so->so_snd); 1245 error = sbready(&so->so_snd, m, count); 1246 SOCKBUF_UNLOCK(&so->so_snd); 1247 if (error) { 1248 INP_WUNLOCK(inp); 1249 return (error); 1250 } 1251 NET_EPOCH_ENTER(et); 1252 error = tcp_output_unlock(tp); 1253 NET_EPOCH_EXIT(et); 1254 1255 return (error); 1256 } 1257 1258 /* 1259 * Abort the TCP. Drop the connection abruptly. 1260 */ 1261 static void 1262 tcp_usr_abort(struct socket *so) 1263 { 1264 struct inpcb *inp; 1265 struct tcpcb *tp; 1266 struct epoch_tracker et; 1267 1268 inp = sotoinpcb(so); 1269 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1270 1271 NET_EPOCH_ENTER(et); 1272 INP_WLOCK(inp); 1273 KASSERT(inp->inp_socket != NULL, 1274 ("tcp_usr_abort: inp_socket == NULL")); 1275 1276 /* 1277 * If we still have full TCP state, and we're not dropped, drop. 1278 */ 1279 if (!(inp->inp_flags & INP_DROPPED)) { 1280 tp = intotcpcb(inp); 1281 tp = tcp_drop(tp, ECONNABORTED); 1282 if (tp == NULL) 1283 goto dropped; 1284 tcp_bblog_pru(tp, PRU_ABORT, 0); 1285 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1286 } 1287 if (!(inp->inp_flags & INP_DROPPED)) { 1288 soref(so); 1289 inp->inp_flags |= INP_SOCKREF; 1290 } 1291 INP_WUNLOCK(inp); 1292 dropped: 1293 NET_EPOCH_EXIT(et); 1294 } 1295 1296 /* 1297 * TCP socket is closed. Start friendly disconnect. 1298 */ 1299 static void 1300 tcp_usr_close(struct socket *so) 1301 { 1302 struct inpcb *inp; 1303 struct tcpcb *tp; 1304 struct epoch_tracker et; 1305 1306 inp = sotoinpcb(so); 1307 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1308 1309 NET_EPOCH_ENTER(et); 1310 INP_WLOCK(inp); 1311 KASSERT(inp->inp_socket != NULL, 1312 ("tcp_usr_close: inp_socket == NULL")); 1313 1314 /* 1315 * If we are still connected and we're not dropped, initiate 1316 * a disconnect. 1317 */ 1318 if (!(inp->inp_flags & INP_DROPPED)) { 1319 tp = intotcpcb(inp); 1320 if (tp->t_state != TCPS_TIME_WAIT) { 1321 tp->t_flags |= TF_CLOSED; 1322 tcp_disconnect(tp); 1323 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1324 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1325 } 1326 } 1327 if (!(inp->inp_flags & INP_DROPPED)) { 1328 soref(so); 1329 inp->inp_flags |= INP_SOCKREF; 1330 } 1331 INP_WUNLOCK(inp); 1332 NET_EPOCH_EXIT(et); 1333 } 1334 1335 static int 1336 tcp_pru_options_support(struct tcpcb *tp, int flags) 1337 { 1338 /* 1339 * If the specific TCP stack has a pru_options 1340 * specified then it does not always support 1341 * all the PRU_XX options and we must ask it. 1342 * If the function is not specified then all 1343 * of the PRU_XX options are supported. 1344 */ 1345 int ret = 0; 1346 1347 if (tp->t_fb->tfb_pru_options) { 1348 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1349 } 1350 return (ret); 1351 } 1352 1353 /* 1354 * Receive out-of-band data. 1355 */ 1356 static int 1357 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1358 { 1359 int error = 0; 1360 struct inpcb *inp; 1361 struct tcpcb *tp; 1362 1363 inp = sotoinpcb(so); 1364 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1365 INP_WLOCK(inp); 1366 if (inp->inp_flags & INP_DROPPED) { 1367 INP_WUNLOCK(inp); 1368 return (ECONNRESET); 1369 } 1370 tp = intotcpcb(inp); 1371 1372 error = tcp_pru_options_support(tp, PRUS_OOB); 1373 if (error) { 1374 goto out; 1375 } 1376 if ((so->so_oobmark == 0 && 1377 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1378 so->so_options & SO_OOBINLINE || 1379 tp->t_oobflags & TCPOOB_HADDATA) { 1380 error = EINVAL; 1381 goto out; 1382 } 1383 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1384 error = EWOULDBLOCK; 1385 goto out; 1386 } 1387 m->m_len = 1; 1388 *mtod(m, caddr_t) = tp->t_iobc; 1389 if ((flags & MSG_PEEK) == 0) 1390 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1391 1392 out: 1393 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1394 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1395 INP_WUNLOCK(inp); 1396 return (error); 1397 } 1398 1399 #ifdef INET 1400 struct protosw tcp_protosw = { 1401 .pr_type = SOCK_STREAM, 1402 .pr_protocol = IPPROTO_TCP, 1403 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1404 PR_CAPATTACH, 1405 .pr_ctloutput = tcp_ctloutput, 1406 .pr_abort = tcp_usr_abort, 1407 .pr_accept = tcp_usr_accept, 1408 .pr_attach = tcp_usr_attach, 1409 .pr_bind = tcp_usr_bind, 1410 .pr_connect = tcp_usr_connect, 1411 .pr_control = in_control, 1412 .pr_detach = tcp_usr_detach, 1413 .pr_disconnect = tcp_usr_disconnect, 1414 .pr_listen = tcp_usr_listen, 1415 .pr_peeraddr = in_getpeeraddr, 1416 .pr_rcvd = tcp_usr_rcvd, 1417 .pr_rcvoob = tcp_usr_rcvoob, 1418 .pr_send = tcp_usr_send, 1419 .pr_ready = tcp_usr_ready, 1420 .pr_shutdown = tcp_usr_shutdown, 1421 .pr_sockaddr = in_getsockaddr, 1422 .pr_sosetlabel = in_pcbsosetlabel, 1423 .pr_close = tcp_usr_close, 1424 }; 1425 #endif /* INET */ 1426 1427 #ifdef INET6 1428 struct protosw tcp6_protosw = { 1429 .pr_type = SOCK_STREAM, 1430 .pr_protocol = IPPROTO_TCP, 1431 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1432 PR_CAPATTACH, 1433 .pr_ctloutput = tcp_ctloutput, 1434 .pr_abort = tcp_usr_abort, 1435 .pr_accept = tcp6_usr_accept, 1436 .pr_attach = tcp_usr_attach, 1437 .pr_bind = tcp6_usr_bind, 1438 .pr_connect = tcp6_usr_connect, 1439 .pr_control = in6_control, 1440 .pr_detach = tcp_usr_detach, 1441 .pr_disconnect = tcp_usr_disconnect, 1442 .pr_listen = tcp6_usr_listen, 1443 .pr_peeraddr = in6_mapped_peeraddr, 1444 .pr_rcvd = tcp_usr_rcvd, 1445 .pr_rcvoob = tcp_usr_rcvoob, 1446 .pr_send = tcp_usr_send, 1447 .pr_ready = tcp_usr_ready, 1448 .pr_shutdown = tcp_usr_shutdown, 1449 .pr_sockaddr = in6_mapped_sockaddr, 1450 .pr_sosetlabel = in_pcbsosetlabel, 1451 .pr_close = tcp_usr_close, 1452 }; 1453 #endif /* INET6 */ 1454 1455 #ifdef INET 1456 /* 1457 * Common subroutine to open a TCP connection to remote host specified 1458 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1459 * and assign a local port number and install the inpcb into the hash. 1460 * Initialize connection parameters and enter SYN-SENT state. 1461 */ 1462 static int 1463 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1464 { 1465 struct inpcb *inp = tptoinpcb(tp); 1466 struct socket *so = tptosocket(tp); 1467 int error; 1468 1469 NET_EPOCH_ASSERT(); 1470 INP_WLOCK_ASSERT(inp); 1471 1472 if (__predict_false((so->so_state & 1473 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1474 SS_ISDISCONNECTED)) != 0)) 1475 return (EISCONN); 1476 1477 INP_HASH_WLOCK(&V_tcbinfo); 1478 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1479 INP_HASH_WUNLOCK(&V_tcbinfo); 1480 if (error != 0) 1481 return (error); 1482 1483 /* 1484 * Compute window scaling to request: 1485 * Scale to fit into sweet spot. See tcp_syncache.c. 1486 * XXX: This should move to tcp_output(). 1487 */ 1488 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1489 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1490 tp->request_r_scale++; 1491 1492 soisconnecting(so); 1493 TCPSTAT_INC(tcps_connattempt); 1494 tcp_state_change(tp, TCPS_SYN_SENT); 1495 tp->iss = tcp_new_isn(&inp->inp_inc); 1496 if (tp->t_flags & TF_REQ_TSTMP) 1497 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1498 tcp_sendseqinit(tp); 1499 1500 return (0); 1501 } 1502 #endif /* INET */ 1503 1504 #ifdef INET6 1505 static int 1506 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1507 { 1508 struct inpcb *inp = tptoinpcb(tp); 1509 struct socket *so = tptosocket(tp); 1510 int error; 1511 1512 NET_EPOCH_ASSERT(); 1513 INP_WLOCK_ASSERT(inp); 1514 1515 if (__predict_false((so->so_state & 1516 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1517 return (EISCONN); 1518 1519 INP_HASH_WLOCK(&V_tcbinfo); 1520 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1521 INP_HASH_WUNLOCK(&V_tcbinfo); 1522 if (error != 0) 1523 return (error); 1524 1525 /* Compute window scaling to request. */ 1526 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1527 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1528 tp->request_r_scale++; 1529 1530 soisconnecting(so); 1531 TCPSTAT_INC(tcps_connattempt); 1532 tcp_state_change(tp, TCPS_SYN_SENT); 1533 tp->iss = tcp_new_isn(&inp->inp_inc); 1534 if (tp->t_flags & TF_REQ_TSTMP) 1535 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1536 tcp_sendseqinit(tp); 1537 1538 return (0); 1539 } 1540 #endif /* INET6 */ 1541 1542 /* 1543 * Export TCP internal state information via a struct tcp_info, based on the 1544 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1545 * (TCP state machine, etc). We export all information using FreeBSD-native 1546 * constants -- for example, the numeric values for tcpi_state will differ 1547 * from Linux. 1548 */ 1549 void 1550 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1551 { 1552 1553 INP_LOCK_ASSERT(tptoinpcb(tp)); 1554 bzero(ti, sizeof(*ti)); 1555 1556 ti->tcpi_state = tp->t_state; 1557 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1558 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1559 if (tp->t_flags & TF_SACK_PERMIT) 1560 ti->tcpi_options |= TCPI_OPT_SACK; 1561 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1562 ti->tcpi_options |= TCPI_OPT_WSCALE; 1563 ti->tcpi_snd_wscale = tp->snd_scale; 1564 ti->tcpi_rcv_wscale = tp->rcv_scale; 1565 } 1566 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1567 case TF2_ECN_PERMIT: 1568 ti->tcpi_options |= TCPI_OPT_ECN; 1569 break; 1570 case TF2_ACE_PERMIT: 1571 /* FALLTHROUGH */ 1572 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1573 ti->tcpi_options |= TCPI_OPT_ACE; 1574 break; 1575 default: 1576 break; 1577 } 1578 if (IS_FASTOPEN(tp->t_flags)) 1579 ti->tcpi_options |= TCPI_OPT_TFO; 1580 1581 ti->tcpi_rto = tp->t_rxtcur * tick; 1582 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1583 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1584 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1585 1586 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1587 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1588 1589 /* 1590 * FreeBSD-specific extension fields for tcp_info. 1591 */ 1592 ti->tcpi_rcv_space = tp->rcv_wnd; 1593 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1594 ti->tcpi_snd_wnd = tp->snd_wnd; 1595 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1596 ti->tcpi_snd_nxt = tp->snd_nxt; 1597 ti->tcpi_snd_mss = tp->t_maxseg; 1598 ti->tcpi_rcv_mss = tp->t_maxseg; 1599 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1600 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1601 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1602 ti->tcpi_snd_una = tp->snd_una; 1603 ti->tcpi_snd_max = tp->snd_max; 1604 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1605 ti->tcpi_rcv_adv = tp->rcv_adv; 1606 ti->tcpi_dupacks = tp->t_dupacks; 1607 #ifdef TCP_OFFLOAD 1608 if (tp->t_flags & TF_TOE) { 1609 ti->tcpi_options |= TCPI_OPT_TOE; 1610 tcp_offload_tcp_info(tp, ti); 1611 } 1612 #endif 1613 /* 1614 * AccECN related counters. 1615 */ 1616 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1617 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1618 /* 1619 * Internal counter starts at 5 for AccECN 1620 * but 0 for RFC3168 ECN. 1621 */ 1622 ti->tcpi_delivered_ce = tp->t_scep - 5; 1623 else 1624 ti->tcpi_delivered_ce = tp->t_scep; 1625 ti->tcpi_received_ce = tp->t_rcep; 1626 } 1627 1628 /* 1629 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1630 * socket option arguments. When it re-acquires the lock after the copy, it 1631 * has to revalidate that the connection is still valid for the socket 1632 * option. 1633 */ 1634 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1635 INP_WLOCK(inp); \ 1636 if (inp->inp_flags & INP_DROPPED) { \ 1637 INP_WUNLOCK(inp); \ 1638 cleanup; \ 1639 return (ECONNRESET); \ 1640 } \ 1641 tp = intotcpcb(inp); \ 1642 } while(0) 1643 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1644 1645 int 1646 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1647 { 1648 struct socket *so = inp->inp_socket; 1649 struct tcpcb *tp = intotcpcb(inp); 1650 int error = 0; 1651 1652 MPASS(sopt->sopt_dir == SOPT_SET); 1653 INP_WLOCK_ASSERT(inp); 1654 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1655 ("inp_flags == %x", inp->inp_flags)); 1656 KASSERT(so != NULL, ("inp_socket == NULL")); 1657 1658 if (sopt->sopt_level != IPPROTO_TCP) { 1659 INP_WUNLOCK(inp); 1660 #ifdef INET6 1661 if (inp->inp_vflag & INP_IPV6PROTO) 1662 error = ip6_ctloutput(so, sopt); 1663 #endif 1664 #if defined(INET6) && defined(INET) 1665 else 1666 #endif 1667 #ifdef INET 1668 error = ip_ctloutput(so, sopt); 1669 #endif 1670 /* 1671 * When an IP-level socket option affects TCP, pass control 1672 * down to stack tfb_tcp_ctloutput, otherwise return what 1673 * IP level returned. 1674 */ 1675 switch (sopt->sopt_level) { 1676 #ifdef INET6 1677 case IPPROTO_IPV6: 1678 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1679 return (error); 1680 switch (sopt->sopt_name) { 1681 case IPV6_TCLASS: 1682 /* Notify tcp stacks that care (e.g. RACK). */ 1683 break; 1684 case IPV6_USE_MIN_MTU: 1685 /* Update t_maxseg accordingly. */ 1686 break; 1687 default: 1688 return (error); 1689 } 1690 break; 1691 #endif 1692 #ifdef INET 1693 case IPPROTO_IP: 1694 switch (sopt->sopt_name) { 1695 case IP_TOS: 1696 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1697 break; 1698 case IP_TTL: 1699 /* Notify tcp stacks that care (e.g. RACK). */ 1700 break; 1701 default: 1702 return (error); 1703 } 1704 break; 1705 #endif 1706 default: 1707 return (error); 1708 } 1709 INP_WLOCK(inp); 1710 if (inp->inp_flags & INP_DROPPED) { 1711 INP_WUNLOCK(inp); 1712 return (ECONNRESET); 1713 } 1714 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1715 /* 1716 * Protect the TCP option TCP_FUNCTION_BLK so 1717 * that a sub-function can *never* overwrite this. 1718 */ 1719 struct tcp_function_set fsn; 1720 struct tcp_function_block *blk; 1721 void *ptr = NULL; 1722 1723 INP_WUNLOCK(inp); 1724 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1725 if (error) 1726 return (error); 1727 1728 INP_WLOCK(inp); 1729 tp = intotcpcb(inp); 1730 1731 blk = find_and_ref_tcp_functions(&fsn); 1732 if (blk == NULL) { 1733 INP_WUNLOCK(inp); 1734 return (ENOENT); 1735 } 1736 if (tp->t_fb == blk) { 1737 /* You already have this */ 1738 refcount_release(&blk->tfb_refcnt); 1739 INP_WUNLOCK(inp); 1740 return (0); 1741 } 1742 if (tp->t_state != TCPS_CLOSED) { 1743 /* 1744 * The user has advanced the state 1745 * past the initial point, we may not 1746 * be able to switch. 1747 */ 1748 if (blk->tfb_tcp_handoff_ok != NULL) { 1749 /* 1750 * Does the stack provide a 1751 * query mechanism, if so it may 1752 * still be possible? 1753 */ 1754 error = (*blk->tfb_tcp_handoff_ok)(tp); 1755 } else 1756 error = EINVAL; 1757 if (error) { 1758 refcount_release(&blk->tfb_refcnt); 1759 INP_WUNLOCK(inp); 1760 return(error); 1761 } 1762 } 1763 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1764 refcount_release(&blk->tfb_refcnt); 1765 INP_WUNLOCK(inp); 1766 return (ENOENT); 1767 } 1768 /* 1769 * Ensure the new stack takes ownership with a 1770 * clean slate on peak rate threshold. 1771 */ 1772 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 1773 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1774 if (blk->tfb_tcp_fb_init) { 1775 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1776 if (error) { 1777 /* 1778 * Release the ref count the lookup 1779 * acquired. 1780 */ 1781 refcount_release(&blk->tfb_refcnt); 1782 /* 1783 * Now there is a chance that the 1784 * init() function mucked with some 1785 * things before it failed, such as 1786 * hpts or inp_flags2 or timer granularity. 1787 * It should not of, but lets give the old 1788 * stack a chance to reset to a known good state. 1789 */ 1790 if (tp->t_fb->tfb_switch_failed) { 1791 (*tp->t_fb->tfb_switch_failed)(tp); 1792 } 1793 goto err_out; 1794 } 1795 } 1796 if (tp->t_fb->tfb_tcp_fb_fini) { 1797 struct epoch_tracker et; 1798 /* 1799 * Tell the stack to cleanup with 0 i.e. 1800 * the tcb is not going away. 1801 */ 1802 NET_EPOCH_ENTER(et); 1803 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1804 NET_EPOCH_EXIT(et); 1805 } 1806 /* 1807 * Release the old refcnt, the 1808 * lookup acquired a ref on the 1809 * new one already. 1810 */ 1811 refcount_release(&tp->t_fb->tfb_refcnt); 1812 /* 1813 * Set in the new stack. 1814 */ 1815 tp->t_fb = blk; 1816 tp->t_fb_ptr = ptr; 1817 #ifdef TCP_OFFLOAD 1818 if (tp->t_flags & TF_TOE) { 1819 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1820 sopt->sopt_name); 1821 } 1822 #endif 1823 err_out: 1824 INP_WUNLOCK(inp); 1825 return (error); 1826 1827 } 1828 1829 /* Pass in the INP locked, callee must unlock it. */ 1830 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1831 } 1832 1833 static int 1834 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1835 { 1836 struct socket *so = inp->inp_socket; 1837 struct tcpcb *tp = intotcpcb(inp); 1838 int error = 0; 1839 1840 MPASS(sopt->sopt_dir == SOPT_GET); 1841 INP_WLOCK_ASSERT(inp); 1842 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1843 ("inp_flags == %x", inp->inp_flags)); 1844 KASSERT(so != NULL, ("inp_socket == NULL")); 1845 1846 if (sopt->sopt_level != IPPROTO_TCP) { 1847 INP_WUNLOCK(inp); 1848 #ifdef INET6 1849 if (inp->inp_vflag & INP_IPV6PROTO) 1850 error = ip6_ctloutput(so, sopt); 1851 #endif /* INET6 */ 1852 #if defined(INET6) && defined(INET) 1853 else 1854 #endif 1855 #ifdef INET 1856 error = ip_ctloutput(so, sopt); 1857 #endif 1858 return (error); 1859 } 1860 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1861 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1862 struct tcp_function_set fsn; 1863 1864 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1865 memset(&fsn, 0, sizeof(fsn)); 1866 find_tcp_function_alias(tp->t_fb, &fsn); 1867 } else { 1868 strncpy(fsn.function_set_name, 1869 tp->t_fb->tfb_tcp_block_name, 1870 TCP_FUNCTION_NAME_LEN_MAX); 1871 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1872 } 1873 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1874 INP_WUNLOCK(inp); 1875 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1876 return (error); 1877 } 1878 1879 /* Pass in the INP locked, callee must unlock it. */ 1880 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1881 } 1882 1883 int 1884 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1885 { 1886 struct inpcb *inp; 1887 1888 inp = sotoinpcb(so); 1889 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1890 1891 INP_WLOCK(inp); 1892 if (inp->inp_flags & INP_DROPPED) { 1893 INP_WUNLOCK(inp); 1894 return (ECONNRESET); 1895 } 1896 if (sopt->sopt_dir == SOPT_SET) 1897 return (tcp_ctloutput_set(inp, sopt)); 1898 else if (sopt->sopt_dir == SOPT_GET) 1899 return (tcp_ctloutput_get(inp, sopt)); 1900 else 1901 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1902 } 1903 1904 /* 1905 * If this assert becomes untrue, we need to change the size of the buf 1906 * variable in tcp_default_ctloutput(). 1907 */ 1908 #ifdef CTASSERT 1909 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1910 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1911 #endif 1912 1913 #ifdef KERN_TLS 1914 static int 1915 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1916 { 1917 struct tls_enable_v0 tls_v0; 1918 int error; 1919 1920 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1921 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1922 sizeof(tls_v0)); 1923 if (error) 1924 return (error); 1925 memset(tls, 0, sizeof(*tls)); 1926 tls->cipher_key = tls_v0.cipher_key; 1927 tls->iv = tls_v0.iv; 1928 tls->auth_key = tls_v0.auth_key; 1929 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1930 tls->cipher_key_len = tls_v0.cipher_key_len; 1931 tls->iv_len = tls_v0.iv_len; 1932 tls->auth_algorithm = tls_v0.auth_algorithm; 1933 tls->auth_key_len = tls_v0.auth_key_len; 1934 tls->flags = tls_v0.flags; 1935 tls->tls_vmajor = tls_v0.tls_vmajor; 1936 tls->tls_vminor = tls_v0.tls_vminor; 1937 return (0); 1938 } 1939 1940 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1941 } 1942 #endif 1943 1944 extern struct cc_algo newreno_cc_algo; 1945 1946 static int 1947 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1948 { 1949 struct cc_algo *algo; 1950 void *ptr = NULL; 1951 struct tcpcb *tp; 1952 struct cc_var cc_mem; 1953 char buf[TCP_CA_NAME_MAX]; 1954 size_t mem_sz; 1955 int error; 1956 1957 INP_WUNLOCK(inp); 1958 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1959 if (error) 1960 return(error); 1961 buf[sopt->sopt_valsize] = '\0'; 1962 CC_LIST_RLOCK(); 1963 STAILQ_FOREACH(algo, &cc_list, entries) { 1964 if (strncmp(buf, algo->name, 1965 TCP_CA_NAME_MAX) == 0) { 1966 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1967 /* We can't "see" modules being unloaded */ 1968 continue; 1969 } 1970 break; 1971 } 1972 } 1973 if (algo == NULL) { 1974 CC_LIST_RUNLOCK(); 1975 return(ESRCH); 1976 } 1977 /* 1978 * With a reference the algorithm cannot be removed 1979 * so we hold a reference through the change process. 1980 */ 1981 cc_refer(algo); 1982 CC_LIST_RUNLOCK(); 1983 if (algo->cb_init != NULL) { 1984 /* We can now pre-get the memory for the CC */ 1985 mem_sz = (*algo->cc_data_sz)(); 1986 if (mem_sz == 0) { 1987 goto no_mem_needed; 1988 } 1989 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1990 } else { 1991 no_mem_needed: 1992 mem_sz = 0; 1993 ptr = NULL; 1994 } 1995 /* 1996 * Make sure its all clean and zero and also get 1997 * back the inplock. 1998 */ 1999 memset(&cc_mem, 0, sizeof(cc_mem)); 2000 INP_WLOCK(inp); 2001 if (inp->inp_flags & INP_DROPPED) { 2002 INP_WUNLOCK(inp); 2003 if (ptr) 2004 free(ptr, M_CC_MEM); 2005 /* Release our temp reference */ 2006 CC_LIST_RLOCK(); 2007 cc_release(algo); 2008 CC_LIST_RUNLOCK(); 2009 return (ECONNRESET); 2010 } 2011 tp = intotcpcb(inp); 2012 if (ptr != NULL) 2013 memset(ptr, 0, mem_sz); 2014 cc_mem.ccvc.tcp = tp; 2015 /* 2016 * We once again hold a write lock over the tcb so it's 2017 * safe to do these things without ordering concerns. 2018 * Note here we init into stack memory. 2019 */ 2020 if (algo->cb_init != NULL) 2021 error = algo->cb_init(&cc_mem, ptr); 2022 else 2023 error = 0; 2024 /* 2025 * The CC algorithms, when given their memory 2026 * should not fail we could in theory have a 2027 * KASSERT here. 2028 */ 2029 if (error == 0) { 2030 /* 2031 * Touchdown, lets go ahead and move the 2032 * connection to the new CC module by 2033 * copying in the cc_mem after we call 2034 * the old ones cleanup (if any). 2035 */ 2036 if (CC_ALGO(tp)->cb_destroy != NULL) 2037 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2038 /* Detach the old CC from the tcpcb */ 2039 cc_detach(tp); 2040 /* Copy in our temp memory that was inited */ 2041 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2042 /* Now attach the new, which takes a reference */ 2043 cc_attach(tp, algo); 2044 /* Ok now are we where we have gotten past any conn_init? */ 2045 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2046 /* Yep run the connection init for the new CC */ 2047 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2048 } 2049 } else if (ptr) 2050 free(ptr, M_CC_MEM); 2051 INP_WUNLOCK(inp); 2052 /* Now lets release our temp reference */ 2053 CC_LIST_RLOCK(); 2054 cc_release(algo); 2055 CC_LIST_RUNLOCK(); 2056 return (error); 2057 } 2058 2059 int 2060 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2061 { 2062 struct inpcb *inp = tptoinpcb(tp); 2063 int error, opt, optval; 2064 u_int ui; 2065 struct tcp_info ti; 2066 #ifdef KERN_TLS 2067 struct tls_enable tls; 2068 struct socket *so = inp->inp_socket; 2069 #endif 2070 char *pbuf, buf[TCP_LOG_ID_LEN]; 2071 #ifdef STATS 2072 struct statsblob *sbp; 2073 #endif 2074 size_t len; 2075 2076 INP_WLOCK_ASSERT(inp); 2077 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2078 ("inp_flags == %x", inp->inp_flags)); 2079 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2080 2081 switch (sopt->sopt_level) { 2082 #ifdef INET6 2083 case IPPROTO_IPV6: 2084 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2085 switch (sopt->sopt_name) { 2086 case IPV6_USE_MIN_MTU: 2087 tcp6_use_min_mtu(tp); 2088 /* FALLTHROUGH */ 2089 } 2090 INP_WUNLOCK(inp); 2091 return (0); 2092 #endif 2093 #ifdef INET 2094 case IPPROTO_IP: 2095 INP_WUNLOCK(inp); 2096 return (0); 2097 #endif 2098 } 2099 2100 /* 2101 * For TCP_CCALGOOPT forward the control to CC module, for both 2102 * SOPT_SET and SOPT_GET. 2103 */ 2104 switch (sopt->sopt_name) { 2105 case TCP_CCALGOOPT: 2106 INP_WUNLOCK(inp); 2107 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2108 return (EINVAL); 2109 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2110 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2111 sopt->sopt_valsize); 2112 if (error) { 2113 free(pbuf, M_TEMP); 2114 return (error); 2115 } 2116 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2117 if (CC_ALGO(tp)->ctl_output != NULL) 2118 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2119 else 2120 error = ENOENT; 2121 INP_WUNLOCK(inp); 2122 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2123 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2124 free(pbuf, M_TEMP); 2125 return (error); 2126 } 2127 2128 switch (sopt->sopt_dir) { 2129 case SOPT_SET: 2130 switch (sopt->sopt_name) { 2131 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2132 case TCP_MD5SIG: 2133 INP_WUNLOCK(inp); 2134 if (!TCPMD5_ENABLED()) 2135 return (ENOPROTOOPT); 2136 error = TCPMD5_PCBCTL(inp, sopt); 2137 if (error) 2138 return (error); 2139 INP_WLOCK_RECHECK(inp); 2140 goto unlock_and_done; 2141 #endif /* IPSEC */ 2142 2143 case TCP_NODELAY: 2144 case TCP_NOOPT: 2145 INP_WUNLOCK(inp); 2146 error = sooptcopyin(sopt, &optval, sizeof optval, 2147 sizeof optval); 2148 if (error) 2149 return (error); 2150 2151 INP_WLOCK_RECHECK(inp); 2152 switch (sopt->sopt_name) { 2153 case TCP_NODELAY: 2154 opt = TF_NODELAY; 2155 break; 2156 case TCP_NOOPT: 2157 opt = TF_NOOPT; 2158 break; 2159 default: 2160 opt = 0; /* dead code to fool gcc */ 2161 break; 2162 } 2163 2164 if (optval) 2165 tp->t_flags |= opt; 2166 else 2167 tp->t_flags &= ~opt; 2168 unlock_and_done: 2169 #ifdef TCP_OFFLOAD 2170 if (tp->t_flags & TF_TOE) { 2171 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2172 sopt->sopt_name); 2173 } 2174 #endif 2175 INP_WUNLOCK(inp); 2176 break; 2177 2178 case TCP_NOPUSH: 2179 INP_WUNLOCK(inp); 2180 error = sooptcopyin(sopt, &optval, sizeof optval, 2181 sizeof optval); 2182 if (error) 2183 return (error); 2184 2185 INP_WLOCK_RECHECK(inp); 2186 if (optval) 2187 tp->t_flags |= TF_NOPUSH; 2188 else if (tp->t_flags & TF_NOPUSH) { 2189 tp->t_flags &= ~TF_NOPUSH; 2190 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2191 struct epoch_tracker et; 2192 2193 NET_EPOCH_ENTER(et); 2194 error = tcp_output_nodrop(tp); 2195 NET_EPOCH_EXIT(et); 2196 } 2197 } 2198 goto unlock_and_done; 2199 2200 case TCP_REMOTE_UDP_ENCAPS_PORT: 2201 INP_WUNLOCK(inp); 2202 error = sooptcopyin(sopt, &optval, sizeof optval, 2203 sizeof optval); 2204 if (error) 2205 return (error); 2206 if ((optval < TCP_TUNNELING_PORT_MIN) || 2207 (optval > TCP_TUNNELING_PORT_MAX)) { 2208 /* Its got to be in range */ 2209 return (EINVAL); 2210 } 2211 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2212 /* You have to have enabled a UDP tunneling port first */ 2213 return (EINVAL); 2214 } 2215 INP_WLOCK_RECHECK(inp); 2216 if (tp->t_state != TCPS_CLOSED) { 2217 /* You can't change after you are connected */ 2218 error = EINVAL; 2219 } else { 2220 /* Ok we are all good set the port */ 2221 tp->t_port = htons(optval); 2222 } 2223 goto unlock_and_done; 2224 2225 case TCP_MAXSEG: 2226 INP_WUNLOCK(inp); 2227 error = sooptcopyin(sopt, &optval, sizeof optval, 2228 sizeof optval); 2229 if (error) 2230 return (error); 2231 2232 INP_WLOCK_RECHECK(inp); 2233 if (optval > 0 && optval <= tp->t_maxseg && 2234 optval + 40 >= V_tcp_minmss) 2235 tp->t_maxseg = optval; 2236 else 2237 error = EINVAL; 2238 goto unlock_and_done; 2239 2240 case TCP_INFO: 2241 INP_WUNLOCK(inp); 2242 error = EINVAL; 2243 break; 2244 2245 case TCP_STATS: 2246 INP_WUNLOCK(inp); 2247 #ifdef STATS 2248 error = sooptcopyin(sopt, &optval, sizeof optval, 2249 sizeof optval); 2250 if (error) 2251 return (error); 2252 2253 if (optval > 0) 2254 sbp = stats_blob_alloc( 2255 V_tcp_perconn_stats_dflt_tpl, 0); 2256 else 2257 sbp = NULL; 2258 2259 INP_WLOCK_RECHECK(inp); 2260 if ((tp->t_stats != NULL && sbp == NULL) || 2261 (tp->t_stats == NULL && sbp != NULL)) { 2262 struct statsblob *t = tp->t_stats; 2263 tp->t_stats = sbp; 2264 sbp = t; 2265 } 2266 INP_WUNLOCK(inp); 2267 2268 stats_blob_destroy(sbp); 2269 #else 2270 return (EOPNOTSUPP); 2271 #endif /* !STATS */ 2272 break; 2273 2274 case TCP_CONGESTION: 2275 error = tcp_set_cc_mod(inp, sopt); 2276 break; 2277 2278 case TCP_REUSPORT_LB_NUMA: 2279 INP_WUNLOCK(inp); 2280 error = sooptcopyin(sopt, &optval, sizeof(optval), 2281 sizeof(optval)); 2282 INP_WLOCK_RECHECK(inp); 2283 if (!error) 2284 error = in_pcblbgroup_numa(inp, optval); 2285 INP_WUNLOCK(inp); 2286 break; 2287 2288 #ifdef KERN_TLS 2289 case TCP_TXTLS_ENABLE: 2290 INP_WUNLOCK(inp); 2291 error = copyin_tls_enable(sopt, &tls); 2292 if (error) 2293 break; 2294 error = ktls_enable_tx(so, &tls); 2295 break; 2296 case TCP_TXTLS_MODE: 2297 INP_WUNLOCK(inp); 2298 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2299 if (error) 2300 return (error); 2301 2302 INP_WLOCK_RECHECK(inp); 2303 error = ktls_set_tx_mode(so, ui); 2304 INP_WUNLOCK(inp); 2305 break; 2306 case TCP_RXTLS_ENABLE: 2307 INP_WUNLOCK(inp); 2308 error = sooptcopyin(sopt, &tls, sizeof(tls), 2309 sizeof(tls)); 2310 if (error) 2311 break; 2312 error = ktls_enable_rx(so, &tls); 2313 break; 2314 #endif 2315 case TCP_MAXUNACKTIME: 2316 case TCP_KEEPIDLE: 2317 case TCP_KEEPINTVL: 2318 case TCP_KEEPINIT: 2319 INP_WUNLOCK(inp); 2320 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2321 if (error) 2322 return (error); 2323 2324 if (ui > (UINT_MAX / hz)) { 2325 error = EINVAL; 2326 break; 2327 } 2328 ui *= hz; 2329 2330 INP_WLOCK_RECHECK(inp); 2331 switch (sopt->sopt_name) { 2332 case TCP_MAXUNACKTIME: 2333 tp->t_maxunacktime = ui; 2334 break; 2335 2336 case TCP_KEEPIDLE: 2337 tp->t_keepidle = ui; 2338 /* 2339 * XXX: better check current remaining 2340 * timeout and "merge" it with new value. 2341 */ 2342 if ((tp->t_state > TCPS_LISTEN) && 2343 (tp->t_state <= TCPS_CLOSING)) 2344 tcp_timer_activate(tp, TT_KEEP, 2345 TP_KEEPIDLE(tp)); 2346 break; 2347 case TCP_KEEPINTVL: 2348 tp->t_keepintvl = ui; 2349 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2350 (TP_MAXIDLE(tp) > 0)) 2351 tcp_timer_activate(tp, TT_2MSL, 2352 TP_MAXIDLE(tp)); 2353 break; 2354 case TCP_KEEPINIT: 2355 tp->t_keepinit = ui; 2356 if (tp->t_state == TCPS_SYN_RECEIVED || 2357 tp->t_state == TCPS_SYN_SENT) 2358 tcp_timer_activate(tp, TT_KEEP, 2359 TP_KEEPINIT(tp)); 2360 break; 2361 } 2362 goto unlock_and_done; 2363 2364 case TCP_KEEPCNT: 2365 INP_WUNLOCK(inp); 2366 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2367 if (error) 2368 return (error); 2369 2370 INP_WLOCK_RECHECK(inp); 2371 tp->t_keepcnt = ui; 2372 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2373 (TP_MAXIDLE(tp) > 0)) 2374 tcp_timer_activate(tp, TT_2MSL, 2375 TP_MAXIDLE(tp)); 2376 goto unlock_and_done; 2377 2378 #ifdef TCPPCAP 2379 case TCP_PCAP_OUT: 2380 case TCP_PCAP_IN: 2381 INP_WUNLOCK(inp); 2382 error = sooptcopyin(sopt, &optval, sizeof optval, 2383 sizeof optval); 2384 if (error) 2385 return (error); 2386 2387 INP_WLOCK_RECHECK(inp); 2388 if (optval >= 0) 2389 tcp_pcap_set_sock_max( 2390 (sopt->sopt_name == TCP_PCAP_OUT) ? 2391 &(tp->t_outpkts) : &(tp->t_inpkts), 2392 optval); 2393 else 2394 error = EINVAL; 2395 goto unlock_and_done; 2396 #endif 2397 2398 case TCP_FASTOPEN: { 2399 struct tcp_fastopen tfo_optval; 2400 2401 INP_WUNLOCK(inp); 2402 if (!V_tcp_fastopen_client_enable && 2403 !V_tcp_fastopen_server_enable) 2404 return (EPERM); 2405 2406 error = sooptcopyin(sopt, &tfo_optval, 2407 sizeof(tfo_optval), sizeof(int)); 2408 if (error) 2409 return (error); 2410 2411 INP_WLOCK_RECHECK(inp); 2412 if ((tp->t_state != TCPS_CLOSED) && 2413 (tp->t_state != TCPS_LISTEN)) { 2414 error = EINVAL; 2415 goto unlock_and_done; 2416 } 2417 if (tfo_optval.enable) { 2418 if (tp->t_state == TCPS_LISTEN) { 2419 if (!V_tcp_fastopen_server_enable) { 2420 error = EPERM; 2421 goto unlock_and_done; 2422 } 2423 2424 if (tp->t_tfo_pending == NULL) 2425 tp->t_tfo_pending = 2426 tcp_fastopen_alloc_counter(); 2427 } else { 2428 /* 2429 * If a pre-shared key was provided, 2430 * stash it in the client cookie 2431 * field of the tcpcb for use during 2432 * connect. 2433 */ 2434 if (sopt->sopt_valsize == 2435 sizeof(tfo_optval)) { 2436 memcpy(tp->t_tfo_cookie.client, 2437 tfo_optval.psk, 2438 TCP_FASTOPEN_PSK_LEN); 2439 tp->t_tfo_client_cookie_len = 2440 TCP_FASTOPEN_PSK_LEN; 2441 } 2442 } 2443 tp->t_flags |= TF_FASTOPEN; 2444 } else 2445 tp->t_flags &= ~TF_FASTOPEN; 2446 goto unlock_and_done; 2447 } 2448 2449 #ifdef TCP_BLACKBOX 2450 case TCP_LOG: 2451 INP_WUNLOCK(inp); 2452 error = sooptcopyin(sopt, &optval, sizeof optval, 2453 sizeof optval); 2454 if (error) 2455 return (error); 2456 2457 INP_WLOCK_RECHECK(inp); 2458 error = tcp_log_state_change(tp, optval); 2459 goto unlock_and_done; 2460 2461 case TCP_LOGBUF: 2462 INP_WUNLOCK(inp); 2463 error = EINVAL; 2464 break; 2465 2466 case TCP_LOGID: 2467 INP_WUNLOCK(inp); 2468 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2469 if (error) 2470 break; 2471 buf[sopt->sopt_valsize] = '\0'; 2472 INP_WLOCK_RECHECK(inp); 2473 error = tcp_log_set_id(tp, buf); 2474 /* tcp_log_set_id() unlocks the INP. */ 2475 break; 2476 2477 case TCP_LOGDUMP: 2478 case TCP_LOGDUMPID: 2479 INP_WUNLOCK(inp); 2480 error = 2481 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2482 if (error) 2483 break; 2484 buf[sopt->sopt_valsize] = '\0'; 2485 INP_WLOCK_RECHECK(inp); 2486 if (sopt->sopt_name == TCP_LOGDUMP) { 2487 error = tcp_log_dump_tp_logbuf(tp, buf, 2488 M_WAITOK, true); 2489 INP_WUNLOCK(inp); 2490 } else { 2491 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2492 /* 2493 * tcp_log_dump_tp_bucket_logbufs() drops the 2494 * INP lock. 2495 */ 2496 } 2497 break; 2498 #endif 2499 2500 default: 2501 INP_WUNLOCK(inp); 2502 error = ENOPROTOOPT; 2503 break; 2504 } 2505 break; 2506 2507 case SOPT_GET: 2508 tp = intotcpcb(inp); 2509 switch (sopt->sopt_name) { 2510 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2511 case TCP_MD5SIG: 2512 INP_WUNLOCK(inp); 2513 if (!TCPMD5_ENABLED()) 2514 return (ENOPROTOOPT); 2515 error = TCPMD5_PCBCTL(inp, sopt); 2516 break; 2517 #endif 2518 2519 case TCP_NODELAY: 2520 optval = tp->t_flags & TF_NODELAY; 2521 INP_WUNLOCK(inp); 2522 error = sooptcopyout(sopt, &optval, sizeof optval); 2523 break; 2524 case TCP_MAXSEG: 2525 optval = tp->t_maxseg; 2526 INP_WUNLOCK(inp); 2527 error = sooptcopyout(sopt, &optval, sizeof optval); 2528 break; 2529 case TCP_REMOTE_UDP_ENCAPS_PORT: 2530 optval = ntohs(tp->t_port); 2531 INP_WUNLOCK(inp); 2532 error = sooptcopyout(sopt, &optval, sizeof optval); 2533 break; 2534 case TCP_NOOPT: 2535 optval = tp->t_flags & TF_NOOPT; 2536 INP_WUNLOCK(inp); 2537 error = sooptcopyout(sopt, &optval, sizeof optval); 2538 break; 2539 case TCP_NOPUSH: 2540 optval = tp->t_flags & TF_NOPUSH; 2541 INP_WUNLOCK(inp); 2542 error = sooptcopyout(sopt, &optval, sizeof optval); 2543 break; 2544 case TCP_INFO: 2545 tcp_fill_info(tp, &ti); 2546 INP_WUNLOCK(inp); 2547 error = sooptcopyout(sopt, &ti, sizeof ti); 2548 break; 2549 case TCP_STATS: 2550 { 2551 #ifdef STATS 2552 int nheld; 2553 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2554 2555 error = 0; 2556 socklen_t outsbsz = sopt->sopt_valsize; 2557 if (tp->t_stats == NULL) 2558 error = ENOENT; 2559 else if (outsbsz >= tp->t_stats->cursz) 2560 outsbsz = tp->t_stats->cursz; 2561 else if (outsbsz >= sizeof(struct statsblob)) 2562 outsbsz = sizeof(struct statsblob); 2563 else 2564 error = EINVAL; 2565 INP_WUNLOCK(inp); 2566 if (error) 2567 break; 2568 2569 sbp = sopt->sopt_val; 2570 nheld = atop(round_page(((vm_offset_t)sbp) + 2571 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2572 vm_page_t ma[nheld]; 2573 if (vm_fault_quick_hold_pages( 2574 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2575 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2576 nheld) < 0) { 2577 error = EFAULT; 2578 break; 2579 } 2580 2581 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2582 SIZEOF_MEMBER(struct statsblob, flags)))) 2583 goto unhold; 2584 2585 INP_WLOCK_RECHECK(inp); 2586 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2587 sbflags | SB_CLONE_USRDSTNOFAULT); 2588 INP_WUNLOCK(inp); 2589 sopt->sopt_valsize = outsbsz; 2590 unhold: 2591 vm_page_unhold_pages(ma, nheld); 2592 #else 2593 INP_WUNLOCK(inp); 2594 error = EOPNOTSUPP; 2595 #endif /* !STATS */ 2596 break; 2597 } 2598 case TCP_CONGESTION: 2599 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2600 INP_WUNLOCK(inp); 2601 error = sooptcopyout(sopt, buf, len + 1); 2602 break; 2603 case TCP_MAXUNACKTIME: 2604 case TCP_KEEPIDLE: 2605 case TCP_KEEPINTVL: 2606 case TCP_KEEPINIT: 2607 case TCP_KEEPCNT: 2608 switch (sopt->sopt_name) { 2609 case TCP_MAXUNACKTIME: 2610 ui = TP_MAXUNACKTIME(tp) / hz; 2611 break; 2612 case TCP_KEEPIDLE: 2613 ui = TP_KEEPIDLE(tp) / hz; 2614 break; 2615 case TCP_KEEPINTVL: 2616 ui = TP_KEEPINTVL(tp) / hz; 2617 break; 2618 case TCP_KEEPINIT: 2619 ui = TP_KEEPINIT(tp) / hz; 2620 break; 2621 case TCP_KEEPCNT: 2622 ui = TP_KEEPCNT(tp); 2623 break; 2624 } 2625 INP_WUNLOCK(inp); 2626 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2627 break; 2628 #ifdef TCPPCAP 2629 case TCP_PCAP_OUT: 2630 case TCP_PCAP_IN: 2631 optval = tcp_pcap_get_sock_max( 2632 (sopt->sopt_name == TCP_PCAP_OUT) ? 2633 &(tp->t_outpkts) : &(tp->t_inpkts)); 2634 INP_WUNLOCK(inp); 2635 error = sooptcopyout(sopt, &optval, sizeof optval); 2636 break; 2637 #endif 2638 case TCP_FASTOPEN: 2639 optval = tp->t_flags & TF_FASTOPEN; 2640 INP_WUNLOCK(inp); 2641 error = sooptcopyout(sopt, &optval, sizeof optval); 2642 break; 2643 #ifdef TCP_BLACKBOX 2644 case TCP_LOG: 2645 optval = tcp_get_bblog_state(tp); 2646 INP_WUNLOCK(inp); 2647 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2648 break; 2649 case TCP_LOGBUF: 2650 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2651 error = tcp_log_getlogbuf(sopt, tp); 2652 break; 2653 case TCP_LOGID: 2654 len = tcp_log_get_id(tp, buf); 2655 INP_WUNLOCK(inp); 2656 error = sooptcopyout(sopt, buf, len + 1); 2657 break; 2658 case TCP_LOGDUMP: 2659 case TCP_LOGDUMPID: 2660 INP_WUNLOCK(inp); 2661 error = EINVAL; 2662 break; 2663 #endif 2664 #ifdef KERN_TLS 2665 case TCP_TXTLS_MODE: 2666 error = ktls_get_tx_mode(so, &optval); 2667 INP_WUNLOCK(inp); 2668 if (error == 0) 2669 error = sooptcopyout(sopt, &optval, 2670 sizeof(optval)); 2671 break; 2672 case TCP_RXTLS_MODE: 2673 error = ktls_get_rx_mode(so, &optval); 2674 INP_WUNLOCK(inp); 2675 if (error == 0) 2676 error = sooptcopyout(sopt, &optval, 2677 sizeof(optval)); 2678 break; 2679 #endif 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