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