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