1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 #include <sys/cdefs.h> 39 #include "opt_ddb.h" 40 #include "opt_inet.h" 41 #include "opt_inet6.h" 42 #include "opt_ipsec.h" 43 #include "opt_kern_tls.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/arb.h> 48 #include <sys/limits.h> 49 #include <sys/malloc.h> 50 #include <sys/refcount.h> 51 #include <sys/kernel.h> 52 #include <sys/ktls.h> 53 #include <sys/qmath.h> 54 #include <sys/sysctl.h> 55 #include <sys/mbuf.h> 56 #ifdef INET6 57 #include <sys/domain.h> 58 #endif /* INET6 */ 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/protosw.h> 62 #include <sys/proc.h> 63 #include <sys/jail.h> 64 #include <sys/stats.h> 65 66 #ifdef DDB 67 #include <ddb/ddb.h> 68 #endif 69 70 #include <net/if.h> 71 #include <net/if_var.h> 72 #include <net/route.h> 73 #include <net/vnet.h> 74 75 #include <netinet/in.h> 76 #include <netinet/in_kdtrace.h> 77 #include <netinet/in_pcb.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/in_var.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 #ifdef INET6 83 #include <netinet/ip6.h> 84 #include <netinet6/in6_pcb.h> 85 #include <netinet6/ip6_var.h> 86 #include <netinet6/scope6_var.h> 87 #endif 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/tcp_log_buf.h> 94 #include <netinet/tcpip.h> 95 #include <netinet/cc/cc.h> 96 #include <netinet/tcp_fastopen.h> 97 #include <netinet/tcp_hpts.h> 98 #ifdef TCPPCAP 99 #include <netinet/tcp_pcap.h> 100 #endif 101 #ifdef TCP_OFFLOAD 102 #include <netinet/tcp_offload.h> 103 #endif 104 #include <netipsec/ipsec_support.h> 105 106 #include <vm/vm.h> 107 #include <vm/vm_param.h> 108 #include <vm/pmap.h> 109 #include <vm/vm_extern.h> 110 #include <vm/vm_map.h> 111 #include <vm/vm_page.h> 112 113 /* 114 * TCP protocol interface to socket abstraction. 115 */ 116 #ifdef INET 117 static int tcp_connect(struct tcpcb *, struct sockaddr_in *, 118 struct thread *td); 119 #endif /* INET */ 120 #ifdef INET6 121 static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *, 122 struct thread *td); 123 #endif /* INET6 */ 124 static void tcp_disconnect(struct tcpcb *); 125 static void tcp_usrclosed(struct tcpcb *); 126 static void tcp_fill_info(const struct tcpcb *, struct tcp_info *); 127 128 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 129 130 static void 131 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error) 132 { 133 struct tcp_log_buffer *lgb; 134 135 KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL")); 136 INP_WLOCK_ASSERT(tptoinpcb(tp)); 137 if (tcp_bblogging_on(tp)) { 138 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error, 139 0, NULL, false, NULL, NULL, 0, NULL); 140 } else { 141 lgb = NULL; 142 } 143 if (lgb != NULL) { 144 if (error >= 0) { 145 lgb->tlb_errno = (uint32_t)error; 146 } 147 lgb->tlb_flex1 = pru; 148 } 149 } 150 151 /* 152 * TCP attaches to socket via pru_attach(), reserving space, 153 * and an internet control block. 154 */ 155 static int 156 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 157 { 158 struct inpcb *inp; 159 struct tcpcb *tp = NULL; 160 int error; 161 162 inp = sotoinpcb(so); 163 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 164 165 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 166 if (error) 167 goto out; 168 169 so->so_rcv.sb_flags |= SB_AUTOSIZE; 170 so->so_snd.sb_flags |= SB_AUTOSIZE; 171 error = in_pcballoc(so, &V_tcbinfo); 172 if (error) 173 goto out; 174 inp = sotoinpcb(so); 175 tp = tcp_newtcpcb(inp); 176 if (tp == NULL) { 177 error = ENOBUFS; 178 in_pcbfree(inp); 179 goto out; 180 } 181 tp->t_state = TCPS_CLOSED; 182 tcp_bblog_pru(tp, PRU_ATTACH, error); 183 INP_WUNLOCK(inp); 184 TCPSTATES_INC(TCPS_CLOSED); 185 out: 186 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 187 return (error); 188 } 189 190 /* 191 * tcp_usr_detach is called when the socket layer loses its final reference 192 * to the socket, be it a file descriptor reference, a reference from TCP, 193 * etc. At this point, there is only one case in which we will keep around 194 * inpcb state: time wait. 195 */ 196 static void 197 tcp_usr_detach(struct socket *so) 198 { 199 struct inpcb *inp; 200 struct tcpcb *tp; 201 202 inp = sotoinpcb(so); 203 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 204 INP_WLOCK(inp); 205 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 206 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 207 208 tp = intotcpcb(inp); 209 210 KASSERT(inp->inp_flags & INP_DROPPED || 211 tp->t_state < TCPS_SYN_SENT, 212 ("%s: inp %p not dropped or embryonic", __func__, inp)); 213 214 tcp_discardcb(tp); 215 in_pcbfree(inp); 216 } 217 218 #ifdef INET 219 /* 220 * Give the socket an address. 221 */ 222 static int 223 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 224 { 225 int error = 0; 226 struct inpcb *inp; 227 struct tcpcb *tp; 228 struct sockaddr_in *sinp; 229 230 inp = sotoinpcb(so); 231 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 232 INP_WLOCK(inp); 233 if (inp->inp_flags & INP_DROPPED) { 234 INP_WUNLOCK(inp); 235 return (EINVAL); 236 } 237 tp = intotcpcb(inp); 238 239 sinp = (struct sockaddr_in *)nam; 240 if (nam->sa_family != AF_INET) { 241 /* 242 * Preserve compatibility with old programs. 243 */ 244 if (nam->sa_family != AF_UNSPEC || 245 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 246 sinp->sin_addr.s_addr != INADDR_ANY) { 247 error = EAFNOSUPPORT; 248 goto out; 249 } 250 nam->sa_family = AF_INET; 251 } 252 if (nam->sa_len != sizeof(*sinp)) { 253 error = EINVAL; 254 goto out; 255 } 256 /* 257 * Must check for multicast addresses and disallow binding 258 * to them. 259 */ 260 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 261 error = EAFNOSUPPORT; 262 goto out; 263 } 264 INP_HASH_WLOCK(&V_tcbinfo); 265 error = in_pcbbind(inp, sinp, td->td_ucred); 266 INP_HASH_WUNLOCK(&V_tcbinfo); 267 out: 268 tcp_bblog_pru(tp, PRU_BIND, error); 269 TCP_PROBE2(debug__user, tp, PRU_BIND); 270 INP_WUNLOCK(inp); 271 272 return (error); 273 } 274 #endif /* INET */ 275 276 #ifdef INET6 277 static int 278 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 279 { 280 int error = 0; 281 struct inpcb *inp; 282 struct tcpcb *tp; 283 struct sockaddr_in6 *sin6; 284 u_char vflagsav; 285 286 inp = sotoinpcb(so); 287 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 288 INP_WLOCK(inp); 289 if (inp->inp_flags & INP_DROPPED) { 290 INP_WUNLOCK(inp); 291 return (EINVAL); 292 } 293 tp = intotcpcb(inp); 294 295 vflagsav = inp->inp_vflag; 296 297 sin6 = (struct sockaddr_in6 *)nam; 298 if (nam->sa_family != AF_INET6) { 299 error = EAFNOSUPPORT; 300 goto out; 301 } 302 if (nam->sa_len != sizeof(*sin6)) { 303 error = EINVAL; 304 goto out; 305 } 306 /* 307 * Must check for multicast addresses and disallow binding 308 * to them. 309 */ 310 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 311 error = EAFNOSUPPORT; 312 goto out; 313 } 314 315 INP_HASH_WLOCK(&V_tcbinfo); 316 inp->inp_vflag &= ~INP_IPV4; 317 inp->inp_vflag |= INP_IPV6; 318 #ifdef INET 319 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 320 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 321 inp->inp_vflag |= INP_IPV4; 322 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 323 struct sockaddr_in sin; 324 325 in6_sin6_2_sin(&sin, sin6); 326 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 327 error = EAFNOSUPPORT; 328 INP_HASH_WUNLOCK(&V_tcbinfo); 329 goto out; 330 } 331 inp->inp_vflag |= INP_IPV4; 332 inp->inp_vflag &= ~INP_IPV6; 333 error = in_pcbbind(inp, &sin, td->td_ucred); 334 INP_HASH_WUNLOCK(&V_tcbinfo); 335 goto out; 336 } 337 } 338 #endif 339 error = in6_pcbbind(inp, sin6, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 out: 342 if (error != 0) 343 inp->inp_vflag = vflagsav; 344 tcp_bblog_pru(tp, PRU_BIND, error); 345 TCP_PROBE2(debug__user, tp, PRU_BIND); 346 INP_WUNLOCK(inp); 347 return (error); 348 } 349 #endif /* INET6 */ 350 351 #ifdef INET 352 /* 353 * Prepare to accept connections. 354 */ 355 static int 356 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 357 { 358 int error = 0; 359 struct inpcb *inp; 360 struct tcpcb *tp; 361 362 inp = sotoinpcb(so); 363 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 364 INP_WLOCK(inp); 365 if (inp->inp_flags & INP_DROPPED) { 366 INP_WUNLOCK(inp); 367 return (EINVAL); 368 } 369 tp = intotcpcb(inp); 370 371 SOCK_LOCK(so); 372 error = solisten_proto_check(so); 373 if (error != 0) { 374 SOCK_UNLOCK(so); 375 goto out; 376 } 377 if (inp->inp_lport == 0) { 378 INP_HASH_WLOCK(&V_tcbinfo); 379 error = in_pcbbind(inp, NULL, td->td_ucred); 380 INP_HASH_WUNLOCK(&V_tcbinfo); 381 } 382 if (error == 0) { 383 tcp_state_change(tp, TCPS_LISTEN); 384 solisten_proto(so, backlog); 385 #ifdef TCP_OFFLOAD 386 if ((so->so_options & SO_NO_OFFLOAD) == 0) 387 tcp_offload_listen_start(tp); 388 #endif 389 } else { 390 solisten_proto_abort(so); 391 } 392 SOCK_UNLOCK(so); 393 394 if (tp->t_flags & TF_FASTOPEN) 395 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 396 397 out: 398 tcp_bblog_pru(tp, PRU_LISTEN, error); 399 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 400 INP_WUNLOCK(inp); 401 return (error); 402 } 403 #endif /* INET */ 404 405 #ifdef INET6 406 static int 407 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 408 { 409 int error = 0; 410 struct inpcb *inp; 411 struct tcpcb *tp; 412 u_char vflagsav; 413 414 inp = sotoinpcb(so); 415 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 416 INP_WLOCK(inp); 417 if (inp->inp_flags & INP_DROPPED) { 418 INP_WUNLOCK(inp); 419 return (EINVAL); 420 } 421 tp = intotcpcb(inp); 422 423 vflagsav = inp->inp_vflag; 424 425 SOCK_LOCK(so); 426 error = solisten_proto_check(so); 427 if (error != 0) { 428 SOCK_UNLOCK(so); 429 goto out; 430 } 431 INP_HASH_WLOCK(&V_tcbinfo); 432 if (inp->inp_lport == 0) { 433 inp->inp_vflag &= ~INP_IPV4; 434 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 435 inp->inp_vflag |= INP_IPV4; 436 error = in6_pcbbind(inp, NULL, td->td_ucred); 437 } 438 INP_HASH_WUNLOCK(&V_tcbinfo); 439 if (error == 0) { 440 tcp_state_change(tp, TCPS_LISTEN); 441 solisten_proto(so, backlog); 442 #ifdef TCP_OFFLOAD 443 if ((so->so_options & SO_NO_OFFLOAD) == 0) 444 tcp_offload_listen_start(tp); 445 #endif 446 } else { 447 solisten_proto_abort(so); 448 } 449 SOCK_UNLOCK(so); 450 451 if (tp->t_flags & TF_FASTOPEN) 452 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 453 454 if (error != 0) 455 inp->inp_vflag = vflagsav; 456 457 out: 458 tcp_bblog_pru(tp, PRU_LISTEN, error); 459 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 460 INP_WUNLOCK(inp); 461 return (error); 462 } 463 #endif /* INET6 */ 464 465 #ifdef INET 466 /* 467 * Initiate connection to peer. 468 * Create a template for use in transmissions on this connection. 469 * Enter SYN_SENT state, and mark socket as connecting. 470 * Start keep-alive timer, and seed output sequence space. 471 * Send initial segment on connection. 472 */ 473 static int 474 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 475 { 476 struct epoch_tracker et; 477 int error = 0; 478 struct inpcb *inp; 479 struct tcpcb *tp; 480 struct sockaddr_in *sinp; 481 482 inp = sotoinpcb(so); 483 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 484 INP_WLOCK(inp); 485 if (inp->inp_flags & INP_DROPPED) { 486 INP_WUNLOCK(inp); 487 return (ECONNREFUSED); 488 } 489 tp = intotcpcb(inp); 490 491 sinp = (struct sockaddr_in *)nam; 492 if (nam->sa_family != AF_INET) { 493 error = EAFNOSUPPORT; 494 goto out; 495 } 496 if (nam->sa_len != sizeof (*sinp)) { 497 error = EINVAL; 498 goto out; 499 } 500 /* 501 * Must disallow TCP ``connections'' to multicast addresses. 502 */ 503 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 504 error = EAFNOSUPPORT; 505 goto out; 506 } 507 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 508 error = EACCES; 509 goto out; 510 } 511 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 512 goto out; 513 if (SOLISTENING(so)) { 514 error = EOPNOTSUPP; 515 goto out; 516 } 517 NET_EPOCH_ENTER(et); 518 if ((error = tcp_connect(tp, sinp, td)) != 0) 519 goto out_in_epoch; 520 #ifdef TCP_OFFLOAD 521 if (registered_toedevs > 0 && 522 (so->so_options & SO_NO_OFFLOAD) == 0 && 523 (error = tcp_offload_connect(so, nam)) == 0) 524 goto out_in_epoch; 525 #endif 526 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 527 error = tcp_output(tp); 528 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 529 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 530 out_in_epoch: 531 NET_EPOCH_EXIT(et); 532 out: 533 tcp_bblog_pru(tp, PRU_CONNECT, error); 534 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 535 INP_WUNLOCK(inp); 536 return (error); 537 } 538 #endif /* INET */ 539 540 #ifdef INET6 541 static int 542 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 543 { 544 struct epoch_tracker et; 545 int error = 0; 546 struct inpcb *inp; 547 struct tcpcb *tp; 548 struct sockaddr_in6 *sin6; 549 u_int8_t incflagsav; 550 u_char vflagsav; 551 552 inp = sotoinpcb(so); 553 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 554 INP_WLOCK(inp); 555 if (inp->inp_flags & INP_DROPPED) { 556 INP_WUNLOCK(inp); 557 return (ECONNREFUSED); 558 } 559 tp = intotcpcb(inp); 560 561 vflagsav = inp->inp_vflag; 562 incflagsav = inp->inp_inc.inc_flags; 563 564 sin6 = (struct sockaddr_in6 *)nam; 565 if (nam->sa_family != AF_INET6) { 566 error = EAFNOSUPPORT; 567 goto out; 568 } 569 if (nam->sa_len != sizeof (*sin6)) { 570 error = EINVAL; 571 goto out; 572 } 573 /* 574 * Must disallow TCP ``connections'' to multicast addresses. 575 */ 576 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 577 error = EAFNOSUPPORT; 578 goto out; 579 } 580 if (SOLISTENING(so)) { 581 error = EINVAL; 582 goto out; 583 } 584 #ifdef INET 585 /* 586 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 587 * therefore probably require the hash lock, which isn't held here. 588 * Is this a significant problem? 589 */ 590 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 591 struct sockaddr_in sin; 592 593 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 594 error = EINVAL; 595 goto out; 596 } 597 if ((inp->inp_vflag & INP_IPV4) == 0) { 598 error = EAFNOSUPPORT; 599 goto out; 600 } 601 602 in6_sin6_2_sin(&sin, sin6); 603 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 604 error = EAFNOSUPPORT; 605 goto out; 606 } 607 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 608 error = EACCES; 609 goto out; 610 } 611 if ((error = prison_remote_ip4(td->td_ucred, 612 &sin.sin_addr)) != 0) 613 goto out; 614 inp->inp_vflag |= INP_IPV4; 615 inp->inp_vflag &= ~INP_IPV6; 616 NET_EPOCH_ENTER(et); 617 if ((error = tcp_connect(tp, &sin, td)) != 0) 618 goto out_in_epoch; 619 #ifdef TCP_OFFLOAD 620 if (registered_toedevs > 0 && 621 (so->so_options & SO_NO_OFFLOAD) == 0 && 622 (error = tcp_offload_connect(so, nam)) == 0) 623 goto out_in_epoch; 624 #endif 625 error = tcp_output(tp); 626 goto out_in_epoch; 627 } else { 628 if ((inp->inp_vflag & INP_IPV6) == 0) { 629 error = EAFNOSUPPORT; 630 goto out; 631 } 632 } 633 #endif 634 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 635 goto out; 636 inp->inp_vflag &= ~INP_IPV4; 637 inp->inp_vflag |= INP_IPV6; 638 inp->inp_inc.inc_flags |= INC_ISIPV6; 639 NET_EPOCH_ENTER(et); 640 if ((error = tcp6_connect(tp, sin6, td)) != 0) 641 goto out_in_epoch; 642 #ifdef TCP_OFFLOAD 643 if (registered_toedevs > 0 && 644 (so->so_options & SO_NO_OFFLOAD) == 0 && 645 (error = tcp_offload_connect(so, nam)) == 0) 646 goto out_in_epoch; 647 #endif 648 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 649 error = tcp_output(tp); 650 out_in_epoch: 651 NET_EPOCH_EXIT(et); 652 out: 653 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 654 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 655 /* 656 * If the implicit bind in the connect call fails, restore 657 * the flags we modified. 658 */ 659 if (error != 0 && inp->inp_lport == 0) { 660 inp->inp_vflag = vflagsav; 661 inp->inp_inc.inc_flags = incflagsav; 662 } 663 664 tcp_bblog_pru(tp, PRU_CONNECT, error); 665 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 666 INP_WUNLOCK(inp); 667 return (error); 668 } 669 #endif /* INET6 */ 670 671 /* 672 * Initiate disconnect from peer. 673 * If connection never passed embryonic stage, just drop; 674 * else if don't need to let data drain, then can just drop anyways, 675 * else have to begin TCP shutdown process: mark socket disconnecting, 676 * drain unread data, state switch to reflect user close, and 677 * send segment (e.g. FIN) to peer. Socket will be really disconnected 678 * when peer sends FIN and acks ours. 679 * 680 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 681 */ 682 static int 683 tcp_usr_disconnect(struct socket *so) 684 { 685 struct inpcb *inp; 686 struct tcpcb *tp = NULL; 687 struct epoch_tracker et; 688 int error = 0; 689 690 NET_EPOCH_ENTER(et); 691 inp = sotoinpcb(so); 692 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 693 INP_WLOCK(inp); 694 if (inp->inp_flags & INP_DROPPED) { 695 INP_WUNLOCK(inp); 696 NET_EPOCH_EXIT(et); 697 return (ECONNRESET); 698 } 699 tp = intotcpcb(inp); 700 701 if (tp->t_state == TCPS_TIME_WAIT) 702 goto out; 703 tcp_disconnect(tp); 704 out: 705 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 706 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 707 INP_WUNLOCK(inp); 708 NET_EPOCH_EXIT(et); 709 return (error); 710 } 711 712 #ifdef INET 713 /* 714 * Accept a connection. Essentially all the work is done at higher levels; 715 * just return the address of the peer, storing through addr. 716 */ 717 static int 718 tcp_usr_accept(struct socket *so, struct sockaddr *sa) 719 { 720 struct inpcb *inp; 721 struct tcpcb *tp; 722 int error = 0; 723 724 inp = sotoinpcb(so); 725 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 726 INP_WLOCK(inp); 727 if (inp->inp_flags & INP_DROPPED) { 728 INP_WUNLOCK(inp); 729 return (ECONNABORTED); 730 } 731 tp = intotcpcb(inp); 732 733 if (so->so_state & SS_ISDISCONNECTED) 734 error = ECONNABORTED; 735 else 736 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 737 .sin_family = AF_INET, 738 .sin_len = sizeof(struct sockaddr_in), 739 .sin_port = inp->inp_fport, 740 .sin_addr = inp->inp_faddr, 741 }; 742 tcp_bblog_pru(tp, PRU_ACCEPT, error); 743 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 744 INP_WUNLOCK(inp); 745 746 return (error); 747 } 748 #endif /* INET */ 749 750 #ifdef INET6 751 static int 752 tcp6_usr_accept(struct socket *so, struct sockaddr *sa) 753 { 754 struct inpcb *inp; 755 struct tcpcb *tp; 756 int error = 0; 757 758 inp = sotoinpcb(so); 759 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 760 INP_WLOCK(inp); 761 if (inp->inp_flags & INP_DROPPED) { 762 INP_WUNLOCK(inp); 763 return (ECONNABORTED); 764 } 765 tp = intotcpcb(inp); 766 767 if (so->so_state & SS_ISDISCONNECTED) { 768 error = ECONNABORTED; 769 } else { 770 if (inp->inp_vflag & INP_IPV4) { 771 struct sockaddr_in sin = { 772 .sin_family = AF_INET, 773 .sin_len = sizeof(struct sockaddr_in), 774 .sin_port = inp->inp_fport, 775 .sin_addr = inp->inp_faddr, 776 }; 777 in6_sin_2_v4mapsin6(&sin, (struct sockaddr_in6 *)sa); 778 } else { 779 *(struct sockaddr_in6 *)sa = (struct sockaddr_in6 ){ 780 .sin6_family = AF_INET6, 781 .sin6_len = sizeof(struct sockaddr_in6), 782 .sin6_port = inp->inp_fport, 783 .sin6_addr = inp->in6p_faddr, 784 }; 785 /* XXX: should catch errors */ 786 (void)sa6_recoverscope((struct sockaddr_in6 *)sa); 787 } 788 } 789 790 tcp_bblog_pru(tp, PRU_ACCEPT, error); 791 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 792 INP_WUNLOCK(inp); 793 794 return (error); 795 } 796 #endif /* INET6 */ 797 798 /* 799 * Mark the connection as being incapable of further output. 800 */ 801 static int 802 tcp_usr_shutdown(struct socket *so, enum shutdown_how how) 803 { 804 struct epoch_tracker et; 805 struct inpcb *inp = sotoinpcb(so); 806 struct tcpcb *tp = intotcpcb(inp); 807 int error = 0; 808 809 SOCK_LOCK(so); 810 if (SOLISTENING(so)) { 811 if (how != SHUT_WR) { 812 so->so_error = ECONNABORTED; 813 solisten_wakeup(so); /* unlocks so */ 814 } else 815 SOCK_UNLOCK(so); 816 return (ENOTCONN); 817 } else if ((so->so_state & 818 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 819 SOCK_UNLOCK(so); 820 return (ENOTCONN); 821 } 822 SOCK_UNLOCK(so); 823 824 switch (how) { 825 case SHUT_RD: 826 sorflush(so); 827 break; 828 case SHUT_RDWR: 829 sorflush(so); 830 /* FALLTHROUGH */ 831 case SHUT_WR: 832 /* 833 * XXXGL: mimicing old soshutdown() here. But shouldn't we 834 * return ECONNRESEST for SHUT_RD as well? 835 */ 836 INP_WLOCK(inp); 837 if (inp->inp_flags & INP_DROPPED) { 838 INP_WUNLOCK(inp); 839 return (ECONNRESET); 840 } 841 842 socantsendmore(so); 843 NET_EPOCH_ENTER(et); 844 tcp_usrclosed(tp); 845 error = tcp_output_nodrop(tp); 846 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 847 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 848 error = tcp_unlock_or_drop(tp, error); 849 NET_EPOCH_EXIT(et); 850 } 851 wakeup(&so->so_timeo); 852 853 return (error); 854 } 855 856 /* 857 * After a receive, possibly send window update to peer. 858 */ 859 static int 860 tcp_usr_rcvd(struct socket *so, int flags) 861 { 862 struct epoch_tracker et; 863 struct inpcb *inp; 864 struct tcpcb *tp; 865 int outrv = 0, error = 0; 866 867 inp = sotoinpcb(so); 868 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 869 INP_WLOCK(inp); 870 if (inp->inp_flags & INP_DROPPED) { 871 INP_WUNLOCK(inp); 872 return (ECONNRESET); 873 } 874 tp = intotcpcb(inp); 875 876 NET_EPOCH_ENTER(et); 877 /* 878 * For passively-created TFO connections, don't attempt a window 879 * update while still in SYN_RECEIVED as this may trigger an early 880 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 881 * application response data, or failing that, when the DELACK timer 882 * expires. 883 */ 884 if ((tp->t_flags & TF_FASTOPEN) && (tp->t_state == TCPS_SYN_RECEIVED)) 885 goto out; 886 #ifdef TCP_OFFLOAD 887 if (tp->t_flags & TF_TOE) 888 tcp_offload_rcvd(tp); 889 else 890 #endif 891 outrv = tcp_output_nodrop(tp); 892 out: 893 tcp_bblog_pru(tp, PRU_RCVD, error); 894 TCP_PROBE2(debug__user, tp, PRU_RCVD); 895 (void) tcp_unlock_or_drop(tp, outrv); 896 NET_EPOCH_EXIT(et); 897 return (error); 898 } 899 900 /* 901 * Do a send by putting data in output queue and updating urgent 902 * marker if URG set. Possibly send more data. Unlike the other 903 * pru_*() routines, the mbuf chains are our responsibility. We 904 * must either enqueue them or free them. The other pru_* routines 905 * generally are caller-frees. 906 */ 907 static int 908 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 909 struct sockaddr *nam, struct mbuf *control, struct thread *td) 910 { 911 struct epoch_tracker et; 912 int error = 0; 913 struct inpcb *inp; 914 struct tcpcb *tp; 915 #ifdef INET 916 #ifdef INET6 917 struct sockaddr_in sin; 918 #endif 919 struct sockaddr_in *sinp; 920 #endif 921 #ifdef INET6 922 struct sockaddr_in6 *sin6; 923 int isipv6; 924 #endif 925 u_int8_t incflagsav; 926 u_char vflagsav; 927 bool restoreflags; 928 929 inp = sotoinpcb(so); 930 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 931 INP_WLOCK(inp); 932 if (inp->inp_flags & INP_DROPPED) { 933 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 934 m_freem(m); 935 INP_WUNLOCK(inp); 936 return (ECONNRESET); 937 } 938 tp = intotcpcb(inp); 939 940 vflagsav = inp->inp_vflag; 941 incflagsav = inp->inp_inc.inc_flags; 942 restoreflags = false; 943 944 NET_EPOCH_ENTER(et); 945 if (control != NULL) { 946 /* TCP doesn't do control messages (rights, creds, etc) */ 947 if (control->m_len > 0) { 948 m_freem(control); 949 error = EINVAL; 950 goto out; 951 } 952 m_freem(control); /* empty control, just free it */ 953 } 954 955 if ((flags & PRUS_OOB) != 0 && 956 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 957 goto out; 958 959 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 960 if (tp->t_state == TCPS_LISTEN) { 961 error = EINVAL; 962 goto out; 963 } 964 switch (nam->sa_family) { 965 #ifdef INET 966 case AF_INET: 967 sinp = (struct sockaddr_in *)nam; 968 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 969 error = EINVAL; 970 goto out; 971 } 972 if ((inp->inp_vflag & INP_IPV6) != 0) { 973 error = EAFNOSUPPORT; 974 goto out; 975 } 976 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 977 error = EAFNOSUPPORT; 978 goto out; 979 } 980 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 981 error = EACCES; 982 goto out; 983 } 984 if ((error = prison_remote_ip4(td->td_ucred, 985 &sinp->sin_addr))) 986 goto out; 987 #ifdef INET6 988 isipv6 = 0; 989 #endif 990 break; 991 #endif /* INET */ 992 #ifdef INET6 993 case AF_INET6: 994 sin6 = (struct sockaddr_in6 *)nam; 995 if (sin6->sin6_len != sizeof(*sin6)) { 996 error = EINVAL; 997 goto out; 998 } 999 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1000 error = EAFNOSUPPORT; 1001 goto out; 1002 } 1003 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1004 error = EAFNOSUPPORT; 1005 goto out; 1006 } 1007 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1008 #ifdef INET 1009 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1010 error = EINVAL; 1011 goto out; 1012 } 1013 if ((inp->inp_vflag & INP_IPV4) == 0) { 1014 error = EAFNOSUPPORT; 1015 goto out; 1016 } 1017 restoreflags = true; 1018 inp->inp_vflag &= ~INP_IPV6; 1019 sinp = &sin; 1020 in6_sin6_2_sin(sinp, sin6); 1021 if (IN_MULTICAST( 1022 ntohl(sinp->sin_addr.s_addr))) { 1023 error = EAFNOSUPPORT; 1024 goto out; 1025 } 1026 if ((error = prison_remote_ip4(td->td_ucred, 1027 &sinp->sin_addr))) 1028 goto out; 1029 isipv6 = 0; 1030 #else /* !INET */ 1031 error = EAFNOSUPPORT; 1032 goto out; 1033 #endif /* INET */ 1034 } else { 1035 if ((inp->inp_vflag & INP_IPV6) == 0) { 1036 error = EAFNOSUPPORT; 1037 goto out; 1038 } 1039 restoreflags = true; 1040 inp->inp_vflag &= ~INP_IPV4; 1041 inp->inp_inc.inc_flags |= INC_ISIPV6; 1042 if ((error = prison_remote_ip6(td->td_ucred, 1043 &sin6->sin6_addr))) 1044 goto out; 1045 isipv6 = 1; 1046 } 1047 break; 1048 #endif /* INET6 */ 1049 default: 1050 error = EAFNOSUPPORT; 1051 goto out; 1052 } 1053 } 1054 if (!(flags & PRUS_OOB)) { 1055 if (tp->t_acktime == 0) 1056 tp->t_acktime = ticks; 1057 sbappendstream(&so->so_snd, m, flags); 1058 m = NULL; 1059 if (nam && tp->t_state < TCPS_SYN_SENT) { 1060 KASSERT(tp->t_state == TCPS_CLOSED, 1061 ("%s: tp %p is listening", __func__, tp)); 1062 1063 /* 1064 * Do implied connect if not yet connected, 1065 * initialize window to default value, and 1066 * initialize maxseg using peer's cached MSS. 1067 */ 1068 #ifdef INET6 1069 if (isipv6) 1070 error = tcp6_connect(tp, sin6, td); 1071 #endif /* INET6 */ 1072 #if defined(INET6) && defined(INET) 1073 else 1074 #endif 1075 #ifdef INET 1076 error = tcp_connect(tp, sinp, td); 1077 #endif 1078 /* 1079 * The bind operation in tcp_connect succeeded. We 1080 * no longer want to restore the flags if later 1081 * operations fail. 1082 */ 1083 if (error == 0 || inp->inp_lport != 0) 1084 restoreflags = false; 1085 1086 if (error) { 1087 /* m is freed if PRUS_NOTREADY is unset. */ 1088 sbflush(&so->so_snd); 1089 goto out; 1090 } 1091 if (tp->t_flags & TF_FASTOPEN) 1092 tcp_fastopen_connect(tp); 1093 else { 1094 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1095 tcp_mss(tp, -1); 1096 } 1097 } 1098 if (flags & PRUS_EOF) { 1099 /* 1100 * Close the send side of the connection after 1101 * the data is sent. 1102 */ 1103 socantsendmore(so); 1104 tcp_usrclosed(tp); 1105 } 1106 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1107 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1108 (tp->t_fbyte_out == 0) && 1109 (so->so_snd.sb_ccc > 0)) { 1110 tp->t_fbyte_out = ticks; 1111 if (tp->t_fbyte_out == 0) 1112 tp->t_fbyte_out = 1; 1113 if (tp->t_fbyte_out && tp->t_fbyte_in) 1114 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1115 } 1116 if (!(inp->inp_flags & INP_DROPPED) && 1117 !(flags & PRUS_NOTREADY)) { 1118 if (flags & PRUS_MORETOCOME) 1119 tp->t_flags |= TF_MORETOCOME; 1120 error = tcp_output_nodrop(tp); 1121 if (flags & PRUS_MORETOCOME) 1122 tp->t_flags &= ~TF_MORETOCOME; 1123 } 1124 } else { 1125 /* 1126 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1127 */ 1128 SOCKBUF_LOCK(&so->so_snd); 1129 if (sbspace(&so->so_snd) < -512) { 1130 SOCKBUF_UNLOCK(&so->so_snd); 1131 error = ENOBUFS; 1132 goto out; 1133 } 1134 /* 1135 * According to RFC961 (Assigned Protocols), 1136 * the urgent pointer points to the last octet 1137 * of urgent data. We continue, however, 1138 * to consider it to indicate the first octet 1139 * of data past the urgent section. 1140 * Otherwise, snd_up should be one lower. 1141 */ 1142 if (tp->t_acktime == 0) 1143 tp->t_acktime = ticks; 1144 sbappendstream_locked(&so->so_snd, m, flags); 1145 SOCKBUF_UNLOCK(&so->so_snd); 1146 m = NULL; 1147 if (nam && tp->t_state < TCPS_SYN_SENT) { 1148 /* 1149 * Do implied connect if not yet connected, 1150 * initialize window to default value, and 1151 * initialize maxseg using peer's cached MSS. 1152 */ 1153 1154 /* 1155 * Not going to contemplate SYN|URG 1156 */ 1157 if (tp->t_flags & TF_FASTOPEN) 1158 tp->t_flags &= ~TF_FASTOPEN; 1159 #ifdef INET6 1160 if (isipv6) 1161 error = tcp6_connect(tp, sin6, td); 1162 #endif /* INET6 */ 1163 #if defined(INET6) && defined(INET) 1164 else 1165 #endif 1166 #ifdef INET 1167 error = tcp_connect(tp, sinp, td); 1168 #endif 1169 /* 1170 * The bind operation in tcp_connect succeeded. We 1171 * no longer want to restore the flags if later 1172 * operations fail. 1173 */ 1174 if (error == 0 || inp->inp_lport != 0) 1175 restoreflags = false; 1176 1177 if (error != 0) { 1178 /* m is freed if PRUS_NOTREADY is unset. */ 1179 sbflush(&so->so_snd); 1180 goto out; 1181 } 1182 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1183 tcp_mss(tp, -1); 1184 } 1185 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1186 if ((flags & PRUS_NOTREADY) == 0) { 1187 tp->t_flags |= TF_FORCEDATA; 1188 error = tcp_output_nodrop(tp); 1189 tp->t_flags &= ~TF_FORCEDATA; 1190 } 1191 } 1192 TCP_LOG_EVENT(tp, NULL, 1193 &inp->inp_socket->so_rcv, 1194 &inp->inp_socket->so_snd, 1195 TCP_LOG_USERSEND, error, 1196 0, NULL, false); 1197 1198 out: 1199 /* 1200 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1201 * responsible for freeing memory. 1202 */ 1203 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1204 m_freem(m); 1205 1206 /* 1207 * If the request was unsuccessful and we changed flags, 1208 * restore the original flags. 1209 */ 1210 if (error != 0 && restoreflags) { 1211 inp->inp_vflag = vflagsav; 1212 inp->inp_inc.inc_flags = incflagsav; 1213 } 1214 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1215 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1216 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1217 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1218 error = tcp_unlock_or_drop(tp, error); 1219 NET_EPOCH_EXIT(et); 1220 return (error); 1221 } 1222 1223 static int 1224 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1225 { 1226 struct epoch_tracker et; 1227 struct inpcb *inp; 1228 struct tcpcb *tp; 1229 int error; 1230 1231 inp = sotoinpcb(so); 1232 INP_WLOCK(inp); 1233 if (inp->inp_flags & INP_DROPPED) { 1234 INP_WUNLOCK(inp); 1235 mb_free_notready(m, count); 1236 return (ECONNRESET); 1237 } 1238 tp = intotcpcb(inp); 1239 1240 SOCKBUF_LOCK(&so->so_snd); 1241 error = sbready(&so->so_snd, m, count); 1242 SOCKBUF_UNLOCK(&so->so_snd); 1243 if (error) { 1244 INP_WUNLOCK(inp); 1245 return (error); 1246 } 1247 NET_EPOCH_ENTER(et); 1248 error = tcp_output_unlock(tp); 1249 NET_EPOCH_EXIT(et); 1250 1251 return (error); 1252 } 1253 1254 /* 1255 * Abort the TCP. Drop the connection abruptly. 1256 */ 1257 static void 1258 tcp_usr_abort(struct socket *so) 1259 { 1260 struct inpcb *inp; 1261 struct tcpcb *tp; 1262 struct epoch_tracker et; 1263 1264 inp = sotoinpcb(so); 1265 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1266 1267 NET_EPOCH_ENTER(et); 1268 INP_WLOCK(inp); 1269 KASSERT(inp->inp_socket != NULL, 1270 ("tcp_usr_abort: inp_socket == NULL")); 1271 1272 /* 1273 * If we still have full TCP state, and we're not dropped, drop. 1274 */ 1275 if (!(inp->inp_flags & INP_DROPPED)) { 1276 tp = intotcpcb(inp); 1277 tp = tcp_drop(tp, ECONNABORTED); 1278 if (tp == NULL) 1279 goto dropped; 1280 tcp_bblog_pru(tp, PRU_ABORT, 0); 1281 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1282 } 1283 if (!(inp->inp_flags & INP_DROPPED)) { 1284 soref(so); 1285 inp->inp_flags |= INP_SOCKREF; 1286 } 1287 INP_WUNLOCK(inp); 1288 dropped: 1289 NET_EPOCH_EXIT(et); 1290 } 1291 1292 /* 1293 * TCP socket is closed. Start friendly disconnect. 1294 */ 1295 static void 1296 tcp_usr_close(struct socket *so) 1297 { 1298 struct inpcb *inp; 1299 struct tcpcb *tp; 1300 struct epoch_tracker et; 1301 1302 inp = sotoinpcb(so); 1303 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1304 1305 NET_EPOCH_ENTER(et); 1306 INP_WLOCK(inp); 1307 KASSERT(inp->inp_socket != NULL, 1308 ("tcp_usr_close: inp_socket == NULL")); 1309 1310 /* 1311 * If we are still connected and we're not dropped, initiate 1312 * a disconnect. 1313 */ 1314 if (!(inp->inp_flags & INP_DROPPED)) { 1315 tp = intotcpcb(inp); 1316 if (tp->t_state != TCPS_TIME_WAIT) { 1317 tp->t_flags |= TF_CLOSED; 1318 tcp_disconnect(tp); 1319 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1320 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1321 } 1322 } 1323 if (!(inp->inp_flags & INP_DROPPED)) { 1324 soref(so); 1325 inp->inp_flags |= INP_SOCKREF; 1326 } 1327 INP_WUNLOCK(inp); 1328 NET_EPOCH_EXIT(et); 1329 } 1330 1331 static int 1332 tcp_pru_options_support(struct tcpcb *tp, int flags) 1333 { 1334 /* 1335 * If the specific TCP stack has a pru_options 1336 * specified then it does not always support 1337 * all the PRU_XX options and we must ask it. 1338 * If the function is not specified then all 1339 * of the PRU_XX options are supported. 1340 */ 1341 int ret = 0; 1342 1343 if (tp->t_fb->tfb_pru_options) { 1344 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1345 } 1346 return (ret); 1347 } 1348 1349 /* 1350 * Receive out-of-band data. 1351 */ 1352 static int 1353 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1354 { 1355 int error = 0; 1356 struct inpcb *inp; 1357 struct tcpcb *tp; 1358 1359 inp = sotoinpcb(so); 1360 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1361 INP_WLOCK(inp); 1362 if (inp->inp_flags & INP_DROPPED) { 1363 INP_WUNLOCK(inp); 1364 return (ECONNRESET); 1365 } 1366 tp = intotcpcb(inp); 1367 1368 error = tcp_pru_options_support(tp, PRUS_OOB); 1369 if (error) { 1370 goto out; 1371 } 1372 if ((so->so_oobmark == 0 && 1373 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1374 so->so_options & SO_OOBINLINE || 1375 tp->t_oobflags & TCPOOB_HADDATA) { 1376 error = EINVAL; 1377 goto out; 1378 } 1379 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1380 error = EWOULDBLOCK; 1381 goto out; 1382 } 1383 m->m_len = 1; 1384 *mtod(m, caddr_t) = tp->t_iobc; 1385 if ((flags & MSG_PEEK) == 0) 1386 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1387 1388 out: 1389 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1390 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1391 INP_WUNLOCK(inp); 1392 return (error); 1393 } 1394 1395 #ifdef INET 1396 struct protosw tcp_protosw = { 1397 .pr_type = SOCK_STREAM, 1398 .pr_protocol = IPPROTO_TCP, 1399 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1400 PR_CAPATTACH, 1401 .pr_ctloutput = tcp_ctloutput, 1402 .pr_abort = tcp_usr_abort, 1403 .pr_accept = tcp_usr_accept, 1404 .pr_attach = tcp_usr_attach, 1405 .pr_bind = tcp_usr_bind, 1406 .pr_connect = tcp_usr_connect, 1407 .pr_control = in_control, 1408 .pr_detach = tcp_usr_detach, 1409 .pr_disconnect = tcp_usr_disconnect, 1410 .pr_listen = tcp_usr_listen, 1411 .pr_peeraddr = in_getpeeraddr, 1412 .pr_rcvd = tcp_usr_rcvd, 1413 .pr_rcvoob = tcp_usr_rcvoob, 1414 .pr_send = tcp_usr_send, 1415 .pr_ready = tcp_usr_ready, 1416 .pr_shutdown = tcp_usr_shutdown, 1417 .pr_sockaddr = in_getsockaddr, 1418 .pr_sosetlabel = in_pcbsosetlabel, 1419 .pr_close = tcp_usr_close, 1420 }; 1421 #endif /* INET */ 1422 1423 #ifdef INET6 1424 struct protosw tcp6_protosw = { 1425 .pr_type = SOCK_STREAM, 1426 .pr_protocol = IPPROTO_TCP, 1427 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1428 PR_CAPATTACH, 1429 .pr_ctloutput = tcp_ctloutput, 1430 .pr_abort = tcp_usr_abort, 1431 .pr_accept = tcp6_usr_accept, 1432 .pr_attach = tcp_usr_attach, 1433 .pr_bind = tcp6_usr_bind, 1434 .pr_connect = tcp6_usr_connect, 1435 .pr_control = in6_control, 1436 .pr_detach = tcp_usr_detach, 1437 .pr_disconnect = tcp_usr_disconnect, 1438 .pr_listen = tcp6_usr_listen, 1439 .pr_peeraddr = in6_mapped_peeraddr, 1440 .pr_rcvd = tcp_usr_rcvd, 1441 .pr_rcvoob = tcp_usr_rcvoob, 1442 .pr_send = tcp_usr_send, 1443 .pr_ready = tcp_usr_ready, 1444 .pr_shutdown = tcp_usr_shutdown, 1445 .pr_sockaddr = in6_mapped_sockaddr, 1446 .pr_sosetlabel = in_pcbsosetlabel, 1447 .pr_close = tcp_usr_close, 1448 }; 1449 #endif /* INET6 */ 1450 1451 #ifdef INET 1452 /* 1453 * Common subroutine to open a TCP connection to remote host specified 1454 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1455 * and assign a local port number and install the inpcb into the hash. 1456 * Initialize connection parameters and enter SYN-SENT state. 1457 */ 1458 static int 1459 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1460 { 1461 struct inpcb *inp = tptoinpcb(tp); 1462 struct socket *so = tptosocket(tp); 1463 int error; 1464 1465 NET_EPOCH_ASSERT(); 1466 INP_WLOCK_ASSERT(inp); 1467 1468 if (__predict_false((so->so_state & 1469 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1470 SS_ISDISCONNECTED)) != 0)) 1471 return (EISCONN); 1472 1473 INP_HASH_WLOCK(&V_tcbinfo); 1474 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1475 INP_HASH_WUNLOCK(&V_tcbinfo); 1476 if (error != 0) 1477 return (error); 1478 1479 /* 1480 * Compute window scaling to request: 1481 * Scale to fit into sweet spot. See tcp_syncache.c. 1482 * XXX: This should move to tcp_output(). 1483 */ 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 /* INET */ 1499 1500 #ifdef INET6 1501 static int 1502 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1503 { 1504 struct inpcb *inp = tptoinpcb(tp); 1505 struct socket *so = tptosocket(tp); 1506 int error; 1507 1508 NET_EPOCH_ASSERT(); 1509 INP_WLOCK_ASSERT(inp); 1510 1511 if (__predict_false((so->so_state & 1512 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1513 return (EISCONN); 1514 1515 INP_HASH_WLOCK(&V_tcbinfo); 1516 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1517 INP_HASH_WUNLOCK(&V_tcbinfo); 1518 if (error != 0) 1519 return (error); 1520 1521 /* Compute window scaling to request. */ 1522 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1523 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1524 tp->request_r_scale++; 1525 1526 soisconnecting(so); 1527 TCPSTAT_INC(tcps_connattempt); 1528 tcp_state_change(tp, TCPS_SYN_SENT); 1529 tp->iss = tcp_new_isn(&inp->inp_inc); 1530 if (tp->t_flags & TF_REQ_TSTMP) 1531 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1532 tcp_sendseqinit(tp); 1533 1534 return (0); 1535 } 1536 #endif /* INET6 */ 1537 1538 /* 1539 * Export TCP internal state information via a struct tcp_info, based on the 1540 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1541 * (TCP state machine, etc). We export all information using FreeBSD-native 1542 * constants -- for example, the numeric values for tcpi_state will differ 1543 * from Linux. 1544 */ 1545 void 1546 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1547 { 1548 1549 INP_LOCK_ASSERT(tptoinpcb(tp)); 1550 bzero(ti, sizeof(*ti)); 1551 1552 ti->tcpi_state = tp->t_state; 1553 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1554 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1555 if (tp->t_flags & TF_SACK_PERMIT) 1556 ti->tcpi_options |= TCPI_OPT_SACK; 1557 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1558 ti->tcpi_options |= TCPI_OPT_WSCALE; 1559 ti->tcpi_snd_wscale = tp->snd_scale; 1560 ti->tcpi_rcv_wscale = tp->rcv_scale; 1561 } 1562 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1563 case TF2_ECN_PERMIT: 1564 ti->tcpi_options |= TCPI_OPT_ECN; 1565 break; 1566 case TF2_ACE_PERMIT: 1567 /* FALLTHROUGH */ 1568 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1569 ti->tcpi_options |= TCPI_OPT_ACE; 1570 break; 1571 default: 1572 break; 1573 } 1574 if (tp->t_flags & TF_FASTOPEN) 1575 ti->tcpi_options |= TCPI_OPT_TFO; 1576 1577 ti->tcpi_rto = tp->t_rxtcur * tick; 1578 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1579 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1580 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1581 1582 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1583 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1584 1585 /* 1586 * FreeBSD-specific extension fields for tcp_info. 1587 */ 1588 ti->tcpi_rcv_space = tp->rcv_wnd; 1589 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1590 ti->tcpi_snd_wnd = tp->snd_wnd; 1591 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1592 ti->tcpi_snd_nxt = tp->snd_nxt; 1593 ti->tcpi_snd_mss = tp->t_maxseg; 1594 ti->tcpi_rcv_mss = tp->t_maxseg; 1595 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1596 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1597 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1598 ti->tcpi_snd_una = tp->snd_una; 1599 ti->tcpi_snd_max = tp->snd_max; 1600 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1601 ti->tcpi_rcv_adv = tp->rcv_adv; 1602 ti->tcpi_dupacks = tp->t_dupacks; 1603 ti->tcpi_rttmin = tp->t_rttlow; 1604 #ifdef TCP_OFFLOAD 1605 if (tp->t_flags & TF_TOE) { 1606 ti->tcpi_options |= TCPI_OPT_TOE; 1607 tcp_offload_tcp_info(tp, ti); 1608 } 1609 #endif 1610 /* 1611 * AccECN related counters. 1612 */ 1613 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1614 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1615 /* 1616 * Internal counter starts at 5 for AccECN 1617 * but 0 for RFC3168 ECN. 1618 */ 1619 ti->tcpi_delivered_ce = tp->t_scep - 5; 1620 else 1621 ti->tcpi_delivered_ce = tp->t_scep; 1622 ti->tcpi_received_ce = tp->t_rcep; 1623 } 1624 1625 /* 1626 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1627 * socket option arguments. When it re-acquires the lock after the copy, it 1628 * has to revalidate that the connection is still valid for the socket 1629 * option. 1630 */ 1631 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1632 INP_WLOCK(inp); \ 1633 if (inp->inp_flags & INP_DROPPED) { \ 1634 INP_WUNLOCK(inp); \ 1635 cleanup; \ 1636 return (ECONNRESET); \ 1637 } \ 1638 tp = intotcpcb(inp); \ 1639 } while(0) 1640 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1641 1642 int 1643 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1644 { 1645 struct socket *so = inp->inp_socket; 1646 struct tcpcb *tp = intotcpcb(inp); 1647 int error = 0; 1648 1649 MPASS(sopt->sopt_dir == SOPT_SET); 1650 INP_WLOCK_ASSERT(inp); 1651 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1652 ("inp_flags == %x", inp->inp_flags)); 1653 KASSERT(so != NULL, ("inp_socket == NULL")); 1654 1655 if (sopt->sopt_level != IPPROTO_TCP) { 1656 INP_WUNLOCK(inp); 1657 #ifdef INET6 1658 if (inp->inp_vflag & INP_IPV6PROTO) 1659 error = ip6_ctloutput(so, sopt); 1660 #endif 1661 #if defined(INET6) && defined(INET) 1662 else 1663 #endif 1664 #ifdef INET 1665 error = ip_ctloutput(so, sopt); 1666 #endif 1667 /* 1668 * When an IP-level socket option affects TCP, pass control 1669 * down to stack tfb_tcp_ctloutput, otherwise return what 1670 * IP level returned. 1671 */ 1672 switch (sopt->sopt_level) { 1673 #ifdef INET6 1674 case IPPROTO_IPV6: 1675 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1676 return (error); 1677 switch (sopt->sopt_name) { 1678 case IPV6_TCLASS: 1679 /* Notify tcp stacks that care (e.g. RACK). */ 1680 break; 1681 case IPV6_USE_MIN_MTU: 1682 /* Update t_maxseg accordingly. */ 1683 break; 1684 default: 1685 return (error); 1686 } 1687 break; 1688 #endif 1689 #ifdef INET 1690 case IPPROTO_IP: 1691 switch (sopt->sopt_name) { 1692 case IP_TOS: 1693 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1694 break; 1695 case IP_TTL: 1696 /* Notify tcp stacks that care (e.g. RACK). */ 1697 break; 1698 default: 1699 return (error); 1700 } 1701 break; 1702 #endif 1703 default: 1704 return (error); 1705 } 1706 INP_WLOCK_RECHECK(inp); 1707 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1708 /* 1709 * Protect the TCP option TCP_FUNCTION_BLK so 1710 * that a sub-function can *never* overwrite this. 1711 */ 1712 struct tcp_function_set fsn; 1713 struct tcp_function_block *blk; 1714 void *ptr = NULL; 1715 1716 INP_WUNLOCK(inp); 1717 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1718 if (error) 1719 return (error); 1720 1721 INP_WLOCK_RECHECK(inp); 1722 1723 blk = find_and_ref_tcp_functions(&fsn); 1724 if (blk == NULL) { 1725 INP_WUNLOCK(inp); 1726 return (ENOENT); 1727 } 1728 if (tp->t_fb == blk) { 1729 /* You already have this */ 1730 refcount_release(&blk->tfb_refcnt); 1731 INP_WUNLOCK(inp); 1732 return (0); 1733 } 1734 if (tp->t_state != TCPS_CLOSED) { 1735 /* 1736 * The user has advanced the state 1737 * past the initial point, we may not 1738 * be able to switch. 1739 */ 1740 if (blk->tfb_tcp_handoff_ok != NULL) { 1741 /* 1742 * Does the stack provide a 1743 * query mechanism, if so it may 1744 * still be possible? 1745 */ 1746 error = (*blk->tfb_tcp_handoff_ok)(tp); 1747 } else 1748 error = EINVAL; 1749 if (error) { 1750 refcount_release(&blk->tfb_refcnt); 1751 INP_WUNLOCK(inp); 1752 return(error); 1753 } 1754 } 1755 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1756 refcount_release(&blk->tfb_refcnt); 1757 INP_WUNLOCK(inp); 1758 return (ENOENT); 1759 } 1760 /* 1761 * Ensure the new stack takes ownership with a 1762 * clean slate on peak rate threshold. 1763 */ 1764 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 1765 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1766 if (blk->tfb_tcp_fb_init) { 1767 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1768 if (error) { 1769 /* 1770 * Release the ref count the lookup 1771 * acquired. 1772 */ 1773 refcount_release(&blk->tfb_refcnt); 1774 /* 1775 * Now there is a chance that the 1776 * init() function mucked with some 1777 * things before it failed, such as 1778 * hpts or inp_flags2 or timer granularity. 1779 * It should not of, but lets give the old 1780 * stack a chance to reset to a known good state. 1781 */ 1782 if (tp->t_fb->tfb_switch_failed) { 1783 (*tp->t_fb->tfb_switch_failed)(tp); 1784 } 1785 goto err_out; 1786 } 1787 } 1788 if (tp->t_fb->tfb_tcp_fb_fini) { 1789 struct epoch_tracker et; 1790 /* 1791 * Tell the stack to cleanup with 0 i.e. 1792 * the tcb is not going away. 1793 */ 1794 NET_EPOCH_ENTER(et); 1795 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1796 NET_EPOCH_EXIT(et); 1797 } 1798 /* 1799 * Release the old refcnt, the 1800 * lookup acquired a ref on the 1801 * new one already. 1802 */ 1803 refcount_release(&tp->t_fb->tfb_refcnt); 1804 /* 1805 * Set in the new stack. 1806 */ 1807 tp->t_fb = blk; 1808 tp->t_fb_ptr = ptr; 1809 #ifdef TCP_OFFLOAD 1810 if (tp->t_flags & TF_TOE) { 1811 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1812 sopt->sopt_name); 1813 } 1814 #endif 1815 err_out: 1816 INP_WUNLOCK(inp); 1817 return (error); 1818 1819 } 1820 1821 /* Pass in the INP locked, callee must unlock it. */ 1822 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1823 } 1824 1825 static int 1826 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1827 { 1828 struct socket *so = inp->inp_socket; 1829 struct tcpcb *tp = intotcpcb(inp); 1830 int error = 0; 1831 1832 MPASS(sopt->sopt_dir == SOPT_GET); 1833 INP_WLOCK_ASSERT(inp); 1834 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1835 ("inp_flags == %x", inp->inp_flags)); 1836 KASSERT(so != NULL, ("inp_socket == NULL")); 1837 1838 if (sopt->sopt_level != IPPROTO_TCP) { 1839 INP_WUNLOCK(inp); 1840 #ifdef INET6 1841 if (inp->inp_vflag & INP_IPV6PROTO) 1842 error = ip6_ctloutput(so, sopt); 1843 #endif /* INET6 */ 1844 #if defined(INET6) && defined(INET) 1845 else 1846 #endif 1847 #ifdef INET 1848 error = ip_ctloutput(so, sopt); 1849 #endif 1850 return (error); 1851 } 1852 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1853 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1854 struct tcp_function_set fsn; 1855 1856 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1857 memset(&fsn, 0, sizeof(fsn)); 1858 find_tcp_function_alias(tp->t_fb, &fsn); 1859 } else { 1860 strncpy(fsn.function_set_name, 1861 tp->t_fb->tfb_tcp_block_name, 1862 TCP_FUNCTION_NAME_LEN_MAX); 1863 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1864 } 1865 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1866 INP_WUNLOCK(inp); 1867 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1868 return (error); 1869 } 1870 1871 /* Pass in the INP locked, callee must unlock it. */ 1872 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1873 } 1874 1875 int 1876 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1877 { 1878 struct inpcb *inp; 1879 1880 inp = sotoinpcb(so); 1881 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1882 1883 INP_WLOCK(inp); 1884 if (inp->inp_flags & INP_DROPPED) { 1885 INP_WUNLOCK(inp); 1886 return (ECONNRESET); 1887 } 1888 if (sopt->sopt_dir == SOPT_SET) 1889 return (tcp_ctloutput_set(inp, sopt)); 1890 else if (sopt->sopt_dir == SOPT_GET) 1891 return (tcp_ctloutput_get(inp, sopt)); 1892 else 1893 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1894 } 1895 1896 /* 1897 * If this assert becomes untrue, we need to change the size of the buf 1898 * variable in tcp_default_ctloutput(). 1899 */ 1900 #ifdef CTASSERT 1901 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1902 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1903 #endif 1904 1905 extern struct cc_algo newreno_cc_algo; 1906 1907 static int 1908 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1909 { 1910 struct cc_algo *algo; 1911 void *ptr = NULL; 1912 struct tcpcb *tp; 1913 struct cc_var cc_mem; 1914 char buf[TCP_CA_NAME_MAX]; 1915 size_t mem_sz; 1916 int error; 1917 1918 INP_WUNLOCK(inp); 1919 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1920 if (error) 1921 return(error); 1922 buf[sopt->sopt_valsize] = '\0'; 1923 CC_LIST_RLOCK(); 1924 STAILQ_FOREACH(algo, &cc_list, entries) { 1925 if (strncmp(buf, algo->name, 1926 TCP_CA_NAME_MAX) == 0) { 1927 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1928 /* We can't "see" modules being unloaded */ 1929 continue; 1930 } 1931 break; 1932 } 1933 } 1934 if (algo == NULL) { 1935 CC_LIST_RUNLOCK(); 1936 return(ESRCH); 1937 } 1938 /* 1939 * With a reference the algorithm cannot be removed 1940 * so we hold a reference through the change process. 1941 */ 1942 cc_refer(algo); 1943 CC_LIST_RUNLOCK(); 1944 if (algo->cb_init != NULL) { 1945 /* We can now pre-get the memory for the CC */ 1946 mem_sz = (*algo->cc_data_sz)(); 1947 if (mem_sz == 0) { 1948 goto no_mem_needed; 1949 } 1950 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1951 } else { 1952 no_mem_needed: 1953 mem_sz = 0; 1954 ptr = NULL; 1955 } 1956 /* 1957 * Make sure its all clean and zero and also get 1958 * back the inplock. 1959 */ 1960 memset(&cc_mem, 0, sizeof(cc_mem)); 1961 INP_WLOCK(inp); 1962 if (inp->inp_flags & INP_DROPPED) { 1963 INP_WUNLOCK(inp); 1964 if (ptr) 1965 free(ptr, M_CC_MEM); 1966 /* Release our temp reference */ 1967 CC_LIST_RLOCK(); 1968 cc_release(algo); 1969 CC_LIST_RUNLOCK(); 1970 return (ECONNRESET); 1971 } 1972 tp = intotcpcb(inp); 1973 if (ptr != NULL) 1974 memset(ptr, 0, mem_sz); 1975 cc_mem.ccvc.tcp = tp; 1976 /* 1977 * We once again hold a write lock over the tcb so it's 1978 * safe to do these things without ordering concerns. 1979 * Note here we init into stack memory. 1980 */ 1981 if (algo->cb_init != NULL) 1982 error = algo->cb_init(&cc_mem, ptr); 1983 else 1984 error = 0; 1985 /* 1986 * The CC algorithms, when given their memory 1987 * should not fail we could in theory have a 1988 * KASSERT here. 1989 */ 1990 if (error == 0) { 1991 /* 1992 * Touchdown, lets go ahead and move the 1993 * connection to the new CC module by 1994 * copying in the cc_mem after we call 1995 * the old ones cleanup (if any). 1996 */ 1997 if (CC_ALGO(tp)->cb_destroy != NULL) 1998 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 1999 /* Detach the old CC from the tcpcb */ 2000 cc_detach(tp); 2001 /* Copy in our temp memory that was inited */ 2002 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2003 /* Now attach the new, which takes a reference */ 2004 cc_attach(tp, algo); 2005 /* Ok now are we where we have gotten past any conn_init? */ 2006 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2007 /* Yep run the connection init for the new CC */ 2008 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2009 } 2010 } else if (ptr) 2011 free(ptr, M_CC_MEM); 2012 INP_WUNLOCK(inp); 2013 /* Now lets release our temp reference */ 2014 CC_LIST_RLOCK(); 2015 cc_release(algo); 2016 CC_LIST_RUNLOCK(); 2017 return (error); 2018 } 2019 2020 int 2021 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2022 { 2023 struct inpcb *inp = tptoinpcb(tp); 2024 int error, opt, optval; 2025 u_int ui; 2026 struct tcp_info ti; 2027 #ifdef KERN_TLS 2028 struct tls_enable tls; 2029 struct socket *so = inp->inp_socket; 2030 #endif 2031 char *pbuf, buf[TCP_LOG_ID_LEN]; 2032 #ifdef STATS 2033 struct statsblob *sbp; 2034 #endif 2035 size_t len; 2036 2037 INP_WLOCK_ASSERT(inp); 2038 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2039 ("inp_flags == %x", inp->inp_flags)); 2040 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2041 2042 switch (sopt->sopt_level) { 2043 #ifdef INET6 2044 case IPPROTO_IPV6: 2045 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2046 switch (sopt->sopt_name) { 2047 case IPV6_USE_MIN_MTU: 2048 tcp6_use_min_mtu(tp); 2049 /* FALLTHROUGH */ 2050 } 2051 INP_WUNLOCK(inp); 2052 return (0); 2053 #endif 2054 #ifdef INET 2055 case IPPROTO_IP: 2056 INP_WUNLOCK(inp); 2057 return (0); 2058 #endif 2059 } 2060 2061 /* 2062 * For TCP_CCALGOOPT forward the control to CC module, for both 2063 * SOPT_SET and SOPT_GET. 2064 */ 2065 switch (sopt->sopt_name) { 2066 case TCP_CCALGOOPT: 2067 INP_WUNLOCK(inp); 2068 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2069 return (EINVAL); 2070 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2071 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2072 sopt->sopt_valsize); 2073 if (error) { 2074 free(pbuf, M_TEMP); 2075 return (error); 2076 } 2077 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2078 if (CC_ALGO(tp)->ctl_output != NULL) 2079 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2080 else 2081 error = ENOENT; 2082 INP_WUNLOCK(inp); 2083 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2084 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2085 free(pbuf, M_TEMP); 2086 return (error); 2087 } 2088 2089 switch (sopt->sopt_dir) { 2090 case SOPT_SET: 2091 switch (sopt->sopt_name) { 2092 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2093 case TCP_MD5SIG: 2094 INP_WUNLOCK(inp); 2095 if (!TCPMD5_ENABLED()) 2096 return (ENOPROTOOPT); 2097 error = TCPMD5_PCBCTL(inp, sopt); 2098 if (error) 2099 return (error); 2100 INP_WLOCK_RECHECK(inp); 2101 goto unlock_and_done; 2102 #endif /* IPSEC */ 2103 2104 case TCP_NODELAY: 2105 case TCP_NOOPT: 2106 INP_WUNLOCK(inp); 2107 error = sooptcopyin(sopt, &optval, sizeof optval, 2108 sizeof optval); 2109 if (error) 2110 return (error); 2111 2112 INP_WLOCK_RECHECK(inp); 2113 switch (sopt->sopt_name) { 2114 case TCP_NODELAY: 2115 opt = TF_NODELAY; 2116 break; 2117 case TCP_NOOPT: 2118 opt = TF_NOOPT; 2119 break; 2120 default: 2121 opt = 0; /* dead code to fool gcc */ 2122 break; 2123 } 2124 2125 if (optval) 2126 tp->t_flags |= opt; 2127 else 2128 tp->t_flags &= ~opt; 2129 unlock_and_done: 2130 #ifdef TCP_OFFLOAD 2131 if (tp->t_flags & TF_TOE) { 2132 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2133 sopt->sopt_name); 2134 } 2135 #endif 2136 INP_WUNLOCK(inp); 2137 break; 2138 2139 case TCP_NOPUSH: 2140 INP_WUNLOCK(inp); 2141 error = sooptcopyin(sopt, &optval, sizeof optval, 2142 sizeof optval); 2143 if (error) 2144 return (error); 2145 2146 INP_WLOCK_RECHECK(inp); 2147 if (optval) 2148 tp->t_flags |= TF_NOPUSH; 2149 else if (tp->t_flags & TF_NOPUSH) { 2150 tp->t_flags &= ~TF_NOPUSH; 2151 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2152 struct epoch_tracker et; 2153 2154 NET_EPOCH_ENTER(et); 2155 error = tcp_output_nodrop(tp); 2156 NET_EPOCH_EXIT(et); 2157 } 2158 } 2159 goto unlock_and_done; 2160 2161 case TCP_REMOTE_UDP_ENCAPS_PORT: 2162 INP_WUNLOCK(inp); 2163 error = sooptcopyin(sopt, &optval, sizeof optval, 2164 sizeof optval); 2165 if (error) 2166 return (error); 2167 if ((optval < TCP_TUNNELING_PORT_MIN) || 2168 (optval > TCP_TUNNELING_PORT_MAX)) { 2169 /* Its got to be in range */ 2170 return (EINVAL); 2171 } 2172 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2173 /* You have to have enabled a UDP tunneling port first */ 2174 return (EINVAL); 2175 } 2176 INP_WLOCK_RECHECK(inp); 2177 if (tp->t_state != TCPS_CLOSED) { 2178 /* You can't change after you are connected */ 2179 error = EINVAL; 2180 } else { 2181 /* Ok we are all good set the port */ 2182 tp->t_port = htons(optval); 2183 } 2184 goto unlock_and_done; 2185 2186 case TCP_MAXSEG: 2187 INP_WUNLOCK(inp); 2188 error = sooptcopyin(sopt, &optval, sizeof optval, 2189 sizeof optval); 2190 if (error) 2191 return (error); 2192 2193 INP_WLOCK_RECHECK(inp); 2194 if (optval > 0 && optval <= tp->t_maxseg && 2195 optval + 40 >= V_tcp_minmss) { 2196 tp->t_maxseg = optval; 2197 if (tp->t_maxseg < V_tcp_mssdflt) { 2198 /* 2199 * The MSS is so small we should not process incoming 2200 * SACK's since we are subject to attack in such a 2201 * case. 2202 */ 2203 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 2204 } else { 2205 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 2206 } 2207 } else 2208 error = EINVAL; 2209 goto unlock_and_done; 2210 2211 case TCP_INFO: 2212 INP_WUNLOCK(inp); 2213 error = EINVAL; 2214 break; 2215 2216 case TCP_STATS: 2217 INP_WUNLOCK(inp); 2218 #ifdef STATS 2219 error = sooptcopyin(sopt, &optval, sizeof optval, 2220 sizeof optval); 2221 if (error) 2222 return (error); 2223 2224 if (optval > 0) 2225 sbp = stats_blob_alloc( 2226 V_tcp_perconn_stats_dflt_tpl, 0); 2227 else 2228 sbp = NULL; 2229 2230 INP_WLOCK_RECHECK(inp); 2231 if ((tp->t_stats != NULL && sbp == NULL) || 2232 (tp->t_stats == NULL && sbp != NULL)) { 2233 struct statsblob *t = tp->t_stats; 2234 tp->t_stats = sbp; 2235 sbp = t; 2236 } 2237 INP_WUNLOCK(inp); 2238 2239 stats_blob_destroy(sbp); 2240 #else 2241 return (EOPNOTSUPP); 2242 #endif /* !STATS */ 2243 break; 2244 2245 case TCP_CONGESTION: 2246 error = tcp_set_cc_mod(inp, sopt); 2247 break; 2248 2249 case TCP_REUSPORT_LB_NUMA: 2250 INP_WUNLOCK(inp); 2251 error = sooptcopyin(sopt, &optval, sizeof(optval), 2252 sizeof(optval)); 2253 INP_WLOCK_RECHECK(inp); 2254 if (!error) 2255 error = in_pcblbgroup_numa(inp, optval); 2256 INP_WUNLOCK(inp); 2257 break; 2258 2259 #ifdef KERN_TLS 2260 case TCP_TXTLS_ENABLE: 2261 INP_WUNLOCK(inp); 2262 error = ktls_copyin_tls_enable(sopt, &tls); 2263 if (error != 0) 2264 break; 2265 error = ktls_enable_tx(so, &tls); 2266 ktls_cleanup_tls_enable(&tls); 2267 break; 2268 case TCP_TXTLS_MODE: 2269 INP_WUNLOCK(inp); 2270 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2271 if (error != 0) 2272 return (error); 2273 2274 INP_WLOCK_RECHECK(inp); 2275 error = ktls_set_tx_mode(so, ui); 2276 INP_WUNLOCK(inp); 2277 break; 2278 case TCP_RXTLS_ENABLE: 2279 INP_WUNLOCK(inp); 2280 error = ktls_copyin_tls_enable(sopt, &tls); 2281 if (error != 0) 2282 break; 2283 error = ktls_enable_rx(so, &tls); 2284 ktls_cleanup_tls_enable(&tls); 2285 break; 2286 #endif 2287 case TCP_MAXUNACKTIME: 2288 case TCP_KEEPIDLE: 2289 case TCP_KEEPINTVL: 2290 case TCP_KEEPINIT: 2291 INP_WUNLOCK(inp); 2292 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2293 if (error) 2294 return (error); 2295 2296 if (ui > (UINT_MAX / hz)) { 2297 error = EINVAL; 2298 break; 2299 } 2300 ui *= hz; 2301 2302 INP_WLOCK_RECHECK(inp); 2303 switch (sopt->sopt_name) { 2304 case TCP_MAXUNACKTIME: 2305 tp->t_maxunacktime = ui; 2306 break; 2307 2308 case TCP_KEEPIDLE: 2309 tp->t_keepidle = ui; 2310 /* 2311 * XXX: better check current remaining 2312 * timeout and "merge" it with new value. 2313 */ 2314 if ((tp->t_state > TCPS_LISTEN) && 2315 (tp->t_state <= TCPS_CLOSING)) 2316 tcp_timer_activate(tp, TT_KEEP, 2317 TP_KEEPIDLE(tp)); 2318 break; 2319 case TCP_KEEPINTVL: 2320 tp->t_keepintvl = ui; 2321 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2322 (TP_MAXIDLE(tp) > 0)) 2323 tcp_timer_activate(tp, TT_2MSL, 2324 TP_MAXIDLE(tp)); 2325 break; 2326 case TCP_KEEPINIT: 2327 tp->t_keepinit = ui; 2328 if (tp->t_state == TCPS_SYN_RECEIVED || 2329 tp->t_state == TCPS_SYN_SENT) 2330 tcp_timer_activate(tp, TT_KEEP, 2331 TP_KEEPINIT(tp)); 2332 break; 2333 } 2334 goto unlock_and_done; 2335 2336 case TCP_KEEPCNT: 2337 INP_WUNLOCK(inp); 2338 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2339 if (error) 2340 return (error); 2341 2342 INP_WLOCK_RECHECK(inp); 2343 tp->t_keepcnt = ui; 2344 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2345 (TP_MAXIDLE(tp) > 0)) 2346 tcp_timer_activate(tp, TT_2MSL, 2347 TP_MAXIDLE(tp)); 2348 goto unlock_and_done; 2349 2350 #ifdef TCPPCAP 2351 case TCP_PCAP_OUT: 2352 case TCP_PCAP_IN: 2353 INP_WUNLOCK(inp); 2354 error = sooptcopyin(sopt, &optval, sizeof optval, 2355 sizeof optval); 2356 if (error) 2357 return (error); 2358 2359 INP_WLOCK_RECHECK(inp); 2360 if (optval >= 0) 2361 tcp_pcap_set_sock_max( 2362 (sopt->sopt_name == TCP_PCAP_OUT) ? 2363 &(tp->t_outpkts) : &(tp->t_inpkts), 2364 optval); 2365 else 2366 error = EINVAL; 2367 goto unlock_and_done; 2368 #endif 2369 2370 case TCP_FASTOPEN: { 2371 struct tcp_fastopen tfo_optval; 2372 2373 INP_WUNLOCK(inp); 2374 if (!V_tcp_fastopen_client_enable && 2375 !V_tcp_fastopen_server_enable) 2376 return (EPERM); 2377 2378 error = sooptcopyin(sopt, &tfo_optval, 2379 sizeof(tfo_optval), sizeof(int)); 2380 if (error) 2381 return (error); 2382 2383 INP_WLOCK_RECHECK(inp); 2384 if ((tp->t_state != TCPS_CLOSED) && 2385 (tp->t_state != TCPS_LISTEN)) { 2386 error = EINVAL; 2387 goto unlock_and_done; 2388 } 2389 if (tfo_optval.enable) { 2390 if (tp->t_state == TCPS_LISTEN) { 2391 if (!V_tcp_fastopen_server_enable) { 2392 error = EPERM; 2393 goto unlock_and_done; 2394 } 2395 2396 if (tp->t_tfo_pending == NULL) 2397 tp->t_tfo_pending = 2398 tcp_fastopen_alloc_counter(); 2399 } else { 2400 /* 2401 * If a pre-shared key was provided, 2402 * stash it in the client cookie 2403 * field of the tcpcb for use during 2404 * connect. 2405 */ 2406 if (sopt->sopt_valsize == 2407 sizeof(tfo_optval)) { 2408 memcpy(tp->t_tfo_cookie.client, 2409 tfo_optval.psk, 2410 TCP_FASTOPEN_PSK_LEN); 2411 tp->t_tfo_client_cookie_len = 2412 TCP_FASTOPEN_PSK_LEN; 2413 } 2414 } 2415 tp->t_flags |= TF_FASTOPEN; 2416 } else 2417 tp->t_flags &= ~TF_FASTOPEN; 2418 goto unlock_and_done; 2419 } 2420 2421 #ifdef TCP_BLACKBOX 2422 case TCP_LOG: 2423 INP_WUNLOCK(inp); 2424 error = sooptcopyin(sopt, &optval, sizeof optval, 2425 sizeof optval); 2426 if (error) 2427 return (error); 2428 2429 INP_WLOCK_RECHECK(inp); 2430 error = tcp_log_state_change(tp, optval); 2431 goto unlock_and_done; 2432 2433 case TCP_LOGBUF: 2434 INP_WUNLOCK(inp); 2435 error = EINVAL; 2436 break; 2437 2438 case TCP_LOGID: 2439 INP_WUNLOCK(inp); 2440 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2441 if (error) 2442 break; 2443 buf[sopt->sopt_valsize] = '\0'; 2444 INP_WLOCK_RECHECK(inp); 2445 error = tcp_log_set_id(tp, buf); 2446 /* tcp_log_set_id() unlocks the INP. */ 2447 break; 2448 2449 case TCP_LOGDUMP: 2450 case TCP_LOGDUMPID: 2451 INP_WUNLOCK(inp); 2452 error = 2453 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2454 if (error) 2455 break; 2456 buf[sopt->sopt_valsize] = '\0'; 2457 INP_WLOCK_RECHECK(inp); 2458 if (sopt->sopt_name == TCP_LOGDUMP) { 2459 error = tcp_log_dump_tp_logbuf(tp, buf, 2460 M_WAITOK, true); 2461 INP_WUNLOCK(inp); 2462 } else { 2463 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2464 /* 2465 * tcp_log_dump_tp_bucket_logbufs() drops the 2466 * INP lock. 2467 */ 2468 } 2469 break; 2470 #endif 2471 2472 default: 2473 INP_WUNLOCK(inp); 2474 error = ENOPROTOOPT; 2475 break; 2476 } 2477 break; 2478 2479 case SOPT_GET: 2480 tp = intotcpcb(inp); 2481 switch (sopt->sopt_name) { 2482 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2483 case TCP_MD5SIG: 2484 INP_WUNLOCK(inp); 2485 if (!TCPMD5_ENABLED()) 2486 return (ENOPROTOOPT); 2487 error = TCPMD5_PCBCTL(inp, sopt); 2488 break; 2489 #endif 2490 2491 case TCP_NODELAY: 2492 optval = tp->t_flags & TF_NODELAY; 2493 INP_WUNLOCK(inp); 2494 error = sooptcopyout(sopt, &optval, sizeof optval); 2495 break; 2496 case TCP_MAXSEG: 2497 optval = tp->t_maxseg; 2498 INP_WUNLOCK(inp); 2499 error = sooptcopyout(sopt, &optval, sizeof optval); 2500 break; 2501 case TCP_REMOTE_UDP_ENCAPS_PORT: 2502 optval = ntohs(tp->t_port); 2503 INP_WUNLOCK(inp); 2504 error = sooptcopyout(sopt, &optval, sizeof optval); 2505 break; 2506 case TCP_NOOPT: 2507 optval = tp->t_flags & TF_NOOPT; 2508 INP_WUNLOCK(inp); 2509 error = sooptcopyout(sopt, &optval, sizeof optval); 2510 break; 2511 case TCP_NOPUSH: 2512 optval = tp->t_flags & TF_NOPUSH; 2513 INP_WUNLOCK(inp); 2514 error = sooptcopyout(sopt, &optval, sizeof optval); 2515 break; 2516 case TCP_INFO: 2517 tcp_fill_info(tp, &ti); 2518 INP_WUNLOCK(inp); 2519 error = sooptcopyout(sopt, &ti, sizeof ti); 2520 break; 2521 case TCP_STATS: 2522 { 2523 #ifdef STATS 2524 int nheld; 2525 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2526 2527 error = 0; 2528 socklen_t outsbsz = sopt->sopt_valsize; 2529 if (tp->t_stats == NULL) 2530 error = ENOENT; 2531 else if (outsbsz >= tp->t_stats->cursz) 2532 outsbsz = tp->t_stats->cursz; 2533 else if (outsbsz >= sizeof(struct statsblob)) 2534 outsbsz = sizeof(struct statsblob); 2535 else 2536 error = EINVAL; 2537 INP_WUNLOCK(inp); 2538 if (error) 2539 break; 2540 2541 sbp = sopt->sopt_val; 2542 nheld = atop(round_page(((vm_offset_t)sbp) + 2543 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2544 vm_page_t ma[nheld]; 2545 if (vm_fault_quick_hold_pages( 2546 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2547 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2548 nheld) < 0) { 2549 error = EFAULT; 2550 break; 2551 } 2552 2553 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2554 SIZEOF_MEMBER(struct statsblob, flags)))) 2555 goto unhold; 2556 2557 INP_WLOCK_RECHECK(inp); 2558 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2559 sbflags | SB_CLONE_USRDSTNOFAULT); 2560 INP_WUNLOCK(inp); 2561 sopt->sopt_valsize = outsbsz; 2562 unhold: 2563 vm_page_unhold_pages(ma, nheld); 2564 #else 2565 INP_WUNLOCK(inp); 2566 error = EOPNOTSUPP; 2567 #endif /* !STATS */ 2568 break; 2569 } 2570 case TCP_CONGESTION: 2571 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2572 INP_WUNLOCK(inp); 2573 error = sooptcopyout(sopt, buf, len + 1); 2574 break; 2575 case TCP_MAXUNACKTIME: 2576 case TCP_KEEPIDLE: 2577 case TCP_KEEPINTVL: 2578 case TCP_KEEPINIT: 2579 case TCP_KEEPCNT: 2580 switch (sopt->sopt_name) { 2581 case TCP_MAXUNACKTIME: 2582 ui = TP_MAXUNACKTIME(tp) / hz; 2583 break; 2584 case TCP_KEEPIDLE: 2585 ui = TP_KEEPIDLE(tp) / hz; 2586 break; 2587 case TCP_KEEPINTVL: 2588 ui = TP_KEEPINTVL(tp) / hz; 2589 break; 2590 case TCP_KEEPINIT: 2591 ui = TP_KEEPINIT(tp) / hz; 2592 break; 2593 case TCP_KEEPCNT: 2594 ui = TP_KEEPCNT(tp); 2595 break; 2596 } 2597 INP_WUNLOCK(inp); 2598 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2599 break; 2600 #ifdef TCPPCAP 2601 case TCP_PCAP_OUT: 2602 case TCP_PCAP_IN: 2603 optval = tcp_pcap_get_sock_max( 2604 (sopt->sopt_name == TCP_PCAP_OUT) ? 2605 &(tp->t_outpkts) : &(tp->t_inpkts)); 2606 INP_WUNLOCK(inp); 2607 error = sooptcopyout(sopt, &optval, sizeof optval); 2608 break; 2609 #endif 2610 case TCP_FASTOPEN: 2611 optval = tp->t_flags & TF_FASTOPEN; 2612 INP_WUNLOCK(inp); 2613 error = sooptcopyout(sopt, &optval, sizeof optval); 2614 break; 2615 #ifdef TCP_BLACKBOX 2616 case TCP_LOG: 2617 optval = tcp_get_bblog_state(tp); 2618 INP_WUNLOCK(inp); 2619 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2620 break; 2621 case TCP_LOGBUF: 2622 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2623 error = tcp_log_getlogbuf(sopt, tp); 2624 break; 2625 case TCP_LOGID: 2626 len = tcp_log_get_id(tp, buf); 2627 INP_WUNLOCK(inp); 2628 error = sooptcopyout(sopt, buf, len + 1); 2629 break; 2630 case TCP_LOGDUMP: 2631 case TCP_LOGDUMPID: 2632 INP_WUNLOCK(inp); 2633 error = EINVAL; 2634 break; 2635 #endif 2636 #ifdef KERN_TLS 2637 case TCP_TXTLS_MODE: 2638 error = ktls_get_tx_mode(so, &optval); 2639 INP_WUNLOCK(inp); 2640 if (error == 0) 2641 error = sooptcopyout(sopt, &optval, 2642 sizeof(optval)); 2643 break; 2644 case TCP_RXTLS_MODE: 2645 error = ktls_get_rx_mode(so, &optval); 2646 INP_WUNLOCK(inp); 2647 if (error == 0) 2648 error = sooptcopyout(sopt, &optval, 2649 sizeof(optval)); 2650 break; 2651 #endif 2652 default: 2653 INP_WUNLOCK(inp); 2654 error = ENOPROTOOPT; 2655 break; 2656 } 2657 break; 2658 } 2659 return (error); 2660 } 2661 #undef INP_WLOCK_RECHECK 2662 #undef INP_WLOCK_RECHECK_CLEANUP 2663 2664 /* 2665 * Initiate (or continue) disconnect. 2666 * If embryonic state, just send reset (once). 2667 * If in ``let data drain'' option and linger null, just drop. 2668 * Otherwise (hard), mark socket disconnecting and drop 2669 * current input data; switch states based on user close, and 2670 * send segment to peer (with FIN). 2671 */ 2672 static void 2673 tcp_disconnect(struct tcpcb *tp) 2674 { 2675 struct inpcb *inp = tptoinpcb(tp); 2676 struct socket *so = tptosocket(tp); 2677 2678 NET_EPOCH_ASSERT(); 2679 INP_WLOCK_ASSERT(inp); 2680 2681 /* 2682 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2683 * socket is still open. 2684 */ 2685 if (tp->t_state < TCPS_ESTABLISHED && 2686 !(tp->t_state > TCPS_LISTEN && (tp->t_flags & TF_FASTOPEN))) { 2687 tp = tcp_close(tp); 2688 KASSERT(tp != NULL, 2689 ("tcp_disconnect: tcp_close() returned NULL")); 2690 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2691 tp = tcp_drop(tp, 0); 2692 KASSERT(tp != NULL, 2693 ("tcp_disconnect: tcp_drop() returned NULL")); 2694 } else { 2695 soisdisconnecting(so); 2696 sbflush(&so->so_rcv); 2697 tcp_usrclosed(tp); 2698 if (!(inp->inp_flags & INP_DROPPED)) 2699 /* Ignore stack's drop request, we already at it. */ 2700 (void)tcp_output_nodrop(tp); 2701 } 2702 } 2703 2704 /* 2705 * User issued close, and wish to trail through shutdown states: 2706 * if never received SYN, just forget it. If got a SYN from peer, 2707 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2708 * If already got a FIN from peer, then almost done; go to LAST_ACK 2709 * state. In all other cases, have already sent FIN to peer (e.g. 2710 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2711 * for peer to send FIN or not respond to keep-alives, etc. 2712 * We can let the user exit from the close as soon as the FIN is acked. 2713 */ 2714 static void 2715 tcp_usrclosed(struct tcpcb *tp) 2716 { 2717 2718 NET_EPOCH_ASSERT(); 2719 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2720 2721 switch (tp->t_state) { 2722 case TCPS_LISTEN: 2723 #ifdef TCP_OFFLOAD 2724 tcp_offload_listen_stop(tp); 2725 #endif 2726 tcp_state_change(tp, TCPS_CLOSED); 2727 /* FALLTHROUGH */ 2728 case TCPS_CLOSED: 2729 tp = tcp_close(tp); 2730 /* 2731 * tcp_close() should never return NULL here as the socket is 2732 * still open. 2733 */ 2734 KASSERT(tp != NULL, 2735 ("tcp_usrclosed: tcp_close() returned NULL")); 2736 break; 2737 2738 case TCPS_SYN_SENT: 2739 case TCPS_SYN_RECEIVED: 2740 tp->t_flags |= TF_NEEDFIN; 2741 break; 2742 2743 case TCPS_ESTABLISHED: 2744 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2745 break; 2746 2747 case TCPS_CLOSE_WAIT: 2748 tcp_state_change(tp, TCPS_LAST_ACK); 2749 break; 2750 } 2751 if (tp->t_acktime == 0) 2752 tp->t_acktime = ticks; 2753 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2754 tcp_free_sackholes(tp); 2755 soisdisconnected(tptosocket(tp)); 2756 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2757 if (tp->t_state == TCPS_FIN_WAIT_2) { 2758 int timeout; 2759 2760 timeout = (tcp_fast_finwait2_recycle) ? 2761 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2762 tcp_timer_activate(tp, TT_2MSL, timeout); 2763 } 2764 } 2765 } 2766 2767 #ifdef DDB 2768 static void 2769 db_print_indent(int indent) 2770 { 2771 int i; 2772 2773 for (i = 0; i < indent; i++) 2774 db_printf(" "); 2775 } 2776 2777 static void 2778 db_print_tstate(int t_state) 2779 { 2780 2781 switch (t_state) { 2782 case TCPS_CLOSED: 2783 db_printf("TCPS_CLOSED"); 2784 return; 2785 2786 case TCPS_LISTEN: 2787 db_printf("TCPS_LISTEN"); 2788 return; 2789 2790 case TCPS_SYN_SENT: 2791 db_printf("TCPS_SYN_SENT"); 2792 return; 2793 2794 case TCPS_SYN_RECEIVED: 2795 db_printf("TCPS_SYN_RECEIVED"); 2796 return; 2797 2798 case TCPS_ESTABLISHED: 2799 db_printf("TCPS_ESTABLISHED"); 2800 return; 2801 2802 case TCPS_CLOSE_WAIT: 2803 db_printf("TCPS_CLOSE_WAIT"); 2804 return; 2805 2806 case TCPS_FIN_WAIT_1: 2807 db_printf("TCPS_FIN_WAIT_1"); 2808 return; 2809 2810 case TCPS_CLOSING: 2811 db_printf("TCPS_CLOSING"); 2812 return; 2813 2814 case TCPS_LAST_ACK: 2815 db_printf("TCPS_LAST_ACK"); 2816 return; 2817 2818 case TCPS_FIN_WAIT_2: 2819 db_printf("TCPS_FIN_WAIT_2"); 2820 return; 2821 2822 case TCPS_TIME_WAIT: 2823 db_printf("TCPS_TIME_WAIT"); 2824 return; 2825 2826 default: 2827 db_printf("unknown"); 2828 return; 2829 } 2830 } 2831 2832 static void 2833 db_print_tflags(u_int t_flags) 2834 { 2835 int comma; 2836 2837 comma = 0; 2838 if (t_flags & TF_ACKNOW) { 2839 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2840 comma = 1; 2841 } 2842 if (t_flags & TF_DELACK) { 2843 db_printf("%sTF_DELACK", comma ? ", " : ""); 2844 comma = 1; 2845 } 2846 if (t_flags & TF_NODELAY) { 2847 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2848 comma = 1; 2849 } 2850 if (t_flags & TF_NOOPT) { 2851 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2852 comma = 1; 2853 } 2854 if (t_flags & TF_SENTFIN) { 2855 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2856 comma = 1; 2857 } 2858 if (t_flags & TF_REQ_SCALE) { 2859 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2860 comma = 1; 2861 } 2862 if (t_flags & TF_RCVD_SCALE) { 2863 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2864 comma = 1; 2865 } 2866 if (t_flags & TF_REQ_TSTMP) { 2867 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2868 comma = 1; 2869 } 2870 if (t_flags & TF_RCVD_TSTMP) { 2871 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2872 comma = 1; 2873 } 2874 if (t_flags & TF_SACK_PERMIT) { 2875 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2876 comma = 1; 2877 } 2878 if (t_flags & TF_NEEDSYN) { 2879 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2880 comma = 1; 2881 } 2882 if (t_flags & TF_NEEDFIN) { 2883 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2884 comma = 1; 2885 } 2886 if (t_flags & TF_NOPUSH) { 2887 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2888 comma = 1; 2889 } 2890 if (t_flags & TF_PREVVALID) { 2891 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2892 comma = 1; 2893 } 2894 if (t_flags & TF_MORETOCOME) { 2895 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2896 comma = 1; 2897 } 2898 if (t_flags & TF_SONOTCONN) { 2899 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2900 comma = 1; 2901 } 2902 if (t_flags & TF_LASTIDLE) { 2903 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2904 comma = 1; 2905 } 2906 if (t_flags & TF_RXWIN0SENT) { 2907 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2908 comma = 1; 2909 } 2910 if (t_flags & TF_FASTRECOVERY) { 2911 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2912 comma = 1; 2913 } 2914 if (t_flags & TF_CONGRECOVERY) { 2915 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2916 comma = 1; 2917 } 2918 if (t_flags & TF_WASFRECOVERY) { 2919 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2920 comma = 1; 2921 } 2922 if (t_flags & TF_WASCRECOVERY) { 2923 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2924 comma = 1; 2925 } 2926 if (t_flags & TF_SIGNATURE) { 2927 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2928 comma = 1; 2929 } 2930 if (t_flags & TF_FORCEDATA) { 2931 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2932 comma = 1; 2933 } 2934 if (t_flags & TF_TSO) { 2935 db_printf("%sTF_TSO", comma ? ", " : ""); 2936 comma = 1; 2937 } 2938 if (t_flags & TF_FASTOPEN) { 2939 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2940 comma = 1; 2941 } 2942 } 2943 2944 static void 2945 db_print_tflags2(u_int t_flags2) 2946 { 2947 int comma; 2948 2949 comma = 0; 2950 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2951 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2952 comma = 1; 2953 } 2954 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2955 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2956 comma = 1; 2957 } 2958 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2959 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2960 comma = 1; 2961 } 2962 if (t_flags2 & TF2_LOG_AUTO) { 2963 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2964 comma = 1; 2965 } 2966 if (t_flags2 & TF2_DROP_AF_DATA) { 2967 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2968 comma = 1; 2969 } 2970 if (t_flags2 & TF2_ECN_PERMIT) { 2971 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2972 comma = 1; 2973 } 2974 if (t_flags2 & TF2_ECN_SND_CWR) { 2975 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 2976 comma = 1; 2977 } 2978 if (t_flags2 & TF2_ECN_SND_ECE) { 2979 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 2980 comma = 1; 2981 } 2982 if (t_flags2 & TF2_ACE_PERMIT) { 2983 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 2984 comma = 1; 2985 } 2986 if (t_flags2 & TF2_FBYTES_COMPLETE) { 2987 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 2988 comma = 1; 2989 } 2990 } 2991 2992 static void 2993 db_print_toobflags(char t_oobflags) 2994 { 2995 int comma; 2996 2997 comma = 0; 2998 if (t_oobflags & TCPOOB_HAVEDATA) { 2999 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3000 comma = 1; 3001 } 3002 if (t_oobflags & TCPOOB_HADDATA) { 3003 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3004 comma = 1; 3005 } 3006 } 3007 3008 static void 3009 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3010 { 3011 3012 db_print_indent(indent); 3013 db_printf("%s at %p\n", name, tp); 3014 3015 indent += 2; 3016 3017 db_print_indent(indent); 3018 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3019 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3020 3021 db_print_indent(indent); 3022 db_printf("t_callout: %p t_timers: %p\n", 3023 &tp->t_callout, &tp->t_timers); 3024 3025 db_print_indent(indent); 3026 db_printf("t_state: %d (", tp->t_state); 3027 db_print_tstate(tp->t_state); 3028 db_printf(")\n"); 3029 3030 db_print_indent(indent); 3031 db_printf("t_flags: 0x%x (", tp->t_flags); 3032 db_print_tflags(tp->t_flags); 3033 db_printf(")\n"); 3034 3035 db_print_indent(indent); 3036 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3037 db_print_tflags2(tp->t_flags2); 3038 db_printf(")\n"); 3039 3040 db_print_indent(indent); 3041 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 3042 tp->snd_una, tp->snd_max, tp->snd_nxt); 3043 3044 db_print_indent(indent); 3045 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3046 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3047 3048 db_print_indent(indent); 3049 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3050 tp->iss, tp->irs, tp->rcv_nxt); 3051 3052 db_print_indent(indent); 3053 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3054 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3055 3056 db_print_indent(indent); 3057 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3058 tp->snd_wnd, tp->snd_cwnd); 3059 3060 db_print_indent(indent); 3061 db_printf("snd_ssthresh: %u snd_recover: " 3062 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3063 3064 db_print_indent(indent); 3065 db_printf("t_rcvtime: %u t_startime: %u\n", 3066 tp->t_rcvtime, tp->t_starttime); 3067 3068 db_print_indent(indent); 3069 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3070 tp->t_rtttime, tp->t_rtseq); 3071 3072 db_print_indent(indent); 3073 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3074 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3075 3076 db_print_indent(indent); 3077 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3078 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3079 3080 db_print_indent(indent); 3081 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3082 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3083 3084 db_print_indent(indent); 3085 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3086 db_print_toobflags(tp->t_oobflags); 3087 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3088 3089 db_print_indent(indent); 3090 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3091 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3092 3093 db_print_indent(indent); 3094 db_printf("ts_recent: %u ts_recent_age: %u\n", 3095 tp->ts_recent, tp->ts_recent_age); 3096 3097 db_print_indent(indent); 3098 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3099 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3100 3101 db_print_indent(indent); 3102 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3103 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3104 tp->snd_recover_prev, tp->t_badrxtwin); 3105 3106 db_print_indent(indent); 3107 db_printf("snd_numholes: %d snd_holes first: %p\n", 3108 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3109 3110 db_print_indent(indent); 3111 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3112 tp->snd_fack, tp->rcv_numsacks); 3113 3114 /* Skip sackblks, sackhint. */ 3115 3116 db_print_indent(indent); 3117 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3118 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3119 } 3120 3121 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3122 { 3123 struct tcpcb *tp; 3124 3125 if (!have_addr) { 3126 db_printf("usage: show tcpcb <addr>\n"); 3127 return; 3128 } 3129 tp = (struct tcpcb *)addr; 3130 3131 db_print_tcpcb(tp, "tcpcb", 0); 3132 } 3133 #endif 3134