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 /* Can we inherit anything from the listener? */ 183 if ((so->so_listen != NULL) && 184 (so->so_listen->so_pcb != NULL) && 185 (tp->t_fb->tfb_inherit != NULL)) { 186 (*tp->t_fb->tfb_inherit)(tp, sotoinpcb(so->so_listen)); 187 } 188 tcp_bblog_pru(tp, PRU_ATTACH, error); 189 INP_WUNLOCK(inp); 190 TCPSTATES_INC(TCPS_CLOSED); 191 out: 192 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 193 return (error); 194 } 195 196 /* 197 * tcp_usr_detach is called when the socket layer loses its final reference 198 * to the socket, be it a file descriptor reference, a reference from TCP, 199 * etc. At this point, there is only one case in which we will keep around 200 * inpcb state: time wait. 201 */ 202 static void 203 tcp_usr_detach(struct socket *so) 204 { 205 struct inpcb *inp; 206 struct tcpcb *tp; 207 208 inp = sotoinpcb(so); 209 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 210 INP_WLOCK(inp); 211 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 212 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 213 214 tp = intotcpcb(inp); 215 216 KASSERT(inp->inp_flags & INP_DROPPED || 217 tp->t_state < TCPS_SYN_SENT, 218 ("%s: inp %p not dropped or embryonic", __func__, inp)); 219 220 tcp_discardcb(tp); 221 in_pcbfree(inp); 222 } 223 224 #ifdef INET 225 /* 226 * Give the socket an address. 227 */ 228 static int 229 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 230 { 231 int error = 0; 232 struct inpcb *inp; 233 struct tcpcb *tp; 234 struct sockaddr_in *sinp; 235 236 inp = sotoinpcb(so); 237 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 238 INP_WLOCK(inp); 239 if (inp->inp_flags & INP_DROPPED) { 240 INP_WUNLOCK(inp); 241 return (EINVAL); 242 } 243 tp = intotcpcb(inp); 244 245 sinp = (struct sockaddr_in *)nam; 246 if (nam->sa_family != AF_INET) { 247 /* 248 * Preserve compatibility with old programs. 249 */ 250 if (nam->sa_family != AF_UNSPEC || 251 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 252 sinp->sin_addr.s_addr != INADDR_ANY) { 253 error = EAFNOSUPPORT; 254 goto out; 255 } 256 nam->sa_family = AF_INET; 257 } 258 if (nam->sa_len != sizeof(*sinp)) { 259 error = EINVAL; 260 goto out; 261 } 262 /* 263 * Must check for multicast addresses and disallow binding 264 * to them. 265 */ 266 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 267 error = EAFNOSUPPORT; 268 goto out; 269 } 270 INP_HASH_WLOCK(&V_tcbinfo); 271 error = in_pcbbind(inp, sinp, td->td_ucred); 272 INP_HASH_WUNLOCK(&V_tcbinfo); 273 out: 274 tcp_bblog_pru(tp, PRU_BIND, error); 275 TCP_PROBE2(debug__user, tp, PRU_BIND); 276 INP_WUNLOCK(inp); 277 278 return (error); 279 } 280 #endif /* INET */ 281 282 #ifdef INET6 283 static int 284 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 285 { 286 int error = 0; 287 struct inpcb *inp; 288 struct tcpcb *tp; 289 struct sockaddr_in6 *sin6; 290 u_char vflagsav; 291 292 inp = sotoinpcb(so); 293 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 294 INP_WLOCK(inp); 295 if (inp->inp_flags & INP_DROPPED) { 296 INP_WUNLOCK(inp); 297 return (EINVAL); 298 } 299 tp = intotcpcb(inp); 300 301 vflagsav = inp->inp_vflag; 302 303 sin6 = (struct sockaddr_in6 *)nam; 304 if (nam->sa_family != AF_INET6) { 305 error = EAFNOSUPPORT; 306 goto out; 307 } 308 if (nam->sa_len != sizeof(*sin6)) { 309 error = EINVAL; 310 goto out; 311 } 312 /* 313 * Must check for multicast addresses and disallow binding 314 * to them. 315 */ 316 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 317 error = EAFNOSUPPORT; 318 goto out; 319 } 320 321 INP_HASH_WLOCK(&V_tcbinfo); 322 inp->inp_vflag &= ~INP_IPV4; 323 inp->inp_vflag |= INP_IPV6; 324 #ifdef INET 325 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 326 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 327 inp->inp_vflag |= INP_IPV4; 328 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 329 struct sockaddr_in sin; 330 331 in6_sin6_2_sin(&sin, sin6); 332 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 333 error = EAFNOSUPPORT; 334 INP_HASH_WUNLOCK(&V_tcbinfo); 335 goto out; 336 } 337 inp->inp_vflag |= INP_IPV4; 338 inp->inp_vflag &= ~INP_IPV6; 339 error = in_pcbbind(inp, &sin, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 goto out; 342 } 343 } 344 #endif 345 error = in6_pcbbind(inp, sin6, td->td_ucred); 346 INP_HASH_WUNLOCK(&V_tcbinfo); 347 out: 348 if (error != 0) 349 inp->inp_vflag = vflagsav; 350 tcp_bblog_pru(tp, PRU_BIND, error); 351 TCP_PROBE2(debug__user, tp, PRU_BIND); 352 INP_WUNLOCK(inp); 353 return (error); 354 } 355 #endif /* INET6 */ 356 357 #ifdef INET 358 /* 359 * Prepare to accept connections. 360 */ 361 static int 362 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 363 { 364 int error = 0; 365 struct inpcb *inp; 366 struct tcpcb *tp; 367 368 inp = sotoinpcb(so); 369 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 370 INP_WLOCK(inp); 371 if (inp->inp_flags & INP_DROPPED) { 372 INP_WUNLOCK(inp); 373 return (EINVAL); 374 } 375 tp = intotcpcb(inp); 376 377 SOCK_LOCK(so); 378 error = solisten_proto_check(so); 379 if (error != 0) { 380 SOCK_UNLOCK(so); 381 goto out; 382 } 383 if (inp->inp_lport == 0) { 384 INP_HASH_WLOCK(&V_tcbinfo); 385 error = in_pcbbind(inp, NULL, td->td_ucred); 386 INP_HASH_WUNLOCK(&V_tcbinfo); 387 } 388 if (error == 0) { 389 tcp_state_change(tp, TCPS_LISTEN); 390 solisten_proto(so, backlog); 391 #ifdef TCP_OFFLOAD 392 if ((so->so_options & SO_NO_OFFLOAD) == 0) 393 tcp_offload_listen_start(tp); 394 #endif 395 } else { 396 solisten_proto_abort(so); 397 } 398 SOCK_UNLOCK(so); 399 400 if (tp->t_flags & TF_FASTOPEN) 401 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 402 403 out: 404 tcp_bblog_pru(tp, PRU_LISTEN, error); 405 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 406 INP_WUNLOCK(inp); 407 return (error); 408 } 409 #endif /* INET */ 410 411 #ifdef INET6 412 static int 413 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 414 { 415 int error = 0; 416 struct inpcb *inp; 417 struct tcpcb *tp; 418 u_char vflagsav; 419 420 inp = sotoinpcb(so); 421 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 422 INP_WLOCK(inp); 423 if (inp->inp_flags & INP_DROPPED) { 424 INP_WUNLOCK(inp); 425 return (EINVAL); 426 } 427 tp = intotcpcb(inp); 428 429 vflagsav = inp->inp_vflag; 430 431 SOCK_LOCK(so); 432 error = solisten_proto_check(so); 433 if (error != 0) { 434 SOCK_UNLOCK(so); 435 goto out; 436 } 437 INP_HASH_WLOCK(&V_tcbinfo); 438 if (inp->inp_lport == 0) { 439 inp->inp_vflag &= ~INP_IPV4; 440 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 441 inp->inp_vflag |= INP_IPV4; 442 error = in6_pcbbind(inp, NULL, td->td_ucred); 443 } 444 INP_HASH_WUNLOCK(&V_tcbinfo); 445 if (error == 0) { 446 tcp_state_change(tp, TCPS_LISTEN); 447 solisten_proto(so, backlog); 448 #ifdef TCP_OFFLOAD 449 if ((so->so_options & SO_NO_OFFLOAD) == 0) 450 tcp_offload_listen_start(tp); 451 #endif 452 } else { 453 solisten_proto_abort(so); 454 } 455 SOCK_UNLOCK(so); 456 457 if (tp->t_flags & TF_FASTOPEN) 458 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 459 460 if (error != 0) 461 inp->inp_vflag = vflagsav; 462 463 out: 464 tcp_bblog_pru(tp, PRU_LISTEN, error); 465 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 466 INP_WUNLOCK(inp); 467 return (error); 468 } 469 #endif /* INET6 */ 470 471 #ifdef INET 472 /* 473 * Initiate connection to peer. 474 * Create a template for use in transmissions on this connection. 475 * Enter SYN_SENT state, and mark socket as connecting. 476 * Start keep-alive timer, and seed output sequence space. 477 * Send initial segment on connection. 478 */ 479 static int 480 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 481 { 482 struct epoch_tracker et; 483 int error = 0; 484 struct inpcb *inp; 485 struct tcpcb *tp; 486 struct sockaddr_in *sinp; 487 488 inp = sotoinpcb(so); 489 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 490 INP_WLOCK(inp); 491 if (inp->inp_flags & INP_DROPPED) { 492 INP_WUNLOCK(inp); 493 return (ECONNREFUSED); 494 } 495 tp = intotcpcb(inp); 496 497 sinp = (struct sockaddr_in *)nam; 498 if (nam->sa_family != AF_INET) { 499 error = EAFNOSUPPORT; 500 goto out; 501 } 502 if (nam->sa_len != sizeof (*sinp)) { 503 error = EINVAL; 504 goto out; 505 } 506 /* 507 * Must disallow TCP ``connections'' to multicast addresses. 508 */ 509 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 510 error = EAFNOSUPPORT; 511 goto out; 512 } 513 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 514 error = EACCES; 515 goto out; 516 } 517 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 518 goto out; 519 if (SOLISTENING(so)) { 520 error = EOPNOTSUPP; 521 goto out; 522 } 523 NET_EPOCH_ENTER(et); 524 if ((error = tcp_connect(tp, sinp, td)) != 0) 525 goto out_in_epoch; 526 #ifdef TCP_OFFLOAD 527 if (registered_toedevs > 0 && 528 (so->so_options & SO_NO_OFFLOAD) == 0 && 529 (error = tcp_offload_connect(so, nam)) == 0) 530 goto out_in_epoch; 531 #endif 532 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 533 error = tcp_output(tp); 534 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 535 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 536 out_in_epoch: 537 NET_EPOCH_EXIT(et); 538 out: 539 tcp_bblog_pru(tp, PRU_CONNECT, error); 540 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 541 INP_WUNLOCK(inp); 542 return (error); 543 } 544 #endif /* INET */ 545 546 #ifdef INET6 547 static int 548 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 549 { 550 struct epoch_tracker et; 551 int error = 0; 552 struct inpcb *inp; 553 struct tcpcb *tp; 554 struct sockaddr_in6 *sin6; 555 u_int8_t incflagsav; 556 u_char vflagsav; 557 558 inp = sotoinpcb(so); 559 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 560 INP_WLOCK(inp); 561 if (inp->inp_flags & INP_DROPPED) { 562 INP_WUNLOCK(inp); 563 return (ECONNREFUSED); 564 } 565 tp = intotcpcb(inp); 566 567 vflagsav = inp->inp_vflag; 568 incflagsav = inp->inp_inc.inc_flags; 569 570 sin6 = (struct sockaddr_in6 *)nam; 571 if (nam->sa_family != AF_INET6) { 572 error = EAFNOSUPPORT; 573 goto out; 574 } 575 if (nam->sa_len != sizeof (*sin6)) { 576 error = EINVAL; 577 goto out; 578 } 579 /* 580 * Must disallow TCP ``connections'' to multicast addresses. 581 */ 582 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 583 error = EAFNOSUPPORT; 584 goto out; 585 } 586 if (SOLISTENING(so)) { 587 error = EINVAL; 588 goto out; 589 } 590 #ifdef INET 591 /* 592 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 593 * therefore probably require the hash lock, which isn't held here. 594 * Is this a significant problem? 595 */ 596 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 597 struct sockaddr_in sin; 598 599 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 600 error = EINVAL; 601 goto out; 602 } 603 if ((inp->inp_vflag & INP_IPV4) == 0) { 604 error = EAFNOSUPPORT; 605 goto out; 606 } 607 608 in6_sin6_2_sin(&sin, sin6); 609 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 610 error = EAFNOSUPPORT; 611 goto out; 612 } 613 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 614 error = EACCES; 615 goto out; 616 } 617 if ((error = prison_remote_ip4(td->td_ucred, 618 &sin.sin_addr)) != 0) 619 goto out; 620 inp->inp_vflag |= INP_IPV4; 621 inp->inp_vflag &= ~INP_IPV6; 622 NET_EPOCH_ENTER(et); 623 if ((error = tcp_connect(tp, &sin, td)) != 0) 624 goto out_in_epoch; 625 #ifdef TCP_OFFLOAD 626 if (registered_toedevs > 0 && 627 (so->so_options & SO_NO_OFFLOAD) == 0 && 628 (error = tcp_offload_connect(so, nam)) == 0) 629 goto out_in_epoch; 630 #endif 631 error = tcp_output(tp); 632 goto out_in_epoch; 633 } else { 634 if ((inp->inp_vflag & INP_IPV6) == 0) { 635 error = EAFNOSUPPORT; 636 goto out; 637 } 638 } 639 #endif 640 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 641 goto out; 642 inp->inp_vflag &= ~INP_IPV4; 643 inp->inp_vflag |= INP_IPV6; 644 inp->inp_inc.inc_flags |= INC_ISIPV6; 645 NET_EPOCH_ENTER(et); 646 if ((error = tcp6_connect(tp, sin6, td)) != 0) 647 goto out_in_epoch; 648 #ifdef TCP_OFFLOAD 649 if (registered_toedevs > 0 && 650 (so->so_options & SO_NO_OFFLOAD) == 0 && 651 (error = tcp_offload_connect(so, nam)) == 0) 652 goto out_in_epoch; 653 #endif 654 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 655 error = tcp_output(tp); 656 out_in_epoch: 657 NET_EPOCH_EXIT(et); 658 out: 659 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 660 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 661 /* 662 * If the implicit bind in the connect call fails, restore 663 * the flags we modified. 664 */ 665 if (error != 0 && inp->inp_lport == 0) { 666 inp->inp_vflag = vflagsav; 667 inp->inp_inc.inc_flags = incflagsav; 668 } 669 670 tcp_bblog_pru(tp, PRU_CONNECT, error); 671 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 672 INP_WUNLOCK(inp); 673 return (error); 674 } 675 #endif /* INET6 */ 676 677 /* 678 * Initiate disconnect from peer. 679 * If connection never passed embryonic stage, just drop; 680 * else if don't need to let data drain, then can just drop anyways, 681 * else have to begin TCP shutdown process: mark socket disconnecting, 682 * drain unread data, state switch to reflect user close, and 683 * send segment (e.g. FIN) to peer. Socket will be really disconnected 684 * when peer sends FIN and acks ours. 685 * 686 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 687 */ 688 static int 689 tcp_usr_disconnect(struct socket *so) 690 { 691 struct inpcb *inp; 692 struct tcpcb *tp = NULL; 693 struct epoch_tracker et; 694 int error = 0; 695 696 NET_EPOCH_ENTER(et); 697 inp = sotoinpcb(so); 698 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 699 INP_WLOCK(inp); 700 if (inp->inp_flags & INP_DROPPED) { 701 INP_WUNLOCK(inp); 702 NET_EPOCH_EXIT(et); 703 return (ECONNRESET); 704 } 705 tp = intotcpcb(inp); 706 707 if (tp->t_state == TCPS_TIME_WAIT) 708 goto out; 709 tcp_disconnect(tp); 710 out: 711 tcp_bblog_pru(tp, PRU_DISCONNECT, error); 712 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 713 INP_WUNLOCK(inp); 714 NET_EPOCH_EXIT(et); 715 return (error); 716 } 717 718 #ifdef INET 719 /* 720 * Accept a connection. Essentially all the work is done at higher levels; 721 * just return the address of the peer, storing through addr. 722 */ 723 static int 724 tcp_usr_accept(struct socket *so, struct sockaddr *sa) 725 { 726 struct inpcb *inp; 727 struct tcpcb *tp; 728 int error = 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 else 742 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 743 .sin_family = AF_INET, 744 .sin_len = sizeof(struct sockaddr_in), 745 .sin_port = inp->inp_fport, 746 .sin_addr = inp->inp_faddr, 747 }; 748 tcp_bblog_pru(tp, PRU_ACCEPT, error); 749 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 750 INP_WUNLOCK(inp); 751 752 return (error); 753 } 754 #endif /* INET */ 755 756 #ifdef INET6 757 static int 758 tcp6_usr_accept(struct socket *so, struct sockaddr *sa) 759 { 760 struct inpcb *inp; 761 struct tcpcb *tp; 762 int error = 0; 763 764 inp = sotoinpcb(so); 765 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 766 INP_WLOCK(inp); 767 if (inp->inp_flags & INP_DROPPED) { 768 INP_WUNLOCK(inp); 769 return (ECONNABORTED); 770 } 771 tp = intotcpcb(inp); 772 773 if (so->so_state & SS_ISDISCONNECTED) { 774 error = ECONNABORTED; 775 } else { 776 if (inp->inp_vflag & INP_IPV4) { 777 struct sockaddr_in sin = { 778 .sin_family = AF_INET, 779 .sin_len = sizeof(struct sockaddr_in), 780 .sin_port = inp->inp_fport, 781 .sin_addr = inp->inp_faddr, 782 }; 783 in6_sin_2_v4mapsin6(&sin, (struct sockaddr_in6 *)sa); 784 } else { 785 *(struct sockaddr_in6 *)sa = (struct sockaddr_in6 ){ 786 .sin6_family = AF_INET6, 787 .sin6_len = sizeof(struct sockaddr_in6), 788 .sin6_port = inp->inp_fport, 789 .sin6_addr = inp->in6p_faddr, 790 }; 791 /* XXX: should catch errors */ 792 (void)sa6_recoverscope((struct sockaddr_in6 *)sa); 793 } 794 } 795 796 tcp_bblog_pru(tp, PRU_ACCEPT, error); 797 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 798 INP_WUNLOCK(inp); 799 800 return (error); 801 } 802 #endif /* INET6 */ 803 804 /* 805 * Mark the connection as being incapable of further output. 806 */ 807 static int 808 tcp_usr_shutdown(struct socket *so, enum shutdown_how how) 809 { 810 struct epoch_tracker et; 811 struct inpcb *inp = sotoinpcb(so); 812 struct tcpcb *tp = intotcpcb(inp); 813 int error = 0; 814 815 SOCK_LOCK(so); 816 if (SOLISTENING(so)) { 817 if (how != SHUT_WR) { 818 so->so_error = ECONNABORTED; 819 solisten_wakeup(so); /* unlocks so */ 820 } else 821 SOCK_UNLOCK(so); 822 return (ENOTCONN); 823 } else if ((so->so_state & 824 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 825 SOCK_UNLOCK(so); 826 return (ENOTCONN); 827 } 828 SOCK_UNLOCK(so); 829 830 switch (how) { 831 case SHUT_RD: 832 sorflush(so); 833 break; 834 case SHUT_RDWR: 835 sorflush(so); 836 /* FALLTHROUGH */ 837 case SHUT_WR: 838 /* 839 * XXXGL: mimicing old soshutdown() here. But shouldn't we 840 * return ECONNRESEST for SHUT_RD as well? 841 */ 842 INP_WLOCK(inp); 843 if (inp->inp_flags & INP_DROPPED) { 844 INP_WUNLOCK(inp); 845 return (ECONNRESET); 846 } 847 848 socantsendmore(so); 849 NET_EPOCH_ENTER(et); 850 tcp_usrclosed(tp); 851 error = tcp_output_nodrop(tp); 852 tcp_bblog_pru(tp, PRU_SHUTDOWN, error); 853 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 854 error = tcp_unlock_or_drop(tp, error); 855 NET_EPOCH_EXIT(et); 856 } 857 wakeup(&so->so_timeo); 858 859 return (error); 860 } 861 862 /* 863 * After a receive, possibly send window update to peer. 864 */ 865 static int 866 tcp_usr_rcvd(struct socket *so, int flags) 867 { 868 struct epoch_tracker et; 869 struct inpcb *inp; 870 struct tcpcb *tp; 871 int outrv = 0, error = 0; 872 873 inp = sotoinpcb(so); 874 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 875 INP_WLOCK(inp); 876 if (inp->inp_flags & INP_DROPPED) { 877 INP_WUNLOCK(inp); 878 return (ECONNRESET); 879 } 880 tp = intotcpcb(inp); 881 882 NET_EPOCH_ENTER(et); 883 /* 884 * For passively-created TFO connections, don't attempt a window 885 * update while still in SYN_RECEIVED as this may trigger an early 886 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 887 * application response data, or failing that, when the DELACK timer 888 * expires. 889 */ 890 if ((tp->t_flags & TF_FASTOPEN) && (tp->t_state == TCPS_SYN_RECEIVED)) 891 goto out; 892 #ifdef TCP_OFFLOAD 893 if (tp->t_flags & TF_TOE) 894 tcp_offload_rcvd(tp); 895 else 896 #endif 897 outrv = tcp_output_nodrop(tp); 898 out: 899 tcp_bblog_pru(tp, PRU_RCVD, error); 900 TCP_PROBE2(debug__user, tp, PRU_RCVD); 901 (void) tcp_unlock_or_drop(tp, outrv); 902 NET_EPOCH_EXIT(et); 903 return (error); 904 } 905 906 /* 907 * Do a send by putting data in output queue and updating urgent 908 * marker if URG set. Possibly send more data. Unlike the other 909 * pru_*() routines, the mbuf chains are our responsibility. We 910 * must either enqueue them or free them. The other pru_* routines 911 * generally are caller-frees. 912 */ 913 static int 914 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 915 struct sockaddr *nam, struct mbuf *control, struct thread *td) 916 { 917 struct epoch_tracker et; 918 int error = 0; 919 struct inpcb *inp; 920 struct tcpcb *tp; 921 #ifdef INET 922 #ifdef INET6 923 struct sockaddr_in sin; 924 #endif 925 struct sockaddr_in *sinp; 926 #endif 927 #ifdef INET6 928 struct sockaddr_in6 *sin6; 929 int isipv6; 930 #endif 931 u_int8_t incflagsav; 932 u_char vflagsav; 933 bool restoreflags; 934 935 inp = sotoinpcb(so); 936 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 937 INP_WLOCK(inp); 938 if (inp->inp_flags & INP_DROPPED) { 939 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 940 m_freem(m); 941 INP_WUNLOCK(inp); 942 return (ECONNRESET); 943 } 944 tp = intotcpcb(inp); 945 946 vflagsav = inp->inp_vflag; 947 incflagsav = inp->inp_inc.inc_flags; 948 restoreflags = false; 949 950 NET_EPOCH_ENTER(et); 951 if (control != NULL) { 952 /* TCP doesn't do control messages (rights, creds, etc) */ 953 if (control->m_len > 0) { 954 m_freem(control); 955 error = EINVAL; 956 goto out; 957 } 958 m_freem(control); /* empty control, just free it */ 959 } 960 961 if ((flags & PRUS_OOB) != 0 && 962 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 963 goto out; 964 965 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 966 if (tp->t_state == TCPS_LISTEN) { 967 error = EINVAL; 968 goto out; 969 } 970 switch (nam->sa_family) { 971 #ifdef INET 972 case AF_INET: 973 sinp = (struct sockaddr_in *)nam; 974 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 975 error = EINVAL; 976 goto out; 977 } 978 if ((inp->inp_vflag & INP_IPV6) != 0) { 979 error = EAFNOSUPPORT; 980 goto out; 981 } 982 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 983 error = EAFNOSUPPORT; 984 goto out; 985 } 986 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 987 error = EACCES; 988 goto out; 989 } 990 if ((error = prison_remote_ip4(td->td_ucred, 991 &sinp->sin_addr))) 992 goto out; 993 #ifdef INET6 994 isipv6 = 0; 995 #endif 996 break; 997 #endif /* INET */ 998 #ifdef INET6 999 case AF_INET6: 1000 sin6 = (struct sockaddr_in6 *)nam; 1001 if (sin6->sin6_len != sizeof(*sin6)) { 1002 error = EINVAL; 1003 goto out; 1004 } 1005 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1006 error = EAFNOSUPPORT; 1007 goto out; 1008 } 1009 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1010 error = EAFNOSUPPORT; 1011 goto out; 1012 } 1013 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1014 #ifdef INET 1015 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1016 error = EINVAL; 1017 goto out; 1018 } 1019 if ((inp->inp_vflag & INP_IPV4) == 0) { 1020 error = EAFNOSUPPORT; 1021 goto out; 1022 } 1023 restoreflags = true; 1024 inp->inp_vflag &= ~INP_IPV6; 1025 sinp = &sin; 1026 in6_sin6_2_sin(sinp, sin6); 1027 if (IN_MULTICAST( 1028 ntohl(sinp->sin_addr.s_addr))) { 1029 error = EAFNOSUPPORT; 1030 goto out; 1031 } 1032 if ((error = prison_remote_ip4(td->td_ucred, 1033 &sinp->sin_addr))) 1034 goto out; 1035 isipv6 = 0; 1036 #else /* !INET */ 1037 error = EAFNOSUPPORT; 1038 goto out; 1039 #endif /* INET */ 1040 } else { 1041 if ((inp->inp_vflag & INP_IPV6) == 0) { 1042 error = EAFNOSUPPORT; 1043 goto out; 1044 } 1045 restoreflags = true; 1046 inp->inp_vflag &= ~INP_IPV4; 1047 inp->inp_inc.inc_flags |= INC_ISIPV6; 1048 if ((error = prison_remote_ip6(td->td_ucred, 1049 &sin6->sin6_addr))) 1050 goto out; 1051 isipv6 = 1; 1052 } 1053 break; 1054 #endif /* INET6 */ 1055 default: 1056 error = EAFNOSUPPORT; 1057 goto out; 1058 } 1059 } 1060 if (!(flags & PRUS_OOB)) { 1061 if (tp->t_acktime == 0) 1062 tp->t_acktime = ticks; 1063 sbappendstream(&so->so_snd, m, flags); 1064 m = NULL; 1065 if (nam && tp->t_state < TCPS_SYN_SENT) { 1066 KASSERT(tp->t_state == TCPS_CLOSED, 1067 ("%s: tp %p is listening", __func__, tp)); 1068 1069 /* 1070 * Do implied connect if not yet connected, 1071 * initialize window to default value, and 1072 * initialize maxseg using peer's cached MSS. 1073 */ 1074 #ifdef INET6 1075 if (isipv6) 1076 error = tcp6_connect(tp, sin6, td); 1077 #endif /* INET6 */ 1078 #if defined(INET6) && defined(INET) 1079 else 1080 #endif 1081 #ifdef INET 1082 error = tcp_connect(tp, sinp, td); 1083 #endif 1084 /* 1085 * The bind operation in tcp_connect succeeded. We 1086 * no longer want to restore the flags if later 1087 * operations fail. 1088 */ 1089 if (error == 0 || inp->inp_lport != 0) 1090 restoreflags = false; 1091 1092 if (error) { 1093 /* m is freed if PRUS_NOTREADY is unset. */ 1094 sbflush(&so->so_snd); 1095 goto out; 1096 } 1097 if (tp->t_flags & TF_FASTOPEN) 1098 tcp_fastopen_connect(tp); 1099 else { 1100 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1101 tcp_mss(tp, -1); 1102 } 1103 } 1104 if (flags & PRUS_EOF) { 1105 /* 1106 * Close the send side of the connection after 1107 * the data is sent. 1108 */ 1109 socantsendmore(so); 1110 tcp_usrclosed(tp); 1111 } 1112 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1113 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1114 (tp->t_fbyte_out == 0) && 1115 (so->so_snd.sb_ccc > 0)) { 1116 tp->t_fbyte_out = ticks; 1117 if (tp->t_fbyte_out == 0) 1118 tp->t_fbyte_out = 1; 1119 if (tp->t_fbyte_out && tp->t_fbyte_in) 1120 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1121 } 1122 if (!(inp->inp_flags & INP_DROPPED) && 1123 !(flags & PRUS_NOTREADY)) { 1124 if (flags & PRUS_MORETOCOME) 1125 tp->t_flags |= TF_MORETOCOME; 1126 error = tcp_output_nodrop(tp); 1127 if (flags & PRUS_MORETOCOME) 1128 tp->t_flags &= ~TF_MORETOCOME; 1129 } 1130 } else { 1131 /* 1132 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1133 */ 1134 SOCKBUF_LOCK(&so->so_snd); 1135 if (sbspace(&so->so_snd) < -512) { 1136 SOCKBUF_UNLOCK(&so->so_snd); 1137 error = ENOBUFS; 1138 goto out; 1139 } 1140 /* 1141 * According to RFC961 (Assigned Protocols), 1142 * the urgent pointer points to the last octet 1143 * of urgent data. We continue, however, 1144 * to consider it to indicate the first octet 1145 * of data past the urgent section. 1146 * Otherwise, snd_up should be one lower. 1147 */ 1148 if (tp->t_acktime == 0) 1149 tp->t_acktime = ticks; 1150 sbappendstream_locked(&so->so_snd, m, flags); 1151 SOCKBUF_UNLOCK(&so->so_snd); 1152 m = NULL; 1153 if (nam && tp->t_state < TCPS_SYN_SENT) { 1154 /* 1155 * Do implied connect if not yet connected, 1156 * initialize window to default value, and 1157 * initialize maxseg using peer's cached MSS. 1158 */ 1159 1160 /* 1161 * Not going to contemplate SYN|URG 1162 */ 1163 if (tp->t_flags & TF_FASTOPEN) 1164 tp->t_flags &= ~TF_FASTOPEN; 1165 #ifdef INET6 1166 if (isipv6) 1167 error = tcp6_connect(tp, sin6, td); 1168 #endif /* INET6 */ 1169 #if defined(INET6) && defined(INET) 1170 else 1171 #endif 1172 #ifdef INET 1173 error = tcp_connect(tp, sinp, td); 1174 #endif 1175 /* 1176 * The bind operation in tcp_connect succeeded. We 1177 * no longer want to restore the flags if later 1178 * operations fail. 1179 */ 1180 if (error == 0 || inp->inp_lport != 0) 1181 restoreflags = false; 1182 1183 if (error != 0) { 1184 /* m is freed if PRUS_NOTREADY is unset. */ 1185 sbflush(&so->so_snd); 1186 goto out; 1187 } 1188 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1189 tcp_mss(tp, -1); 1190 } 1191 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1192 if ((flags & PRUS_NOTREADY) == 0) { 1193 tp->t_flags |= TF_FORCEDATA; 1194 error = tcp_output_nodrop(tp); 1195 tp->t_flags &= ~TF_FORCEDATA; 1196 } 1197 } 1198 TCP_LOG_EVENT(tp, NULL, 1199 &inp->inp_socket->so_rcv, 1200 &inp->inp_socket->so_snd, 1201 TCP_LOG_USERSEND, error, 1202 0, NULL, false); 1203 1204 out: 1205 /* 1206 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1207 * responsible for freeing memory. 1208 */ 1209 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1210 m_freem(m); 1211 1212 /* 1213 * If the request was unsuccessful and we changed flags, 1214 * restore the original flags. 1215 */ 1216 if (error != 0 && restoreflags) { 1217 inp->inp_vflag = vflagsav; 1218 inp->inp_inc.inc_flags = incflagsav; 1219 } 1220 tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1221 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error); 1222 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1223 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1224 error = tcp_unlock_or_drop(tp, error); 1225 NET_EPOCH_EXIT(et); 1226 return (error); 1227 } 1228 1229 static int 1230 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1231 { 1232 struct epoch_tracker et; 1233 struct inpcb *inp; 1234 struct tcpcb *tp; 1235 int error; 1236 1237 inp = sotoinpcb(so); 1238 INP_WLOCK(inp); 1239 if (inp->inp_flags & INP_DROPPED) { 1240 INP_WUNLOCK(inp); 1241 mb_free_notready(m, count); 1242 return (ECONNRESET); 1243 } 1244 tp = intotcpcb(inp); 1245 1246 SOCKBUF_LOCK(&so->so_snd); 1247 error = sbready(&so->so_snd, m, count); 1248 SOCKBUF_UNLOCK(&so->so_snd); 1249 if (error) { 1250 INP_WUNLOCK(inp); 1251 return (error); 1252 } 1253 NET_EPOCH_ENTER(et); 1254 error = tcp_output_unlock(tp); 1255 NET_EPOCH_EXIT(et); 1256 1257 return (error); 1258 } 1259 1260 /* 1261 * Abort the TCP. Drop the connection abruptly. 1262 */ 1263 static void 1264 tcp_usr_abort(struct socket *so) 1265 { 1266 struct inpcb *inp; 1267 struct tcpcb *tp; 1268 struct epoch_tracker et; 1269 1270 inp = sotoinpcb(so); 1271 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1272 1273 NET_EPOCH_ENTER(et); 1274 INP_WLOCK(inp); 1275 KASSERT(inp->inp_socket != NULL, 1276 ("tcp_usr_abort: inp_socket == NULL")); 1277 1278 /* 1279 * If we still have full TCP state, and we're not dropped, drop. 1280 */ 1281 if (!(inp->inp_flags & INP_DROPPED)) { 1282 tp = intotcpcb(inp); 1283 tp = tcp_drop(tp, ECONNABORTED); 1284 if (tp == NULL) 1285 goto dropped; 1286 tcp_bblog_pru(tp, PRU_ABORT, 0); 1287 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1288 } 1289 if (!(inp->inp_flags & INP_DROPPED)) { 1290 soref(so); 1291 inp->inp_flags |= INP_SOCKREF; 1292 } 1293 INP_WUNLOCK(inp); 1294 dropped: 1295 NET_EPOCH_EXIT(et); 1296 } 1297 1298 /* 1299 * TCP socket is closed. Start friendly disconnect. 1300 */ 1301 static void 1302 tcp_usr_close(struct socket *so) 1303 { 1304 struct inpcb *inp; 1305 struct tcpcb *tp; 1306 struct epoch_tracker et; 1307 1308 inp = sotoinpcb(so); 1309 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1310 1311 NET_EPOCH_ENTER(et); 1312 INP_WLOCK(inp); 1313 KASSERT(inp->inp_socket != NULL, 1314 ("tcp_usr_close: inp_socket == NULL")); 1315 1316 /* 1317 * If we are still connected and we're not dropped, initiate 1318 * a disconnect. 1319 */ 1320 if (!(inp->inp_flags & INP_DROPPED)) { 1321 tp = intotcpcb(inp); 1322 if (tp->t_state != TCPS_TIME_WAIT) { 1323 tp->t_flags |= TF_CLOSED; 1324 tcp_disconnect(tp); 1325 tcp_bblog_pru(tp, PRU_CLOSE, 0); 1326 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1327 } 1328 } 1329 if (!(inp->inp_flags & INP_DROPPED)) { 1330 soref(so); 1331 inp->inp_flags |= INP_SOCKREF; 1332 } 1333 INP_WUNLOCK(inp); 1334 NET_EPOCH_EXIT(et); 1335 } 1336 1337 static int 1338 tcp_pru_options_support(struct tcpcb *tp, int flags) 1339 { 1340 /* 1341 * If the specific TCP stack has a pru_options 1342 * specified then it does not always support 1343 * all the PRU_XX options and we must ask it. 1344 * If the function is not specified then all 1345 * of the PRU_XX options are supported. 1346 */ 1347 int ret = 0; 1348 1349 if (tp->t_fb->tfb_pru_options) { 1350 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1351 } 1352 return (ret); 1353 } 1354 1355 /* 1356 * Receive out-of-band data. 1357 */ 1358 static int 1359 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1360 { 1361 int error = 0; 1362 struct inpcb *inp; 1363 struct tcpcb *tp; 1364 1365 inp = sotoinpcb(so); 1366 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1367 INP_WLOCK(inp); 1368 if (inp->inp_flags & INP_DROPPED) { 1369 INP_WUNLOCK(inp); 1370 return (ECONNRESET); 1371 } 1372 tp = intotcpcb(inp); 1373 1374 error = tcp_pru_options_support(tp, PRUS_OOB); 1375 if (error) { 1376 goto out; 1377 } 1378 if ((so->so_oobmark == 0 && 1379 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1380 so->so_options & SO_OOBINLINE || 1381 tp->t_oobflags & TCPOOB_HADDATA) { 1382 error = EINVAL; 1383 goto out; 1384 } 1385 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1386 error = EWOULDBLOCK; 1387 goto out; 1388 } 1389 m->m_len = 1; 1390 *mtod(m, caddr_t) = tp->t_iobc; 1391 if ((flags & MSG_PEEK) == 0) 1392 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1393 1394 out: 1395 tcp_bblog_pru(tp, PRU_RCVOOB, error); 1396 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1397 INP_WUNLOCK(inp); 1398 return (error); 1399 } 1400 1401 #ifdef INET 1402 struct protosw tcp_protosw = { 1403 .pr_type = SOCK_STREAM, 1404 .pr_protocol = IPPROTO_TCP, 1405 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD | 1406 PR_CAPATTACH, 1407 .pr_ctloutput = tcp_ctloutput, 1408 .pr_abort = tcp_usr_abort, 1409 .pr_accept = tcp_usr_accept, 1410 .pr_attach = tcp_usr_attach, 1411 .pr_bind = tcp_usr_bind, 1412 .pr_connect = tcp_usr_connect, 1413 .pr_control = in_control, 1414 .pr_detach = tcp_usr_detach, 1415 .pr_disconnect = tcp_usr_disconnect, 1416 .pr_listen = tcp_usr_listen, 1417 .pr_peeraddr = in_getpeeraddr, 1418 .pr_rcvd = tcp_usr_rcvd, 1419 .pr_rcvoob = tcp_usr_rcvoob, 1420 .pr_send = tcp_usr_send, 1421 .pr_ready = tcp_usr_ready, 1422 .pr_shutdown = tcp_usr_shutdown, 1423 .pr_sockaddr = in_getsockaddr, 1424 .pr_sosetlabel = in_pcbsosetlabel, 1425 .pr_close = tcp_usr_close, 1426 }; 1427 #endif /* INET */ 1428 1429 #ifdef INET6 1430 struct protosw tcp6_protosw = { 1431 .pr_type = SOCK_STREAM, 1432 .pr_protocol = IPPROTO_TCP, 1433 .pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD | 1434 PR_CAPATTACH, 1435 .pr_ctloutput = tcp_ctloutput, 1436 .pr_abort = tcp_usr_abort, 1437 .pr_accept = tcp6_usr_accept, 1438 .pr_attach = tcp_usr_attach, 1439 .pr_bind = tcp6_usr_bind, 1440 .pr_connect = tcp6_usr_connect, 1441 .pr_control = in6_control, 1442 .pr_detach = tcp_usr_detach, 1443 .pr_disconnect = tcp_usr_disconnect, 1444 .pr_listen = tcp6_usr_listen, 1445 .pr_peeraddr = in6_mapped_peeraddr, 1446 .pr_rcvd = tcp_usr_rcvd, 1447 .pr_rcvoob = tcp_usr_rcvoob, 1448 .pr_send = tcp_usr_send, 1449 .pr_ready = tcp_usr_ready, 1450 .pr_shutdown = tcp_usr_shutdown, 1451 .pr_sockaddr = in6_mapped_sockaddr, 1452 .pr_sosetlabel = in_pcbsosetlabel, 1453 .pr_close = tcp_usr_close, 1454 }; 1455 #endif /* INET6 */ 1456 1457 #ifdef INET 1458 /* 1459 * Common subroutine to open a TCP connection to remote host specified 1460 * by struct sockaddr_in. Call in_pcbconnect() to choose local host address 1461 * and assign a local port number and install the inpcb into the hash. 1462 * Initialize connection parameters and enter SYN-SENT state. 1463 */ 1464 static int 1465 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td) 1466 { 1467 struct inpcb *inp = tptoinpcb(tp); 1468 struct socket *so = tptosocket(tp); 1469 int error; 1470 1471 NET_EPOCH_ASSERT(); 1472 INP_WLOCK_ASSERT(inp); 1473 1474 if (__predict_false((so->so_state & 1475 (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | 1476 SS_ISDISCONNECTED)) != 0)) 1477 return (EISCONN); 1478 1479 INP_HASH_WLOCK(&V_tcbinfo); 1480 error = in_pcbconnect(inp, sin, td->td_ucred, true); 1481 INP_HASH_WUNLOCK(&V_tcbinfo); 1482 if (error != 0) 1483 return (error); 1484 1485 /* 1486 * Compute window scaling to request: 1487 * Scale to fit into sweet spot. See tcp_syncache.c. 1488 * XXX: This should move to tcp_output(). 1489 */ 1490 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1491 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1492 tp->request_r_scale++; 1493 1494 soisconnecting(so); 1495 TCPSTAT_INC(tcps_connattempt); 1496 tcp_state_change(tp, TCPS_SYN_SENT); 1497 tp->iss = tcp_new_isn(&inp->inp_inc); 1498 if (tp->t_flags & TF_REQ_TSTMP) 1499 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1500 tcp_sendseqinit(tp); 1501 1502 return (0); 1503 } 1504 #endif /* INET */ 1505 1506 #ifdef INET6 1507 static int 1508 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td) 1509 { 1510 struct inpcb *inp = tptoinpcb(tp); 1511 struct socket *so = tptosocket(tp); 1512 int error; 1513 1514 NET_EPOCH_ASSERT(); 1515 INP_WLOCK_ASSERT(inp); 1516 1517 if (__predict_false((so->so_state & 1518 (SS_ISCONNECTING | SS_ISCONNECTED)) != 0)) 1519 return (EISCONN); 1520 1521 INP_HASH_WLOCK(&V_tcbinfo); 1522 error = in6_pcbconnect(inp, sin6, td->td_ucred, true); 1523 INP_HASH_WUNLOCK(&V_tcbinfo); 1524 if (error != 0) 1525 return (error); 1526 1527 /* Compute window scaling to request. */ 1528 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1529 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1530 tp->request_r_scale++; 1531 1532 soisconnecting(so); 1533 TCPSTAT_INC(tcps_connattempt); 1534 tcp_state_change(tp, TCPS_SYN_SENT); 1535 tp->iss = tcp_new_isn(&inp->inp_inc); 1536 if (tp->t_flags & TF_REQ_TSTMP) 1537 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1538 tcp_sendseqinit(tp); 1539 1540 return (0); 1541 } 1542 #endif /* INET6 */ 1543 1544 /* 1545 * Export TCP internal state information via a struct tcp_info, based on the 1546 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1547 * (TCP state machine, etc). We export all information using FreeBSD-native 1548 * constants -- for example, the numeric values for tcpi_state will differ 1549 * from Linux. 1550 */ 1551 void 1552 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti) 1553 { 1554 1555 INP_LOCK_ASSERT(tptoinpcb(tp)); 1556 bzero(ti, sizeof(*ti)); 1557 1558 ti->tcpi_state = tp->t_state; 1559 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1560 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1561 if (tp->t_flags & TF_SACK_PERMIT) 1562 ti->tcpi_options |= TCPI_OPT_SACK; 1563 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1564 ti->tcpi_options |= TCPI_OPT_WSCALE; 1565 ti->tcpi_snd_wscale = tp->snd_scale; 1566 ti->tcpi_rcv_wscale = tp->rcv_scale; 1567 } 1568 switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { 1569 case TF2_ECN_PERMIT: 1570 ti->tcpi_options |= TCPI_OPT_ECN; 1571 break; 1572 case TF2_ACE_PERMIT: 1573 /* FALLTHROUGH */ 1574 case TF2_ECN_PERMIT | TF2_ACE_PERMIT: 1575 ti->tcpi_options |= TCPI_OPT_ACE; 1576 break; 1577 default: 1578 break; 1579 } 1580 if (tp->t_flags & TF_FASTOPEN) 1581 ti->tcpi_options |= TCPI_OPT_TFO; 1582 1583 ti->tcpi_rto = tp->t_rxtcur * tick; 1584 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1585 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1586 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1587 1588 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1589 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1590 1591 /* 1592 * FreeBSD-specific extension fields for tcp_info. 1593 */ 1594 ti->tcpi_rcv_space = tp->rcv_wnd; 1595 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1596 ti->tcpi_snd_wnd = tp->snd_wnd; 1597 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1598 ti->tcpi_snd_nxt = tp->snd_nxt; 1599 ti->tcpi_snd_mss = tp->t_maxseg; 1600 ti->tcpi_rcv_mss = tp->t_maxseg; 1601 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1602 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1603 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1604 ti->tcpi_snd_una = tp->snd_una; 1605 ti->tcpi_snd_max = tp->snd_max; 1606 ti->tcpi_rcv_numsacks = tp->rcv_numsacks; 1607 ti->tcpi_rcv_adv = tp->rcv_adv; 1608 ti->tcpi_dupacks = tp->t_dupacks; 1609 ti->tcpi_rttmin = tp->t_rttlow; 1610 #ifdef TCP_OFFLOAD 1611 if (tp->t_flags & TF_TOE) { 1612 ti->tcpi_options |= TCPI_OPT_TOE; 1613 tcp_offload_tcp_info(tp, ti); 1614 } 1615 #endif 1616 /* 1617 * AccECN related counters. 1618 */ 1619 if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) == 1620 (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) 1621 /* 1622 * Internal counter starts at 5 for AccECN 1623 * but 0 for RFC3168 ECN. 1624 */ 1625 ti->tcpi_delivered_ce = tp->t_scep - 5; 1626 else 1627 ti->tcpi_delivered_ce = tp->t_scep; 1628 ti->tcpi_received_ce = tp->t_rcep; 1629 } 1630 1631 /* 1632 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1633 * socket option arguments. When it re-acquires the lock after the copy, it 1634 * has to revalidate that the connection is still valid for the socket 1635 * option. 1636 */ 1637 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1638 INP_WLOCK(inp); \ 1639 if (inp->inp_flags & INP_DROPPED) { \ 1640 INP_WUNLOCK(inp); \ 1641 cleanup; \ 1642 return (ECONNRESET); \ 1643 } \ 1644 tp = intotcpcb(inp); \ 1645 } while(0) 1646 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1647 1648 int 1649 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1650 { 1651 struct socket *so = inp->inp_socket; 1652 struct tcpcb *tp = intotcpcb(inp); 1653 int error = 0; 1654 1655 MPASS(sopt->sopt_dir == SOPT_SET); 1656 INP_WLOCK_ASSERT(inp); 1657 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1658 ("inp_flags == %x", inp->inp_flags)); 1659 KASSERT(so != NULL, ("inp_socket == NULL")); 1660 1661 if (sopt->sopt_level != IPPROTO_TCP) { 1662 INP_WUNLOCK(inp); 1663 #ifdef INET6 1664 if (inp->inp_vflag & INP_IPV6PROTO) 1665 error = ip6_ctloutput(so, sopt); 1666 #endif 1667 #if defined(INET6) && defined(INET) 1668 else 1669 #endif 1670 #ifdef INET 1671 error = ip_ctloutput(so, sopt); 1672 #endif 1673 /* 1674 * When an IP-level socket option affects TCP, pass control 1675 * down to stack tfb_tcp_ctloutput, otherwise return what 1676 * IP level returned. 1677 */ 1678 switch (sopt->sopt_level) { 1679 #ifdef INET6 1680 case IPPROTO_IPV6: 1681 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1682 return (error); 1683 switch (sopt->sopt_name) { 1684 case IPV6_TCLASS: 1685 /* Notify tcp stacks that care (e.g. RACK). */ 1686 break; 1687 case IPV6_USE_MIN_MTU: 1688 /* Update t_maxseg accordingly. */ 1689 break; 1690 default: 1691 return (error); 1692 } 1693 break; 1694 #endif 1695 #ifdef INET 1696 case IPPROTO_IP: 1697 switch (sopt->sopt_name) { 1698 case IP_TOS: 1699 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1700 break; 1701 case IP_TTL: 1702 /* Notify tcp stacks that care (e.g. RACK). */ 1703 break; 1704 default: 1705 return (error); 1706 } 1707 break; 1708 #endif 1709 default: 1710 return (error); 1711 } 1712 INP_WLOCK_RECHECK(inp); 1713 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1714 /* 1715 * Protect the TCP option TCP_FUNCTION_BLK so 1716 * that a sub-function can *never* overwrite this. 1717 */ 1718 struct tcp_function_set fsn; 1719 struct tcp_function_block *blk; 1720 void *ptr = NULL; 1721 1722 INP_WUNLOCK(inp); 1723 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1724 if (error) 1725 return (error); 1726 1727 INP_WLOCK_RECHECK(inp); 1728 1729 blk = find_and_ref_tcp_functions(&fsn); 1730 if (blk == NULL) { 1731 INP_WUNLOCK(inp); 1732 return (ENOENT); 1733 } 1734 if (tp->t_fb == blk) { 1735 /* You already have this */ 1736 refcount_release(&blk->tfb_refcnt); 1737 INP_WUNLOCK(inp); 1738 return (0); 1739 } 1740 if (tp->t_state != TCPS_CLOSED) { 1741 /* 1742 * The user has advanced the state 1743 * past the initial point, we may not 1744 * be able to switch. 1745 */ 1746 if (blk->tfb_tcp_handoff_ok != NULL) { 1747 /* 1748 * Does the stack provide a 1749 * query mechanism, if so it may 1750 * still be possible? 1751 */ 1752 error = (*blk->tfb_tcp_handoff_ok)(tp); 1753 } else 1754 error = EINVAL; 1755 if (error) { 1756 refcount_release(&blk->tfb_refcnt); 1757 INP_WUNLOCK(inp); 1758 return(error); 1759 } 1760 } 1761 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1762 refcount_release(&blk->tfb_refcnt); 1763 INP_WUNLOCK(inp); 1764 return (ENOENT); 1765 } 1766 /* 1767 * Ensure the new stack takes ownership with a 1768 * clean slate on peak rate threshold. 1769 */ 1770 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 1771 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1772 if (blk->tfb_tcp_fb_init) { 1773 error = (*blk->tfb_tcp_fb_init)(tp, &ptr); 1774 if (error) { 1775 /* 1776 * Release the ref count the lookup 1777 * acquired. 1778 */ 1779 refcount_release(&blk->tfb_refcnt); 1780 /* 1781 * Now there is a chance that the 1782 * init() function mucked with some 1783 * things before it failed, such as 1784 * hpts or inp_flags2 or timer granularity. 1785 * It should not of, but lets give the old 1786 * stack a chance to reset to a known good state. 1787 */ 1788 if (tp->t_fb->tfb_switch_failed) { 1789 (*tp->t_fb->tfb_switch_failed)(tp); 1790 } 1791 goto err_out; 1792 } 1793 } 1794 if (tp->t_fb->tfb_tcp_fb_fini) { 1795 struct epoch_tracker et; 1796 /* 1797 * Tell the stack to cleanup with 0 i.e. 1798 * the tcb is not going away. 1799 */ 1800 NET_EPOCH_ENTER(et); 1801 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1802 NET_EPOCH_EXIT(et); 1803 } 1804 /* 1805 * Release the old refcnt, the 1806 * lookup acquired a ref on the 1807 * new one already. 1808 */ 1809 refcount_release(&tp->t_fb->tfb_refcnt); 1810 /* 1811 * Set in the new stack. 1812 */ 1813 tp->t_fb = blk; 1814 tp->t_fb_ptr = ptr; 1815 #ifdef TCP_OFFLOAD 1816 if (tp->t_flags & TF_TOE) { 1817 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1818 sopt->sopt_name); 1819 } 1820 #endif 1821 err_out: 1822 INP_WUNLOCK(inp); 1823 return (error); 1824 1825 } 1826 1827 /* Pass in the INP locked, callee must unlock it. */ 1828 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1829 } 1830 1831 static int 1832 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1833 { 1834 struct socket *so = inp->inp_socket; 1835 struct tcpcb *tp = intotcpcb(inp); 1836 int error = 0; 1837 1838 MPASS(sopt->sopt_dir == SOPT_GET); 1839 INP_WLOCK_ASSERT(inp); 1840 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 1841 ("inp_flags == %x", inp->inp_flags)); 1842 KASSERT(so != NULL, ("inp_socket == NULL")); 1843 1844 if (sopt->sopt_level != IPPROTO_TCP) { 1845 INP_WUNLOCK(inp); 1846 #ifdef INET6 1847 if (inp->inp_vflag & INP_IPV6PROTO) 1848 error = ip6_ctloutput(so, sopt); 1849 #endif /* INET6 */ 1850 #if defined(INET6) && defined(INET) 1851 else 1852 #endif 1853 #ifdef INET 1854 error = ip_ctloutput(so, sopt); 1855 #endif 1856 return (error); 1857 } 1858 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1859 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1860 struct tcp_function_set fsn; 1861 1862 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1863 memset(&fsn, 0, sizeof(fsn)); 1864 find_tcp_function_alias(tp->t_fb, &fsn); 1865 } else { 1866 strncpy(fsn.function_set_name, 1867 tp->t_fb->tfb_tcp_block_name, 1868 TCP_FUNCTION_NAME_LEN_MAX); 1869 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1870 } 1871 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1872 INP_WUNLOCK(inp); 1873 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1874 return (error); 1875 } 1876 1877 /* Pass in the INP locked, callee must unlock it. */ 1878 return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt)); 1879 } 1880 1881 int 1882 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1883 { 1884 struct inpcb *inp; 1885 1886 inp = sotoinpcb(so); 1887 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1888 1889 INP_WLOCK(inp); 1890 if (inp->inp_flags & INP_DROPPED) { 1891 INP_WUNLOCK(inp); 1892 return (ECONNRESET); 1893 } 1894 if (sopt->sopt_dir == SOPT_SET) 1895 return (tcp_ctloutput_set(inp, sopt)); 1896 else if (sopt->sopt_dir == SOPT_GET) 1897 return (tcp_ctloutput_get(inp, sopt)); 1898 else 1899 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1900 } 1901 1902 /* 1903 * If this assert becomes untrue, we need to change the size of the buf 1904 * variable in tcp_default_ctloutput(). 1905 */ 1906 #ifdef CTASSERT 1907 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1908 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1909 #endif 1910 1911 extern struct cc_algo newreno_cc_algo; 1912 1913 static int 1914 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt) 1915 { 1916 struct cc_algo *algo; 1917 void *ptr = NULL; 1918 struct tcpcb *tp; 1919 struct cc_var cc_mem; 1920 char buf[TCP_CA_NAME_MAX]; 1921 size_t mem_sz; 1922 int error; 1923 1924 INP_WUNLOCK(inp); 1925 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 1926 if (error) 1927 return(error); 1928 buf[sopt->sopt_valsize] = '\0'; 1929 CC_LIST_RLOCK(); 1930 STAILQ_FOREACH(algo, &cc_list, entries) { 1931 if (strncmp(buf, algo->name, 1932 TCP_CA_NAME_MAX) == 0) { 1933 if (algo->flags & CC_MODULE_BEING_REMOVED) { 1934 /* We can't "see" modules being unloaded */ 1935 continue; 1936 } 1937 break; 1938 } 1939 } 1940 if (algo == NULL) { 1941 CC_LIST_RUNLOCK(); 1942 return(ESRCH); 1943 } 1944 /* 1945 * With a reference the algorithm cannot be removed 1946 * so we hold a reference through the change process. 1947 */ 1948 cc_refer(algo); 1949 CC_LIST_RUNLOCK(); 1950 if (algo->cb_init != NULL) { 1951 /* We can now pre-get the memory for the CC */ 1952 mem_sz = (*algo->cc_data_sz)(); 1953 if (mem_sz == 0) { 1954 goto no_mem_needed; 1955 } 1956 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 1957 } else { 1958 no_mem_needed: 1959 mem_sz = 0; 1960 ptr = NULL; 1961 } 1962 /* 1963 * Make sure its all clean and zero and also get 1964 * back the inplock. 1965 */ 1966 memset(&cc_mem, 0, sizeof(cc_mem)); 1967 INP_WLOCK(inp); 1968 if (inp->inp_flags & INP_DROPPED) { 1969 INP_WUNLOCK(inp); 1970 if (ptr) 1971 free(ptr, M_CC_MEM); 1972 /* Release our temp reference */ 1973 CC_LIST_RLOCK(); 1974 cc_release(algo); 1975 CC_LIST_RUNLOCK(); 1976 return (ECONNRESET); 1977 } 1978 tp = intotcpcb(inp); 1979 if (ptr != NULL) 1980 memset(ptr, 0, mem_sz); 1981 cc_mem.ccvc.tcp = tp; 1982 /* 1983 * We once again hold a write lock over the tcb so it's 1984 * safe to do these things without ordering concerns. 1985 * Note here we init into stack memory. 1986 */ 1987 if (algo->cb_init != NULL) 1988 error = algo->cb_init(&cc_mem, ptr); 1989 else 1990 error = 0; 1991 /* 1992 * The CC algorithms, when given their memory 1993 * should not fail we could in theory have a 1994 * KASSERT here. 1995 */ 1996 if (error == 0) { 1997 /* 1998 * Touchdown, lets go ahead and move the 1999 * connection to the new CC module by 2000 * copying in the cc_mem after we call 2001 * the old ones cleanup (if any). 2002 */ 2003 if (CC_ALGO(tp)->cb_destroy != NULL) 2004 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2005 /* Detach the old CC from the tcpcb */ 2006 cc_detach(tp); 2007 /* Copy in our temp memory that was inited */ 2008 memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var)); 2009 /* Now attach the new, which takes a reference */ 2010 cc_attach(tp, algo); 2011 /* Ok now are we where we have gotten past any conn_init? */ 2012 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2013 /* Yep run the connection init for the new CC */ 2014 CC_ALGO(tp)->conn_init(&tp->t_ccv); 2015 } 2016 } else if (ptr) 2017 free(ptr, M_CC_MEM); 2018 INP_WUNLOCK(inp); 2019 /* Now lets release our temp reference */ 2020 CC_LIST_RLOCK(); 2021 cc_release(algo); 2022 CC_LIST_RUNLOCK(); 2023 return (error); 2024 } 2025 2026 int 2027 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt) 2028 { 2029 struct inpcb *inp = tptoinpcb(tp); 2030 int error, opt, optval; 2031 u_int ui; 2032 struct tcp_info ti; 2033 #ifdef KERN_TLS 2034 struct tls_enable tls; 2035 struct socket *so = inp->inp_socket; 2036 #endif 2037 char *pbuf, buf[TCP_LOG_ID_LEN]; 2038 #ifdef STATS 2039 struct statsblob *sbp; 2040 #endif 2041 size_t len; 2042 2043 INP_WLOCK_ASSERT(inp); 2044 KASSERT((inp->inp_flags & INP_DROPPED) == 0, 2045 ("inp_flags == %x", inp->inp_flags)); 2046 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL")); 2047 2048 switch (sopt->sopt_level) { 2049 #ifdef INET6 2050 case IPPROTO_IPV6: 2051 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2052 switch (sopt->sopt_name) { 2053 case IPV6_USE_MIN_MTU: 2054 tcp6_use_min_mtu(tp); 2055 /* FALLTHROUGH */ 2056 } 2057 INP_WUNLOCK(inp); 2058 return (0); 2059 #endif 2060 #ifdef INET 2061 case IPPROTO_IP: 2062 INP_WUNLOCK(inp); 2063 return (0); 2064 #endif 2065 } 2066 2067 /* 2068 * For TCP_CCALGOOPT forward the control to CC module, for both 2069 * SOPT_SET and SOPT_GET. 2070 */ 2071 switch (sopt->sopt_name) { 2072 case TCP_CCALGOOPT: 2073 INP_WUNLOCK(inp); 2074 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2075 return (EINVAL); 2076 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2077 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2078 sopt->sopt_valsize); 2079 if (error) { 2080 free(pbuf, M_TEMP); 2081 return (error); 2082 } 2083 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2084 if (CC_ALGO(tp)->ctl_output != NULL) 2085 error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf); 2086 else 2087 error = ENOENT; 2088 INP_WUNLOCK(inp); 2089 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2090 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2091 free(pbuf, M_TEMP); 2092 return (error); 2093 } 2094 2095 switch (sopt->sopt_dir) { 2096 case SOPT_SET: 2097 switch (sopt->sopt_name) { 2098 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2099 case TCP_MD5SIG: 2100 INP_WUNLOCK(inp); 2101 if (!TCPMD5_ENABLED()) 2102 return (ENOPROTOOPT); 2103 error = TCPMD5_PCBCTL(inp, sopt); 2104 if (error) 2105 return (error); 2106 INP_WLOCK_RECHECK(inp); 2107 goto unlock_and_done; 2108 #endif /* IPSEC */ 2109 2110 case TCP_NODELAY: 2111 case TCP_NOOPT: 2112 INP_WUNLOCK(inp); 2113 error = sooptcopyin(sopt, &optval, sizeof optval, 2114 sizeof optval); 2115 if (error) 2116 return (error); 2117 2118 INP_WLOCK_RECHECK(inp); 2119 switch (sopt->sopt_name) { 2120 case TCP_NODELAY: 2121 opt = TF_NODELAY; 2122 break; 2123 case TCP_NOOPT: 2124 opt = TF_NOOPT; 2125 break; 2126 default: 2127 opt = 0; /* dead code to fool gcc */ 2128 break; 2129 } 2130 2131 if (optval) 2132 tp->t_flags |= opt; 2133 else 2134 tp->t_flags &= ~opt; 2135 unlock_and_done: 2136 #ifdef TCP_OFFLOAD 2137 if (tp->t_flags & TF_TOE) { 2138 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2139 sopt->sopt_name); 2140 } 2141 #endif 2142 INP_WUNLOCK(inp); 2143 break; 2144 2145 case TCP_NOPUSH: 2146 INP_WUNLOCK(inp); 2147 error = sooptcopyin(sopt, &optval, sizeof optval, 2148 sizeof optval); 2149 if (error) 2150 return (error); 2151 2152 INP_WLOCK_RECHECK(inp); 2153 if (optval) 2154 tp->t_flags |= TF_NOPUSH; 2155 else if (tp->t_flags & TF_NOPUSH) { 2156 tp->t_flags &= ~TF_NOPUSH; 2157 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2158 struct epoch_tracker et; 2159 2160 NET_EPOCH_ENTER(et); 2161 error = tcp_output_nodrop(tp); 2162 NET_EPOCH_EXIT(et); 2163 } 2164 } 2165 goto unlock_and_done; 2166 2167 case TCP_REMOTE_UDP_ENCAPS_PORT: 2168 INP_WUNLOCK(inp); 2169 error = sooptcopyin(sopt, &optval, sizeof optval, 2170 sizeof optval); 2171 if (error) 2172 return (error); 2173 if ((optval < TCP_TUNNELING_PORT_MIN) || 2174 (optval > TCP_TUNNELING_PORT_MAX)) { 2175 /* Its got to be in range */ 2176 return (EINVAL); 2177 } 2178 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2179 /* You have to have enabled a UDP tunneling port first */ 2180 return (EINVAL); 2181 } 2182 INP_WLOCK_RECHECK(inp); 2183 if (tp->t_state != TCPS_CLOSED) { 2184 /* You can't change after you are connected */ 2185 error = EINVAL; 2186 } else { 2187 /* Ok we are all good set the port */ 2188 tp->t_port = htons(optval); 2189 } 2190 goto unlock_and_done; 2191 2192 case TCP_MAXSEG: 2193 INP_WUNLOCK(inp); 2194 error = sooptcopyin(sopt, &optval, sizeof optval, 2195 sizeof optval); 2196 if (error) 2197 return (error); 2198 2199 INP_WLOCK_RECHECK(inp); 2200 if (optval > 0 && optval <= tp->t_maxseg && 2201 optval + 40 >= V_tcp_minmss) { 2202 tp->t_maxseg = optval; 2203 if (tp->t_maxseg < V_tcp_mssdflt) { 2204 /* 2205 * The MSS is so small we should not process incoming 2206 * SACK's since we are subject to attack in such a 2207 * case. 2208 */ 2209 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 2210 } else { 2211 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 2212 } 2213 } else 2214 error = EINVAL; 2215 goto unlock_and_done; 2216 2217 case TCP_INFO: 2218 INP_WUNLOCK(inp); 2219 error = EINVAL; 2220 break; 2221 2222 case TCP_STATS: 2223 INP_WUNLOCK(inp); 2224 #ifdef STATS 2225 error = sooptcopyin(sopt, &optval, sizeof optval, 2226 sizeof optval); 2227 if (error) 2228 return (error); 2229 2230 if (optval > 0) 2231 sbp = stats_blob_alloc( 2232 V_tcp_perconn_stats_dflt_tpl, 0); 2233 else 2234 sbp = NULL; 2235 2236 INP_WLOCK_RECHECK(inp); 2237 if ((tp->t_stats != NULL && sbp == NULL) || 2238 (tp->t_stats == NULL && sbp != NULL)) { 2239 struct statsblob *t = tp->t_stats; 2240 tp->t_stats = sbp; 2241 sbp = t; 2242 } 2243 INP_WUNLOCK(inp); 2244 2245 stats_blob_destroy(sbp); 2246 #else 2247 return (EOPNOTSUPP); 2248 #endif /* !STATS */ 2249 break; 2250 2251 case TCP_CONGESTION: 2252 error = tcp_set_cc_mod(inp, sopt); 2253 break; 2254 2255 case TCP_REUSPORT_LB_NUMA: 2256 INP_WUNLOCK(inp); 2257 error = sooptcopyin(sopt, &optval, sizeof(optval), 2258 sizeof(optval)); 2259 INP_WLOCK_RECHECK(inp); 2260 if (!error) 2261 error = in_pcblbgroup_numa(inp, optval); 2262 INP_WUNLOCK(inp); 2263 break; 2264 2265 #ifdef KERN_TLS 2266 case TCP_TXTLS_ENABLE: 2267 INP_WUNLOCK(inp); 2268 error = ktls_copyin_tls_enable(sopt, &tls); 2269 if (error != 0) 2270 break; 2271 error = ktls_enable_tx(so, &tls); 2272 ktls_cleanup_tls_enable(&tls); 2273 break; 2274 case TCP_TXTLS_MODE: 2275 INP_WUNLOCK(inp); 2276 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2277 if (error != 0) 2278 return (error); 2279 2280 INP_WLOCK_RECHECK(inp); 2281 error = ktls_set_tx_mode(so, ui); 2282 INP_WUNLOCK(inp); 2283 break; 2284 case TCP_RXTLS_ENABLE: 2285 INP_WUNLOCK(inp); 2286 error = ktls_copyin_tls_enable(sopt, &tls); 2287 if (error != 0) 2288 break; 2289 error = ktls_enable_rx(so, &tls); 2290 ktls_cleanup_tls_enable(&tls); 2291 break; 2292 #endif 2293 case TCP_MAXUNACKTIME: 2294 case TCP_KEEPIDLE: 2295 case TCP_KEEPINTVL: 2296 case TCP_KEEPINIT: 2297 INP_WUNLOCK(inp); 2298 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2299 if (error) 2300 return (error); 2301 2302 if (ui > (UINT_MAX / hz)) { 2303 error = EINVAL; 2304 break; 2305 } 2306 ui *= hz; 2307 2308 INP_WLOCK_RECHECK(inp); 2309 switch (sopt->sopt_name) { 2310 case TCP_MAXUNACKTIME: 2311 tp->t_maxunacktime = ui; 2312 break; 2313 2314 case TCP_KEEPIDLE: 2315 tp->t_keepidle = ui; 2316 /* 2317 * XXX: better check current remaining 2318 * timeout and "merge" it with new value. 2319 */ 2320 if ((tp->t_state > TCPS_LISTEN) && 2321 (tp->t_state <= TCPS_CLOSING)) 2322 tcp_timer_activate(tp, TT_KEEP, 2323 TP_KEEPIDLE(tp)); 2324 break; 2325 case TCP_KEEPINTVL: 2326 tp->t_keepintvl = ui; 2327 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2328 (TP_MAXIDLE(tp) > 0)) 2329 tcp_timer_activate(tp, TT_2MSL, 2330 TP_MAXIDLE(tp)); 2331 break; 2332 case TCP_KEEPINIT: 2333 tp->t_keepinit = ui; 2334 if (tp->t_state == TCPS_SYN_RECEIVED || 2335 tp->t_state == TCPS_SYN_SENT) 2336 tcp_timer_activate(tp, TT_KEEP, 2337 TP_KEEPINIT(tp)); 2338 break; 2339 } 2340 goto unlock_and_done; 2341 2342 case TCP_KEEPCNT: 2343 INP_WUNLOCK(inp); 2344 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2345 if (error) 2346 return (error); 2347 2348 INP_WLOCK_RECHECK(inp); 2349 tp->t_keepcnt = ui; 2350 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2351 (TP_MAXIDLE(tp) > 0)) 2352 tcp_timer_activate(tp, TT_2MSL, 2353 TP_MAXIDLE(tp)); 2354 goto unlock_and_done; 2355 2356 #ifdef TCPPCAP 2357 case TCP_PCAP_OUT: 2358 case TCP_PCAP_IN: 2359 INP_WUNLOCK(inp); 2360 error = sooptcopyin(sopt, &optval, sizeof optval, 2361 sizeof optval); 2362 if (error) 2363 return (error); 2364 2365 INP_WLOCK_RECHECK(inp); 2366 if (optval >= 0) 2367 tcp_pcap_set_sock_max( 2368 (sopt->sopt_name == TCP_PCAP_OUT) ? 2369 &(tp->t_outpkts) : &(tp->t_inpkts), 2370 optval); 2371 else 2372 error = EINVAL; 2373 goto unlock_and_done; 2374 #endif 2375 2376 case TCP_FASTOPEN: { 2377 struct tcp_fastopen tfo_optval; 2378 2379 INP_WUNLOCK(inp); 2380 if (!V_tcp_fastopen_client_enable && 2381 !V_tcp_fastopen_server_enable) 2382 return (EPERM); 2383 2384 error = sooptcopyin(sopt, &tfo_optval, 2385 sizeof(tfo_optval), sizeof(int)); 2386 if (error) 2387 return (error); 2388 2389 INP_WLOCK_RECHECK(inp); 2390 if ((tp->t_state != TCPS_CLOSED) && 2391 (tp->t_state != TCPS_LISTEN)) { 2392 error = EINVAL; 2393 goto unlock_and_done; 2394 } 2395 if (tfo_optval.enable) { 2396 if (tp->t_state == TCPS_LISTEN) { 2397 if (!V_tcp_fastopen_server_enable) { 2398 error = EPERM; 2399 goto unlock_and_done; 2400 } 2401 2402 if (tp->t_tfo_pending == NULL) 2403 tp->t_tfo_pending = 2404 tcp_fastopen_alloc_counter(); 2405 } else { 2406 /* 2407 * If a pre-shared key was provided, 2408 * stash it in the client cookie 2409 * field of the tcpcb for use during 2410 * connect. 2411 */ 2412 if (sopt->sopt_valsize == 2413 sizeof(tfo_optval)) { 2414 memcpy(tp->t_tfo_cookie.client, 2415 tfo_optval.psk, 2416 TCP_FASTOPEN_PSK_LEN); 2417 tp->t_tfo_client_cookie_len = 2418 TCP_FASTOPEN_PSK_LEN; 2419 } 2420 } 2421 tp->t_flags |= TF_FASTOPEN; 2422 } else 2423 tp->t_flags &= ~TF_FASTOPEN; 2424 goto unlock_and_done; 2425 } 2426 2427 #ifdef TCP_BLACKBOX 2428 case TCP_LOG: 2429 INP_WUNLOCK(inp); 2430 error = sooptcopyin(sopt, &optval, sizeof optval, 2431 sizeof optval); 2432 if (error) 2433 return (error); 2434 2435 INP_WLOCK_RECHECK(inp); 2436 error = tcp_log_state_change(tp, optval); 2437 goto unlock_and_done; 2438 2439 case TCP_LOGBUF: 2440 INP_WUNLOCK(inp); 2441 error = EINVAL; 2442 break; 2443 2444 case TCP_LOGID: 2445 INP_WUNLOCK(inp); 2446 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2447 if (error) 2448 break; 2449 buf[sopt->sopt_valsize] = '\0'; 2450 INP_WLOCK_RECHECK(inp); 2451 error = tcp_log_set_id(tp, buf); 2452 /* tcp_log_set_id() unlocks the INP. */ 2453 break; 2454 2455 case TCP_LOGDUMP: 2456 case TCP_LOGDUMPID: 2457 INP_WUNLOCK(inp); 2458 error = 2459 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2460 if (error) 2461 break; 2462 buf[sopt->sopt_valsize] = '\0'; 2463 INP_WLOCK_RECHECK(inp); 2464 if (sopt->sopt_name == TCP_LOGDUMP) { 2465 error = tcp_log_dump_tp_logbuf(tp, buf, 2466 M_WAITOK, true); 2467 INP_WUNLOCK(inp); 2468 } else { 2469 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2470 /* 2471 * tcp_log_dump_tp_bucket_logbufs() drops the 2472 * INP lock. 2473 */ 2474 } 2475 break; 2476 #endif 2477 2478 default: 2479 INP_WUNLOCK(inp); 2480 error = ENOPROTOOPT; 2481 break; 2482 } 2483 break; 2484 2485 case SOPT_GET: 2486 tp = intotcpcb(inp); 2487 switch (sopt->sopt_name) { 2488 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2489 case TCP_MD5SIG: 2490 INP_WUNLOCK(inp); 2491 if (!TCPMD5_ENABLED()) 2492 return (ENOPROTOOPT); 2493 error = TCPMD5_PCBCTL(inp, sopt); 2494 break; 2495 #endif 2496 2497 case TCP_NODELAY: 2498 optval = tp->t_flags & TF_NODELAY; 2499 INP_WUNLOCK(inp); 2500 error = sooptcopyout(sopt, &optval, sizeof optval); 2501 break; 2502 case TCP_MAXSEG: 2503 optval = tp->t_maxseg; 2504 INP_WUNLOCK(inp); 2505 error = sooptcopyout(sopt, &optval, sizeof optval); 2506 break; 2507 case TCP_REMOTE_UDP_ENCAPS_PORT: 2508 optval = ntohs(tp->t_port); 2509 INP_WUNLOCK(inp); 2510 error = sooptcopyout(sopt, &optval, sizeof optval); 2511 break; 2512 case TCP_NOOPT: 2513 optval = tp->t_flags & TF_NOOPT; 2514 INP_WUNLOCK(inp); 2515 error = sooptcopyout(sopt, &optval, sizeof optval); 2516 break; 2517 case TCP_NOPUSH: 2518 optval = tp->t_flags & TF_NOPUSH; 2519 INP_WUNLOCK(inp); 2520 error = sooptcopyout(sopt, &optval, sizeof optval); 2521 break; 2522 case TCP_INFO: 2523 tcp_fill_info(tp, &ti); 2524 INP_WUNLOCK(inp); 2525 error = sooptcopyout(sopt, &ti, sizeof ti); 2526 break; 2527 case TCP_STATS: 2528 { 2529 #ifdef STATS 2530 int nheld; 2531 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2532 2533 error = 0; 2534 socklen_t outsbsz = sopt->sopt_valsize; 2535 if (tp->t_stats == NULL) 2536 error = ENOENT; 2537 else if (outsbsz >= tp->t_stats->cursz) 2538 outsbsz = tp->t_stats->cursz; 2539 else if (outsbsz >= sizeof(struct statsblob)) 2540 outsbsz = sizeof(struct statsblob); 2541 else 2542 error = EINVAL; 2543 INP_WUNLOCK(inp); 2544 if (error) 2545 break; 2546 2547 sbp = sopt->sopt_val; 2548 nheld = atop(round_page(((vm_offset_t)sbp) + 2549 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2550 vm_page_t ma[nheld]; 2551 if (vm_fault_quick_hold_pages( 2552 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2553 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2554 nheld) < 0) { 2555 error = EFAULT; 2556 break; 2557 } 2558 2559 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2560 SIZEOF_MEMBER(struct statsblob, flags)))) 2561 goto unhold; 2562 2563 INP_WLOCK_RECHECK(inp); 2564 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2565 sbflags | SB_CLONE_USRDSTNOFAULT); 2566 INP_WUNLOCK(inp); 2567 sopt->sopt_valsize = outsbsz; 2568 unhold: 2569 vm_page_unhold_pages(ma, nheld); 2570 #else 2571 INP_WUNLOCK(inp); 2572 error = EOPNOTSUPP; 2573 #endif /* !STATS */ 2574 break; 2575 } 2576 case TCP_CONGESTION: 2577 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2578 INP_WUNLOCK(inp); 2579 error = sooptcopyout(sopt, buf, len + 1); 2580 break; 2581 case TCP_MAXUNACKTIME: 2582 case TCP_KEEPIDLE: 2583 case TCP_KEEPINTVL: 2584 case TCP_KEEPINIT: 2585 case TCP_KEEPCNT: 2586 switch (sopt->sopt_name) { 2587 case TCP_MAXUNACKTIME: 2588 ui = TP_MAXUNACKTIME(tp) / hz; 2589 break; 2590 case TCP_KEEPIDLE: 2591 ui = TP_KEEPIDLE(tp) / hz; 2592 break; 2593 case TCP_KEEPINTVL: 2594 ui = TP_KEEPINTVL(tp) / hz; 2595 break; 2596 case TCP_KEEPINIT: 2597 ui = TP_KEEPINIT(tp) / hz; 2598 break; 2599 case TCP_KEEPCNT: 2600 ui = TP_KEEPCNT(tp); 2601 break; 2602 } 2603 INP_WUNLOCK(inp); 2604 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2605 break; 2606 #ifdef TCPPCAP 2607 case TCP_PCAP_OUT: 2608 case TCP_PCAP_IN: 2609 optval = tcp_pcap_get_sock_max( 2610 (sopt->sopt_name == TCP_PCAP_OUT) ? 2611 &(tp->t_outpkts) : &(tp->t_inpkts)); 2612 INP_WUNLOCK(inp); 2613 error = sooptcopyout(sopt, &optval, sizeof optval); 2614 break; 2615 #endif 2616 case TCP_FASTOPEN: 2617 optval = tp->t_flags & TF_FASTOPEN; 2618 INP_WUNLOCK(inp); 2619 error = sooptcopyout(sopt, &optval, sizeof optval); 2620 break; 2621 #ifdef TCP_BLACKBOX 2622 case TCP_LOG: 2623 optval = tcp_get_bblog_state(tp); 2624 INP_WUNLOCK(inp); 2625 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2626 break; 2627 case TCP_LOGBUF: 2628 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2629 error = tcp_log_getlogbuf(sopt, tp); 2630 break; 2631 case TCP_LOGID: 2632 len = tcp_log_get_id(tp, buf); 2633 INP_WUNLOCK(inp); 2634 error = sooptcopyout(sopt, buf, len + 1); 2635 break; 2636 case TCP_LOGDUMP: 2637 case TCP_LOGDUMPID: 2638 INP_WUNLOCK(inp); 2639 error = EINVAL; 2640 break; 2641 #endif 2642 #ifdef KERN_TLS 2643 case TCP_TXTLS_MODE: 2644 error = ktls_get_tx_mode(so, &optval); 2645 INP_WUNLOCK(inp); 2646 if (error == 0) 2647 error = sooptcopyout(sopt, &optval, 2648 sizeof(optval)); 2649 break; 2650 case TCP_RXTLS_MODE: 2651 error = ktls_get_rx_mode(so, &optval); 2652 INP_WUNLOCK(inp); 2653 if (error == 0) 2654 error = sooptcopyout(sopt, &optval, 2655 sizeof(optval)); 2656 break; 2657 #endif 2658 default: 2659 INP_WUNLOCK(inp); 2660 error = ENOPROTOOPT; 2661 break; 2662 } 2663 break; 2664 } 2665 return (error); 2666 } 2667 #undef INP_WLOCK_RECHECK 2668 #undef INP_WLOCK_RECHECK_CLEANUP 2669 2670 /* 2671 * Initiate (or continue) disconnect. 2672 * If embryonic state, just send reset (once). 2673 * If in ``let data drain'' option and linger null, just drop. 2674 * Otherwise (hard), mark socket disconnecting and drop 2675 * current input data; switch states based on user close, and 2676 * send segment to peer (with FIN). 2677 */ 2678 static void 2679 tcp_disconnect(struct tcpcb *tp) 2680 { 2681 struct inpcb *inp = tptoinpcb(tp); 2682 struct socket *so = tptosocket(tp); 2683 2684 NET_EPOCH_ASSERT(); 2685 INP_WLOCK_ASSERT(inp); 2686 2687 /* 2688 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2689 * socket is still open. 2690 */ 2691 if (tp->t_state < TCPS_ESTABLISHED && 2692 !(tp->t_state > TCPS_LISTEN && (tp->t_flags & TF_FASTOPEN))) { 2693 tp = tcp_close(tp); 2694 KASSERT(tp != NULL, 2695 ("tcp_disconnect: tcp_close() returned NULL")); 2696 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2697 tp = tcp_drop(tp, 0); 2698 KASSERT(tp != NULL, 2699 ("tcp_disconnect: tcp_drop() returned NULL")); 2700 } else { 2701 soisdisconnecting(so); 2702 sbflush(&so->so_rcv); 2703 tcp_usrclosed(tp); 2704 if (!(inp->inp_flags & INP_DROPPED)) 2705 /* Ignore stack's drop request, we already at it. */ 2706 (void)tcp_output_nodrop(tp); 2707 } 2708 } 2709 2710 /* 2711 * User issued close, and wish to trail through shutdown states: 2712 * if never received SYN, just forget it. If got a SYN from peer, 2713 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2714 * If already got a FIN from peer, then almost done; go to LAST_ACK 2715 * state. In all other cases, have already sent FIN to peer (e.g. 2716 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2717 * for peer to send FIN or not respond to keep-alives, etc. 2718 * We can let the user exit from the close as soon as the FIN is acked. 2719 */ 2720 static void 2721 tcp_usrclosed(struct tcpcb *tp) 2722 { 2723 2724 NET_EPOCH_ASSERT(); 2725 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2726 2727 switch (tp->t_state) { 2728 case TCPS_LISTEN: 2729 #ifdef TCP_OFFLOAD 2730 tcp_offload_listen_stop(tp); 2731 #endif 2732 tcp_state_change(tp, TCPS_CLOSED); 2733 /* FALLTHROUGH */ 2734 case TCPS_CLOSED: 2735 tp = tcp_close(tp); 2736 /* 2737 * tcp_close() should never return NULL here as the socket is 2738 * still open. 2739 */ 2740 KASSERT(tp != NULL, 2741 ("tcp_usrclosed: tcp_close() returned NULL")); 2742 break; 2743 2744 case TCPS_SYN_SENT: 2745 case TCPS_SYN_RECEIVED: 2746 tp->t_flags |= TF_NEEDFIN; 2747 break; 2748 2749 case TCPS_ESTABLISHED: 2750 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2751 break; 2752 2753 case TCPS_CLOSE_WAIT: 2754 tcp_state_change(tp, TCPS_LAST_ACK); 2755 break; 2756 } 2757 if (tp->t_acktime == 0) 2758 tp->t_acktime = ticks; 2759 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2760 tcp_free_sackholes(tp); 2761 soisdisconnected(tptosocket(tp)); 2762 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2763 if (tp->t_state == TCPS_FIN_WAIT_2) { 2764 int timeout; 2765 2766 timeout = (tcp_fast_finwait2_recycle) ? 2767 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2768 tcp_timer_activate(tp, TT_2MSL, timeout); 2769 } 2770 } 2771 } 2772 2773 #ifdef DDB 2774 static void 2775 db_print_indent(int indent) 2776 { 2777 int i; 2778 2779 for (i = 0; i < indent; i++) 2780 db_printf(" "); 2781 } 2782 2783 static void 2784 db_print_tstate(int t_state) 2785 { 2786 2787 switch (t_state) { 2788 case TCPS_CLOSED: 2789 db_printf("TCPS_CLOSED"); 2790 return; 2791 2792 case TCPS_LISTEN: 2793 db_printf("TCPS_LISTEN"); 2794 return; 2795 2796 case TCPS_SYN_SENT: 2797 db_printf("TCPS_SYN_SENT"); 2798 return; 2799 2800 case TCPS_SYN_RECEIVED: 2801 db_printf("TCPS_SYN_RECEIVED"); 2802 return; 2803 2804 case TCPS_ESTABLISHED: 2805 db_printf("TCPS_ESTABLISHED"); 2806 return; 2807 2808 case TCPS_CLOSE_WAIT: 2809 db_printf("TCPS_CLOSE_WAIT"); 2810 return; 2811 2812 case TCPS_FIN_WAIT_1: 2813 db_printf("TCPS_FIN_WAIT_1"); 2814 return; 2815 2816 case TCPS_CLOSING: 2817 db_printf("TCPS_CLOSING"); 2818 return; 2819 2820 case TCPS_LAST_ACK: 2821 db_printf("TCPS_LAST_ACK"); 2822 return; 2823 2824 case TCPS_FIN_WAIT_2: 2825 db_printf("TCPS_FIN_WAIT_2"); 2826 return; 2827 2828 case TCPS_TIME_WAIT: 2829 db_printf("TCPS_TIME_WAIT"); 2830 return; 2831 2832 default: 2833 db_printf("unknown"); 2834 return; 2835 } 2836 } 2837 2838 static void 2839 db_print_tflags(u_int t_flags) 2840 { 2841 int comma; 2842 2843 comma = 0; 2844 if (t_flags & TF_ACKNOW) { 2845 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2846 comma = 1; 2847 } 2848 if (t_flags & TF_DELACK) { 2849 db_printf("%sTF_DELACK", comma ? ", " : ""); 2850 comma = 1; 2851 } 2852 if (t_flags & TF_NODELAY) { 2853 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2854 comma = 1; 2855 } 2856 if (t_flags & TF_NOOPT) { 2857 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2858 comma = 1; 2859 } 2860 if (t_flags & TF_SENTFIN) { 2861 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2862 comma = 1; 2863 } 2864 if (t_flags & TF_REQ_SCALE) { 2865 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2866 comma = 1; 2867 } 2868 if (t_flags & TF_RCVD_SCALE) { 2869 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2870 comma = 1; 2871 } 2872 if (t_flags & TF_REQ_TSTMP) { 2873 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2874 comma = 1; 2875 } 2876 if (t_flags & TF_RCVD_TSTMP) { 2877 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2878 comma = 1; 2879 } 2880 if (t_flags & TF_SACK_PERMIT) { 2881 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2882 comma = 1; 2883 } 2884 if (t_flags & TF_NEEDSYN) { 2885 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2886 comma = 1; 2887 } 2888 if (t_flags & TF_NEEDFIN) { 2889 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2890 comma = 1; 2891 } 2892 if (t_flags & TF_NOPUSH) { 2893 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2894 comma = 1; 2895 } 2896 if (t_flags & TF_PREVVALID) { 2897 db_printf("%sTF_PREVVALID", comma ? ", " : ""); 2898 comma = 1; 2899 } 2900 if (t_flags & TF_MORETOCOME) { 2901 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2902 comma = 1; 2903 } 2904 if (t_flags & TF_SONOTCONN) { 2905 db_printf("%sTF_SONOTCONN", comma ? ", " : ""); 2906 comma = 1; 2907 } 2908 if (t_flags & TF_LASTIDLE) { 2909 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2910 comma = 1; 2911 } 2912 if (t_flags & TF_RXWIN0SENT) { 2913 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2914 comma = 1; 2915 } 2916 if (t_flags & TF_FASTRECOVERY) { 2917 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2918 comma = 1; 2919 } 2920 if (t_flags & TF_CONGRECOVERY) { 2921 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2922 comma = 1; 2923 } 2924 if (t_flags & TF_WASFRECOVERY) { 2925 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2926 comma = 1; 2927 } 2928 if (t_flags & TF_WASCRECOVERY) { 2929 db_printf("%sTF_WASCRECOVERY", comma ? ", " : ""); 2930 comma = 1; 2931 } 2932 if (t_flags & TF_SIGNATURE) { 2933 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2934 comma = 1; 2935 } 2936 if (t_flags & TF_FORCEDATA) { 2937 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2938 comma = 1; 2939 } 2940 if (t_flags & TF_TSO) { 2941 db_printf("%sTF_TSO", comma ? ", " : ""); 2942 comma = 1; 2943 } 2944 if (t_flags & TF_FASTOPEN) { 2945 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2946 comma = 1; 2947 } 2948 } 2949 2950 static void 2951 db_print_tflags2(u_int t_flags2) 2952 { 2953 int comma; 2954 2955 comma = 0; 2956 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) { 2957 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : ""); 2958 comma = 1; 2959 } 2960 if (t_flags2 & TF2_PLPMTU_PMTUD) { 2961 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : ""); 2962 comma = 1; 2963 } 2964 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) { 2965 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : ""); 2966 comma = 1; 2967 } 2968 if (t_flags2 & TF2_LOG_AUTO) { 2969 db_printf("%sTF2_LOG_AUTO", comma ? ", " : ""); 2970 comma = 1; 2971 } 2972 if (t_flags2 & TF2_DROP_AF_DATA) { 2973 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : ""); 2974 comma = 1; 2975 } 2976 if (t_flags2 & TF2_ECN_PERMIT) { 2977 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2978 comma = 1; 2979 } 2980 if (t_flags2 & TF2_ECN_SND_CWR) { 2981 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : ""); 2982 comma = 1; 2983 } 2984 if (t_flags2 & TF2_ECN_SND_ECE) { 2985 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : ""); 2986 comma = 1; 2987 } 2988 if (t_flags2 & TF2_ACE_PERMIT) { 2989 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : ""); 2990 comma = 1; 2991 } 2992 if (t_flags2 & TF2_FBYTES_COMPLETE) { 2993 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : ""); 2994 comma = 1; 2995 } 2996 } 2997 2998 static void 2999 db_print_toobflags(char t_oobflags) 3000 { 3001 int comma; 3002 3003 comma = 0; 3004 if (t_oobflags & TCPOOB_HAVEDATA) { 3005 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3006 comma = 1; 3007 } 3008 if (t_oobflags & TCPOOB_HADDATA) { 3009 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3010 comma = 1; 3011 } 3012 } 3013 3014 static void 3015 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3016 { 3017 3018 db_print_indent(indent); 3019 db_printf("%s at %p\n", name, tp); 3020 3021 indent += 2; 3022 3023 db_print_indent(indent); 3024 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3025 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3026 3027 db_print_indent(indent); 3028 db_printf("t_callout: %p t_timers: %p\n", 3029 &tp->t_callout, &tp->t_timers); 3030 3031 db_print_indent(indent); 3032 db_printf("t_state: %d (", tp->t_state); 3033 db_print_tstate(tp->t_state); 3034 db_printf(")\n"); 3035 3036 db_print_indent(indent); 3037 db_printf("t_flags: 0x%x (", tp->t_flags); 3038 db_print_tflags(tp->t_flags); 3039 db_printf(")\n"); 3040 3041 db_print_indent(indent); 3042 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3043 db_print_tflags2(tp->t_flags2); 3044 db_printf(")\n"); 3045 3046 db_print_indent(indent); 3047 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n", 3048 tp->snd_una, tp->snd_max, tp->snd_nxt); 3049 3050 db_print_indent(indent); 3051 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3052 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3053 3054 db_print_indent(indent); 3055 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3056 tp->iss, tp->irs, tp->rcv_nxt); 3057 3058 db_print_indent(indent); 3059 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3060 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3061 3062 db_print_indent(indent); 3063 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3064 tp->snd_wnd, tp->snd_cwnd); 3065 3066 db_print_indent(indent); 3067 db_printf("snd_ssthresh: %u snd_recover: " 3068 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3069 3070 db_print_indent(indent); 3071 db_printf("t_rcvtime: %u t_startime: %u\n", 3072 tp->t_rcvtime, tp->t_starttime); 3073 3074 db_print_indent(indent); 3075 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3076 tp->t_rtttime, tp->t_rtseq); 3077 3078 db_print_indent(indent); 3079 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3080 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3081 3082 db_print_indent(indent); 3083 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n", 3084 tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin); 3085 3086 db_print_indent(indent); 3087 db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n", 3088 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3089 3090 db_print_indent(indent); 3091 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3092 db_print_toobflags(tp->t_oobflags); 3093 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3094 3095 db_print_indent(indent); 3096 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3097 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3098 3099 db_print_indent(indent); 3100 db_printf("ts_recent: %u ts_recent_age: %u\n", 3101 tp->ts_recent, tp->ts_recent_age); 3102 3103 db_print_indent(indent); 3104 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3105 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3106 3107 db_print_indent(indent); 3108 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3109 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3110 tp->snd_recover_prev, tp->t_badrxtwin); 3111 3112 db_print_indent(indent); 3113 db_printf("snd_numholes: %d snd_holes first: %p\n", 3114 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3115 3116 db_print_indent(indent); 3117 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3118 tp->snd_fack, tp->rcv_numsacks); 3119 3120 /* Skip sackblks, sackhint. */ 3121 3122 db_print_indent(indent); 3123 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3124 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3125 } 3126 3127 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3128 { 3129 struct tcpcb *tp; 3130 3131 if (!have_addr) { 3132 db_printf("usage: show tcpcb <addr>\n"); 3133 return; 3134 } 3135 tp = (struct tcpcb *)addr; 3136 3137 db_print_tcpcb(tp, "tcpcb", 0); 3138 } 3139 #endif 3140