1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2006-2007 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_ipsec.h" 47 #include "opt_kern_tls.h" 48 #include "opt_tcpdebug.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/arb.h> 53 #include <sys/limits.h> 54 #include <sys/malloc.h> 55 #include <sys/refcount.h> 56 #include <sys/kernel.h> 57 #include <sys/ktls.h> 58 #include <sys/qmath.h> 59 #include <sys/sysctl.h> 60 #include <sys/mbuf.h> 61 #ifdef INET6 62 #include <sys/domain.h> 63 #endif /* INET6 */ 64 #include <sys/socket.h> 65 #include <sys/socketvar.h> 66 #include <sys/protosw.h> 67 #include <sys/proc.h> 68 #include <sys/jail.h> 69 #include <sys/syslog.h> 70 #include <sys/stats.h> 71 72 #ifdef DDB 73 #include <ddb/ddb.h> 74 #endif 75 76 #include <net/if.h> 77 #include <net/if_var.h> 78 #include <net/route.h> 79 #include <net/vnet.h> 80 81 #include <netinet/in.h> 82 #include <netinet/in_kdtrace.h> 83 #include <netinet/in_pcb.h> 84 #include <netinet/in_systm.h> 85 #include <netinet/in_var.h> 86 #include <netinet/ip_var.h> 87 #ifdef INET6 88 #include <netinet/ip6.h> 89 #include <netinet6/in6_pcb.h> 90 #include <netinet6/ip6_var.h> 91 #include <netinet6/scope6_var.h> 92 #endif 93 #include <netinet/tcp.h> 94 #include <netinet/tcp_fsm.h> 95 #include <netinet/tcp_seq.h> 96 #include <netinet/tcp_timer.h> 97 #include <netinet/tcp_var.h> 98 #include <netinet/tcp_log_buf.h> 99 #include <netinet/tcpip.h> 100 #include <netinet/cc/cc.h> 101 #include <netinet/tcp_fastopen.h> 102 #include <netinet/tcp_hpts.h> 103 #ifdef TCPPCAP 104 #include <netinet/tcp_pcap.h> 105 #endif 106 #ifdef TCPDEBUG 107 #include <netinet/tcp_debug.h> 108 #endif 109 #ifdef TCP_OFFLOAD 110 #include <netinet/tcp_offload.h> 111 #endif 112 #include <netipsec/ipsec_support.h> 113 114 #include <vm/vm.h> 115 #include <vm/vm_param.h> 116 #include <vm/pmap.h> 117 #include <vm/vm_extern.h> 118 #include <vm/vm_map.h> 119 #include <vm/vm_page.h> 120 121 /* 122 * TCP protocol interface to socket abstraction. 123 */ 124 #ifdef INET 125 static int tcp_connect(struct tcpcb *, struct sockaddr *, 126 struct thread *td); 127 #endif /* INET */ 128 #ifdef INET6 129 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 130 struct thread *td); 131 #endif /* INET6 */ 132 static void tcp_disconnect(struct tcpcb *); 133 static void tcp_usrclosed(struct tcpcb *); 134 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 135 136 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 137 138 #ifdef TCPDEBUG 139 #define TCPDEBUG0 int ostate = 0 140 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 141 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 142 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 143 #else 144 #define TCPDEBUG0 145 #define TCPDEBUG1() 146 #define TCPDEBUG2(req) 147 #endif 148 149 /* 150 * tcp_require_unique port requires a globally-unique source port for each 151 * outgoing connection. The default is to require the 4-tuple to be unique. 152 */ 153 VNET_DEFINE(int, tcp_require_unique_port) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port, 155 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0, 156 "Require globally-unique ephemeral port for outgoing connections"); 157 #define V_tcp_require_unique_port VNET(tcp_require_unique_port) 158 159 /* 160 * TCP attaches to socket via pru_attach(), reserving space, 161 * and an internet control block. 162 */ 163 static int 164 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 165 { 166 struct inpcb *inp; 167 struct tcpcb *tp = NULL; 168 int error; 169 TCPDEBUG0; 170 171 inp = sotoinpcb(so); 172 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 173 TCPDEBUG1(); 174 175 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 176 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 177 if (error) 178 goto out; 179 } 180 181 so->so_rcv.sb_flags |= SB_AUTOSIZE; 182 so->so_snd.sb_flags |= SB_AUTOSIZE; 183 error = in_pcballoc(so, &V_tcbinfo); 184 if (error) 185 goto out; 186 inp = sotoinpcb(so); 187 #ifdef INET6 188 if (inp->inp_vflag & INP_IPV6PROTO) { 189 inp->inp_vflag |= INP_IPV6; 190 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 191 inp->inp_vflag |= INP_IPV4; 192 inp->in6p_hops = -1; /* use kernel default */ 193 } 194 else 195 #endif 196 inp->inp_vflag |= INP_IPV4; 197 tp = tcp_newtcpcb(inp); 198 if (tp == NULL) { 199 error = ENOBUFS; 200 in_pcbdetach(inp); 201 in_pcbfree(inp); 202 goto out; 203 } 204 tp->t_state = TCPS_CLOSED; 205 INP_WUNLOCK(inp); 206 TCPSTATES_INC(TCPS_CLOSED); 207 out: 208 TCPDEBUG2(PRU_ATTACH); 209 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 210 return (error); 211 } 212 213 /* 214 * tcp_usr_detach is called when the socket layer loses its final reference 215 * to the socket, be it a file descriptor reference, a reference from TCP, 216 * etc. At this point, there is only one case in which we will keep around 217 * inpcb state: time wait. 218 */ 219 static void 220 tcp_usr_detach(struct socket *so) 221 { 222 struct inpcb *inp; 223 struct tcpcb *tp; 224 225 inp = sotoinpcb(so); 226 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 227 INP_WLOCK(inp); 228 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 229 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 230 231 tp = intotcpcb(inp); 232 233 if (inp->inp_flags & INP_TIMEWAIT) { 234 /* 235 * There are two cases to handle: one in which the time wait 236 * state is being discarded (INP_DROPPED), and one in which 237 * this connection will remain in timewait. In the former, 238 * it is time to discard all state (except tcptw, which has 239 * already been discarded by the timewait close code, which 240 * should be further up the call stack somewhere). In the 241 * latter case, we detach from the socket, but leave the pcb 242 * present until timewait ends. 243 * 244 * XXXRW: Would it be cleaner to free the tcptw here? 245 * 246 * Astute question indeed, from twtcp perspective there are 247 * four cases to consider: 248 * 249 * #1 tcp_usr_detach is called at tcptw creation time by 250 * tcp_twstart, then do not discard the newly created tcptw 251 * and leave inpcb present until timewait ends 252 * #2 tcp_usr_detach is called at tcptw creation time by 253 * tcp_twstart, but connection is local and tw will be 254 * discarded immediately 255 * #3 tcp_usr_detach is called at timewait end (or reuse) by 256 * tcp_twclose, then the tcptw has already been discarded 257 * (or reused) and inpcb is freed here 258 * #4 tcp_usr_detach is called() after timewait ends (or reuse) 259 * (e.g. by soclose), then tcptw has already been discarded 260 * (or reused) and inpcb is freed here 261 * 262 * In all three cases the tcptw should not be freed here. 263 */ 264 if (inp->inp_flags & INP_DROPPED) { 265 in_pcbdetach(inp); 266 if (__predict_true(tp == NULL)) { 267 in_pcbfree(inp); 268 } else { 269 /* 270 * This case should not happen as in TIMEWAIT 271 * state the inp should not be destroyed before 272 * its tcptw. If INVARIANTS is defined, panic. 273 */ 274 #ifdef INVARIANTS 275 panic("%s: Panic before an inp double-free: " 276 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 277 , __func__); 278 #else 279 log(LOG_ERR, "%s: Avoid an inp double-free: " 280 "INP_TIMEWAIT && INP_DROPPED && tp != NULL" 281 , __func__); 282 #endif 283 INP_WUNLOCK(inp); 284 } 285 } else { 286 in_pcbdetach(inp); 287 INP_WUNLOCK(inp); 288 } 289 } else { 290 /* 291 * If the connection is not in timewait, we consider two 292 * two conditions: one in which no further processing is 293 * necessary (dropped || embryonic), and one in which TCP is 294 * not yet done, but no longer requires the socket, so the 295 * pcb will persist for the time being. 296 * 297 * XXXRW: Does the second case still occur? 298 */ 299 if (inp->inp_flags & INP_DROPPED || 300 tp->t_state < TCPS_SYN_SENT) { 301 tcp_discardcb(tp); 302 in_pcbdetach(inp); 303 in_pcbfree(inp); 304 } else { 305 in_pcbdetach(inp); 306 INP_WUNLOCK(inp); 307 } 308 } 309 } 310 311 #ifdef INET 312 /* 313 * Give the socket an address. 314 */ 315 static int 316 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 317 { 318 int error = 0; 319 struct inpcb *inp; 320 struct tcpcb *tp = NULL; 321 struct sockaddr_in *sinp; 322 323 sinp = (struct sockaddr_in *)nam; 324 if (nam->sa_family != AF_INET) 325 return (EAFNOSUPPORT); 326 if (nam->sa_len != sizeof(*sinp)) 327 return (EINVAL); 328 329 /* 330 * Must check for multicast addresses and disallow binding 331 * to them. 332 */ 333 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 334 return (EAFNOSUPPORT); 335 336 TCPDEBUG0; 337 inp = sotoinpcb(so); 338 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 339 INP_WLOCK(inp); 340 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 341 error = EINVAL; 342 goto out; 343 } 344 tp = intotcpcb(inp); 345 TCPDEBUG1(); 346 INP_HASH_WLOCK(&V_tcbinfo); 347 error = in_pcbbind(inp, nam, td->td_ucred); 348 INP_HASH_WUNLOCK(&V_tcbinfo); 349 out: 350 TCPDEBUG2(PRU_BIND); 351 TCP_PROBE2(debug__user, tp, PRU_BIND); 352 INP_WUNLOCK(inp); 353 354 return (error); 355 } 356 #endif /* INET */ 357 358 #ifdef INET6 359 static int 360 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 361 { 362 int error = 0; 363 struct inpcb *inp; 364 struct tcpcb *tp = NULL; 365 struct sockaddr_in6 *sin6; 366 u_char vflagsav; 367 368 sin6 = (struct sockaddr_in6 *)nam; 369 if (nam->sa_family != AF_INET6) 370 return (EAFNOSUPPORT); 371 if (nam->sa_len != sizeof(*sin6)) 372 return (EINVAL); 373 374 /* 375 * Must check for multicast addresses and disallow binding 376 * to them. 377 */ 378 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 379 return (EAFNOSUPPORT); 380 381 TCPDEBUG0; 382 inp = sotoinpcb(so); 383 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 384 INP_WLOCK(inp); 385 vflagsav = inp->inp_vflag; 386 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 387 error = EINVAL; 388 goto out; 389 } 390 tp = intotcpcb(inp); 391 TCPDEBUG1(); 392 INP_HASH_WLOCK(&V_tcbinfo); 393 inp->inp_vflag &= ~INP_IPV4; 394 inp->inp_vflag |= INP_IPV6; 395 #ifdef INET 396 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 397 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 398 inp->inp_vflag |= INP_IPV4; 399 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 400 struct sockaddr_in sin; 401 402 in6_sin6_2_sin(&sin, sin6); 403 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 404 error = EAFNOSUPPORT; 405 INP_HASH_WUNLOCK(&V_tcbinfo); 406 goto out; 407 } 408 inp->inp_vflag |= INP_IPV4; 409 inp->inp_vflag &= ~INP_IPV6; 410 error = in_pcbbind(inp, (struct sockaddr *)&sin, 411 td->td_ucred); 412 INP_HASH_WUNLOCK(&V_tcbinfo); 413 goto out; 414 } 415 } 416 #endif 417 error = in6_pcbbind(inp, nam, td->td_ucred); 418 INP_HASH_WUNLOCK(&V_tcbinfo); 419 out: 420 if (error != 0) 421 inp->inp_vflag = vflagsav; 422 TCPDEBUG2(PRU_BIND); 423 TCP_PROBE2(debug__user, tp, PRU_BIND); 424 INP_WUNLOCK(inp); 425 return (error); 426 } 427 #endif /* INET6 */ 428 429 #ifdef INET 430 /* 431 * Prepare to accept connections. 432 */ 433 static int 434 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 435 { 436 int error = 0; 437 struct inpcb *inp; 438 struct tcpcb *tp = NULL; 439 440 TCPDEBUG0; 441 inp = sotoinpcb(so); 442 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 443 INP_WLOCK(inp); 444 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 445 error = EINVAL; 446 goto out; 447 } 448 tp = intotcpcb(inp); 449 TCPDEBUG1(); 450 SOCK_LOCK(so); 451 error = solisten_proto_check(so); 452 INP_HASH_WLOCK(&V_tcbinfo); 453 if (error == 0 && inp->inp_lport == 0) 454 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 455 INP_HASH_WUNLOCK(&V_tcbinfo); 456 if (error == 0) { 457 tcp_state_change(tp, TCPS_LISTEN); 458 solisten_proto(so, backlog); 459 #ifdef TCP_OFFLOAD 460 if ((so->so_options & SO_NO_OFFLOAD) == 0) 461 tcp_offload_listen_start(tp); 462 #endif 463 } 464 SOCK_UNLOCK(so); 465 466 if (IS_FASTOPEN(tp->t_flags)) 467 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 468 469 out: 470 TCPDEBUG2(PRU_LISTEN); 471 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 472 INP_WUNLOCK(inp); 473 return (error); 474 } 475 #endif /* INET */ 476 477 #ifdef INET6 478 static int 479 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 480 { 481 int error = 0; 482 struct inpcb *inp; 483 struct tcpcb *tp = NULL; 484 u_char vflagsav; 485 486 TCPDEBUG0; 487 inp = sotoinpcb(so); 488 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 489 INP_WLOCK(inp); 490 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 491 error = EINVAL; 492 goto out; 493 } 494 vflagsav = inp->inp_vflag; 495 tp = intotcpcb(inp); 496 TCPDEBUG1(); 497 SOCK_LOCK(so); 498 error = solisten_proto_check(so); 499 INP_HASH_WLOCK(&V_tcbinfo); 500 if (error == 0 && inp->inp_lport == 0) { 501 inp->inp_vflag &= ~INP_IPV4; 502 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 503 inp->inp_vflag |= INP_IPV4; 504 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 505 } 506 INP_HASH_WUNLOCK(&V_tcbinfo); 507 if (error == 0) { 508 tcp_state_change(tp, TCPS_LISTEN); 509 solisten_proto(so, backlog); 510 #ifdef TCP_OFFLOAD 511 if ((so->so_options & SO_NO_OFFLOAD) == 0) 512 tcp_offload_listen_start(tp); 513 #endif 514 } 515 SOCK_UNLOCK(so); 516 517 if (IS_FASTOPEN(tp->t_flags)) 518 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 519 520 if (error != 0) 521 inp->inp_vflag = vflagsav; 522 523 out: 524 TCPDEBUG2(PRU_LISTEN); 525 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 526 INP_WUNLOCK(inp); 527 return (error); 528 } 529 #endif /* INET6 */ 530 531 #ifdef INET 532 /* 533 * Initiate connection to peer. 534 * Create a template for use in transmissions on this connection. 535 * Enter SYN_SENT state, and mark socket as connecting. 536 * Start keep-alive timer, and seed output sequence space. 537 * Send initial segment on connection. 538 */ 539 static int 540 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 541 { 542 struct epoch_tracker et; 543 int error = 0; 544 struct inpcb *inp; 545 struct tcpcb *tp = NULL; 546 struct sockaddr_in *sinp; 547 548 sinp = (struct sockaddr_in *)nam; 549 if (nam->sa_family != AF_INET) 550 return (EAFNOSUPPORT); 551 if (nam->sa_len != sizeof (*sinp)) 552 return (EINVAL); 553 554 /* 555 * Must disallow TCP ``connections'' to multicast addresses. 556 */ 557 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 558 return (EAFNOSUPPORT); 559 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) 560 return (EACCES); 561 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 562 return (error); 563 564 TCPDEBUG0; 565 inp = sotoinpcb(so); 566 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 567 INP_WLOCK(inp); 568 if (inp->inp_flags & INP_TIMEWAIT) { 569 error = EADDRINUSE; 570 goto out; 571 } 572 if (inp->inp_flags & INP_DROPPED) { 573 error = ECONNREFUSED; 574 goto out; 575 } 576 tp = intotcpcb(inp); 577 TCPDEBUG1(); 578 NET_EPOCH_ENTER(et); 579 if ((error = tcp_connect(tp, nam, td)) != 0) 580 goto out_in_epoch; 581 #ifdef TCP_OFFLOAD 582 if (registered_toedevs > 0 && 583 (so->so_options & SO_NO_OFFLOAD) == 0 && 584 (error = tcp_offload_connect(so, nam)) == 0) 585 goto out_in_epoch; 586 #endif 587 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 588 error = tp->t_fb->tfb_tcp_output(tp); 589 out_in_epoch: 590 NET_EPOCH_EXIT(et); 591 out: 592 TCPDEBUG2(PRU_CONNECT); 593 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 594 INP_WUNLOCK(inp); 595 return (error); 596 } 597 #endif /* INET */ 598 599 #ifdef INET6 600 static int 601 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 602 { 603 struct epoch_tracker et; 604 int error = 0; 605 struct inpcb *inp; 606 struct tcpcb *tp = NULL; 607 struct sockaddr_in6 *sin6; 608 u_int8_t incflagsav; 609 u_char vflagsav; 610 611 TCPDEBUG0; 612 613 sin6 = (struct sockaddr_in6 *)nam; 614 if (nam->sa_family != AF_INET6) 615 return (EAFNOSUPPORT); 616 if (nam->sa_len != sizeof (*sin6)) 617 return (EINVAL); 618 619 /* 620 * Must disallow TCP ``connections'' to multicast addresses. 621 */ 622 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 623 return (EAFNOSUPPORT); 624 625 inp = sotoinpcb(so); 626 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 627 INP_WLOCK(inp); 628 vflagsav = inp->inp_vflag; 629 incflagsav = inp->inp_inc.inc_flags; 630 if (inp->inp_flags & INP_TIMEWAIT) { 631 error = EADDRINUSE; 632 goto out; 633 } 634 if (inp->inp_flags & INP_DROPPED) { 635 error = ECONNREFUSED; 636 goto out; 637 } 638 tp = intotcpcb(inp); 639 TCPDEBUG1(); 640 #ifdef INET 641 /* 642 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 643 * therefore probably require the hash lock, which isn't held here. 644 * Is this a significant problem? 645 */ 646 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 647 struct sockaddr_in sin; 648 649 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 650 error = EINVAL; 651 goto out; 652 } 653 if ((inp->inp_vflag & INP_IPV4) == 0) { 654 error = EAFNOSUPPORT; 655 goto out; 656 } 657 658 in6_sin6_2_sin(&sin, sin6); 659 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 660 error = EAFNOSUPPORT; 661 goto out; 662 } 663 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 664 error = EACCES; 665 goto out; 666 } 667 if ((error = prison_remote_ip4(td->td_ucred, 668 &sin.sin_addr)) != 0) 669 goto out; 670 inp->inp_vflag |= INP_IPV4; 671 inp->inp_vflag &= ~INP_IPV6; 672 NET_EPOCH_ENTER(et); 673 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 674 goto out_in_epoch; 675 #ifdef TCP_OFFLOAD 676 if (registered_toedevs > 0 && 677 (so->so_options & SO_NO_OFFLOAD) == 0 && 678 (error = tcp_offload_connect(so, nam)) == 0) 679 goto out_in_epoch; 680 #endif 681 error = tp->t_fb->tfb_tcp_output(tp); 682 goto out_in_epoch; 683 } else { 684 if ((inp->inp_vflag & INP_IPV6) == 0) { 685 error = EAFNOSUPPORT; 686 goto out; 687 } 688 } 689 #endif 690 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 691 goto out; 692 inp->inp_vflag &= ~INP_IPV4; 693 inp->inp_vflag |= INP_IPV6; 694 inp->inp_inc.inc_flags |= INC_ISIPV6; 695 if ((error = tcp6_connect(tp, nam, td)) != 0) 696 goto out; 697 #ifdef TCP_OFFLOAD 698 if (registered_toedevs > 0 && 699 (so->so_options & SO_NO_OFFLOAD) == 0 && 700 (error = tcp_offload_connect(so, nam)) == 0) 701 goto out; 702 #endif 703 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 704 NET_EPOCH_ENTER(et); 705 error = tp->t_fb->tfb_tcp_output(tp); 706 #ifdef INET 707 out_in_epoch: 708 #endif 709 NET_EPOCH_EXIT(et); 710 out: 711 /* 712 * If the implicit bind in the connect call fails, restore 713 * the flags we modified. 714 */ 715 if (error != 0 && inp->inp_lport == 0) { 716 inp->inp_vflag = vflagsav; 717 inp->inp_inc.inc_flags = incflagsav; 718 } 719 720 TCPDEBUG2(PRU_CONNECT); 721 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 722 INP_WUNLOCK(inp); 723 return (error); 724 } 725 #endif /* INET6 */ 726 727 /* 728 * Initiate disconnect from peer. 729 * If connection never passed embryonic stage, just drop; 730 * else if don't need to let data drain, then can just drop anyways, 731 * else have to begin TCP shutdown process: mark socket disconnecting, 732 * drain unread data, state switch to reflect user close, and 733 * send segment (e.g. FIN) to peer. Socket will be really disconnected 734 * when peer sends FIN and acks ours. 735 * 736 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 737 */ 738 static int 739 tcp_usr_disconnect(struct socket *so) 740 { 741 struct inpcb *inp; 742 struct tcpcb *tp = NULL; 743 struct epoch_tracker et; 744 int error = 0; 745 746 TCPDEBUG0; 747 NET_EPOCH_ENTER(et); 748 inp = sotoinpcb(so); 749 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 750 INP_WLOCK(inp); 751 if (inp->inp_flags & INP_TIMEWAIT) 752 goto out; 753 if (inp->inp_flags & INP_DROPPED) { 754 error = ECONNRESET; 755 goto out; 756 } 757 tp = intotcpcb(inp); 758 TCPDEBUG1(); 759 tcp_disconnect(tp); 760 out: 761 TCPDEBUG2(PRU_DISCONNECT); 762 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 763 INP_WUNLOCK(inp); 764 NET_EPOCH_EXIT(et); 765 return (error); 766 } 767 768 #ifdef INET 769 /* 770 * Accept a connection. Essentially all the work is done at higher levels; 771 * just return the address of the peer, storing through addr. 772 */ 773 static int 774 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 775 { 776 int error = 0; 777 struct inpcb *inp = NULL; 778 struct tcpcb *tp = NULL; 779 struct in_addr addr; 780 in_port_t port = 0; 781 TCPDEBUG0; 782 783 if (so->so_state & SS_ISDISCONNECTED) 784 return (ECONNABORTED); 785 786 inp = sotoinpcb(so); 787 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 788 INP_WLOCK(inp); 789 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 790 error = ECONNABORTED; 791 goto out; 792 } 793 tp = intotcpcb(inp); 794 TCPDEBUG1(); 795 796 /* 797 * We inline in_getpeeraddr and COMMON_END here, so that we can 798 * copy the data of interest and defer the malloc until after we 799 * release the lock. 800 */ 801 port = inp->inp_fport; 802 addr = inp->inp_faddr; 803 804 out: 805 TCPDEBUG2(PRU_ACCEPT); 806 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 807 INP_WUNLOCK(inp); 808 if (error == 0) 809 *nam = in_sockaddr(port, &addr); 810 return error; 811 } 812 #endif /* INET */ 813 814 #ifdef INET6 815 static int 816 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 817 { 818 struct inpcb *inp = NULL; 819 int error = 0; 820 struct tcpcb *tp = NULL; 821 struct in_addr addr; 822 struct in6_addr addr6; 823 struct epoch_tracker et; 824 in_port_t port = 0; 825 int v4 = 0; 826 TCPDEBUG0; 827 828 if (so->so_state & SS_ISDISCONNECTED) 829 return (ECONNABORTED); 830 831 inp = sotoinpcb(so); 832 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 833 NET_EPOCH_ENTER(et); 834 INP_WLOCK(inp); 835 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 836 error = ECONNABORTED; 837 goto out; 838 } 839 tp = intotcpcb(inp); 840 TCPDEBUG1(); 841 842 /* 843 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 844 * copy the data of interest and defer the malloc until after we 845 * release the lock. 846 */ 847 if (inp->inp_vflag & INP_IPV4) { 848 v4 = 1; 849 port = inp->inp_fport; 850 addr = inp->inp_faddr; 851 } else { 852 port = inp->inp_fport; 853 addr6 = inp->in6p_faddr; 854 } 855 856 out: 857 TCPDEBUG2(PRU_ACCEPT); 858 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 859 INP_WUNLOCK(inp); 860 NET_EPOCH_EXIT(et); 861 if (error == 0) { 862 if (v4) 863 *nam = in6_v4mapsin6_sockaddr(port, &addr); 864 else 865 *nam = in6_sockaddr(port, &addr6); 866 } 867 return error; 868 } 869 #endif /* INET6 */ 870 871 /* 872 * Mark the connection as being incapable of further output. 873 */ 874 static int 875 tcp_usr_shutdown(struct socket *so) 876 { 877 int error = 0; 878 struct inpcb *inp; 879 struct tcpcb *tp = NULL; 880 struct epoch_tracker et; 881 882 TCPDEBUG0; 883 NET_EPOCH_ENTER(et); 884 inp = sotoinpcb(so); 885 KASSERT(inp != NULL, ("inp == NULL")); 886 INP_WLOCK(inp); 887 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 888 error = ECONNRESET; 889 goto out; 890 } 891 tp = intotcpcb(inp); 892 TCPDEBUG1(); 893 socantsendmore(so); 894 tcp_usrclosed(tp); 895 if (!(inp->inp_flags & INP_DROPPED)) 896 error = tp->t_fb->tfb_tcp_output(tp); 897 898 out: 899 TCPDEBUG2(PRU_SHUTDOWN); 900 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 901 INP_WUNLOCK(inp); 902 NET_EPOCH_EXIT(et); 903 904 return (error); 905 } 906 907 /* 908 * After a receive, possibly send window update to peer. 909 */ 910 static int 911 tcp_usr_rcvd(struct socket *so, int flags) 912 { 913 struct epoch_tracker et; 914 struct inpcb *inp; 915 struct tcpcb *tp = NULL; 916 int error = 0; 917 918 TCPDEBUG0; 919 inp = sotoinpcb(so); 920 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 921 INP_WLOCK(inp); 922 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 923 error = ECONNRESET; 924 goto out; 925 } 926 tp = intotcpcb(inp); 927 TCPDEBUG1(); 928 /* 929 * For passively-created TFO connections, don't attempt a window 930 * update while still in SYN_RECEIVED as this may trigger an early 931 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 932 * application response data, or failing that, when the DELACK timer 933 * expires. 934 */ 935 if (IS_FASTOPEN(tp->t_flags) && 936 (tp->t_state == TCPS_SYN_RECEIVED)) 937 goto out; 938 NET_EPOCH_ENTER(et); 939 #ifdef TCP_OFFLOAD 940 if (tp->t_flags & TF_TOE) 941 tcp_offload_rcvd(tp); 942 else 943 #endif 944 tp->t_fb->tfb_tcp_output(tp); 945 NET_EPOCH_EXIT(et); 946 out: 947 TCPDEBUG2(PRU_RCVD); 948 TCP_PROBE2(debug__user, tp, PRU_RCVD); 949 INP_WUNLOCK(inp); 950 return (error); 951 } 952 953 /* 954 * Do a send by putting data in output queue and updating urgent 955 * marker if URG set. Possibly send more data. Unlike the other 956 * pru_*() routines, the mbuf chains are our responsibility. We 957 * must either enqueue them or free them. The other pru_* routines 958 * generally are caller-frees. 959 */ 960 static int 961 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 962 struct sockaddr *nam, struct mbuf *control, struct thread *td) 963 { 964 struct epoch_tracker et; 965 int error = 0; 966 struct inpcb *inp; 967 struct tcpcb *tp = NULL; 968 #ifdef INET 969 #ifdef INET6 970 struct sockaddr_in sin; 971 #endif 972 struct sockaddr_in *sinp; 973 #endif 974 #ifdef INET6 975 int isipv6; 976 #endif 977 u_int8_t incflagsav; 978 u_char vflagsav; 979 bool restoreflags; 980 TCPDEBUG0; 981 982 /* 983 * We require the pcbinfo "read lock" if we will close the socket 984 * as part of this call. 985 */ 986 NET_EPOCH_ENTER(et); 987 inp = sotoinpcb(so); 988 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 989 INP_WLOCK(inp); 990 vflagsav = inp->inp_vflag; 991 incflagsav = inp->inp_inc.inc_flags; 992 restoreflags = false; 993 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 994 if (control) 995 m_freem(control); 996 997 /* 998 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible 999 * for freeing memory. 1000 */ 1001 if ((flags & PRUS_NOTREADY) == 0) 1002 m_freem(m); 1003 error = ECONNRESET; 1004 goto out; 1005 } 1006 if (control != NULL) { 1007 /* TCP doesn't do control messages (rights, creds, etc) */ 1008 if (control->m_len) { 1009 m_freem(control); 1010 m_freem(m); 1011 error = EINVAL; 1012 goto out; 1013 } 1014 m_freem(control); /* empty control, just free it */ 1015 control = NULL; 1016 } 1017 tp = intotcpcb(inp); 1018 if (flags & PRUS_OOB) { 1019 if ((error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) { 1020 if ((flags & PRUS_NOTREADY) == 0) 1021 m_freem(m); 1022 goto out; 1023 } 1024 } 1025 TCPDEBUG1(); 1026 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 1027 switch (nam->sa_family) { 1028 #ifdef INET 1029 case AF_INET: 1030 sinp = (struct sockaddr_in *)nam; 1031 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 1032 m_freem(m); 1033 error = EINVAL; 1034 goto out; 1035 } 1036 if ((inp->inp_vflag & INP_IPV6) != 0) { 1037 m_freem(m); 1038 error = EAFNOSUPPORT; 1039 goto out; 1040 } 1041 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 1042 m_freem(m); 1043 error = EAFNOSUPPORT; 1044 goto out; 1045 } 1046 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 1047 m_freem(m); 1048 error = EACCES; 1049 goto out; 1050 } 1051 if ((error = prison_remote_ip4(td->td_ucred, 1052 &sinp->sin_addr))) { 1053 m_freem(m); 1054 goto out; 1055 } 1056 #ifdef INET6 1057 isipv6 = 0; 1058 #endif 1059 break; 1060 #endif /* INET */ 1061 #ifdef INET6 1062 case AF_INET6: 1063 { 1064 struct sockaddr_in6 *sin6; 1065 1066 sin6 = (struct sockaddr_in6 *)nam; 1067 if (sin6->sin6_len != sizeof(*sin6)) { 1068 m_freem(m); 1069 error = EINVAL; 1070 goto out; 1071 } 1072 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1073 m_freem(m); 1074 error = EAFNOSUPPORT; 1075 goto out; 1076 } 1077 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1078 m_freem(m); 1079 error = EAFNOSUPPORT; 1080 goto out; 1081 } 1082 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1083 #ifdef INET 1084 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1085 error = EINVAL; 1086 m_freem(m); 1087 goto out; 1088 } 1089 if ((inp->inp_vflag & INP_IPV4) == 0) { 1090 error = EAFNOSUPPORT; 1091 m_freem(m); 1092 goto out; 1093 } 1094 restoreflags = true; 1095 inp->inp_vflag &= ~INP_IPV6; 1096 sinp = &sin; 1097 in6_sin6_2_sin(sinp, sin6); 1098 if (IN_MULTICAST( 1099 ntohl(sinp->sin_addr.s_addr))) { 1100 error = EAFNOSUPPORT; 1101 m_freem(m); 1102 goto out; 1103 } 1104 if ((error = prison_remote_ip4(td->td_ucred, 1105 &sinp->sin_addr))) { 1106 m_freem(m); 1107 goto out; 1108 } 1109 isipv6 = 0; 1110 #else /* !INET */ 1111 error = EAFNOSUPPORT; 1112 m_freem(m); 1113 goto out; 1114 #endif /* INET */ 1115 } else { 1116 if ((inp->inp_vflag & INP_IPV6) == 0) { 1117 m_freem(m); 1118 error = EAFNOSUPPORT; 1119 goto out; 1120 } 1121 restoreflags = true; 1122 inp->inp_vflag &= ~INP_IPV4; 1123 inp->inp_inc.inc_flags |= INC_ISIPV6; 1124 if ((error = prison_remote_ip6(td->td_ucred, 1125 &sin6->sin6_addr))) { 1126 m_freem(m); 1127 goto out; 1128 } 1129 isipv6 = 1; 1130 } 1131 break; 1132 } 1133 #endif /* INET6 */ 1134 default: 1135 m_freem(m); 1136 error = EAFNOSUPPORT; 1137 goto out; 1138 } 1139 } 1140 if (!(flags & PRUS_OOB)) { 1141 sbappendstream(&so->so_snd, m, flags); 1142 if (nam && tp->t_state < TCPS_SYN_SENT) { 1143 /* 1144 * Do implied connect if not yet connected, 1145 * initialize window to default value, and 1146 * initialize maxseg using peer's cached MSS. 1147 */ 1148 #ifdef INET6 1149 if (isipv6) 1150 error = tcp6_connect(tp, nam, td); 1151 #endif /* INET6 */ 1152 #if defined(INET6) && defined(INET) 1153 else 1154 #endif 1155 #ifdef INET 1156 error = tcp_connect(tp, 1157 (struct sockaddr *)sinp, td); 1158 #endif 1159 /* 1160 * The bind operation in tcp_connect succeeded. We 1161 * no longer want to restore the flags if later 1162 * operations fail. 1163 */ 1164 if (error == 0 || inp->inp_lport != 0) 1165 restoreflags = false; 1166 1167 if (error) 1168 goto out; 1169 if (IS_FASTOPEN(tp->t_flags)) 1170 tcp_fastopen_connect(tp); 1171 else { 1172 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1173 tcp_mss(tp, -1); 1174 } 1175 } 1176 if (flags & PRUS_EOF) { 1177 /* 1178 * Close the send side of the connection after 1179 * the data is sent. 1180 */ 1181 socantsendmore(so); 1182 tcp_usrclosed(tp); 1183 } 1184 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1185 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1186 (tp->t_fbyte_out == 0) && 1187 (so->so_snd.sb_ccc > 0)) { 1188 tp->t_fbyte_out = ticks; 1189 if (tp->t_fbyte_out == 0) 1190 tp->t_fbyte_out = 1; 1191 if (tp->t_fbyte_out && tp->t_fbyte_in) 1192 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1193 } 1194 if (!(inp->inp_flags & INP_DROPPED) && 1195 !(flags & PRUS_NOTREADY)) { 1196 if (flags & PRUS_MORETOCOME) 1197 tp->t_flags |= TF_MORETOCOME; 1198 error = tp->t_fb->tfb_tcp_output(tp); 1199 if (flags & PRUS_MORETOCOME) 1200 tp->t_flags &= ~TF_MORETOCOME; 1201 } 1202 } else { 1203 /* 1204 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1205 */ 1206 SOCKBUF_LOCK(&so->so_snd); 1207 if (sbspace(&so->so_snd) < -512) { 1208 SOCKBUF_UNLOCK(&so->so_snd); 1209 m_freem(m); 1210 error = ENOBUFS; 1211 goto out; 1212 } 1213 /* 1214 * According to RFC961 (Assigned Protocols), 1215 * the urgent pointer points to the last octet 1216 * of urgent data. We continue, however, 1217 * to consider it to indicate the first octet 1218 * of data past the urgent section. 1219 * Otherwise, snd_up should be one lower. 1220 */ 1221 sbappendstream_locked(&so->so_snd, m, flags); 1222 SOCKBUF_UNLOCK(&so->so_snd); 1223 if (nam && tp->t_state < TCPS_SYN_SENT) { 1224 /* 1225 * Do implied connect if not yet connected, 1226 * initialize window to default value, and 1227 * initialize maxseg using peer's cached MSS. 1228 */ 1229 1230 /* 1231 * Not going to contemplate SYN|URG 1232 */ 1233 if (IS_FASTOPEN(tp->t_flags)) 1234 tp->t_flags &= ~TF_FASTOPEN; 1235 #ifdef INET6 1236 if (isipv6) 1237 error = tcp6_connect(tp, nam, td); 1238 #endif /* INET6 */ 1239 #if defined(INET6) && defined(INET) 1240 else 1241 #endif 1242 #ifdef INET 1243 error = tcp_connect(tp, 1244 (struct sockaddr *)sinp, td); 1245 #endif 1246 /* 1247 * The bind operation in tcp_connect succeeded. We 1248 * no longer want to restore the flags if later 1249 * operations fail. 1250 */ 1251 if (error == 0 || inp->inp_lport != 0) 1252 restoreflags = false; 1253 1254 if (error) 1255 goto out; 1256 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1257 tcp_mss(tp, -1); 1258 } 1259 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1260 if (!(flags & PRUS_NOTREADY)) { 1261 tp->t_flags |= TF_FORCEDATA; 1262 error = tp->t_fb->tfb_tcp_output(tp); 1263 tp->t_flags &= ~TF_FORCEDATA; 1264 } 1265 } 1266 TCP_LOG_EVENT(tp, NULL, 1267 &inp->inp_socket->so_rcv, 1268 &inp->inp_socket->so_snd, 1269 TCP_LOG_USERSEND, error, 1270 0, NULL, false); 1271 out: 1272 /* 1273 * If the request was unsuccessful and we changed flags, 1274 * restore the original flags. 1275 */ 1276 if (error != 0 && restoreflags) { 1277 inp->inp_vflag = vflagsav; 1278 inp->inp_inc.inc_flags = incflagsav; 1279 } 1280 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1281 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1282 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1283 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1284 INP_WUNLOCK(inp); 1285 NET_EPOCH_EXIT(et); 1286 return (error); 1287 } 1288 1289 static int 1290 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1291 { 1292 struct epoch_tracker et; 1293 struct inpcb *inp; 1294 struct tcpcb *tp; 1295 int error; 1296 1297 inp = sotoinpcb(so); 1298 INP_WLOCK(inp); 1299 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1300 INP_WUNLOCK(inp); 1301 mb_free_notready(m, count); 1302 return (ECONNRESET); 1303 } 1304 tp = intotcpcb(inp); 1305 1306 SOCKBUF_LOCK(&so->so_snd); 1307 error = sbready(&so->so_snd, m, count); 1308 SOCKBUF_UNLOCK(&so->so_snd); 1309 if (error == 0) { 1310 NET_EPOCH_ENTER(et); 1311 error = tp->t_fb->tfb_tcp_output(tp); 1312 NET_EPOCH_EXIT(et); 1313 } 1314 INP_WUNLOCK(inp); 1315 1316 return (error); 1317 } 1318 1319 /* 1320 * Abort the TCP. Drop the connection abruptly. 1321 */ 1322 static void 1323 tcp_usr_abort(struct socket *so) 1324 { 1325 struct inpcb *inp; 1326 struct tcpcb *tp = NULL; 1327 struct epoch_tracker et; 1328 TCPDEBUG0; 1329 1330 inp = sotoinpcb(so); 1331 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1332 1333 NET_EPOCH_ENTER(et); 1334 INP_WLOCK(inp); 1335 KASSERT(inp->inp_socket != NULL, 1336 ("tcp_usr_abort: inp_socket == NULL")); 1337 1338 /* 1339 * If we still have full TCP state, and we're not dropped, drop. 1340 */ 1341 if (!(inp->inp_flags & INP_TIMEWAIT) && 1342 !(inp->inp_flags & INP_DROPPED)) { 1343 tp = intotcpcb(inp); 1344 TCPDEBUG1(); 1345 tp = tcp_drop(tp, ECONNABORTED); 1346 if (tp == NULL) 1347 goto dropped; 1348 TCPDEBUG2(PRU_ABORT); 1349 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1350 } 1351 if (!(inp->inp_flags & INP_DROPPED)) { 1352 SOCK_LOCK(so); 1353 so->so_state |= SS_PROTOREF; 1354 SOCK_UNLOCK(so); 1355 inp->inp_flags |= INP_SOCKREF; 1356 } 1357 INP_WUNLOCK(inp); 1358 dropped: 1359 NET_EPOCH_EXIT(et); 1360 } 1361 1362 /* 1363 * TCP socket is closed. Start friendly disconnect. 1364 */ 1365 static void 1366 tcp_usr_close(struct socket *so) 1367 { 1368 struct inpcb *inp; 1369 struct tcpcb *tp = NULL; 1370 struct epoch_tracker et; 1371 TCPDEBUG0; 1372 1373 inp = sotoinpcb(so); 1374 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1375 1376 NET_EPOCH_ENTER(et); 1377 INP_WLOCK(inp); 1378 KASSERT(inp->inp_socket != NULL, 1379 ("tcp_usr_close: inp_socket == NULL")); 1380 1381 /* 1382 * If we still have full TCP state, and we're not dropped, initiate 1383 * a disconnect. 1384 */ 1385 if (!(inp->inp_flags & INP_TIMEWAIT) && 1386 !(inp->inp_flags & INP_DROPPED)) { 1387 tp = intotcpcb(inp); 1388 TCPDEBUG1(); 1389 tcp_disconnect(tp); 1390 TCPDEBUG2(PRU_CLOSE); 1391 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1392 } 1393 if (!(inp->inp_flags & INP_DROPPED)) { 1394 SOCK_LOCK(so); 1395 so->so_state |= SS_PROTOREF; 1396 SOCK_UNLOCK(so); 1397 inp->inp_flags |= INP_SOCKREF; 1398 } 1399 INP_WUNLOCK(inp); 1400 NET_EPOCH_EXIT(et); 1401 } 1402 1403 static int 1404 tcp_pru_options_support(struct tcpcb *tp, int flags) 1405 { 1406 /* 1407 * If the specific TCP stack has a pru_options 1408 * specified then it does not always support 1409 * all the PRU_XX options and we must ask it. 1410 * If the function is not specified then all 1411 * of the PRU_XX options are supported. 1412 */ 1413 int ret = 0; 1414 1415 if (tp->t_fb->tfb_pru_options) { 1416 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1417 } 1418 return (ret); 1419 } 1420 1421 /* 1422 * Receive out-of-band data. 1423 */ 1424 static int 1425 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1426 { 1427 int error = 0; 1428 struct inpcb *inp; 1429 struct tcpcb *tp = NULL; 1430 1431 TCPDEBUG0; 1432 inp = sotoinpcb(so); 1433 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1434 INP_WLOCK(inp); 1435 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1436 error = ECONNRESET; 1437 goto out; 1438 } 1439 tp = intotcpcb(inp); 1440 error = tcp_pru_options_support(tp, PRUS_OOB); 1441 if (error) { 1442 goto out; 1443 } 1444 TCPDEBUG1(); 1445 if ((so->so_oobmark == 0 && 1446 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1447 so->so_options & SO_OOBINLINE || 1448 tp->t_oobflags & TCPOOB_HADDATA) { 1449 error = EINVAL; 1450 goto out; 1451 } 1452 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1453 error = EWOULDBLOCK; 1454 goto out; 1455 } 1456 m->m_len = 1; 1457 *mtod(m, caddr_t) = tp->t_iobc; 1458 if ((flags & MSG_PEEK) == 0) 1459 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1460 1461 out: 1462 TCPDEBUG2(PRU_RCVOOB); 1463 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1464 INP_WUNLOCK(inp); 1465 return (error); 1466 } 1467 1468 #ifdef INET 1469 struct pr_usrreqs tcp_usrreqs = { 1470 .pru_abort = tcp_usr_abort, 1471 .pru_accept = tcp_usr_accept, 1472 .pru_attach = tcp_usr_attach, 1473 .pru_bind = tcp_usr_bind, 1474 .pru_connect = tcp_usr_connect, 1475 .pru_control = in_control, 1476 .pru_detach = tcp_usr_detach, 1477 .pru_disconnect = tcp_usr_disconnect, 1478 .pru_listen = tcp_usr_listen, 1479 .pru_peeraddr = in_getpeeraddr, 1480 .pru_rcvd = tcp_usr_rcvd, 1481 .pru_rcvoob = tcp_usr_rcvoob, 1482 .pru_send = tcp_usr_send, 1483 .pru_ready = tcp_usr_ready, 1484 .pru_shutdown = tcp_usr_shutdown, 1485 .pru_sockaddr = in_getsockaddr, 1486 .pru_sosetlabel = in_pcbsosetlabel, 1487 .pru_close = tcp_usr_close, 1488 }; 1489 #endif /* INET */ 1490 1491 #ifdef INET6 1492 struct pr_usrreqs tcp6_usrreqs = { 1493 .pru_abort = tcp_usr_abort, 1494 .pru_accept = tcp6_usr_accept, 1495 .pru_attach = tcp_usr_attach, 1496 .pru_bind = tcp6_usr_bind, 1497 .pru_connect = tcp6_usr_connect, 1498 .pru_control = in6_control, 1499 .pru_detach = tcp_usr_detach, 1500 .pru_disconnect = tcp_usr_disconnect, 1501 .pru_listen = tcp6_usr_listen, 1502 .pru_peeraddr = in6_mapped_peeraddr, 1503 .pru_rcvd = tcp_usr_rcvd, 1504 .pru_rcvoob = tcp_usr_rcvoob, 1505 .pru_send = tcp_usr_send, 1506 .pru_ready = tcp_usr_ready, 1507 .pru_shutdown = tcp_usr_shutdown, 1508 .pru_sockaddr = in6_mapped_sockaddr, 1509 .pru_sosetlabel = in_pcbsosetlabel, 1510 .pru_close = tcp_usr_close, 1511 }; 1512 #endif /* INET6 */ 1513 1514 #ifdef INET 1515 /* 1516 * Common subroutine to open a TCP connection to remote host specified 1517 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1518 * port number if needed. Call in_pcbconnect_setup to do the routing and 1519 * to choose a local host address (interface). If there is an existing 1520 * incarnation of the same connection in TIME-WAIT state and if the remote 1521 * host was sending CC options and if the connection duration was < MSL, then 1522 * truncate the previous TIME-WAIT state and proceed. 1523 * Initialize connection parameters and enter SYN-SENT state. 1524 */ 1525 static int 1526 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1527 { 1528 struct inpcb *inp = tp->t_inpcb, *oinp; 1529 struct socket *so = inp->inp_socket; 1530 struct in_addr laddr; 1531 u_short lport; 1532 int error; 1533 1534 NET_EPOCH_ASSERT(); 1535 INP_WLOCK_ASSERT(inp); 1536 INP_HASH_WLOCK(&V_tcbinfo); 1537 1538 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1539 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1540 if (error) 1541 goto out; 1542 } 1543 1544 /* 1545 * Cannot simply call in_pcbconnect, because there might be an 1546 * earlier incarnation of this same connection still in 1547 * TIME_WAIT state, creating an ADDRINUSE error. 1548 */ 1549 laddr = inp->inp_laddr; 1550 lport = inp->inp_lport; 1551 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1552 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1553 if (error && oinp == NULL) 1554 goto out; 1555 if (oinp) { 1556 error = EADDRINUSE; 1557 goto out; 1558 } 1559 /* Handle initial bind if it hadn't been done in advance. */ 1560 if (inp->inp_lport == 0) { 1561 inp->inp_lport = lport; 1562 if (in_pcbinshash(inp) != 0) { 1563 inp->inp_lport = 0; 1564 error = EAGAIN; 1565 goto out; 1566 } 1567 } 1568 inp->inp_laddr = laddr; 1569 in_pcbrehash(inp); 1570 INP_HASH_WUNLOCK(&V_tcbinfo); 1571 1572 /* 1573 * Compute window scaling to request: 1574 * Scale to fit into sweet spot. See tcp_syncache.c. 1575 * XXX: This should move to tcp_output(). 1576 */ 1577 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1578 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1579 tp->request_r_scale++; 1580 1581 soisconnecting(so); 1582 TCPSTAT_INC(tcps_connattempt); 1583 tcp_state_change(tp, TCPS_SYN_SENT); 1584 tp->iss = tcp_new_isn(&inp->inp_inc); 1585 if (tp->t_flags & TF_REQ_TSTMP) 1586 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1587 tcp_sendseqinit(tp); 1588 1589 return 0; 1590 1591 out: 1592 INP_HASH_WUNLOCK(&V_tcbinfo); 1593 return (error); 1594 } 1595 #endif /* INET */ 1596 1597 #ifdef INET6 1598 static int 1599 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1600 { 1601 struct inpcb *inp = tp->t_inpcb; 1602 int error; 1603 1604 INP_WLOCK_ASSERT(inp); 1605 INP_HASH_WLOCK(&V_tcbinfo); 1606 1607 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1608 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1609 if (error) 1610 goto out; 1611 } 1612 error = in6_pcbconnect(inp, nam, td->td_ucred); 1613 if (error != 0) 1614 goto out; 1615 INP_HASH_WUNLOCK(&V_tcbinfo); 1616 1617 /* Compute window scaling to request. */ 1618 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1619 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1620 tp->request_r_scale++; 1621 1622 soisconnecting(inp->inp_socket); 1623 TCPSTAT_INC(tcps_connattempt); 1624 tcp_state_change(tp, TCPS_SYN_SENT); 1625 tp->iss = tcp_new_isn(&inp->inp_inc); 1626 if (tp->t_flags & TF_REQ_TSTMP) 1627 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1628 tcp_sendseqinit(tp); 1629 1630 return 0; 1631 1632 out: 1633 INP_HASH_WUNLOCK(&V_tcbinfo); 1634 return error; 1635 } 1636 #endif /* INET6 */ 1637 1638 /* 1639 * Export TCP internal state information via a struct tcp_info, based on the 1640 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1641 * (TCP state machine, etc). We export all information using FreeBSD-native 1642 * constants -- for example, the numeric values for tcpi_state will differ 1643 * from Linux. 1644 */ 1645 static void 1646 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1647 { 1648 1649 INP_WLOCK_ASSERT(tp->t_inpcb); 1650 bzero(ti, sizeof(*ti)); 1651 1652 ti->tcpi_state = tp->t_state; 1653 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1654 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1655 if (tp->t_flags & TF_SACK_PERMIT) 1656 ti->tcpi_options |= TCPI_OPT_SACK; 1657 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1658 ti->tcpi_options |= TCPI_OPT_WSCALE; 1659 ti->tcpi_snd_wscale = tp->snd_scale; 1660 ti->tcpi_rcv_wscale = tp->rcv_scale; 1661 } 1662 if (tp->t_flags2 & TF2_ECN_PERMIT) 1663 ti->tcpi_options |= TCPI_OPT_ECN; 1664 1665 ti->tcpi_rto = tp->t_rxtcur * tick; 1666 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1667 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1668 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1669 1670 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1671 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1672 1673 /* 1674 * FreeBSD-specific extension fields for tcp_info. 1675 */ 1676 ti->tcpi_rcv_space = tp->rcv_wnd; 1677 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1678 ti->tcpi_snd_wnd = tp->snd_wnd; 1679 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1680 ti->tcpi_snd_nxt = tp->snd_nxt; 1681 ti->tcpi_snd_mss = tp->t_maxseg; 1682 ti->tcpi_rcv_mss = tp->t_maxseg; 1683 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1684 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1685 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1686 #ifdef TCP_OFFLOAD 1687 if (tp->t_flags & TF_TOE) { 1688 ti->tcpi_options |= TCPI_OPT_TOE; 1689 tcp_offload_tcp_info(tp, ti); 1690 } 1691 #endif 1692 } 1693 1694 /* 1695 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1696 * socket option arguments. When it re-acquires the lock after the copy, it 1697 * has to revalidate that the connection is still valid for the socket 1698 * option. 1699 */ 1700 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1701 INP_WLOCK(inp); \ 1702 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1703 INP_WUNLOCK(inp); \ 1704 cleanup; \ 1705 return (ECONNRESET); \ 1706 } \ 1707 tp = intotcpcb(inp); \ 1708 } while(0) 1709 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1710 1711 int 1712 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1713 { 1714 int error; 1715 struct inpcb *inp; 1716 struct tcpcb *tp; 1717 struct tcp_function_block *blk; 1718 struct tcp_function_set fsn; 1719 1720 error = 0; 1721 inp = sotoinpcb(so); 1722 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1723 if (sopt->sopt_level != IPPROTO_TCP) { 1724 #ifdef INET6 1725 if (inp->inp_vflag & INP_IPV6PROTO) { 1726 error = ip6_ctloutput(so, sopt); 1727 /* 1728 * In case of the IPV6_USE_MIN_MTU socket option, 1729 * the INC_IPV6MINMTU flag to announce a corresponding 1730 * MSS during the initial handshake. 1731 * If the TCP connection is not in the front states, 1732 * just reduce the MSS being used. 1733 * This avoids the sending of TCP segments which will 1734 * be fragmented at the IPv6 layer. 1735 */ 1736 if ((error == 0) && 1737 (sopt->sopt_dir == SOPT_SET) && 1738 (sopt->sopt_level == IPPROTO_IPV6) && 1739 (sopt->sopt_name == IPV6_USE_MIN_MTU)) { 1740 INP_WLOCK(inp); 1741 if ((inp->inp_flags & 1742 (INP_TIMEWAIT | INP_DROPPED))) { 1743 INP_WUNLOCK(inp); 1744 return (ECONNRESET); 1745 } 1746 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 1747 tp = intotcpcb(inp); 1748 if ((tp->t_state >= TCPS_SYN_SENT) && 1749 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 1750 struct ip6_pktopts *opt; 1751 1752 opt = inp->in6p_outputopts; 1753 if ((opt != NULL) && 1754 (opt->ip6po_minmtu == 1755 IP6PO_MINMTU_ALL)) { 1756 if (tp->t_maxseg > TCP6_MSS) { 1757 tp->t_maxseg = TCP6_MSS; 1758 } 1759 } 1760 } 1761 INP_WUNLOCK(inp); 1762 } 1763 } 1764 #endif /* INET6 */ 1765 #if defined(INET6) && defined(INET) 1766 else 1767 #endif 1768 #ifdef INET 1769 { 1770 error = ip_ctloutput(so, sopt); 1771 } 1772 #endif 1773 return (error); 1774 } 1775 INP_WLOCK(inp); 1776 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1777 INP_WUNLOCK(inp); 1778 return (ECONNRESET); 1779 } 1780 tp = intotcpcb(inp); 1781 /* 1782 * Protect the TCP option TCP_FUNCTION_BLK so 1783 * that a sub-function can *never* overwrite this. 1784 */ 1785 if ((sopt->sopt_dir == SOPT_SET) && 1786 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1787 INP_WUNLOCK(inp); 1788 error = sooptcopyin(sopt, &fsn, sizeof fsn, 1789 sizeof fsn); 1790 if (error) 1791 return (error); 1792 INP_WLOCK_RECHECK(inp); 1793 blk = find_and_ref_tcp_functions(&fsn); 1794 if (blk == NULL) { 1795 INP_WUNLOCK(inp); 1796 return (ENOENT); 1797 } 1798 if (tp->t_fb == blk) { 1799 /* You already have this */ 1800 refcount_release(&blk->tfb_refcnt); 1801 INP_WUNLOCK(inp); 1802 return (0); 1803 } 1804 if (tp->t_state != TCPS_CLOSED) { 1805 /* 1806 * The user has advanced the state 1807 * past the initial point, we may not 1808 * be able to switch. 1809 */ 1810 if (blk->tfb_tcp_handoff_ok != NULL) { 1811 /* 1812 * Does the stack provide a 1813 * query mechanism, if so it may 1814 * still be possible? 1815 */ 1816 error = (*blk->tfb_tcp_handoff_ok)(tp); 1817 } else 1818 error = EINVAL; 1819 if (error) { 1820 refcount_release(&blk->tfb_refcnt); 1821 INP_WUNLOCK(inp); 1822 return(error); 1823 } 1824 } 1825 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1826 refcount_release(&blk->tfb_refcnt); 1827 INP_WUNLOCK(inp); 1828 return (ENOENT); 1829 } 1830 /* 1831 * Release the old refcnt, the 1832 * lookup acquired a ref on the 1833 * new one already. 1834 */ 1835 if (tp->t_fb->tfb_tcp_fb_fini) { 1836 /* 1837 * Tell the stack to cleanup with 0 i.e. 1838 * the tcb is not going away. 1839 */ 1840 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1841 } 1842 #ifdef TCPHPTS 1843 /* Assure that we are not on any hpts */ 1844 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL); 1845 #endif 1846 if (blk->tfb_tcp_fb_init) { 1847 error = (*blk->tfb_tcp_fb_init)(tp); 1848 if (error) { 1849 refcount_release(&blk->tfb_refcnt); 1850 if (tp->t_fb->tfb_tcp_fb_init) { 1851 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1852 /* Fall back failed, drop the connection */ 1853 INP_WUNLOCK(inp); 1854 soabort(so); 1855 return(error); 1856 } 1857 } 1858 goto err_out; 1859 } 1860 } 1861 refcount_release(&tp->t_fb->tfb_refcnt); 1862 tp->t_fb = blk; 1863 #ifdef TCP_OFFLOAD 1864 if (tp->t_flags & TF_TOE) { 1865 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1866 sopt->sopt_name); 1867 } 1868 #endif 1869 err_out: 1870 INP_WUNLOCK(inp); 1871 return (error); 1872 } else if ((sopt->sopt_dir == SOPT_GET) && 1873 (sopt->sopt_name == TCP_FUNCTION_BLK)) { 1874 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name, 1875 TCP_FUNCTION_NAME_LEN_MAX); 1876 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1877 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1878 INP_WUNLOCK(inp); 1879 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1880 return (error); 1881 } 1882 /* Pass in the INP locked, called must unlock it */ 1883 return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp)); 1884 } 1885 1886 /* 1887 * If this assert becomes untrue, we need to change the size of the buf 1888 * variable in tcp_default_ctloutput(). 1889 */ 1890 #ifdef CTASSERT 1891 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1892 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1893 #endif 1894 1895 #ifdef KERN_TLS 1896 static int 1897 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1898 { 1899 struct tls_enable_v0 tls_v0; 1900 int error; 1901 1902 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1903 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1904 sizeof(tls_v0)); 1905 if (error) 1906 return (error); 1907 memset(tls, 0, sizeof(*tls)); 1908 tls->cipher_key = tls_v0.cipher_key; 1909 tls->iv = tls_v0.iv; 1910 tls->auth_key = tls_v0.auth_key; 1911 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1912 tls->cipher_key_len = tls_v0.cipher_key_len; 1913 tls->iv_len = tls_v0.iv_len; 1914 tls->auth_algorithm = tls_v0.auth_algorithm; 1915 tls->auth_key_len = tls_v0.auth_key_len; 1916 tls->flags = tls_v0.flags; 1917 tls->tls_vmajor = tls_v0.tls_vmajor; 1918 tls->tls_vminor = tls_v0.tls_vminor; 1919 return (0); 1920 } 1921 1922 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1923 } 1924 #endif 1925 1926 int 1927 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 1928 { 1929 int error, opt, optval; 1930 u_int ui; 1931 struct tcp_info ti; 1932 #ifdef KERN_TLS 1933 struct tls_enable tls; 1934 #endif 1935 struct cc_algo *algo; 1936 char *pbuf, buf[TCP_LOG_ID_LEN]; 1937 #ifdef STATS 1938 struct statsblob *sbp; 1939 #endif 1940 size_t len; 1941 1942 /* 1943 * For TCP_CCALGOOPT forward the control to CC module, for both 1944 * SOPT_SET and SOPT_GET. 1945 */ 1946 switch (sopt->sopt_name) { 1947 case TCP_CCALGOOPT: 1948 INP_WUNLOCK(inp); 1949 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 1950 return (EINVAL); 1951 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 1952 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 1953 sopt->sopt_valsize); 1954 if (error) { 1955 free(pbuf, M_TEMP); 1956 return (error); 1957 } 1958 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 1959 if (CC_ALGO(tp)->ctl_output != NULL) 1960 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 1961 else 1962 error = ENOENT; 1963 INP_WUNLOCK(inp); 1964 if (error == 0 && sopt->sopt_dir == SOPT_GET) 1965 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 1966 free(pbuf, M_TEMP); 1967 return (error); 1968 } 1969 1970 switch (sopt->sopt_dir) { 1971 case SOPT_SET: 1972 switch (sopt->sopt_name) { 1973 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1974 case TCP_MD5SIG: 1975 if (!TCPMD5_ENABLED()) { 1976 INP_WUNLOCK(inp); 1977 return (ENOPROTOOPT); 1978 } 1979 error = TCPMD5_PCBCTL(inp, sopt); 1980 if (error) 1981 return (error); 1982 goto unlock_and_done; 1983 #endif /* IPSEC */ 1984 1985 case TCP_NODELAY: 1986 case TCP_NOOPT: 1987 case TCP_LRD: 1988 INP_WUNLOCK(inp); 1989 error = sooptcopyin(sopt, &optval, sizeof optval, 1990 sizeof optval); 1991 if (error) 1992 return (error); 1993 1994 INP_WLOCK_RECHECK(inp); 1995 switch (sopt->sopt_name) { 1996 case TCP_NODELAY: 1997 opt = TF_NODELAY; 1998 break; 1999 case TCP_NOOPT: 2000 opt = TF_NOOPT; 2001 break; 2002 case TCP_LRD: 2003 opt = TF_LRD; 2004 break; 2005 default: 2006 opt = 0; /* dead code to fool gcc */ 2007 break; 2008 } 2009 2010 if (optval) 2011 tp->t_flags |= opt; 2012 else 2013 tp->t_flags &= ~opt; 2014 unlock_and_done: 2015 #ifdef TCP_OFFLOAD 2016 if (tp->t_flags & TF_TOE) { 2017 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2018 sopt->sopt_name); 2019 } 2020 #endif 2021 INP_WUNLOCK(inp); 2022 break; 2023 2024 case TCP_NOPUSH: 2025 INP_WUNLOCK(inp); 2026 error = sooptcopyin(sopt, &optval, sizeof optval, 2027 sizeof optval); 2028 if (error) 2029 return (error); 2030 2031 INP_WLOCK_RECHECK(inp); 2032 if (optval) 2033 tp->t_flags |= TF_NOPUSH; 2034 else if (tp->t_flags & TF_NOPUSH) { 2035 tp->t_flags &= ~TF_NOPUSH; 2036 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2037 struct epoch_tracker et; 2038 2039 NET_EPOCH_ENTER(et); 2040 error = tp->t_fb->tfb_tcp_output(tp); 2041 NET_EPOCH_EXIT(et); 2042 } 2043 } 2044 goto unlock_and_done; 2045 2046 case TCP_REMOTE_UDP_ENCAPS_PORT: 2047 INP_WUNLOCK(inp); 2048 error = sooptcopyin(sopt, &optval, sizeof optval, 2049 sizeof optval); 2050 if (error) 2051 return (error); 2052 if ((optval < TCP_TUNNELING_PORT_MIN) || 2053 (optval > TCP_TUNNELING_PORT_MAX)) { 2054 /* Its got to be in range */ 2055 return (EINVAL); 2056 } 2057 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2058 /* You have to have enabled a UDP tunneling port first */ 2059 return (EINVAL); 2060 } 2061 INP_WLOCK_RECHECK(inp); 2062 if (tp->t_state != TCPS_CLOSED) { 2063 /* You can't change after you are connected */ 2064 error = EINVAL; 2065 } else { 2066 /* Ok we are all good set the port */ 2067 tp->t_port = htons(optval); 2068 } 2069 goto unlock_and_done; 2070 2071 case TCP_MAXSEG: 2072 INP_WUNLOCK(inp); 2073 error = sooptcopyin(sopt, &optval, sizeof optval, 2074 sizeof optval); 2075 if (error) 2076 return (error); 2077 2078 INP_WLOCK_RECHECK(inp); 2079 if (optval > 0 && optval <= tp->t_maxseg && 2080 optval + 40 >= V_tcp_minmss) 2081 tp->t_maxseg = optval; 2082 else 2083 error = EINVAL; 2084 goto unlock_and_done; 2085 2086 case TCP_INFO: 2087 INP_WUNLOCK(inp); 2088 error = EINVAL; 2089 break; 2090 2091 case TCP_STATS: 2092 INP_WUNLOCK(inp); 2093 #ifdef STATS 2094 error = sooptcopyin(sopt, &optval, sizeof optval, 2095 sizeof optval); 2096 if (error) 2097 return (error); 2098 2099 if (optval > 0) 2100 sbp = stats_blob_alloc( 2101 V_tcp_perconn_stats_dflt_tpl, 0); 2102 else 2103 sbp = NULL; 2104 2105 INP_WLOCK_RECHECK(inp); 2106 if ((tp->t_stats != NULL && sbp == NULL) || 2107 (tp->t_stats == NULL && sbp != NULL)) { 2108 struct statsblob *t = tp->t_stats; 2109 tp->t_stats = sbp; 2110 sbp = t; 2111 } 2112 INP_WUNLOCK(inp); 2113 2114 stats_blob_destroy(sbp); 2115 #else 2116 return (EOPNOTSUPP); 2117 #endif /* !STATS */ 2118 break; 2119 2120 case TCP_CONGESTION: 2121 INP_WUNLOCK(inp); 2122 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 2123 if (error) 2124 break; 2125 buf[sopt->sopt_valsize] = '\0'; 2126 INP_WLOCK_RECHECK(inp); 2127 CC_LIST_RLOCK(); 2128 STAILQ_FOREACH(algo, &cc_list, entries) 2129 if (strncmp(buf, algo->name, 2130 TCP_CA_NAME_MAX) == 0) 2131 break; 2132 CC_LIST_RUNLOCK(); 2133 if (algo == NULL) { 2134 INP_WUNLOCK(inp); 2135 error = EINVAL; 2136 break; 2137 } 2138 /* 2139 * We hold a write lock over the tcb so it's safe to 2140 * do these things without ordering concerns. 2141 */ 2142 if (CC_ALGO(tp)->cb_destroy != NULL) 2143 CC_ALGO(tp)->cb_destroy(tp->ccv); 2144 CC_DATA(tp) = NULL; 2145 CC_ALGO(tp) = algo; 2146 /* 2147 * If something goes pear shaped initialising the new 2148 * algo, fall back to newreno (which does not 2149 * require initialisation). 2150 */ 2151 if (algo->cb_init != NULL && 2152 algo->cb_init(tp->ccv) != 0) { 2153 CC_ALGO(tp) = &newreno_cc_algo; 2154 /* 2155 * The only reason init should fail is 2156 * because of malloc. 2157 */ 2158 error = ENOMEM; 2159 } 2160 INP_WUNLOCK(inp); 2161 break; 2162 2163 case TCP_REUSPORT_LB_NUMA: 2164 INP_WUNLOCK(inp); 2165 error = sooptcopyin(sopt, &optval, sizeof(optval), 2166 sizeof(optval)); 2167 INP_WLOCK_RECHECK(inp); 2168 if (!error) 2169 error = in_pcblbgroup_numa(inp, optval); 2170 INP_WUNLOCK(inp); 2171 break; 2172 2173 #ifdef KERN_TLS 2174 case TCP_TXTLS_ENABLE: 2175 INP_WUNLOCK(inp); 2176 error = copyin_tls_enable(sopt, &tls); 2177 if (error) 2178 break; 2179 error = ktls_enable_tx(so, &tls); 2180 break; 2181 case TCP_TXTLS_MODE: 2182 INP_WUNLOCK(inp); 2183 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2184 if (error) 2185 return (error); 2186 2187 INP_WLOCK_RECHECK(inp); 2188 error = ktls_set_tx_mode(so, ui); 2189 INP_WUNLOCK(inp); 2190 break; 2191 case TCP_RXTLS_ENABLE: 2192 INP_WUNLOCK(inp); 2193 error = sooptcopyin(sopt, &tls, sizeof(tls), 2194 sizeof(tls)); 2195 if (error) 2196 break; 2197 error = ktls_enable_rx(so, &tls); 2198 break; 2199 #endif 2200 2201 case TCP_KEEPIDLE: 2202 case TCP_KEEPINTVL: 2203 case TCP_KEEPINIT: 2204 INP_WUNLOCK(inp); 2205 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2206 if (error) 2207 return (error); 2208 2209 if (ui > (UINT_MAX / hz)) { 2210 error = EINVAL; 2211 break; 2212 } 2213 ui *= hz; 2214 2215 INP_WLOCK_RECHECK(inp); 2216 switch (sopt->sopt_name) { 2217 case TCP_KEEPIDLE: 2218 tp->t_keepidle = ui; 2219 /* 2220 * XXX: better check current remaining 2221 * timeout and "merge" it with new value. 2222 */ 2223 if ((tp->t_state > TCPS_LISTEN) && 2224 (tp->t_state <= TCPS_CLOSING)) 2225 tcp_timer_activate(tp, TT_KEEP, 2226 TP_KEEPIDLE(tp)); 2227 break; 2228 case TCP_KEEPINTVL: 2229 tp->t_keepintvl = ui; 2230 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2231 (TP_MAXIDLE(tp) > 0)) 2232 tcp_timer_activate(tp, TT_2MSL, 2233 TP_MAXIDLE(tp)); 2234 break; 2235 case TCP_KEEPINIT: 2236 tp->t_keepinit = ui; 2237 if (tp->t_state == TCPS_SYN_RECEIVED || 2238 tp->t_state == TCPS_SYN_SENT) 2239 tcp_timer_activate(tp, TT_KEEP, 2240 TP_KEEPINIT(tp)); 2241 break; 2242 } 2243 goto unlock_and_done; 2244 2245 case TCP_KEEPCNT: 2246 INP_WUNLOCK(inp); 2247 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2248 if (error) 2249 return (error); 2250 2251 INP_WLOCK_RECHECK(inp); 2252 tp->t_keepcnt = ui; 2253 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2254 (TP_MAXIDLE(tp) > 0)) 2255 tcp_timer_activate(tp, TT_2MSL, 2256 TP_MAXIDLE(tp)); 2257 goto unlock_and_done; 2258 2259 #ifdef TCPPCAP 2260 case TCP_PCAP_OUT: 2261 case TCP_PCAP_IN: 2262 INP_WUNLOCK(inp); 2263 error = sooptcopyin(sopt, &optval, sizeof optval, 2264 sizeof optval); 2265 if (error) 2266 return (error); 2267 2268 INP_WLOCK_RECHECK(inp); 2269 if (optval >= 0) 2270 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2271 &(tp->t_outpkts) : &(tp->t_inpkts), 2272 optval); 2273 else 2274 error = EINVAL; 2275 goto unlock_and_done; 2276 #endif 2277 2278 case TCP_FASTOPEN: { 2279 struct tcp_fastopen tfo_optval; 2280 2281 INP_WUNLOCK(inp); 2282 if (!V_tcp_fastopen_client_enable && 2283 !V_tcp_fastopen_server_enable) 2284 return (EPERM); 2285 2286 error = sooptcopyin(sopt, &tfo_optval, 2287 sizeof(tfo_optval), sizeof(int)); 2288 if (error) 2289 return (error); 2290 2291 INP_WLOCK_RECHECK(inp); 2292 if ((tp->t_state != TCPS_CLOSED) && 2293 (tp->t_state != TCPS_LISTEN)) { 2294 error = EINVAL; 2295 goto unlock_and_done; 2296 } 2297 if (tfo_optval.enable) { 2298 if (tp->t_state == TCPS_LISTEN) { 2299 if (!V_tcp_fastopen_server_enable) { 2300 error = EPERM; 2301 goto unlock_and_done; 2302 } 2303 2304 if (tp->t_tfo_pending == NULL) 2305 tp->t_tfo_pending = 2306 tcp_fastopen_alloc_counter(); 2307 } else { 2308 /* 2309 * If a pre-shared key was provided, 2310 * stash it in the client cookie 2311 * field of the tcpcb for use during 2312 * connect. 2313 */ 2314 if (sopt->sopt_valsize == 2315 sizeof(tfo_optval)) { 2316 memcpy(tp->t_tfo_cookie.client, 2317 tfo_optval.psk, 2318 TCP_FASTOPEN_PSK_LEN); 2319 tp->t_tfo_client_cookie_len = 2320 TCP_FASTOPEN_PSK_LEN; 2321 } 2322 } 2323 tp->t_flags |= TF_FASTOPEN; 2324 } else 2325 tp->t_flags &= ~TF_FASTOPEN; 2326 goto unlock_and_done; 2327 } 2328 2329 #ifdef TCP_BLACKBOX 2330 case TCP_LOG: 2331 INP_WUNLOCK(inp); 2332 error = sooptcopyin(sopt, &optval, sizeof optval, 2333 sizeof optval); 2334 if (error) 2335 return (error); 2336 2337 INP_WLOCK_RECHECK(inp); 2338 error = tcp_log_state_change(tp, optval); 2339 goto unlock_and_done; 2340 2341 case TCP_LOGBUF: 2342 INP_WUNLOCK(inp); 2343 error = EINVAL; 2344 break; 2345 2346 case TCP_LOGID: 2347 INP_WUNLOCK(inp); 2348 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2349 if (error) 2350 break; 2351 buf[sopt->sopt_valsize] = '\0'; 2352 INP_WLOCK_RECHECK(inp); 2353 error = tcp_log_set_id(tp, buf); 2354 /* tcp_log_set_id() unlocks the INP. */ 2355 break; 2356 2357 case TCP_LOGDUMP: 2358 case TCP_LOGDUMPID: 2359 INP_WUNLOCK(inp); 2360 error = 2361 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2362 if (error) 2363 break; 2364 buf[sopt->sopt_valsize] = '\0'; 2365 INP_WLOCK_RECHECK(inp); 2366 if (sopt->sopt_name == TCP_LOGDUMP) { 2367 error = tcp_log_dump_tp_logbuf(tp, buf, 2368 M_WAITOK, true); 2369 INP_WUNLOCK(inp); 2370 } else { 2371 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2372 /* 2373 * tcp_log_dump_tp_bucket_logbufs() drops the 2374 * INP lock. 2375 */ 2376 } 2377 break; 2378 #endif 2379 2380 default: 2381 INP_WUNLOCK(inp); 2382 error = ENOPROTOOPT; 2383 break; 2384 } 2385 break; 2386 2387 case SOPT_GET: 2388 tp = intotcpcb(inp); 2389 switch (sopt->sopt_name) { 2390 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2391 case TCP_MD5SIG: 2392 if (!TCPMD5_ENABLED()) { 2393 INP_WUNLOCK(inp); 2394 return (ENOPROTOOPT); 2395 } 2396 error = TCPMD5_PCBCTL(inp, sopt); 2397 break; 2398 #endif 2399 2400 case TCP_NODELAY: 2401 optval = tp->t_flags & TF_NODELAY; 2402 INP_WUNLOCK(inp); 2403 error = sooptcopyout(sopt, &optval, sizeof optval); 2404 break; 2405 case TCP_MAXSEG: 2406 optval = tp->t_maxseg; 2407 INP_WUNLOCK(inp); 2408 error = sooptcopyout(sopt, &optval, sizeof optval); 2409 break; 2410 case TCP_REMOTE_UDP_ENCAPS_PORT: 2411 optval = ntohs(tp->t_port); 2412 INP_WUNLOCK(inp); 2413 error = sooptcopyout(sopt, &optval, sizeof optval); 2414 break; 2415 case TCP_NOOPT: 2416 optval = tp->t_flags & TF_NOOPT; 2417 INP_WUNLOCK(inp); 2418 error = sooptcopyout(sopt, &optval, sizeof optval); 2419 break; 2420 case TCP_NOPUSH: 2421 optval = tp->t_flags & TF_NOPUSH; 2422 INP_WUNLOCK(inp); 2423 error = sooptcopyout(sopt, &optval, sizeof optval); 2424 break; 2425 case TCP_INFO: 2426 tcp_fill_info(tp, &ti); 2427 INP_WUNLOCK(inp); 2428 error = sooptcopyout(sopt, &ti, sizeof ti); 2429 break; 2430 case TCP_STATS: 2431 { 2432 #ifdef STATS 2433 int nheld; 2434 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2435 2436 error = 0; 2437 socklen_t outsbsz = sopt->sopt_valsize; 2438 if (tp->t_stats == NULL) 2439 error = ENOENT; 2440 else if (outsbsz >= tp->t_stats->cursz) 2441 outsbsz = tp->t_stats->cursz; 2442 else if (outsbsz >= sizeof(struct statsblob)) 2443 outsbsz = sizeof(struct statsblob); 2444 else 2445 error = EINVAL; 2446 INP_WUNLOCK(inp); 2447 if (error) 2448 break; 2449 2450 sbp = sopt->sopt_val; 2451 nheld = atop(round_page(((vm_offset_t)sbp) + 2452 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2453 vm_page_t ma[nheld]; 2454 if (vm_fault_quick_hold_pages( 2455 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2456 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2457 nheld) < 0) { 2458 error = EFAULT; 2459 break; 2460 } 2461 2462 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2463 SIZEOF_MEMBER(struct statsblob, flags)))) 2464 goto unhold; 2465 2466 INP_WLOCK_RECHECK(inp); 2467 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2468 sbflags | SB_CLONE_USRDSTNOFAULT); 2469 INP_WUNLOCK(inp); 2470 sopt->sopt_valsize = outsbsz; 2471 unhold: 2472 vm_page_unhold_pages(ma, nheld); 2473 #else 2474 INP_WUNLOCK(inp); 2475 error = EOPNOTSUPP; 2476 #endif /* !STATS */ 2477 break; 2478 } 2479 case TCP_CONGESTION: 2480 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2481 INP_WUNLOCK(inp); 2482 error = sooptcopyout(sopt, buf, len + 1); 2483 break; 2484 case TCP_KEEPIDLE: 2485 case TCP_KEEPINTVL: 2486 case TCP_KEEPINIT: 2487 case TCP_KEEPCNT: 2488 switch (sopt->sopt_name) { 2489 case TCP_KEEPIDLE: 2490 ui = TP_KEEPIDLE(tp) / hz; 2491 break; 2492 case TCP_KEEPINTVL: 2493 ui = TP_KEEPINTVL(tp) / hz; 2494 break; 2495 case TCP_KEEPINIT: 2496 ui = TP_KEEPINIT(tp) / hz; 2497 break; 2498 case TCP_KEEPCNT: 2499 ui = TP_KEEPCNT(tp); 2500 break; 2501 } 2502 INP_WUNLOCK(inp); 2503 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2504 break; 2505 #ifdef TCPPCAP 2506 case TCP_PCAP_OUT: 2507 case TCP_PCAP_IN: 2508 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2509 &(tp->t_outpkts) : &(tp->t_inpkts)); 2510 INP_WUNLOCK(inp); 2511 error = sooptcopyout(sopt, &optval, sizeof optval); 2512 break; 2513 #endif 2514 case TCP_FASTOPEN: 2515 optval = tp->t_flags & TF_FASTOPEN; 2516 INP_WUNLOCK(inp); 2517 error = sooptcopyout(sopt, &optval, sizeof optval); 2518 break; 2519 #ifdef TCP_BLACKBOX 2520 case TCP_LOG: 2521 optval = tp->t_logstate; 2522 INP_WUNLOCK(inp); 2523 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2524 break; 2525 case TCP_LOGBUF: 2526 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2527 error = tcp_log_getlogbuf(sopt, tp); 2528 break; 2529 case TCP_LOGID: 2530 len = tcp_log_get_id(tp, buf); 2531 INP_WUNLOCK(inp); 2532 error = sooptcopyout(sopt, buf, len + 1); 2533 break; 2534 case TCP_LOGDUMP: 2535 case TCP_LOGDUMPID: 2536 INP_WUNLOCK(inp); 2537 error = EINVAL; 2538 break; 2539 #endif 2540 #ifdef KERN_TLS 2541 case TCP_TXTLS_MODE: 2542 optval = ktls_get_tx_mode(so); 2543 INP_WUNLOCK(inp); 2544 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2545 break; 2546 case TCP_RXTLS_MODE: 2547 optval = ktls_get_rx_mode(so); 2548 INP_WUNLOCK(inp); 2549 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2550 break; 2551 #endif 2552 case TCP_LRD: 2553 optval = tp->t_flags & TF_LRD; 2554 INP_WUNLOCK(inp); 2555 error = sooptcopyout(sopt, &optval, sizeof optval); 2556 break; 2557 default: 2558 INP_WUNLOCK(inp); 2559 error = ENOPROTOOPT; 2560 break; 2561 } 2562 break; 2563 } 2564 return (error); 2565 } 2566 #undef INP_WLOCK_RECHECK 2567 #undef INP_WLOCK_RECHECK_CLEANUP 2568 2569 /* 2570 * Initiate (or continue) disconnect. 2571 * If embryonic state, just send reset (once). 2572 * If in ``let data drain'' option and linger null, just drop. 2573 * Otherwise (hard), mark socket disconnecting and drop 2574 * current input data; switch states based on user close, and 2575 * send segment to peer (with FIN). 2576 */ 2577 static void 2578 tcp_disconnect(struct tcpcb *tp) 2579 { 2580 struct inpcb *inp = tp->t_inpcb; 2581 struct socket *so = inp->inp_socket; 2582 2583 NET_EPOCH_ASSERT(); 2584 INP_WLOCK_ASSERT(inp); 2585 2586 /* 2587 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2588 * socket is still open. 2589 */ 2590 if (tp->t_state < TCPS_ESTABLISHED && 2591 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2592 tp = tcp_close(tp); 2593 KASSERT(tp != NULL, 2594 ("tcp_disconnect: tcp_close() returned NULL")); 2595 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2596 tp = tcp_drop(tp, 0); 2597 KASSERT(tp != NULL, 2598 ("tcp_disconnect: tcp_drop() returned NULL")); 2599 } else { 2600 soisdisconnecting(so); 2601 sbflush(&so->so_rcv); 2602 tcp_usrclosed(tp); 2603 if (!(inp->inp_flags & INP_DROPPED)) 2604 tp->t_fb->tfb_tcp_output(tp); 2605 } 2606 } 2607 2608 /* 2609 * User issued close, and wish to trail through shutdown states: 2610 * if never received SYN, just forget it. If got a SYN from peer, 2611 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2612 * If already got a FIN from peer, then almost done; go to LAST_ACK 2613 * state. In all other cases, have already sent FIN to peer (e.g. 2614 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2615 * for peer to send FIN or not respond to keep-alives, etc. 2616 * We can let the user exit from the close as soon as the FIN is acked. 2617 */ 2618 static void 2619 tcp_usrclosed(struct tcpcb *tp) 2620 { 2621 2622 NET_EPOCH_ASSERT(); 2623 INP_WLOCK_ASSERT(tp->t_inpcb); 2624 2625 switch (tp->t_state) { 2626 case TCPS_LISTEN: 2627 #ifdef TCP_OFFLOAD 2628 tcp_offload_listen_stop(tp); 2629 #endif 2630 tcp_state_change(tp, TCPS_CLOSED); 2631 /* FALLTHROUGH */ 2632 case TCPS_CLOSED: 2633 tp = tcp_close(tp); 2634 /* 2635 * tcp_close() should never return NULL here as the socket is 2636 * still open. 2637 */ 2638 KASSERT(tp != NULL, 2639 ("tcp_usrclosed: tcp_close() returned NULL")); 2640 break; 2641 2642 case TCPS_SYN_SENT: 2643 case TCPS_SYN_RECEIVED: 2644 tp->t_flags |= TF_NEEDFIN; 2645 break; 2646 2647 case TCPS_ESTABLISHED: 2648 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2649 break; 2650 2651 case TCPS_CLOSE_WAIT: 2652 tcp_state_change(tp, TCPS_LAST_ACK); 2653 break; 2654 } 2655 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2656 soisdisconnected(tp->t_inpcb->inp_socket); 2657 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2658 if (tp->t_state == TCPS_FIN_WAIT_2) { 2659 int timeout; 2660 2661 timeout = (tcp_fast_finwait2_recycle) ? 2662 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2663 tcp_timer_activate(tp, TT_2MSL, timeout); 2664 } 2665 } 2666 } 2667 2668 #ifdef DDB 2669 static void 2670 db_print_indent(int indent) 2671 { 2672 int i; 2673 2674 for (i = 0; i < indent; i++) 2675 db_printf(" "); 2676 } 2677 2678 static void 2679 db_print_tstate(int t_state) 2680 { 2681 2682 switch (t_state) { 2683 case TCPS_CLOSED: 2684 db_printf("TCPS_CLOSED"); 2685 return; 2686 2687 case TCPS_LISTEN: 2688 db_printf("TCPS_LISTEN"); 2689 return; 2690 2691 case TCPS_SYN_SENT: 2692 db_printf("TCPS_SYN_SENT"); 2693 return; 2694 2695 case TCPS_SYN_RECEIVED: 2696 db_printf("TCPS_SYN_RECEIVED"); 2697 return; 2698 2699 case TCPS_ESTABLISHED: 2700 db_printf("TCPS_ESTABLISHED"); 2701 return; 2702 2703 case TCPS_CLOSE_WAIT: 2704 db_printf("TCPS_CLOSE_WAIT"); 2705 return; 2706 2707 case TCPS_FIN_WAIT_1: 2708 db_printf("TCPS_FIN_WAIT_1"); 2709 return; 2710 2711 case TCPS_CLOSING: 2712 db_printf("TCPS_CLOSING"); 2713 return; 2714 2715 case TCPS_LAST_ACK: 2716 db_printf("TCPS_LAST_ACK"); 2717 return; 2718 2719 case TCPS_FIN_WAIT_2: 2720 db_printf("TCPS_FIN_WAIT_2"); 2721 return; 2722 2723 case TCPS_TIME_WAIT: 2724 db_printf("TCPS_TIME_WAIT"); 2725 return; 2726 2727 default: 2728 db_printf("unknown"); 2729 return; 2730 } 2731 } 2732 2733 static void 2734 db_print_tflags(u_int t_flags) 2735 { 2736 int comma; 2737 2738 comma = 0; 2739 if (t_flags & TF_ACKNOW) { 2740 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2741 comma = 1; 2742 } 2743 if (t_flags & TF_DELACK) { 2744 db_printf("%sTF_DELACK", comma ? ", " : ""); 2745 comma = 1; 2746 } 2747 if (t_flags & TF_NODELAY) { 2748 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2749 comma = 1; 2750 } 2751 if (t_flags & TF_NOOPT) { 2752 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2753 comma = 1; 2754 } 2755 if (t_flags & TF_SENTFIN) { 2756 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2757 comma = 1; 2758 } 2759 if (t_flags & TF_REQ_SCALE) { 2760 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2761 comma = 1; 2762 } 2763 if (t_flags & TF_RCVD_SCALE) { 2764 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2765 comma = 1; 2766 } 2767 if (t_flags & TF_REQ_TSTMP) { 2768 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2769 comma = 1; 2770 } 2771 if (t_flags & TF_RCVD_TSTMP) { 2772 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2773 comma = 1; 2774 } 2775 if (t_flags & TF_SACK_PERMIT) { 2776 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2777 comma = 1; 2778 } 2779 if (t_flags & TF_NEEDSYN) { 2780 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2781 comma = 1; 2782 } 2783 if (t_flags & TF_NEEDFIN) { 2784 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2785 comma = 1; 2786 } 2787 if (t_flags & TF_NOPUSH) { 2788 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2789 comma = 1; 2790 } 2791 if (t_flags & TF_MORETOCOME) { 2792 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2793 comma = 1; 2794 } 2795 if (t_flags & TF_LQ_OVERFLOW) { 2796 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2797 comma = 1; 2798 } 2799 if (t_flags & TF_LASTIDLE) { 2800 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2801 comma = 1; 2802 } 2803 if (t_flags & TF_RXWIN0SENT) { 2804 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2805 comma = 1; 2806 } 2807 if (t_flags & TF_FASTRECOVERY) { 2808 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2809 comma = 1; 2810 } 2811 if (t_flags & TF_CONGRECOVERY) { 2812 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2813 comma = 1; 2814 } 2815 if (t_flags & TF_WASFRECOVERY) { 2816 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2817 comma = 1; 2818 } 2819 if (t_flags & TF_SIGNATURE) { 2820 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2821 comma = 1; 2822 } 2823 if (t_flags & TF_FORCEDATA) { 2824 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2825 comma = 1; 2826 } 2827 if (t_flags & TF_TSO) { 2828 db_printf("%sTF_TSO", comma ? ", " : ""); 2829 comma = 1; 2830 } 2831 if (t_flags & TF_FASTOPEN) { 2832 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2833 comma = 1; 2834 } 2835 } 2836 2837 static void 2838 db_print_tflags2(u_int t_flags2) 2839 { 2840 int comma; 2841 2842 comma = 0; 2843 if (t_flags2 & TF2_ECN_PERMIT) { 2844 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 2845 comma = 1; 2846 } 2847 } 2848 2849 static void 2850 db_print_toobflags(char t_oobflags) 2851 { 2852 int comma; 2853 2854 comma = 0; 2855 if (t_oobflags & TCPOOB_HAVEDATA) { 2856 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 2857 comma = 1; 2858 } 2859 if (t_oobflags & TCPOOB_HADDATA) { 2860 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 2861 comma = 1; 2862 } 2863 } 2864 2865 static void 2866 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 2867 { 2868 2869 db_print_indent(indent); 2870 db_printf("%s at %p\n", name, tp); 2871 2872 indent += 2; 2873 2874 db_print_indent(indent); 2875 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 2876 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 2877 2878 db_print_indent(indent); 2879 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 2880 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 2881 2882 db_print_indent(indent); 2883 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 2884 &tp->t_timers->tt_delack, tp->t_inpcb); 2885 2886 db_print_indent(indent); 2887 db_printf("t_state: %d (", tp->t_state); 2888 db_print_tstate(tp->t_state); 2889 db_printf(")\n"); 2890 2891 db_print_indent(indent); 2892 db_printf("t_flags: 0x%x (", tp->t_flags); 2893 db_print_tflags(tp->t_flags); 2894 db_printf(")\n"); 2895 2896 db_print_indent(indent); 2897 db_printf("t_flags2: 0x%x (", tp->t_flags2); 2898 db_print_tflags2(tp->t_flags2); 2899 db_printf(")\n"); 2900 2901 db_print_indent(indent); 2902 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 2903 tp->snd_una, tp->snd_max, tp->snd_nxt); 2904 2905 db_print_indent(indent); 2906 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 2907 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 2908 2909 db_print_indent(indent); 2910 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 2911 tp->iss, tp->irs, tp->rcv_nxt); 2912 2913 db_print_indent(indent); 2914 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 2915 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 2916 2917 db_print_indent(indent); 2918 db_printf("snd_wnd: %u snd_cwnd: %u\n", 2919 tp->snd_wnd, tp->snd_cwnd); 2920 2921 db_print_indent(indent); 2922 db_printf("snd_ssthresh: %u snd_recover: " 2923 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 2924 2925 db_print_indent(indent); 2926 db_printf("t_rcvtime: %u t_startime: %u\n", 2927 tp->t_rcvtime, tp->t_starttime); 2928 2929 db_print_indent(indent); 2930 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 2931 tp->t_rtttime, tp->t_rtseq); 2932 2933 db_print_indent(indent); 2934 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 2935 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 2936 2937 db_print_indent(indent); 2938 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 2939 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 2940 tp->t_rttbest); 2941 2942 db_print_indent(indent); 2943 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 2944 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 2945 2946 db_print_indent(indent); 2947 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 2948 db_print_toobflags(tp->t_oobflags); 2949 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 2950 2951 db_print_indent(indent); 2952 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 2953 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 2954 2955 db_print_indent(indent); 2956 db_printf("ts_recent: %u ts_recent_age: %u\n", 2957 tp->ts_recent, tp->ts_recent_age); 2958 2959 db_print_indent(indent); 2960 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 2961 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 2962 2963 db_print_indent(indent); 2964 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 2965 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 2966 tp->snd_recover_prev, tp->t_badrxtwin); 2967 2968 db_print_indent(indent); 2969 db_printf("snd_numholes: %d snd_holes first: %p\n", 2970 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 2971 2972 db_print_indent(indent); 2973 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 2974 tp->snd_fack, tp->rcv_numsacks); 2975 2976 /* Skip sackblks, sackhint. */ 2977 2978 db_print_indent(indent); 2979 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 2980 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 2981 } 2982 2983 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 2984 { 2985 struct tcpcb *tp; 2986 2987 if (!have_addr) { 2988 db_printf("usage: show tcpcb <addr>\n"); 2989 return; 2990 } 2991 tp = (struct tcpcb *)addr; 2992 2993 db_print_tcpcb(tp, "tcpcb", 0); 2994 } 2995 #endif 2996