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 /* 326 * Preserve compatibility with old programs. 327 */ 328 if (nam->sa_family != AF_UNSPEC || 329 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 330 sinp->sin_addr.s_addr != INADDR_ANY) 331 return (EAFNOSUPPORT); 332 nam->sa_family = AF_INET; 333 } 334 if (nam->sa_len != sizeof(*sinp)) 335 return (EINVAL); 336 337 /* 338 * Must check for multicast addresses and disallow binding 339 * to them. 340 */ 341 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 342 return (EAFNOSUPPORT); 343 344 TCPDEBUG0; 345 inp = sotoinpcb(so); 346 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 347 INP_WLOCK(inp); 348 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 349 error = EINVAL; 350 goto out; 351 } 352 tp = intotcpcb(inp); 353 TCPDEBUG1(); 354 INP_HASH_WLOCK(&V_tcbinfo); 355 error = in_pcbbind(inp, nam, td->td_ucred); 356 INP_HASH_WUNLOCK(&V_tcbinfo); 357 out: 358 TCPDEBUG2(PRU_BIND); 359 TCP_PROBE2(debug__user, tp, PRU_BIND); 360 INP_WUNLOCK(inp); 361 362 return (error); 363 } 364 #endif /* INET */ 365 366 #ifdef INET6 367 static int 368 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 369 { 370 int error = 0; 371 struct inpcb *inp; 372 struct tcpcb *tp = NULL; 373 struct sockaddr_in6 *sin6; 374 u_char vflagsav; 375 376 sin6 = (struct sockaddr_in6 *)nam; 377 if (nam->sa_family != AF_INET6) 378 return (EAFNOSUPPORT); 379 if (nam->sa_len != sizeof(*sin6)) 380 return (EINVAL); 381 382 /* 383 * Must check for multicast addresses and disallow binding 384 * to them. 385 */ 386 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 387 return (EAFNOSUPPORT); 388 389 TCPDEBUG0; 390 inp = sotoinpcb(so); 391 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 392 INP_WLOCK(inp); 393 vflagsav = inp->inp_vflag; 394 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 395 error = EINVAL; 396 goto out; 397 } 398 tp = intotcpcb(inp); 399 TCPDEBUG1(); 400 INP_HASH_WLOCK(&V_tcbinfo); 401 inp->inp_vflag &= ~INP_IPV4; 402 inp->inp_vflag |= INP_IPV6; 403 #ifdef INET 404 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 405 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 406 inp->inp_vflag |= INP_IPV4; 407 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 408 struct sockaddr_in sin; 409 410 in6_sin6_2_sin(&sin, sin6); 411 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 412 error = EAFNOSUPPORT; 413 INP_HASH_WUNLOCK(&V_tcbinfo); 414 goto out; 415 } 416 inp->inp_vflag |= INP_IPV4; 417 inp->inp_vflag &= ~INP_IPV6; 418 error = in_pcbbind(inp, (struct sockaddr *)&sin, 419 td->td_ucred); 420 INP_HASH_WUNLOCK(&V_tcbinfo); 421 goto out; 422 } 423 } 424 #endif 425 error = in6_pcbbind(inp, nam, td->td_ucred); 426 INP_HASH_WUNLOCK(&V_tcbinfo); 427 out: 428 if (error != 0) 429 inp->inp_vflag = vflagsav; 430 TCPDEBUG2(PRU_BIND); 431 TCP_PROBE2(debug__user, tp, PRU_BIND); 432 INP_WUNLOCK(inp); 433 return (error); 434 } 435 #endif /* INET6 */ 436 437 #ifdef INET 438 /* 439 * Prepare to accept connections. 440 */ 441 static int 442 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 443 { 444 int error = 0; 445 struct inpcb *inp; 446 struct tcpcb *tp = NULL; 447 448 TCPDEBUG0; 449 inp = sotoinpcb(so); 450 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 451 INP_WLOCK(inp); 452 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 453 error = EINVAL; 454 goto out; 455 } 456 tp = intotcpcb(inp); 457 TCPDEBUG1(); 458 SOCK_LOCK(so); 459 error = solisten_proto_check(so); 460 if (error != 0) { 461 SOCK_UNLOCK(so); 462 goto out; 463 } 464 if (inp->inp_lport == 0) { 465 INP_HASH_WLOCK(&V_tcbinfo); 466 error = in_pcbbind(inp, NULL, td->td_ucred); 467 INP_HASH_WUNLOCK(&V_tcbinfo); 468 } 469 if (error == 0) { 470 tcp_state_change(tp, TCPS_LISTEN); 471 solisten_proto(so, backlog); 472 #ifdef TCP_OFFLOAD 473 if ((so->so_options & SO_NO_OFFLOAD) == 0) 474 tcp_offload_listen_start(tp); 475 #endif 476 } else { 477 solisten_proto_abort(so); 478 } 479 SOCK_UNLOCK(so); 480 481 if (IS_FASTOPEN(tp->t_flags)) 482 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 483 484 out: 485 TCPDEBUG2(PRU_LISTEN); 486 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 487 INP_WUNLOCK(inp); 488 return (error); 489 } 490 #endif /* INET */ 491 492 #ifdef INET6 493 static int 494 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 495 { 496 int error = 0; 497 struct inpcb *inp; 498 struct tcpcb *tp = NULL; 499 u_char vflagsav; 500 501 TCPDEBUG0; 502 inp = sotoinpcb(so); 503 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 504 INP_WLOCK(inp); 505 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 506 error = EINVAL; 507 goto out; 508 } 509 vflagsav = inp->inp_vflag; 510 tp = intotcpcb(inp); 511 TCPDEBUG1(); 512 SOCK_LOCK(so); 513 error = solisten_proto_check(so); 514 if (error != 0) { 515 SOCK_UNLOCK(so); 516 goto out; 517 } 518 INP_HASH_WLOCK(&V_tcbinfo); 519 if (inp->inp_lport == 0) { 520 inp->inp_vflag &= ~INP_IPV4; 521 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 522 inp->inp_vflag |= INP_IPV4; 523 error = in6_pcbbind(inp, NULL, td->td_ucred); 524 } 525 INP_HASH_WUNLOCK(&V_tcbinfo); 526 if (error == 0) { 527 tcp_state_change(tp, TCPS_LISTEN); 528 solisten_proto(so, backlog); 529 #ifdef TCP_OFFLOAD 530 if ((so->so_options & SO_NO_OFFLOAD) == 0) 531 tcp_offload_listen_start(tp); 532 #endif 533 } else { 534 solisten_proto_abort(so); 535 } 536 SOCK_UNLOCK(so); 537 538 if (IS_FASTOPEN(tp->t_flags)) 539 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 540 541 if (error != 0) 542 inp->inp_vflag = vflagsav; 543 544 out: 545 TCPDEBUG2(PRU_LISTEN); 546 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 547 INP_WUNLOCK(inp); 548 return (error); 549 } 550 #endif /* INET6 */ 551 552 #ifdef INET 553 /* 554 * Initiate connection to peer. 555 * Create a template for use in transmissions on this connection. 556 * Enter SYN_SENT state, and mark socket as connecting. 557 * Start keep-alive timer, and seed output sequence space. 558 * Send initial segment on connection. 559 */ 560 static int 561 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 562 { 563 struct epoch_tracker et; 564 int error = 0; 565 struct inpcb *inp; 566 struct tcpcb *tp = NULL; 567 struct sockaddr_in *sinp; 568 569 sinp = (struct sockaddr_in *)nam; 570 if (nam->sa_family != AF_INET) 571 return (EAFNOSUPPORT); 572 if (nam->sa_len != sizeof (*sinp)) 573 return (EINVAL); 574 575 /* 576 * Must disallow TCP ``connections'' to multicast addresses. 577 */ 578 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 579 return (EAFNOSUPPORT); 580 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) 581 return (EACCES); 582 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 583 return (error); 584 585 TCPDEBUG0; 586 inp = sotoinpcb(so); 587 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 588 INP_WLOCK(inp); 589 if (inp->inp_flags & INP_TIMEWAIT) { 590 error = EADDRINUSE; 591 goto out; 592 } 593 if (inp->inp_flags & INP_DROPPED) { 594 error = ECONNREFUSED; 595 goto out; 596 } 597 if (SOLISTENING(so)) { 598 error = EOPNOTSUPP; 599 goto out; 600 } 601 tp = intotcpcb(inp); 602 TCPDEBUG1(); 603 NET_EPOCH_ENTER(et); 604 if ((error = tcp_connect(tp, nam, td)) != 0) 605 goto out_in_epoch; 606 #ifdef TCP_OFFLOAD 607 if (registered_toedevs > 0 && 608 (so->so_options & SO_NO_OFFLOAD) == 0 && 609 (error = tcp_offload_connect(so, nam)) == 0) 610 goto out_in_epoch; 611 #endif 612 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 613 error = tp->t_fb->tfb_tcp_output(tp); 614 out_in_epoch: 615 NET_EPOCH_EXIT(et); 616 out: 617 TCPDEBUG2(PRU_CONNECT); 618 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 619 INP_WUNLOCK(inp); 620 return (error); 621 } 622 #endif /* INET */ 623 624 #ifdef INET6 625 static int 626 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 627 { 628 struct epoch_tracker et; 629 int error = 0; 630 struct inpcb *inp; 631 struct tcpcb *tp = NULL; 632 struct sockaddr_in6 *sin6; 633 u_int8_t incflagsav; 634 u_char vflagsav; 635 636 TCPDEBUG0; 637 638 sin6 = (struct sockaddr_in6 *)nam; 639 if (nam->sa_family != AF_INET6) 640 return (EAFNOSUPPORT); 641 if (nam->sa_len != sizeof (*sin6)) 642 return (EINVAL); 643 644 /* 645 * Must disallow TCP ``connections'' to multicast addresses. 646 */ 647 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 648 return (EAFNOSUPPORT); 649 650 inp = sotoinpcb(so); 651 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 652 INP_WLOCK(inp); 653 vflagsav = inp->inp_vflag; 654 incflagsav = inp->inp_inc.inc_flags; 655 if (inp->inp_flags & INP_TIMEWAIT) { 656 error = EADDRINUSE; 657 goto out; 658 } 659 if (inp->inp_flags & INP_DROPPED) { 660 error = ECONNREFUSED; 661 goto out; 662 } 663 if (SOLISTENING(so)) { 664 error = EINVAL; 665 goto out; 666 } 667 tp = intotcpcb(inp); 668 TCPDEBUG1(); 669 #ifdef INET 670 /* 671 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 672 * therefore probably require the hash lock, which isn't held here. 673 * Is this a significant problem? 674 */ 675 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 676 struct sockaddr_in sin; 677 678 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 679 error = EINVAL; 680 goto out; 681 } 682 if ((inp->inp_vflag & INP_IPV4) == 0) { 683 error = EAFNOSUPPORT; 684 goto out; 685 } 686 687 in6_sin6_2_sin(&sin, sin6); 688 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 689 error = EAFNOSUPPORT; 690 goto out; 691 } 692 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 693 error = EACCES; 694 goto out; 695 } 696 if ((error = prison_remote_ip4(td->td_ucred, 697 &sin.sin_addr)) != 0) 698 goto out; 699 inp->inp_vflag |= INP_IPV4; 700 inp->inp_vflag &= ~INP_IPV6; 701 NET_EPOCH_ENTER(et); 702 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 703 goto out_in_epoch; 704 #ifdef TCP_OFFLOAD 705 if (registered_toedevs > 0 && 706 (so->so_options & SO_NO_OFFLOAD) == 0 && 707 (error = tcp_offload_connect(so, nam)) == 0) 708 goto out_in_epoch; 709 #endif 710 error = tp->t_fb->tfb_tcp_output(tp); 711 goto out_in_epoch; 712 } else { 713 if ((inp->inp_vflag & INP_IPV6) == 0) { 714 error = EAFNOSUPPORT; 715 goto out; 716 } 717 } 718 #endif 719 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 720 goto out; 721 inp->inp_vflag &= ~INP_IPV4; 722 inp->inp_vflag |= INP_IPV6; 723 inp->inp_inc.inc_flags |= INC_ISIPV6; 724 if ((error = tcp6_connect(tp, nam, td)) != 0) 725 goto out; 726 #ifdef TCP_OFFLOAD 727 if (registered_toedevs > 0 && 728 (so->so_options & SO_NO_OFFLOAD) == 0 && 729 (error = tcp_offload_connect(so, nam)) == 0) 730 goto out; 731 #endif 732 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 733 NET_EPOCH_ENTER(et); 734 error = tp->t_fb->tfb_tcp_output(tp); 735 #ifdef INET 736 out_in_epoch: 737 #endif 738 NET_EPOCH_EXIT(et); 739 out: 740 /* 741 * If the implicit bind in the connect call fails, restore 742 * the flags we modified. 743 */ 744 if (error != 0 && inp->inp_lport == 0) { 745 inp->inp_vflag = vflagsav; 746 inp->inp_inc.inc_flags = incflagsav; 747 } 748 749 TCPDEBUG2(PRU_CONNECT); 750 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 751 INP_WUNLOCK(inp); 752 return (error); 753 } 754 #endif /* INET6 */ 755 756 /* 757 * Initiate disconnect from peer. 758 * If connection never passed embryonic stage, just drop; 759 * else if don't need to let data drain, then can just drop anyways, 760 * else have to begin TCP shutdown process: mark socket disconnecting, 761 * drain unread data, state switch to reflect user close, and 762 * send segment (e.g. FIN) to peer. Socket will be really disconnected 763 * when peer sends FIN and acks ours. 764 * 765 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 766 */ 767 static int 768 tcp_usr_disconnect(struct socket *so) 769 { 770 struct inpcb *inp; 771 struct tcpcb *tp = NULL; 772 struct epoch_tracker et; 773 int error = 0; 774 775 TCPDEBUG0; 776 NET_EPOCH_ENTER(et); 777 inp = sotoinpcb(so); 778 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 779 INP_WLOCK(inp); 780 if (inp->inp_flags & INP_TIMEWAIT) 781 goto out; 782 if (inp->inp_flags & INP_DROPPED) { 783 error = ECONNRESET; 784 goto out; 785 } 786 tp = intotcpcb(inp); 787 TCPDEBUG1(); 788 tcp_disconnect(tp); 789 out: 790 TCPDEBUG2(PRU_DISCONNECT); 791 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 792 INP_WUNLOCK(inp); 793 NET_EPOCH_EXIT(et); 794 return (error); 795 } 796 797 #ifdef INET 798 /* 799 * Accept a connection. Essentially all the work is done at higher levels; 800 * just return the address of the peer, storing through addr. 801 */ 802 static int 803 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 804 { 805 int error = 0; 806 struct inpcb *inp = NULL; 807 struct tcpcb *tp = NULL; 808 struct in_addr addr; 809 in_port_t port = 0; 810 TCPDEBUG0; 811 812 if (so->so_state & SS_ISDISCONNECTED) 813 return (ECONNABORTED); 814 815 inp = sotoinpcb(so); 816 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 817 INP_WLOCK(inp); 818 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 819 error = ECONNABORTED; 820 goto out; 821 } 822 tp = intotcpcb(inp); 823 TCPDEBUG1(); 824 825 /* 826 * We inline in_getpeeraddr and COMMON_END here, so that we can 827 * copy the data of interest and defer the malloc until after we 828 * release the lock. 829 */ 830 port = inp->inp_fport; 831 addr = inp->inp_faddr; 832 833 out: 834 TCPDEBUG2(PRU_ACCEPT); 835 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 836 INP_WUNLOCK(inp); 837 if (error == 0) 838 *nam = in_sockaddr(port, &addr); 839 return error; 840 } 841 #endif /* INET */ 842 843 #ifdef INET6 844 static int 845 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 846 { 847 struct inpcb *inp = NULL; 848 int error = 0; 849 struct tcpcb *tp = NULL; 850 struct in_addr addr; 851 struct in6_addr addr6; 852 struct epoch_tracker et; 853 in_port_t port = 0; 854 int v4 = 0; 855 TCPDEBUG0; 856 857 if (so->so_state & SS_ISDISCONNECTED) 858 return (ECONNABORTED); 859 860 inp = sotoinpcb(so); 861 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 862 NET_EPOCH_ENTER(et); 863 INP_WLOCK(inp); 864 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 865 error = ECONNABORTED; 866 goto out; 867 } 868 tp = intotcpcb(inp); 869 TCPDEBUG1(); 870 871 /* 872 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 873 * copy the data of interest and defer the malloc until after we 874 * release the lock. 875 */ 876 if (inp->inp_vflag & INP_IPV4) { 877 v4 = 1; 878 port = inp->inp_fport; 879 addr = inp->inp_faddr; 880 } else { 881 port = inp->inp_fport; 882 addr6 = inp->in6p_faddr; 883 } 884 885 out: 886 TCPDEBUG2(PRU_ACCEPT); 887 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 888 INP_WUNLOCK(inp); 889 NET_EPOCH_EXIT(et); 890 if (error == 0) { 891 if (v4) 892 *nam = in6_v4mapsin6_sockaddr(port, &addr); 893 else 894 *nam = in6_sockaddr(port, &addr6); 895 } 896 return error; 897 } 898 #endif /* INET6 */ 899 900 /* 901 * Mark the connection as being incapable of further output. 902 */ 903 static int 904 tcp_usr_shutdown(struct socket *so) 905 { 906 int error = 0; 907 struct inpcb *inp; 908 struct tcpcb *tp = NULL; 909 struct epoch_tracker et; 910 911 TCPDEBUG0; 912 NET_EPOCH_ENTER(et); 913 inp = sotoinpcb(so); 914 KASSERT(inp != NULL, ("inp == NULL")); 915 INP_WLOCK(inp); 916 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 917 error = ECONNRESET; 918 goto out; 919 } 920 tp = intotcpcb(inp); 921 TCPDEBUG1(); 922 socantsendmore(so); 923 tcp_usrclosed(tp); 924 if (!(inp->inp_flags & INP_DROPPED)) 925 error = tp->t_fb->tfb_tcp_output(tp); 926 927 out: 928 TCPDEBUG2(PRU_SHUTDOWN); 929 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 930 INP_WUNLOCK(inp); 931 NET_EPOCH_EXIT(et); 932 933 return (error); 934 } 935 936 /* 937 * After a receive, possibly send window update to peer. 938 */ 939 static int 940 tcp_usr_rcvd(struct socket *so, int flags) 941 { 942 struct epoch_tracker et; 943 struct inpcb *inp; 944 struct tcpcb *tp = NULL; 945 int error = 0; 946 947 TCPDEBUG0; 948 inp = sotoinpcb(so); 949 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 950 INP_WLOCK(inp); 951 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 952 error = ECONNRESET; 953 goto out; 954 } 955 tp = intotcpcb(inp); 956 TCPDEBUG1(); 957 /* 958 * For passively-created TFO connections, don't attempt a window 959 * update while still in SYN_RECEIVED as this may trigger an early 960 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 961 * application response data, or failing that, when the DELACK timer 962 * expires. 963 */ 964 if (IS_FASTOPEN(tp->t_flags) && 965 (tp->t_state == TCPS_SYN_RECEIVED)) 966 goto out; 967 NET_EPOCH_ENTER(et); 968 #ifdef TCP_OFFLOAD 969 if (tp->t_flags & TF_TOE) 970 tcp_offload_rcvd(tp); 971 else 972 #endif 973 tp->t_fb->tfb_tcp_output(tp); 974 NET_EPOCH_EXIT(et); 975 out: 976 TCPDEBUG2(PRU_RCVD); 977 TCP_PROBE2(debug__user, tp, PRU_RCVD); 978 INP_WUNLOCK(inp); 979 return (error); 980 } 981 982 /* 983 * Do a send by putting data in output queue and updating urgent 984 * marker if URG set. Possibly send more data. Unlike the other 985 * pru_*() routines, the mbuf chains are our responsibility. We 986 * must either enqueue them or free them. The other pru_* routines 987 * generally are caller-frees. 988 */ 989 static int 990 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 991 struct sockaddr *nam, struct mbuf *control, struct thread *td) 992 { 993 struct epoch_tracker et; 994 int error = 0; 995 struct inpcb *inp; 996 struct tcpcb *tp = NULL; 997 #ifdef INET 998 #ifdef INET6 999 struct sockaddr_in sin; 1000 #endif 1001 struct sockaddr_in *sinp; 1002 #endif 1003 #ifdef INET6 1004 int isipv6; 1005 #endif 1006 u_int8_t incflagsav; 1007 u_char vflagsav; 1008 bool restoreflags; 1009 TCPDEBUG0; 1010 1011 /* 1012 * We require the pcbinfo "read lock" if we will close the socket 1013 * as part of this call. 1014 */ 1015 NET_EPOCH_ENTER(et); 1016 inp = sotoinpcb(so); 1017 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 1018 INP_WLOCK(inp); 1019 vflagsav = inp->inp_vflag; 1020 incflagsav = inp->inp_inc.inc_flags; 1021 restoreflags = false; 1022 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1023 if (control) 1024 m_freem(control); 1025 error = ECONNRESET; 1026 goto out; 1027 } 1028 if (control != NULL) { 1029 /* TCP doesn't do control messages (rights, creds, etc) */ 1030 if (control->m_len) { 1031 m_freem(control); 1032 error = EINVAL; 1033 goto out; 1034 } 1035 m_freem(control); /* empty control, just free it */ 1036 control = NULL; 1037 } 1038 tp = intotcpcb(inp); 1039 if ((flags & PRUS_OOB) != 0 && 1040 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 1041 goto out; 1042 1043 TCPDEBUG1(); 1044 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 1045 if (tp->t_state == TCPS_LISTEN) { 1046 error = EINVAL; 1047 goto out; 1048 } 1049 switch (nam->sa_family) { 1050 #ifdef INET 1051 case AF_INET: 1052 sinp = (struct sockaddr_in *)nam; 1053 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 1054 error = EINVAL; 1055 goto out; 1056 } 1057 if ((inp->inp_vflag & INP_IPV6) != 0) { 1058 error = EAFNOSUPPORT; 1059 goto out; 1060 } 1061 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 1062 error = EAFNOSUPPORT; 1063 goto out; 1064 } 1065 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 1066 error = EACCES; 1067 goto out; 1068 } 1069 if ((error = prison_remote_ip4(td->td_ucred, 1070 &sinp->sin_addr))) 1071 goto out; 1072 #ifdef INET6 1073 isipv6 = 0; 1074 #endif 1075 break; 1076 #endif /* INET */ 1077 #ifdef INET6 1078 case AF_INET6: 1079 { 1080 struct sockaddr_in6 *sin6; 1081 1082 sin6 = (struct sockaddr_in6 *)nam; 1083 if (sin6->sin6_len != sizeof(*sin6)) { 1084 error = EINVAL; 1085 goto out; 1086 } 1087 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1088 error = EAFNOSUPPORT; 1089 goto out; 1090 } 1091 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1092 error = EAFNOSUPPORT; 1093 goto out; 1094 } 1095 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1096 #ifdef INET 1097 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1098 error = EINVAL; 1099 goto out; 1100 } 1101 if ((inp->inp_vflag & INP_IPV4) == 0) { 1102 error = EAFNOSUPPORT; 1103 goto out; 1104 } 1105 restoreflags = true; 1106 inp->inp_vflag &= ~INP_IPV6; 1107 sinp = &sin; 1108 in6_sin6_2_sin(sinp, sin6); 1109 if (IN_MULTICAST( 1110 ntohl(sinp->sin_addr.s_addr))) { 1111 error = EAFNOSUPPORT; 1112 goto out; 1113 } 1114 if ((error = prison_remote_ip4(td->td_ucred, 1115 &sinp->sin_addr))) 1116 goto out; 1117 isipv6 = 0; 1118 #else /* !INET */ 1119 error = EAFNOSUPPORT; 1120 goto out; 1121 #endif /* INET */ 1122 } else { 1123 if ((inp->inp_vflag & INP_IPV6) == 0) { 1124 error = EAFNOSUPPORT; 1125 goto out; 1126 } 1127 restoreflags = true; 1128 inp->inp_vflag &= ~INP_IPV4; 1129 inp->inp_inc.inc_flags |= INC_ISIPV6; 1130 if ((error = prison_remote_ip6(td->td_ucred, 1131 &sin6->sin6_addr))) 1132 goto out; 1133 isipv6 = 1; 1134 } 1135 break; 1136 } 1137 #endif /* INET6 */ 1138 default: 1139 error = EAFNOSUPPORT; 1140 goto out; 1141 } 1142 } 1143 if (!(flags & PRUS_OOB)) { 1144 sbappendstream(&so->so_snd, m, flags); 1145 m = NULL; 1146 if (nam && tp->t_state < TCPS_SYN_SENT) { 1147 KASSERT(tp->t_state == TCPS_CLOSED, 1148 ("%s: tp %p is listening", __func__, tp)); 1149 1150 /* 1151 * Do implied connect if not yet connected, 1152 * initialize window to default value, and 1153 * initialize maxseg using peer's cached MSS. 1154 */ 1155 #ifdef INET6 1156 if (isipv6) 1157 error = tcp6_connect(tp, nam, td); 1158 #endif /* INET6 */ 1159 #if defined(INET6) && defined(INET) 1160 else 1161 #endif 1162 #ifdef INET 1163 error = tcp_connect(tp, 1164 (struct sockaddr *)sinp, td); 1165 #endif 1166 /* 1167 * The bind operation in tcp_connect succeeded. We 1168 * no longer want to restore the flags if later 1169 * operations fail. 1170 */ 1171 if (error == 0 || inp->inp_lport != 0) 1172 restoreflags = false; 1173 1174 if (error) { 1175 /* m is freed if PRUS_NOTREADY is unset. */ 1176 sbflush(&so->so_snd); 1177 goto out; 1178 } 1179 if (IS_FASTOPEN(tp->t_flags)) 1180 tcp_fastopen_connect(tp); 1181 else { 1182 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1183 tcp_mss(tp, -1); 1184 } 1185 } 1186 if (flags & PRUS_EOF) { 1187 /* 1188 * Close the send side of the connection after 1189 * the data is sent. 1190 */ 1191 socantsendmore(so); 1192 tcp_usrclosed(tp); 1193 } 1194 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1195 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1196 (tp->t_fbyte_out == 0) && 1197 (so->so_snd.sb_ccc > 0)) { 1198 tp->t_fbyte_out = ticks; 1199 if (tp->t_fbyte_out == 0) 1200 tp->t_fbyte_out = 1; 1201 if (tp->t_fbyte_out && tp->t_fbyte_in) 1202 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1203 } 1204 if (!(inp->inp_flags & INP_DROPPED) && 1205 !(flags & PRUS_NOTREADY)) { 1206 if (flags & PRUS_MORETOCOME) 1207 tp->t_flags |= TF_MORETOCOME; 1208 error = tp->t_fb->tfb_tcp_output(tp); 1209 if (flags & PRUS_MORETOCOME) 1210 tp->t_flags &= ~TF_MORETOCOME; 1211 } 1212 } else { 1213 /* 1214 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1215 */ 1216 SOCKBUF_LOCK(&so->so_snd); 1217 if (sbspace(&so->so_snd) < -512) { 1218 SOCKBUF_UNLOCK(&so->so_snd); 1219 error = ENOBUFS; 1220 goto out; 1221 } 1222 /* 1223 * According to RFC961 (Assigned Protocols), 1224 * the urgent pointer points to the last octet 1225 * of urgent data. We continue, however, 1226 * to consider it to indicate the first octet 1227 * of data past the urgent section. 1228 * Otherwise, snd_up should be one lower. 1229 */ 1230 sbappendstream_locked(&so->so_snd, m, flags); 1231 SOCKBUF_UNLOCK(&so->so_snd); 1232 m = NULL; 1233 if (nam && tp->t_state < TCPS_SYN_SENT) { 1234 /* 1235 * Do implied connect if not yet connected, 1236 * initialize window to default value, and 1237 * initialize maxseg using peer's cached MSS. 1238 */ 1239 1240 /* 1241 * Not going to contemplate SYN|URG 1242 */ 1243 if (IS_FASTOPEN(tp->t_flags)) 1244 tp->t_flags &= ~TF_FASTOPEN; 1245 #ifdef INET6 1246 if (isipv6) 1247 error = tcp6_connect(tp, nam, td); 1248 #endif /* INET6 */ 1249 #if defined(INET6) && defined(INET) 1250 else 1251 #endif 1252 #ifdef INET 1253 error = tcp_connect(tp, 1254 (struct sockaddr *)sinp, td); 1255 #endif 1256 /* 1257 * The bind operation in tcp_connect succeeded. We 1258 * no longer want to restore the flags if later 1259 * operations fail. 1260 */ 1261 if (error == 0 || inp->inp_lport != 0) 1262 restoreflags = false; 1263 1264 if (error != 0) { 1265 /* m is freed if PRUS_NOTREADY is unset. */ 1266 sbflush(&so->so_snd); 1267 goto out; 1268 } 1269 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1270 tcp_mss(tp, -1); 1271 } 1272 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1273 if ((flags & PRUS_NOTREADY) == 0) { 1274 tp->t_flags |= TF_FORCEDATA; 1275 error = tp->t_fb->tfb_tcp_output(tp); 1276 tp->t_flags &= ~TF_FORCEDATA; 1277 } 1278 } 1279 TCP_LOG_EVENT(tp, NULL, 1280 &inp->inp_socket->so_rcv, 1281 &inp->inp_socket->so_snd, 1282 TCP_LOG_USERSEND, error, 1283 0, NULL, false); 1284 1285 out: 1286 /* 1287 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1288 * responsible for freeing memory. 1289 */ 1290 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1291 m_freem(m); 1292 1293 /* 1294 * If the request was unsuccessful and we changed flags, 1295 * restore the original flags. 1296 */ 1297 if (error != 0 && restoreflags) { 1298 inp->inp_vflag = vflagsav; 1299 inp->inp_inc.inc_flags = incflagsav; 1300 } 1301 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1302 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1303 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1304 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1305 INP_WUNLOCK(inp); 1306 NET_EPOCH_EXIT(et); 1307 return (error); 1308 } 1309 1310 static int 1311 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1312 { 1313 struct epoch_tracker et; 1314 struct inpcb *inp; 1315 struct tcpcb *tp; 1316 int error; 1317 1318 inp = sotoinpcb(so); 1319 INP_WLOCK(inp); 1320 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1321 INP_WUNLOCK(inp); 1322 mb_free_notready(m, count); 1323 return (ECONNRESET); 1324 } 1325 tp = intotcpcb(inp); 1326 1327 SOCKBUF_LOCK(&so->so_snd); 1328 error = sbready(&so->so_snd, m, count); 1329 SOCKBUF_UNLOCK(&so->so_snd); 1330 if (error == 0) { 1331 NET_EPOCH_ENTER(et); 1332 error = tp->t_fb->tfb_tcp_output(tp); 1333 NET_EPOCH_EXIT(et); 1334 } 1335 INP_WUNLOCK(inp); 1336 1337 return (error); 1338 } 1339 1340 /* 1341 * Abort the TCP. Drop the connection abruptly. 1342 */ 1343 static void 1344 tcp_usr_abort(struct socket *so) 1345 { 1346 struct inpcb *inp; 1347 struct tcpcb *tp = NULL; 1348 struct epoch_tracker et; 1349 TCPDEBUG0; 1350 1351 inp = sotoinpcb(so); 1352 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1353 1354 NET_EPOCH_ENTER(et); 1355 INP_WLOCK(inp); 1356 KASSERT(inp->inp_socket != NULL, 1357 ("tcp_usr_abort: inp_socket == NULL")); 1358 1359 /* 1360 * If we still have full TCP state, and we're not dropped, drop. 1361 */ 1362 if (!(inp->inp_flags & INP_TIMEWAIT) && 1363 !(inp->inp_flags & INP_DROPPED)) { 1364 tp = intotcpcb(inp); 1365 TCPDEBUG1(); 1366 tp = tcp_drop(tp, ECONNABORTED); 1367 if (tp == NULL) 1368 goto dropped; 1369 TCPDEBUG2(PRU_ABORT); 1370 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1371 } 1372 if (!(inp->inp_flags & INP_DROPPED)) { 1373 SOCK_LOCK(so); 1374 so->so_state |= SS_PROTOREF; 1375 SOCK_UNLOCK(so); 1376 inp->inp_flags |= INP_SOCKREF; 1377 } 1378 INP_WUNLOCK(inp); 1379 dropped: 1380 NET_EPOCH_EXIT(et); 1381 } 1382 1383 /* 1384 * TCP socket is closed. Start friendly disconnect. 1385 */ 1386 static void 1387 tcp_usr_close(struct socket *so) 1388 { 1389 struct inpcb *inp; 1390 struct tcpcb *tp = NULL; 1391 struct epoch_tracker et; 1392 TCPDEBUG0; 1393 1394 inp = sotoinpcb(so); 1395 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1396 1397 NET_EPOCH_ENTER(et); 1398 INP_WLOCK(inp); 1399 KASSERT(inp->inp_socket != NULL, 1400 ("tcp_usr_close: inp_socket == NULL")); 1401 1402 /* 1403 * If we still have full TCP state, and we're not dropped, initiate 1404 * a disconnect. 1405 */ 1406 if (!(inp->inp_flags & INP_TIMEWAIT) && 1407 !(inp->inp_flags & INP_DROPPED)) { 1408 tp = intotcpcb(inp); 1409 TCPDEBUG1(); 1410 tcp_disconnect(tp); 1411 TCPDEBUG2(PRU_CLOSE); 1412 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1413 } 1414 if (!(inp->inp_flags & INP_DROPPED)) { 1415 SOCK_LOCK(so); 1416 so->so_state |= SS_PROTOREF; 1417 SOCK_UNLOCK(so); 1418 inp->inp_flags |= INP_SOCKREF; 1419 } 1420 INP_WUNLOCK(inp); 1421 NET_EPOCH_EXIT(et); 1422 } 1423 1424 static int 1425 tcp_pru_options_support(struct tcpcb *tp, int flags) 1426 { 1427 /* 1428 * If the specific TCP stack has a pru_options 1429 * specified then it does not always support 1430 * all the PRU_XX options and we must ask it. 1431 * If the function is not specified then all 1432 * of the PRU_XX options are supported. 1433 */ 1434 int ret = 0; 1435 1436 if (tp->t_fb->tfb_pru_options) { 1437 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1438 } 1439 return (ret); 1440 } 1441 1442 /* 1443 * Receive out-of-band data. 1444 */ 1445 static int 1446 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1447 { 1448 int error = 0; 1449 struct inpcb *inp; 1450 struct tcpcb *tp = NULL; 1451 1452 TCPDEBUG0; 1453 inp = sotoinpcb(so); 1454 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1455 INP_WLOCK(inp); 1456 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1457 error = ECONNRESET; 1458 goto out; 1459 } 1460 tp = intotcpcb(inp); 1461 error = tcp_pru_options_support(tp, PRUS_OOB); 1462 if (error) { 1463 goto out; 1464 } 1465 TCPDEBUG1(); 1466 if ((so->so_oobmark == 0 && 1467 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1468 so->so_options & SO_OOBINLINE || 1469 tp->t_oobflags & TCPOOB_HADDATA) { 1470 error = EINVAL; 1471 goto out; 1472 } 1473 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1474 error = EWOULDBLOCK; 1475 goto out; 1476 } 1477 m->m_len = 1; 1478 *mtod(m, caddr_t) = tp->t_iobc; 1479 if ((flags & MSG_PEEK) == 0) 1480 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1481 1482 out: 1483 TCPDEBUG2(PRU_RCVOOB); 1484 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1485 INP_WUNLOCK(inp); 1486 return (error); 1487 } 1488 1489 #ifdef INET 1490 struct pr_usrreqs tcp_usrreqs = { 1491 .pru_abort = tcp_usr_abort, 1492 .pru_accept = tcp_usr_accept, 1493 .pru_attach = tcp_usr_attach, 1494 .pru_bind = tcp_usr_bind, 1495 .pru_connect = tcp_usr_connect, 1496 .pru_control = in_control, 1497 .pru_detach = tcp_usr_detach, 1498 .pru_disconnect = tcp_usr_disconnect, 1499 .pru_listen = tcp_usr_listen, 1500 .pru_peeraddr = in_getpeeraddr, 1501 .pru_rcvd = tcp_usr_rcvd, 1502 .pru_rcvoob = tcp_usr_rcvoob, 1503 .pru_send = tcp_usr_send, 1504 .pru_ready = tcp_usr_ready, 1505 .pru_shutdown = tcp_usr_shutdown, 1506 .pru_sockaddr = in_getsockaddr, 1507 .pru_sosetlabel = in_pcbsosetlabel, 1508 .pru_close = tcp_usr_close, 1509 }; 1510 #endif /* INET */ 1511 1512 #ifdef INET6 1513 struct pr_usrreqs tcp6_usrreqs = { 1514 .pru_abort = tcp_usr_abort, 1515 .pru_accept = tcp6_usr_accept, 1516 .pru_attach = tcp_usr_attach, 1517 .pru_bind = tcp6_usr_bind, 1518 .pru_connect = tcp6_usr_connect, 1519 .pru_control = in6_control, 1520 .pru_detach = tcp_usr_detach, 1521 .pru_disconnect = tcp_usr_disconnect, 1522 .pru_listen = tcp6_usr_listen, 1523 .pru_peeraddr = in6_mapped_peeraddr, 1524 .pru_rcvd = tcp_usr_rcvd, 1525 .pru_rcvoob = tcp_usr_rcvoob, 1526 .pru_send = tcp_usr_send, 1527 .pru_ready = tcp_usr_ready, 1528 .pru_shutdown = tcp_usr_shutdown, 1529 .pru_sockaddr = in6_mapped_sockaddr, 1530 .pru_sosetlabel = in_pcbsosetlabel, 1531 .pru_close = tcp_usr_close, 1532 }; 1533 #endif /* INET6 */ 1534 1535 #ifdef INET 1536 /* 1537 * Common subroutine to open a TCP connection to remote host specified 1538 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1539 * port number if needed. Call in_pcbconnect_setup to do the routing and 1540 * to choose a local host address (interface). If there is an existing 1541 * incarnation of the same connection in TIME-WAIT state and if the remote 1542 * host was sending CC options and if the connection duration was < MSL, then 1543 * truncate the previous TIME-WAIT state and proceed. 1544 * Initialize connection parameters and enter SYN-SENT state. 1545 */ 1546 static int 1547 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1548 { 1549 struct inpcb *inp = tp->t_inpcb, *oinp; 1550 struct socket *so = inp->inp_socket; 1551 struct in_addr laddr; 1552 u_short lport; 1553 int error; 1554 1555 NET_EPOCH_ASSERT(); 1556 INP_WLOCK_ASSERT(inp); 1557 INP_HASH_WLOCK(&V_tcbinfo); 1558 1559 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1560 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1561 if (error) 1562 goto out; 1563 } 1564 1565 /* 1566 * Cannot simply call in_pcbconnect, because there might be an 1567 * earlier incarnation of this same connection still in 1568 * TIME_WAIT state, creating an ADDRINUSE error. 1569 */ 1570 laddr = inp->inp_laddr; 1571 lport = inp->inp_lport; 1572 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1573 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1574 if (error && oinp == NULL) 1575 goto out; 1576 if (oinp) { 1577 error = EADDRINUSE; 1578 goto out; 1579 } 1580 /* Handle initial bind if it hadn't been done in advance. */ 1581 if (inp->inp_lport == 0) { 1582 inp->inp_lport = lport; 1583 if (in_pcbinshash(inp) != 0) { 1584 inp->inp_lport = 0; 1585 error = EAGAIN; 1586 goto out; 1587 } 1588 } 1589 inp->inp_laddr = laddr; 1590 in_pcbrehash(inp); 1591 INP_HASH_WUNLOCK(&V_tcbinfo); 1592 1593 /* 1594 * Compute window scaling to request: 1595 * Scale to fit into sweet spot. See tcp_syncache.c. 1596 * XXX: This should move to tcp_output(). 1597 */ 1598 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1599 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1600 tp->request_r_scale++; 1601 1602 soisconnecting(so); 1603 TCPSTAT_INC(tcps_connattempt); 1604 tcp_state_change(tp, TCPS_SYN_SENT); 1605 tp->iss = tcp_new_isn(&inp->inp_inc); 1606 if (tp->t_flags & TF_REQ_TSTMP) 1607 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1608 tcp_sendseqinit(tp); 1609 1610 return 0; 1611 1612 out: 1613 INP_HASH_WUNLOCK(&V_tcbinfo); 1614 return (error); 1615 } 1616 #endif /* INET */ 1617 1618 #ifdef INET6 1619 static int 1620 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1621 { 1622 struct inpcb *inp = tp->t_inpcb; 1623 int error; 1624 1625 INP_WLOCK_ASSERT(inp); 1626 INP_HASH_WLOCK(&V_tcbinfo); 1627 1628 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1629 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1630 if (error) 1631 goto out; 1632 } 1633 error = in6_pcbconnect(inp, nam, td->td_ucred); 1634 if (error != 0) 1635 goto out; 1636 INP_HASH_WUNLOCK(&V_tcbinfo); 1637 1638 /* Compute window scaling to request. */ 1639 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1640 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1641 tp->request_r_scale++; 1642 1643 soisconnecting(inp->inp_socket); 1644 TCPSTAT_INC(tcps_connattempt); 1645 tcp_state_change(tp, TCPS_SYN_SENT); 1646 tp->iss = tcp_new_isn(&inp->inp_inc); 1647 if (tp->t_flags & TF_REQ_TSTMP) 1648 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1649 tcp_sendseqinit(tp); 1650 1651 return 0; 1652 1653 out: 1654 INP_HASH_WUNLOCK(&V_tcbinfo); 1655 return error; 1656 } 1657 #endif /* INET6 */ 1658 1659 /* 1660 * Export TCP internal state information via a struct tcp_info, based on the 1661 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1662 * (TCP state machine, etc). We export all information using FreeBSD-native 1663 * constants -- for example, the numeric values for tcpi_state will differ 1664 * from Linux. 1665 */ 1666 static void 1667 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1668 { 1669 1670 INP_WLOCK_ASSERT(tp->t_inpcb); 1671 bzero(ti, sizeof(*ti)); 1672 1673 ti->tcpi_state = tp->t_state; 1674 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1675 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1676 if (tp->t_flags & TF_SACK_PERMIT) 1677 ti->tcpi_options |= TCPI_OPT_SACK; 1678 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1679 ti->tcpi_options |= TCPI_OPT_WSCALE; 1680 ti->tcpi_snd_wscale = tp->snd_scale; 1681 ti->tcpi_rcv_wscale = tp->rcv_scale; 1682 } 1683 if (tp->t_flags2 & TF2_ECN_PERMIT) 1684 ti->tcpi_options |= TCPI_OPT_ECN; 1685 1686 ti->tcpi_rto = tp->t_rxtcur * tick; 1687 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1688 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1689 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1690 1691 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1692 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1693 1694 /* 1695 * FreeBSD-specific extension fields for tcp_info. 1696 */ 1697 ti->tcpi_rcv_space = tp->rcv_wnd; 1698 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1699 ti->tcpi_snd_wnd = tp->snd_wnd; 1700 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1701 ti->tcpi_snd_nxt = tp->snd_nxt; 1702 ti->tcpi_snd_mss = tp->t_maxseg; 1703 ti->tcpi_rcv_mss = tp->t_maxseg; 1704 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1705 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1706 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1707 #ifdef TCP_OFFLOAD 1708 if (tp->t_flags & TF_TOE) { 1709 ti->tcpi_options |= TCPI_OPT_TOE; 1710 tcp_offload_tcp_info(tp, ti); 1711 } 1712 #endif 1713 } 1714 1715 /* 1716 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1717 * socket option arguments. When it re-acquires the lock after the copy, it 1718 * has to revalidate that the connection is still valid for the socket 1719 * option. 1720 */ 1721 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1722 INP_WLOCK(inp); \ 1723 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1724 INP_WUNLOCK(inp); \ 1725 cleanup; \ 1726 return (ECONNRESET); \ 1727 } \ 1728 tp = intotcpcb(inp); \ 1729 } while(0) 1730 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1731 1732 static int 1733 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1734 { 1735 struct tcpcb *tp = intotcpcb(inp); 1736 int error = 0; 1737 1738 MPASS(sopt->sopt_dir == SOPT_SET); 1739 1740 if (sopt->sopt_level != IPPROTO_TCP) { 1741 #ifdef INET6 1742 if (inp->inp_vflag & INP_IPV6PROTO) 1743 error = ip6_ctloutput(inp->inp_socket, sopt); 1744 #endif 1745 #if defined(INET6) && defined(INET) 1746 else 1747 #endif 1748 #ifdef INET 1749 error = ip_ctloutput(inp->inp_socket, sopt); 1750 #endif 1751 /* 1752 * When an IP-level socket option affects TCP, pass control 1753 * down to stack tfb_tcp_ctloutput, otherwise return what 1754 * IP level returned. 1755 */ 1756 switch (sopt->sopt_level) { 1757 #ifdef INET6 1758 case IPPROTO_IPV6: 1759 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1760 return (error); 1761 switch (sopt->sopt_name) { 1762 case IPV6_TCLASS: 1763 /* Notify tcp stacks that care (e.g. RACK). */ 1764 break; 1765 case IPV6_USE_MIN_MTU: 1766 /* Update t_maxseg accordingly. */ 1767 break; 1768 default: 1769 return (error); 1770 } 1771 break; 1772 #endif 1773 #ifdef INET 1774 case IPPROTO_IP: 1775 switch (sopt->sopt_name) { 1776 case IP_TOS: 1777 case IP_TTL: 1778 /* Notify tcp stacks that care (e.g. RACK). */ 1779 break; 1780 default: 1781 return (error); 1782 } 1783 break; 1784 #endif 1785 default: 1786 return (error); 1787 } 1788 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1789 /* 1790 * Protect the TCP option TCP_FUNCTION_BLK so 1791 * that a sub-function can *never* overwrite this. 1792 */ 1793 struct tcp_function_set fsn; 1794 struct tcp_function_block *blk; 1795 1796 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1797 if (error) 1798 return (error); 1799 1800 INP_WLOCK(inp); 1801 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1802 INP_WUNLOCK(inp); 1803 return (ECONNRESET); 1804 } 1805 tp = intotcpcb(inp); 1806 1807 blk = find_and_ref_tcp_functions(&fsn); 1808 if (blk == NULL) { 1809 INP_WUNLOCK(inp); 1810 return (ENOENT); 1811 } 1812 if (tp->t_fb == blk) { 1813 /* You already have this */ 1814 refcount_release(&blk->tfb_refcnt); 1815 INP_WUNLOCK(inp); 1816 return (0); 1817 } 1818 if (tp->t_state != TCPS_CLOSED) { 1819 /* 1820 * The user has advanced the state 1821 * past the initial point, we may not 1822 * be able to switch. 1823 */ 1824 if (blk->tfb_tcp_handoff_ok != NULL) { 1825 /* 1826 * Does the stack provide a 1827 * query mechanism, if so it may 1828 * still be possible? 1829 */ 1830 error = (*blk->tfb_tcp_handoff_ok)(tp); 1831 } else 1832 error = EINVAL; 1833 if (error) { 1834 refcount_release(&blk->tfb_refcnt); 1835 INP_WUNLOCK(inp); 1836 return(error); 1837 } 1838 } 1839 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1840 refcount_release(&blk->tfb_refcnt); 1841 INP_WUNLOCK(inp); 1842 return (ENOENT); 1843 } 1844 /* 1845 * Release the old refcnt, the 1846 * lookup acquired a ref on the 1847 * new one already. 1848 */ 1849 if (tp->t_fb->tfb_tcp_fb_fini) { 1850 struct epoch_tracker et; 1851 /* 1852 * Tell the stack to cleanup with 0 i.e. 1853 * the tcb is not going away. 1854 */ 1855 NET_EPOCH_ENTER(et); 1856 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1857 NET_EPOCH_EXIT(et); 1858 } 1859 #ifdef TCPHPTS 1860 /* Assure that we are not on any hpts */ 1861 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL); 1862 #endif 1863 if (blk->tfb_tcp_fb_init) { 1864 error = (*blk->tfb_tcp_fb_init)(tp); 1865 if (error) { 1866 refcount_release(&blk->tfb_refcnt); 1867 if (tp->t_fb->tfb_tcp_fb_init) { 1868 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1869 /* Fall back failed, drop the connection */ 1870 INP_WUNLOCK(inp); 1871 soabort(inp->inp_socket); 1872 return(error); 1873 } 1874 } 1875 goto err_out; 1876 } 1877 } 1878 refcount_release(&tp->t_fb->tfb_refcnt); 1879 tp->t_fb = blk; 1880 #ifdef TCP_OFFLOAD 1881 if (tp->t_flags & TF_TOE) { 1882 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1883 sopt->sopt_name); 1884 } 1885 #endif 1886 err_out: 1887 INP_WUNLOCK(inp); 1888 return (error); 1889 } 1890 1891 INP_WLOCK(inp); 1892 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1893 INP_WUNLOCK(inp); 1894 return (ECONNRESET); 1895 } 1896 tp = intotcpcb(inp); 1897 1898 /* Pass in the INP locked, caller must unlock it. */ 1899 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp)); 1900 } 1901 1902 static int 1903 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1904 { 1905 int error = 0; 1906 struct tcpcb *tp; 1907 1908 MPASS(sopt->sopt_dir == SOPT_GET); 1909 1910 if (sopt->sopt_level != IPPROTO_TCP) { 1911 #ifdef INET6 1912 if (inp->inp_vflag & INP_IPV6PROTO) 1913 error = ip6_ctloutput(inp->inp_socket, sopt); 1914 #endif /* INET6 */ 1915 #if defined(INET6) && defined(INET) 1916 else 1917 #endif 1918 #ifdef INET 1919 error = ip_ctloutput(inp->inp_socket, sopt); 1920 #endif 1921 return (error); 1922 } 1923 INP_WLOCK(inp); 1924 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1925 INP_WUNLOCK(inp); 1926 return (ECONNRESET); 1927 } 1928 tp = intotcpcb(inp); 1929 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1930 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1931 struct tcp_function_set fsn; 1932 1933 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1934 memset(&fsn, 0, sizeof(fsn)); 1935 find_tcp_function_alias(tp->t_fb, &fsn); 1936 } else { 1937 strncpy(fsn.function_set_name, 1938 tp->t_fb->tfb_tcp_block_name, 1939 TCP_FUNCTION_NAME_LEN_MAX); 1940 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1941 } 1942 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1943 INP_WUNLOCK(inp); 1944 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1945 return (error); 1946 } 1947 1948 /* Pass in the INP locked, caller must unlock it. */ 1949 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp)); 1950 } 1951 1952 int 1953 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1954 { 1955 int error; 1956 struct inpcb *inp; 1957 1958 error = 0; 1959 inp = sotoinpcb(so); 1960 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1961 1962 if (sopt->sopt_dir == SOPT_SET) 1963 return (tcp_ctloutput_set(inp, sopt)); 1964 else if (sopt->sopt_dir == SOPT_GET) 1965 return (tcp_ctloutput_get(inp, sopt)); 1966 else 1967 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1968 } 1969 1970 /* 1971 * If this assert becomes untrue, we need to change the size of the buf 1972 * variable in tcp_default_ctloutput(). 1973 */ 1974 #ifdef CTASSERT 1975 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1976 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1977 #endif 1978 1979 #ifdef KERN_TLS 1980 static int 1981 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1982 { 1983 struct tls_enable_v0 tls_v0; 1984 int error; 1985 1986 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1987 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1988 sizeof(tls_v0)); 1989 if (error) 1990 return (error); 1991 memset(tls, 0, sizeof(*tls)); 1992 tls->cipher_key = tls_v0.cipher_key; 1993 tls->iv = tls_v0.iv; 1994 tls->auth_key = tls_v0.auth_key; 1995 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1996 tls->cipher_key_len = tls_v0.cipher_key_len; 1997 tls->iv_len = tls_v0.iv_len; 1998 tls->auth_algorithm = tls_v0.auth_algorithm; 1999 tls->auth_key_len = tls_v0.auth_key_len; 2000 tls->flags = tls_v0.flags; 2001 tls->tls_vmajor = tls_v0.tls_vmajor; 2002 tls->tls_vminor = tls_v0.tls_vminor; 2003 return (0); 2004 } 2005 2006 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 2007 } 2008 #endif 2009 2010 extern struct cc_algo newreno_cc_algo; 2011 2012 static int 2013 tcp_congestion(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 2014 { 2015 struct cc_algo *algo; 2016 void *ptr = NULL; 2017 struct cc_var cc_mem; 2018 char buf[TCP_CA_NAME_MAX]; 2019 size_t mem_sz; 2020 int error; 2021 2022 INP_WUNLOCK(inp); 2023 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 2024 if (error) 2025 return(error); 2026 buf[sopt->sopt_valsize] = '\0'; 2027 CC_LIST_RLOCK(); 2028 STAILQ_FOREACH(algo, &cc_list, entries) 2029 if (strncmp(buf, algo->name, 2030 TCP_CA_NAME_MAX) == 0) { 2031 if (algo->flags & CC_MODULE_BEING_REMOVED) { 2032 /* We can't "see" modules being unloaded */ 2033 continue; 2034 } 2035 break; 2036 } 2037 if (algo == NULL) { 2038 CC_LIST_RUNLOCK(); 2039 return(ESRCH); 2040 } 2041 do_over: 2042 if (algo->cb_init != NULL) { 2043 /* We can now pre-get the memory for the CC */ 2044 mem_sz = (*algo->cc_data_sz)(); 2045 if (mem_sz == 0) { 2046 goto no_mem_needed; 2047 } 2048 CC_LIST_RUNLOCK(); 2049 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 2050 CC_LIST_RLOCK(); 2051 STAILQ_FOREACH(algo, &cc_list, entries) 2052 if (strncmp(buf, algo->name, 2053 TCP_CA_NAME_MAX) == 0) 2054 break; 2055 if (algo == NULL) { 2056 if (ptr) 2057 free(ptr, M_CC_MEM); 2058 CC_LIST_RUNLOCK(); 2059 return(ESRCH); 2060 } 2061 } else { 2062 no_mem_needed: 2063 mem_sz = 0; 2064 ptr = NULL; 2065 } 2066 /* 2067 * Make sure its all clean and zero and also get 2068 * back the inplock. 2069 */ 2070 memset(&cc_mem, 0, sizeof(cc_mem)); 2071 if (mem_sz != (*algo->cc_data_sz)()) { 2072 if (ptr) 2073 free(ptr, M_CC_MEM); 2074 goto do_over; 2075 } 2076 if (ptr) { 2077 memset(ptr, 0, mem_sz); 2078 INP_WLOCK_RECHECK_CLEANUP(inp, free(ptr, M_CC_MEM)); 2079 } else 2080 INP_WLOCK_RECHECK(inp); 2081 CC_LIST_RUNLOCK(); 2082 cc_mem.ccvc.tcp = tp; 2083 /* 2084 * We once again hold a write lock over the tcb so it's 2085 * safe to do these things without ordering concerns. 2086 * Note here we init into stack memory. 2087 */ 2088 if (algo->cb_init != NULL) 2089 error = algo->cb_init(&cc_mem, ptr); 2090 else 2091 error = 0; 2092 /* 2093 * The CC algorithms, when given their memory 2094 * should not fail we could in theory have a 2095 * KASSERT here. 2096 */ 2097 if (error == 0) { 2098 /* 2099 * Touchdown, lets go ahead and move the 2100 * connection to the new CC module by 2101 * copying in the cc_mem after we call 2102 * the old ones cleanup (if any). 2103 */ 2104 if (CC_ALGO(tp)->cb_destroy != NULL) 2105 CC_ALGO(tp)->cb_destroy(tp->ccv); 2106 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2107 tp->cc_algo = algo; 2108 /* Ok now are we where we have gotten past any conn_init? */ 2109 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2110 /* Yep run the connection init for the new CC */ 2111 CC_ALGO(tp)->conn_init(tp->ccv); 2112 } 2113 } else if (ptr) 2114 free(ptr, M_CC_MEM); 2115 INP_WUNLOCK(inp); 2116 return (error); 2117 } 2118 2119 int 2120 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 2121 { 2122 int error, opt, optval; 2123 u_int ui; 2124 struct tcp_info ti; 2125 #ifdef KERN_TLS 2126 struct tls_enable tls; 2127 #endif 2128 char *pbuf, buf[TCP_LOG_ID_LEN]; 2129 #ifdef STATS 2130 struct statsblob *sbp; 2131 #endif 2132 size_t len; 2133 2134 INP_WLOCK_ASSERT(inp); 2135 2136 switch (sopt->sopt_level) { 2137 #ifdef INET6 2138 case IPPROTO_IPV6: 2139 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2140 switch (sopt->sopt_name) { 2141 case IPV6_USE_MIN_MTU: 2142 tcp6_use_min_mtu(tp); 2143 /* FALLTHROUGH */ 2144 } 2145 INP_WUNLOCK(inp); 2146 return (0); 2147 #endif 2148 #ifdef INET 2149 case IPPROTO_IP: 2150 INP_WUNLOCK(inp); 2151 return (0); 2152 #endif 2153 } 2154 2155 /* 2156 * For TCP_CCALGOOPT forward the control to CC module, for both 2157 * SOPT_SET and SOPT_GET. 2158 */ 2159 switch (sopt->sopt_name) { 2160 case TCP_CCALGOOPT: 2161 INP_WUNLOCK(inp); 2162 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2163 return (EINVAL); 2164 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2165 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2166 sopt->sopt_valsize); 2167 if (error) { 2168 free(pbuf, M_TEMP); 2169 return (error); 2170 } 2171 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2172 if (CC_ALGO(tp)->ctl_output != NULL) 2173 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 2174 else 2175 error = ENOENT; 2176 INP_WUNLOCK(inp); 2177 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2178 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2179 free(pbuf, M_TEMP); 2180 return (error); 2181 } 2182 2183 switch (sopt->sopt_dir) { 2184 case SOPT_SET: 2185 switch (sopt->sopt_name) { 2186 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2187 case TCP_MD5SIG: 2188 if (!TCPMD5_ENABLED()) { 2189 INP_WUNLOCK(inp); 2190 return (ENOPROTOOPT); 2191 } 2192 error = TCPMD5_PCBCTL(inp, sopt); 2193 if (error) 2194 return (error); 2195 goto unlock_and_done; 2196 #endif /* IPSEC */ 2197 2198 case TCP_NODELAY: 2199 case TCP_NOOPT: 2200 case TCP_LRD: 2201 INP_WUNLOCK(inp); 2202 error = sooptcopyin(sopt, &optval, sizeof optval, 2203 sizeof optval); 2204 if (error) 2205 return (error); 2206 2207 INP_WLOCK_RECHECK(inp); 2208 switch (sopt->sopt_name) { 2209 case TCP_NODELAY: 2210 opt = TF_NODELAY; 2211 break; 2212 case TCP_NOOPT: 2213 opt = TF_NOOPT; 2214 break; 2215 case TCP_LRD: 2216 opt = TF_LRD; 2217 break; 2218 default: 2219 opt = 0; /* dead code to fool gcc */ 2220 break; 2221 } 2222 2223 if (optval) 2224 tp->t_flags |= opt; 2225 else 2226 tp->t_flags &= ~opt; 2227 unlock_and_done: 2228 #ifdef TCP_OFFLOAD 2229 if (tp->t_flags & TF_TOE) { 2230 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2231 sopt->sopt_name); 2232 } 2233 #endif 2234 INP_WUNLOCK(inp); 2235 break; 2236 2237 case TCP_NOPUSH: 2238 INP_WUNLOCK(inp); 2239 error = sooptcopyin(sopt, &optval, sizeof optval, 2240 sizeof optval); 2241 if (error) 2242 return (error); 2243 2244 INP_WLOCK_RECHECK(inp); 2245 if (optval) 2246 tp->t_flags |= TF_NOPUSH; 2247 else if (tp->t_flags & TF_NOPUSH) { 2248 tp->t_flags &= ~TF_NOPUSH; 2249 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2250 struct epoch_tracker et; 2251 2252 NET_EPOCH_ENTER(et); 2253 error = tp->t_fb->tfb_tcp_output(tp); 2254 NET_EPOCH_EXIT(et); 2255 } 2256 } 2257 goto unlock_and_done; 2258 2259 case TCP_REMOTE_UDP_ENCAPS_PORT: 2260 INP_WUNLOCK(inp); 2261 error = sooptcopyin(sopt, &optval, sizeof optval, 2262 sizeof optval); 2263 if (error) 2264 return (error); 2265 if ((optval < TCP_TUNNELING_PORT_MIN) || 2266 (optval > TCP_TUNNELING_PORT_MAX)) { 2267 /* Its got to be in range */ 2268 return (EINVAL); 2269 } 2270 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2271 /* You have to have enabled a UDP tunneling port first */ 2272 return (EINVAL); 2273 } 2274 INP_WLOCK_RECHECK(inp); 2275 if (tp->t_state != TCPS_CLOSED) { 2276 /* You can't change after you are connected */ 2277 error = EINVAL; 2278 } else { 2279 /* Ok we are all good set the port */ 2280 tp->t_port = htons(optval); 2281 } 2282 goto unlock_and_done; 2283 2284 case TCP_MAXSEG: 2285 INP_WUNLOCK(inp); 2286 error = sooptcopyin(sopt, &optval, sizeof optval, 2287 sizeof optval); 2288 if (error) 2289 return (error); 2290 2291 INP_WLOCK_RECHECK(inp); 2292 if (optval > 0 && optval <= tp->t_maxseg && 2293 optval + 40 >= V_tcp_minmss) 2294 tp->t_maxseg = optval; 2295 else 2296 error = EINVAL; 2297 goto unlock_and_done; 2298 2299 case TCP_INFO: 2300 INP_WUNLOCK(inp); 2301 error = EINVAL; 2302 break; 2303 2304 case TCP_STATS: 2305 INP_WUNLOCK(inp); 2306 #ifdef STATS 2307 error = sooptcopyin(sopt, &optval, sizeof optval, 2308 sizeof optval); 2309 if (error) 2310 return (error); 2311 2312 if (optval > 0) 2313 sbp = stats_blob_alloc( 2314 V_tcp_perconn_stats_dflt_tpl, 0); 2315 else 2316 sbp = NULL; 2317 2318 INP_WLOCK_RECHECK(inp); 2319 if ((tp->t_stats != NULL && sbp == NULL) || 2320 (tp->t_stats == NULL && sbp != NULL)) { 2321 struct statsblob *t = tp->t_stats; 2322 tp->t_stats = sbp; 2323 sbp = t; 2324 } 2325 INP_WUNLOCK(inp); 2326 2327 stats_blob_destroy(sbp); 2328 #else 2329 return (EOPNOTSUPP); 2330 #endif /* !STATS */ 2331 break; 2332 2333 case TCP_CONGESTION: 2334 error = tcp_congestion(so, sopt, inp, tp); 2335 break; 2336 2337 case TCP_REUSPORT_LB_NUMA: 2338 INP_WUNLOCK(inp); 2339 error = sooptcopyin(sopt, &optval, sizeof(optval), 2340 sizeof(optval)); 2341 INP_WLOCK_RECHECK(inp); 2342 if (!error) 2343 error = in_pcblbgroup_numa(inp, optval); 2344 INP_WUNLOCK(inp); 2345 break; 2346 2347 #ifdef KERN_TLS 2348 case TCP_TXTLS_ENABLE: 2349 INP_WUNLOCK(inp); 2350 error = copyin_tls_enable(sopt, &tls); 2351 if (error) 2352 break; 2353 error = ktls_enable_tx(so, &tls); 2354 break; 2355 case TCP_TXTLS_MODE: 2356 INP_WUNLOCK(inp); 2357 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2358 if (error) 2359 return (error); 2360 2361 INP_WLOCK_RECHECK(inp); 2362 error = ktls_set_tx_mode(so, ui); 2363 INP_WUNLOCK(inp); 2364 break; 2365 case TCP_RXTLS_ENABLE: 2366 INP_WUNLOCK(inp); 2367 error = sooptcopyin(sopt, &tls, sizeof(tls), 2368 sizeof(tls)); 2369 if (error) 2370 break; 2371 error = ktls_enable_rx(so, &tls); 2372 break; 2373 #endif 2374 2375 case TCP_KEEPIDLE: 2376 case TCP_KEEPINTVL: 2377 case TCP_KEEPINIT: 2378 INP_WUNLOCK(inp); 2379 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2380 if (error) 2381 return (error); 2382 2383 if (ui > (UINT_MAX / hz)) { 2384 error = EINVAL; 2385 break; 2386 } 2387 ui *= hz; 2388 2389 INP_WLOCK_RECHECK(inp); 2390 switch (sopt->sopt_name) { 2391 case TCP_KEEPIDLE: 2392 tp->t_keepidle = ui; 2393 /* 2394 * XXX: better check current remaining 2395 * timeout and "merge" it with new value. 2396 */ 2397 if ((tp->t_state > TCPS_LISTEN) && 2398 (tp->t_state <= TCPS_CLOSING)) 2399 tcp_timer_activate(tp, TT_KEEP, 2400 TP_KEEPIDLE(tp)); 2401 break; 2402 case TCP_KEEPINTVL: 2403 tp->t_keepintvl = ui; 2404 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2405 (TP_MAXIDLE(tp) > 0)) 2406 tcp_timer_activate(tp, TT_2MSL, 2407 TP_MAXIDLE(tp)); 2408 break; 2409 case TCP_KEEPINIT: 2410 tp->t_keepinit = ui; 2411 if (tp->t_state == TCPS_SYN_RECEIVED || 2412 tp->t_state == TCPS_SYN_SENT) 2413 tcp_timer_activate(tp, TT_KEEP, 2414 TP_KEEPINIT(tp)); 2415 break; 2416 } 2417 goto unlock_and_done; 2418 2419 case TCP_KEEPCNT: 2420 INP_WUNLOCK(inp); 2421 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2422 if (error) 2423 return (error); 2424 2425 INP_WLOCK_RECHECK(inp); 2426 tp->t_keepcnt = ui; 2427 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2428 (TP_MAXIDLE(tp) > 0)) 2429 tcp_timer_activate(tp, TT_2MSL, 2430 TP_MAXIDLE(tp)); 2431 goto unlock_and_done; 2432 2433 #ifdef TCPPCAP 2434 case TCP_PCAP_OUT: 2435 case TCP_PCAP_IN: 2436 INP_WUNLOCK(inp); 2437 error = sooptcopyin(sopt, &optval, sizeof optval, 2438 sizeof optval); 2439 if (error) 2440 return (error); 2441 2442 INP_WLOCK_RECHECK(inp); 2443 if (optval >= 0) 2444 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2445 &(tp->t_outpkts) : &(tp->t_inpkts), 2446 optval); 2447 else 2448 error = EINVAL; 2449 goto unlock_and_done; 2450 #endif 2451 2452 case TCP_FASTOPEN: { 2453 struct tcp_fastopen tfo_optval; 2454 2455 INP_WUNLOCK(inp); 2456 if (!V_tcp_fastopen_client_enable && 2457 !V_tcp_fastopen_server_enable) 2458 return (EPERM); 2459 2460 error = sooptcopyin(sopt, &tfo_optval, 2461 sizeof(tfo_optval), sizeof(int)); 2462 if (error) 2463 return (error); 2464 2465 INP_WLOCK_RECHECK(inp); 2466 if ((tp->t_state != TCPS_CLOSED) && 2467 (tp->t_state != TCPS_LISTEN)) { 2468 error = EINVAL; 2469 goto unlock_and_done; 2470 } 2471 if (tfo_optval.enable) { 2472 if (tp->t_state == TCPS_LISTEN) { 2473 if (!V_tcp_fastopen_server_enable) { 2474 error = EPERM; 2475 goto unlock_and_done; 2476 } 2477 2478 if (tp->t_tfo_pending == NULL) 2479 tp->t_tfo_pending = 2480 tcp_fastopen_alloc_counter(); 2481 } else { 2482 /* 2483 * If a pre-shared key was provided, 2484 * stash it in the client cookie 2485 * field of the tcpcb for use during 2486 * connect. 2487 */ 2488 if (sopt->sopt_valsize == 2489 sizeof(tfo_optval)) { 2490 memcpy(tp->t_tfo_cookie.client, 2491 tfo_optval.psk, 2492 TCP_FASTOPEN_PSK_LEN); 2493 tp->t_tfo_client_cookie_len = 2494 TCP_FASTOPEN_PSK_LEN; 2495 } 2496 } 2497 tp->t_flags |= TF_FASTOPEN; 2498 } else 2499 tp->t_flags &= ~TF_FASTOPEN; 2500 goto unlock_and_done; 2501 } 2502 2503 #ifdef TCP_BLACKBOX 2504 case TCP_LOG: 2505 INP_WUNLOCK(inp); 2506 error = sooptcopyin(sopt, &optval, sizeof optval, 2507 sizeof optval); 2508 if (error) 2509 return (error); 2510 2511 INP_WLOCK_RECHECK(inp); 2512 error = tcp_log_state_change(tp, optval); 2513 goto unlock_and_done; 2514 2515 case TCP_LOGBUF: 2516 INP_WUNLOCK(inp); 2517 error = EINVAL; 2518 break; 2519 2520 case TCP_LOGID: 2521 INP_WUNLOCK(inp); 2522 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2523 if (error) 2524 break; 2525 buf[sopt->sopt_valsize] = '\0'; 2526 INP_WLOCK_RECHECK(inp); 2527 error = tcp_log_set_id(tp, buf); 2528 /* tcp_log_set_id() unlocks the INP. */ 2529 break; 2530 2531 case TCP_LOGDUMP: 2532 case TCP_LOGDUMPID: 2533 INP_WUNLOCK(inp); 2534 error = 2535 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2536 if (error) 2537 break; 2538 buf[sopt->sopt_valsize] = '\0'; 2539 INP_WLOCK_RECHECK(inp); 2540 if (sopt->sopt_name == TCP_LOGDUMP) { 2541 error = tcp_log_dump_tp_logbuf(tp, buf, 2542 M_WAITOK, true); 2543 INP_WUNLOCK(inp); 2544 } else { 2545 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2546 /* 2547 * tcp_log_dump_tp_bucket_logbufs() drops the 2548 * INP lock. 2549 */ 2550 } 2551 break; 2552 #endif 2553 2554 default: 2555 INP_WUNLOCK(inp); 2556 error = ENOPROTOOPT; 2557 break; 2558 } 2559 break; 2560 2561 case SOPT_GET: 2562 tp = intotcpcb(inp); 2563 switch (sopt->sopt_name) { 2564 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2565 case TCP_MD5SIG: 2566 if (!TCPMD5_ENABLED()) { 2567 INP_WUNLOCK(inp); 2568 return (ENOPROTOOPT); 2569 } 2570 error = TCPMD5_PCBCTL(inp, sopt); 2571 break; 2572 #endif 2573 2574 case TCP_NODELAY: 2575 optval = tp->t_flags & TF_NODELAY; 2576 INP_WUNLOCK(inp); 2577 error = sooptcopyout(sopt, &optval, sizeof optval); 2578 break; 2579 case TCP_MAXSEG: 2580 optval = tp->t_maxseg; 2581 INP_WUNLOCK(inp); 2582 error = sooptcopyout(sopt, &optval, sizeof optval); 2583 break; 2584 case TCP_REMOTE_UDP_ENCAPS_PORT: 2585 optval = ntohs(tp->t_port); 2586 INP_WUNLOCK(inp); 2587 error = sooptcopyout(sopt, &optval, sizeof optval); 2588 break; 2589 case TCP_NOOPT: 2590 optval = tp->t_flags & TF_NOOPT; 2591 INP_WUNLOCK(inp); 2592 error = sooptcopyout(sopt, &optval, sizeof optval); 2593 break; 2594 case TCP_NOPUSH: 2595 optval = tp->t_flags & TF_NOPUSH; 2596 INP_WUNLOCK(inp); 2597 error = sooptcopyout(sopt, &optval, sizeof optval); 2598 break; 2599 case TCP_INFO: 2600 tcp_fill_info(tp, &ti); 2601 INP_WUNLOCK(inp); 2602 error = sooptcopyout(sopt, &ti, sizeof ti); 2603 break; 2604 case TCP_STATS: 2605 { 2606 #ifdef STATS 2607 int nheld; 2608 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2609 2610 error = 0; 2611 socklen_t outsbsz = sopt->sopt_valsize; 2612 if (tp->t_stats == NULL) 2613 error = ENOENT; 2614 else if (outsbsz >= tp->t_stats->cursz) 2615 outsbsz = tp->t_stats->cursz; 2616 else if (outsbsz >= sizeof(struct statsblob)) 2617 outsbsz = sizeof(struct statsblob); 2618 else 2619 error = EINVAL; 2620 INP_WUNLOCK(inp); 2621 if (error) 2622 break; 2623 2624 sbp = sopt->sopt_val; 2625 nheld = atop(round_page(((vm_offset_t)sbp) + 2626 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2627 vm_page_t ma[nheld]; 2628 if (vm_fault_quick_hold_pages( 2629 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2630 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2631 nheld) < 0) { 2632 error = EFAULT; 2633 break; 2634 } 2635 2636 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2637 SIZEOF_MEMBER(struct statsblob, flags)))) 2638 goto unhold; 2639 2640 INP_WLOCK_RECHECK(inp); 2641 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2642 sbflags | SB_CLONE_USRDSTNOFAULT); 2643 INP_WUNLOCK(inp); 2644 sopt->sopt_valsize = outsbsz; 2645 unhold: 2646 vm_page_unhold_pages(ma, nheld); 2647 #else 2648 INP_WUNLOCK(inp); 2649 error = EOPNOTSUPP; 2650 #endif /* !STATS */ 2651 break; 2652 } 2653 case TCP_CONGESTION: 2654 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2655 INP_WUNLOCK(inp); 2656 error = sooptcopyout(sopt, buf, len + 1); 2657 break; 2658 case TCP_KEEPIDLE: 2659 case TCP_KEEPINTVL: 2660 case TCP_KEEPINIT: 2661 case TCP_KEEPCNT: 2662 switch (sopt->sopt_name) { 2663 case TCP_KEEPIDLE: 2664 ui = TP_KEEPIDLE(tp) / hz; 2665 break; 2666 case TCP_KEEPINTVL: 2667 ui = TP_KEEPINTVL(tp) / hz; 2668 break; 2669 case TCP_KEEPINIT: 2670 ui = TP_KEEPINIT(tp) / hz; 2671 break; 2672 case TCP_KEEPCNT: 2673 ui = TP_KEEPCNT(tp); 2674 break; 2675 } 2676 INP_WUNLOCK(inp); 2677 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2678 break; 2679 #ifdef TCPPCAP 2680 case TCP_PCAP_OUT: 2681 case TCP_PCAP_IN: 2682 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2683 &(tp->t_outpkts) : &(tp->t_inpkts)); 2684 INP_WUNLOCK(inp); 2685 error = sooptcopyout(sopt, &optval, sizeof optval); 2686 break; 2687 #endif 2688 case TCP_FASTOPEN: 2689 optval = tp->t_flags & TF_FASTOPEN; 2690 INP_WUNLOCK(inp); 2691 error = sooptcopyout(sopt, &optval, sizeof optval); 2692 break; 2693 #ifdef TCP_BLACKBOX 2694 case TCP_LOG: 2695 optval = tp->t_logstate; 2696 INP_WUNLOCK(inp); 2697 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2698 break; 2699 case TCP_LOGBUF: 2700 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2701 error = tcp_log_getlogbuf(sopt, tp); 2702 break; 2703 case TCP_LOGID: 2704 len = tcp_log_get_id(tp, buf); 2705 INP_WUNLOCK(inp); 2706 error = sooptcopyout(sopt, buf, len + 1); 2707 break; 2708 case TCP_LOGDUMP: 2709 case TCP_LOGDUMPID: 2710 INP_WUNLOCK(inp); 2711 error = EINVAL; 2712 break; 2713 #endif 2714 #ifdef KERN_TLS 2715 case TCP_TXTLS_MODE: 2716 error = ktls_get_tx_mode(so, &optval); 2717 INP_WUNLOCK(inp); 2718 if (error == 0) 2719 error = sooptcopyout(sopt, &optval, 2720 sizeof(optval)); 2721 break; 2722 case TCP_RXTLS_MODE: 2723 error = ktls_get_rx_mode(so, &optval); 2724 INP_WUNLOCK(inp); 2725 if (error == 0) 2726 error = sooptcopyout(sopt, &optval, 2727 sizeof(optval)); 2728 break; 2729 #endif 2730 case TCP_LRD: 2731 optval = tp->t_flags & TF_LRD; 2732 INP_WUNLOCK(inp); 2733 error = sooptcopyout(sopt, &optval, sizeof optval); 2734 break; 2735 default: 2736 INP_WUNLOCK(inp); 2737 error = ENOPROTOOPT; 2738 break; 2739 } 2740 break; 2741 } 2742 return (error); 2743 } 2744 #undef INP_WLOCK_RECHECK 2745 #undef INP_WLOCK_RECHECK_CLEANUP 2746 2747 /* 2748 * Initiate (or continue) disconnect. 2749 * If embryonic state, just send reset (once). 2750 * If in ``let data drain'' option and linger null, just drop. 2751 * Otherwise (hard), mark socket disconnecting and drop 2752 * current input data; switch states based on user close, and 2753 * send segment to peer (with FIN). 2754 */ 2755 static void 2756 tcp_disconnect(struct tcpcb *tp) 2757 { 2758 struct inpcb *inp = tp->t_inpcb; 2759 struct socket *so = inp->inp_socket; 2760 2761 NET_EPOCH_ASSERT(); 2762 INP_WLOCK_ASSERT(inp); 2763 2764 /* 2765 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2766 * socket is still open. 2767 */ 2768 if (tp->t_state < TCPS_ESTABLISHED && 2769 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2770 tp = tcp_close(tp); 2771 KASSERT(tp != NULL, 2772 ("tcp_disconnect: tcp_close() returned NULL")); 2773 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2774 tp = tcp_drop(tp, 0); 2775 KASSERT(tp != NULL, 2776 ("tcp_disconnect: tcp_drop() returned NULL")); 2777 } else { 2778 soisdisconnecting(so); 2779 sbflush(&so->so_rcv); 2780 tcp_usrclosed(tp); 2781 if (!(inp->inp_flags & INP_DROPPED)) 2782 tp->t_fb->tfb_tcp_output(tp); 2783 } 2784 } 2785 2786 /* 2787 * User issued close, and wish to trail through shutdown states: 2788 * if never received SYN, just forget it. If got a SYN from peer, 2789 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2790 * If already got a FIN from peer, then almost done; go to LAST_ACK 2791 * state. In all other cases, have already sent FIN to peer (e.g. 2792 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2793 * for peer to send FIN or not respond to keep-alives, etc. 2794 * We can let the user exit from the close as soon as the FIN is acked. 2795 */ 2796 static void 2797 tcp_usrclosed(struct tcpcb *tp) 2798 { 2799 2800 NET_EPOCH_ASSERT(); 2801 INP_WLOCK_ASSERT(tp->t_inpcb); 2802 2803 switch (tp->t_state) { 2804 case TCPS_LISTEN: 2805 #ifdef TCP_OFFLOAD 2806 tcp_offload_listen_stop(tp); 2807 #endif 2808 tcp_state_change(tp, TCPS_CLOSED); 2809 /* FALLTHROUGH */ 2810 case TCPS_CLOSED: 2811 tp = tcp_close(tp); 2812 /* 2813 * tcp_close() should never return NULL here as the socket is 2814 * still open. 2815 */ 2816 KASSERT(tp != NULL, 2817 ("tcp_usrclosed: tcp_close() returned NULL")); 2818 break; 2819 2820 case TCPS_SYN_SENT: 2821 case TCPS_SYN_RECEIVED: 2822 tp->t_flags |= TF_NEEDFIN; 2823 break; 2824 2825 case TCPS_ESTABLISHED: 2826 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2827 break; 2828 2829 case TCPS_CLOSE_WAIT: 2830 tcp_state_change(tp, TCPS_LAST_ACK); 2831 break; 2832 } 2833 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2834 soisdisconnected(tp->t_inpcb->inp_socket); 2835 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2836 if (tp->t_state == TCPS_FIN_WAIT_2) { 2837 int timeout; 2838 2839 timeout = (tcp_fast_finwait2_recycle) ? 2840 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2841 tcp_timer_activate(tp, TT_2MSL, timeout); 2842 } 2843 } 2844 } 2845 2846 #ifdef DDB 2847 static void 2848 db_print_indent(int indent) 2849 { 2850 int i; 2851 2852 for (i = 0; i < indent; i++) 2853 db_printf(" "); 2854 } 2855 2856 static void 2857 db_print_tstate(int t_state) 2858 { 2859 2860 switch (t_state) { 2861 case TCPS_CLOSED: 2862 db_printf("TCPS_CLOSED"); 2863 return; 2864 2865 case TCPS_LISTEN: 2866 db_printf("TCPS_LISTEN"); 2867 return; 2868 2869 case TCPS_SYN_SENT: 2870 db_printf("TCPS_SYN_SENT"); 2871 return; 2872 2873 case TCPS_SYN_RECEIVED: 2874 db_printf("TCPS_SYN_RECEIVED"); 2875 return; 2876 2877 case TCPS_ESTABLISHED: 2878 db_printf("TCPS_ESTABLISHED"); 2879 return; 2880 2881 case TCPS_CLOSE_WAIT: 2882 db_printf("TCPS_CLOSE_WAIT"); 2883 return; 2884 2885 case TCPS_FIN_WAIT_1: 2886 db_printf("TCPS_FIN_WAIT_1"); 2887 return; 2888 2889 case TCPS_CLOSING: 2890 db_printf("TCPS_CLOSING"); 2891 return; 2892 2893 case TCPS_LAST_ACK: 2894 db_printf("TCPS_LAST_ACK"); 2895 return; 2896 2897 case TCPS_FIN_WAIT_2: 2898 db_printf("TCPS_FIN_WAIT_2"); 2899 return; 2900 2901 case TCPS_TIME_WAIT: 2902 db_printf("TCPS_TIME_WAIT"); 2903 return; 2904 2905 default: 2906 db_printf("unknown"); 2907 return; 2908 } 2909 } 2910 2911 static void 2912 db_print_tflags(u_int t_flags) 2913 { 2914 int comma; 2915 2916 comma = 0; 2917 if (t_flags & TF_ACKNOW) { 2918 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2919 comma = 1; 2920 } 2921 if (t_flags & TF_DELACK) { 2922 db_printf("%sTF_DELACK", comma ? ", " : ""); 2923 comma = 1; 2924 } 2925 if (t_flags & TF_NODELAY) { 2926 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2927 comma = 1; 2928 } 2929 if (t_flags & TF_NOOPT) { 2930 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2931 comma = 1; 2932 } 2933 if (t_flags & TF_SENTFIN) { 2934 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2935 comma = 1; 2936 } 2937 if (t_flags & TF_REQ_SCALE) { 2938 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2939 comma = 1; 2940 } 2941 if (t_flags & TF_RCVD_SCALE) { 2942 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2943 comma = 1; 2944 } 2945 if (t_flags & TF_REQ_TSTMP) { 2946 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2947 comma = 1; 2948 } 2949 if (t_flags & TF_RCVD_TSTMP) { 2950 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2951 comma = 1; 2952 } 2953 if (t_flags & TF_SACK_PERMIT) { 2954 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2955 comma = 1; 2956 } 2957 if (t_flags & TF_NEEDSYN) { 2958 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2959 comma = 1; 2960 } 2961 if (t_flags & TF_NEEDFIN) { 2962 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2963 comma = 1; 2964 } 2965 if (t_flags & TF_NOPUSH) { 2966 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2967 comma = 1; 2968 } 2969 if (t_flags & TF_MORETOCOME) { 2970 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2971 comma = 1; 2972 } 2973 if (t_flags & TF_LQ_OVERFLOW) { 2974 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2975 comma = 1; 2976 } 2977 if (t_flags & TF_LASTIDLE) { 2978 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2979 comma = 1; 2980 } 2981 if (t_flags & TF_RXWIN0SENT) { 2982 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2983 comma = 1; 2984 } 2985 if (t_flags & TF_FASTRECOVERY) { 2986 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2987 comma = 1; 2988 } 2989 if (t_flags & TF_CONGRECOVERY) { 2990 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2991 comma = 1; 2992 } 2993 if (t_flags & TF_WASFRECOVERY) { 2994 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2995 comma = 1; 2996 } 2997 if (t_flags & TF_SIGNATURE) { 2998 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2999 comma = 1; 3000 } 3001 if (t_flags & TF_FORCEDATA) { 3002 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 3003 comma = 1; 3004 } 3005 if (t_flags & TF_TSO) { 3006 db_printf("%sTF_TSO", comma ? ", " : ""); 3007 comma = 1; 3008 } 3009 if (t_flags & TF_FASTOPEN) { 3010 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 3011 comma = 1; 3012 } 3013 } 3014 3015 static void 3016 db_print_tflags2(u_int t_flags2) 3017 { 3018 int comma; 3019 3020 comma = 0; 3021 if (t_flags2 & TF2_ECN_PERMIT) { 3022 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 3023 comma = 1; 3024 } 3025 } 3026 3027 static void 3028 db_print_toobflags(char t_oobflags) 3029 { 3030 int comma; 3031 3032 comma = 0; 3033 if (t_oobflags & TCPOOB_HAVEDATA) { 3034 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3035 comma = 1; 3036 } 3037 if (t_oobflags & TCPOOB_HADDATA) { 3038 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3039 comma = 1; 3040 } 3041 } 3042 3043 static void 3044 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3045 { 3046 3047 db_print_indent(indent); 3048 db_printf("%s at %p\n", name, tp); 3049 3050 indent += 2; 3051 3052 db_print_indent(indent); 3053 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3054 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3055 3056 db_print_indent(indent); 3057 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 3058 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 3059 3060 db_print_indent(indent); 3061 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 3062 &tp->t_timers->tt_delack, tp->t_inpcb); 3063 3064 db_print_indent(indent); 3065 db_printf("t_state: %d (", tp->t_state); 3066 db_print_tstate(tp->t_state); 3067 db_printf(")\n"); 3068 3069 db_print_indent(indent); 3070 db_printf("t_flags: 0x%x (", tp->t_flags); 3071 db_print_tflags(tp->t_flags); 3072 db_printf(")\n"); 3073 3074 db_print_indent(indent); 3075 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3076 db_print_tflags2(tp->t_flags2); 3077 db_printf(")\n"); 3078 3079 db_print_indent(indent); 3080 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 3081 tp->snd_una, tp->snd_max, tp->snd_nxt); 3082 3083 db_print_indent(indent); 3084 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3085 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3086 3087 db_print_indent(indent); 3088 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3089 tp->iss, tp->irs, tp->rcv_nxt); 3090 3091 db_print_indent(indent); 3092 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3093 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3094 3095 db_print_indent(indent); 3096 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3097 tp->snd_wnd, tp->snd_cwnd); 3098 3099 db_print_indent(indent); 3100 db_printf("snd_ssthresh: %u snd_recover: " 3101 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3102 3103 db_print_indent(indent); 3104 db_printf("t_rcvtime: %u t_startime: %u\n", 3105 tp->t_rcvtime, tp->t_starttime); 3106 3107 db_print_indent(indent); 3108 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3109 tp->t_rtttime, tp->t_rtseq); 3110 3111 db_print_indent(indent); 3112 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3113 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3114 3115 db_print_indent(indent); 3116 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 3117 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 3118 tp->t_rttbest); 3119 3120 db_print_indent(indent); 3121 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 3122 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3123 3124 db_print_indent(indent); 3125 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3126 db_print_toobflags(tp->t_oobflags); 3127 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3128 3129 db_print_indent(indent); 3130 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3131 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3132 3133 db_print_indent(indent); 3134 db_printf("ts_recent: %u ts_recent_age: %u\n", 3135 tp->ts_recent, tp->ts_recent_age); 3136 3137 db_print_indent(indent); 3138 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3139 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3140 3141 db_print_indent(indent); 3142 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3143 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3144 tp->snd_recover_prev, tp->t_badrxtwin); 3145 3146 db_print_indent(indent); 3147 db_printf("snd_numholes: %d snd_holes first: %p\n", 3148 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3149 3150 db_print_indent(indent); 3151 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3152 tp->snd_fack, tp->rcv_numsacks); 3153 3154 /* Skip sackblks, sackhint. */ 3155 3156 db_print_indent(indent); 3157 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3158 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3159 } 3160 3161 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3162 { 3163 struct tcpcb *tp; 3164 3165 if (!have_addr) { 3166 db_printf("usage: show tcpcb <addr>\n"); 3167 return; 3168 } 3169 tp = (struct tcpcb *)addr; 3170 3171 db_print_tcpcb(tp, "tcpcb", 0); 3172 } 3173 #endif 3174