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/stats.h> 70 71 #ifdef DDB 72 #include <ddb/ddb.h> 73 #endif 74 75 #include <net/if.h> 76 #include <net/if_var.h> 77 #include <net/route.h> 78 #include <net/vnet.h> 79 80 #include <netinet/in.h> 81 #include <netinet/in_kdtrace.h> 82 #include <netinet/in_pcb.h> 83 #include <netinet/in_systm.h> 84 #include <netinet/in_var.h> 85 #include <netinet/ip.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 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 266 "INP_DROPPED && tp != NULL")); 267 in_pcbdetach(inp); 268 in_pcbfree(inp); 269 } else { 270 in_pcbdetach(inp); 271 INP_WUNLOCK(inp); 272 } 273 } else { 274 /* 275 * If the connection is not in timewait, we consider two 276 * two conditions: one in which no further processing is 277 * necessary (dropped || embryonic), and one in which TCP is 278 * not yet done, but no longer requires the socket, so the 279 * pcb will persist for the time being. 280 * 281 * XXXRW: Does the second case still occur? 282 */ 283 if (inp->inp_flags & INP_DROPPED || 284 tp->t_state < TCPS_SYN_SENT) { 285 tcp_discardcb(tp); 286 in_pcbdetach(inp); 287 in_pcbfree(inp); 288 } else { 289 in_pcbdetach(inp); 290 INP_WUNLOCK(inp); 291 } 292 } 293 } 294 295 #ifdef INET 296 /* 297 * Give the socket an address. 298 */ 299 static int 300 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 301 { 302 int error = 0; 303 struct inpcb *inp; 304 struct tcpcb *tp = NULL; 305 struct sockaddr_in *sinp; 306 307 sinp = (struct sockaddr_in *)nam; 308 if (nam->sa_family != AF_INET) { 309 /* 310 * Preserve compatibility with old programs. 311 */ 312 if (nam->sa_family != AF_UNSPEC || 313 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 314 sinp->sin_addr.s_addr != INADDR_ANY) 315 return (EAFNOSUPPORT); 316 nam->sa_family = AF_INET; 317 } 318 if (nam->sa_len != sizeof(*sinp)) 319 return (EINVAL); 320 321 /* 322 * Must check for multicast addresses and disallow binding 323 * to them. 324 */ 325 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 326 return (EAFNOSUPPORT); 327 328 TCPDEBUG0; 329 inp = sotoinpcb(so); 330 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 331 INP_WLOCK(inp); 332 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 333 error = EINVAL; 334 goto out; 335 } 336 tp = intotcpcb(inp); 337 TCPDEBUG1(); 338 INP_HASH_WLOCK(&V_tcbinfo); 339 error = in_pcbbind(inp, nam, td->td_ucred); 340 INP_HASH_WUNLOCK(&V_tcbinfo); 341 out: 342 TCPDEBUG2(PRU_BIND); 343 TCP_PROBE2(debug__user, tp, PRU_BIND); 344 INP_WUNLOCK(inp); 345 346 return (error); 347 } 348 #endif /* INET */ 349 350 #ifdef INET6 351 static int 352 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 353 { 354 int error = 0; 355 struct inpcb *inp; 356 struct tcpcb *tp = NULL; 357 struct sockaddr_in6 *sin6; 358 u_char vflagsav; 359 360 sin6 = (struct sockaddr_in6 *)nam; 361 if (nam->sa_family != AF_INET6) 362 return (EAFNOSUPPORT); 363 if (nam->sa_len != sizeof(*sin6)) 364 return (EINVAL); 365 366 /* 367 * Must check for multicast addresses and disallow binding 368 * to them. 369 */ 370 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 371 return (EAFNOSUPPORT); 372 373 TCPDEBUG0; 374 inp = sotoinpcb(so); 375 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 376 INP_WLOCK(inp); 377 vflagsav = inp->inp_vflag; 378 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 379 error = EINVAL; 380 goto out; 381 } 382 tp = intotcpcb(inp); 383 TCPDEBUG1(); 384 INP_HASH_WLOCK(&V_tcbinfo); 385 inp->inp_vflag &= ~INP_IPV4; 386 inp->inp_vflag |= INP_IPV6; 387 #ifdef INET 388 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 389 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 390 inp->inp_vflag |= INP_IPV4; 391 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 392 struct sockaddr_in sin; 393 394 in6_sin6_2_sin(&sin, sin6); 395 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 396 error = EAFNOSUPPORT; 397 INP_HASH_WUNLOCK(&V_tcbinfo); 398 goto out; 399 } 400 inp->inp_vflag |= INP_IPV4; 401 inp->inp_vflag &= ~INP_IPV6; 402 error = in_pcbbind(inp, (struct sockaddr *)&sin, 403 td->td_ucred); 404 INP_HASH_WUNLOCK(&V_tcbinfo); 405 goto out; 406 } 407 } 408 #endif 409 error = in6_pcbbind(inp, nam, td->td_ucred); 410 INP_HASH_WUNLOCK(&V_tcbinfo); 411 out: 412 if (error != 0) 413 inp->inp_vflag = vflagsav; 414 TCPDEBUG2(PRU_BIND); 415 TCP_PROBE2(debug__user, tp, PRU_BIND); 416 INP_WUNLOCK(inp); 417 return (error); 418 } 419 #endif /* INET6 */ 420 421 #ifdef INET 422 /* 423 * Prepare to accept connections. 424 */ 425 static int 426 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 427 { 428 int error = 0; 429 struct inpcb *inp; 430 struct tcpcb *tp = NULL; 431 432 TCPDEBUG0; 433 inp = sotoinpcb(so); 434 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 435 INP_WLOCK(inp); 436 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 437 error = EINVAL; 438 goto out; 439 } 440 tp = intotcpcb(inp); 441 TCPDEBUG1(); 442 SOCK_LOCK(so); 443 error = solisten_proto_check(so); 444 if (error != 0) { 445 SOCK_UNLOCK(so); 446 goto out; 447 } 448 if (inp->inp_lport == 0) { 449 INP_HASH_WLOCK(&V_tcbinfo); 450 error = in_pcbbind(inp, NULL, td->td_ucred); 451 INP_HASH_WUNLOCK(&V_tcbinfo); 452 } 453 if (error == 0) { 454 tcp_state_change(tp, TCPS_LISTEN); 455 solisten_proto(so, backlog); 456 #ifdef TCP_OFFLOAD 457 if ((so->so_options & SO_NO_OFFLOAD) == 0) 458 tcp_offload_listen_start(tp); 459 #endif 460 } else { 461 solisten_proto_abort(so); 462 } 463 SOCK_UNLOCK(so); 464 465 if (IS_FASTOPEN(tp->t_flags)) 466 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 467 468 out: 469 TCPDEBUG2(PRU_LISTEN); 470 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 471 INP_WUNLOCK(inp); 472 return (error); 473 } 474 #endif /* INET */ 475 476 #ifdef INET6 477 static int 478 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 479 { 480 int error = 0; 481 struct inpcb *inp; 482 struct tcpcb *tp = NULL; 483 u_char vflagsav; 484 485 TCPDEBUG0; 486 inp = sotoinpcb(so); 487 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 488 INP_WLOCK(inp); 489 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 490 error = EINVAL; 491 goto out; 492 } 493 vflagsav = inp->inp_vflag; 494 tp = intotcpcb(inp); 495 TCPDEBUG1(); 496 SOCK_LOCK(so); 497 error = solisten_proto_check(so); 498 if (error != 0) { 499 SOCK_UNLOCK(so); 500 goto out; 501 } 502 INP_HASH_WLOCK(&V_tcbinfo); 503 if (inp->inp_lport == 0) { 504 inp->inp_vflag &= ~INP_IPV4; 505 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 506 inp->inp_vflag |= INP_IPV4; 507 error = in6_pcbbind(inp, NULL, td->td_ucred); 508 } 509 INP_HASH_WUNLOCK(&V_tcbinfo); 510 if (error == 0) { 511 tcp_state_change(tp, TCPS_LISTEN); 512 solisten_proto(so, backlog); 513 #ifdef TCP_OFFLOAD 514 if ((so->so_options & SO_NO_OFFLOAD) == 0) 515 tcp_offload_listen_start(tp); 516 #endif 517 } else { 518 solisten_proto_abort(so); 519 } 520 SOCK_UNLOCK(so); 521 522 if (IS_FASTOPEN(tp->t_flags)) 523 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 524 525 if (error != 0) 526 inp->inp_vflag = vflagsav; 527 528 out: 529 TCPDEBUG2(PRU_LISTEN); 530 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 531 INP_WUNLOCK(inp); 532 return (error); 533 } 534 #endif /* INET6 */ 535 536 #ifdef INET 537 /* 538 * Initiate connection to peer. 539 * Create a template for use in transmissions on this connection. 540 * Enter SYN_SENT state, and mark socket as connecting. 541 * Start keep-alive timer, and seed output sequence space. 542 * Send initial segment on connection. 543 */ 544 static int 545 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 546 { 547 struct epoch_tracker et; 548 int error = 0; 549 struct inpcb *inp; 550 struct tcpcb *tp = NULL; 551 struct sockaddr_in *sinp; 552 553 sinp = (struct sockaddr_in *)nam; 554 if (nam->sa_family != AF_INET) 555 return (EAFNOSUPPORT); 556 if (nam->sa_len != sizeof (*sinp)) 557 return (EINVAL); 558 559 /* 560 * Must disallow TCP ``connections'' to multicast addresses. 561 */ 562 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 563 return (EAFNOSUPPORT); 564 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) 565 return (EACCES); 566 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 567 return (error); 568 569 TCPDEBUG0; 570 inp = sotoinpcb(so); 571 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 572 INP_WLOCK(inp); 573 if (inp->inp_flags & INP_TIMEWAIT) { 574 error = EADDRINUSE; 575 goto out; 576 } 577 if (inp->inp_flags & INP_DROPPED) { 578 error = ECONNREFUSED; 579 goto out; 580 } 581 if (SOLISTENING(so)) { 582 error = EOPNOTSUPP; 583 goto out; 584 } 585 tp = intotcpcb(inp); 586 TCPDEBUG1(); 587 NET_EPOCH_ENTER(et); 588 if ((error = tcp_connect(tp, nam, td)) != 0) 589 goto out_in_epoch; 590 #ifdef TCP_OFFLOAD 591 if (registered_toedevs > 0 && 592 (so->so_options & SO_NO_OFFLOAD) == 0 && 593 (error = tcp_offload_connect(so, nam)) == 0) 594 goto out_in_epoch; 595 #endif 596 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 597 error = tcp_output(tp); 598 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 599 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 600 out_in_epoch: 601 NET_EPOCH_EXIT(et); 602 out: 603 TCPDEBUG2(PRU_CONNECT); 604 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 605 INP_WUNLOCK(inp); 606 return (error); 607 } 608 #endif /* INET */ 609 610 #ifdef INET6 611 static int 612 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 613 { 614 struct epoch_tracker et; 615 int error = 0; 616 struct inpcb *inp; 617 struct tcpcb *tp = NULL; 618 struct sockaddr_in6 *sin6; 619 u_int8_t incflagsav; 620 u_char vflagsav; 621 622 TCPDEBUG0; 623 624 sin6 = (struct sockaddr_in6 *)nam; 625 if (nam->sa_family != AF_INET6) 626 return (EAFNOSUPPORT); 627 if (nam->sa_len != sizeof (*sin6)) 628 return (EINVAL); 629 630 /* 631 * Must disallow TCP ``connections'' to multicast addresses. 632 */ 633 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 634 return (EAFNOSUPPORT); 635 636 inp = sotoinpcb(so); 637 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 638 INP_WLOCK(inp); 639 vflagsav = inp->inp_vflag; 640 incflagsav = inp->inp_inc.inc_flags; 641 if (inp->inp_flags & INP_TIMEWAIT) { 642 error = EADDRINUSE; 643 goto out; 644 } 645 if (inp->inp_flags & INP_DROPPED) { 646 error = ECONNREFUSED; 647 goto out; 648 } 649 if (SOLISTENING(so)) { 650 error = EINVAL; 651 goto out; 652 } 653 tp = intotcpcb(inp); 654 TCPDEBUG1(); 655 #ifdef INET 656 /* 657 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 658 * therefore probably require the hash lock, which isn't held here. 659 * Is this a significant problem? 660 */ 661 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 662 struct sockaddr_in sin; 663 664 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 665 error = EINVAL; 666 goto out; 667 } 668 if ((inp->inp_vflag & INP_IPV4) == 0) { 669 error = EAFNOSUPPORT; 670 goto out; 671 } 672 673 in6_sin6_2_sin(&sin, sin6); 674 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 675 error = EAFNOSUPPORT; 676 goto out; 677 } 678 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 679 error = EACCES; 680 goto out; 681 } 682 if ((error = prison_remote_ip4(td->td_ucred, 683 &sin.sin_addr)) != 0) 684 goto out; 685 inp->inp_vflag |= INP_IPV4; 686 inp->inp_vflag &= ~INP_IPV6; 687 NET_EPOCH_ENTER(et); 688 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 689 goto out_in_epoch; 690 #ifdef TCP_OFFLOAD 691 if (registered_toedevs > 0 && 692 (so->so_options & SO_NO_OFFLOAD) == 0 && 693 (error = tcp_offload_connect(so, nam)) == 0) 694 goto out_in_epoch; 695 #endif 696 error = tcp_output(tp); 697 goto out_in_epoch; 698 } else { 699 if ((inp->inp_vflag & INP_IPV6) == 0) { 700 error = EAFNOSUPPORT; 701 goto out; 702 } 703 } 704 #endif 705 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 706 goto out; 707 inp->inp_vflag &= ~INP_IPV4; 708 inp->inp_vflag |= INP_IPV6; 709 inp->inp_inc.inc_flags |= INC_ISIPV6; 710 if ((error = tcp6_connect(tp, nam, td)) != 0) 711 goto out; 712 #ifdef TCP_OFFLOAD 713 if (registered_toedevs > 0 && 714 (so->so_options & SO_NO_OFFLOAD) == 0 && 715 (error = tcp_offload_connect(so, nam)) == 0) 716 goto out; 717 #endif 718 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 719 NET_EPOCH_ENTER(et); 720 error = tcp_output(tp); 721 #ifdef INET 722 out_in_epoch: 723 #endif 724 NET_EPOCH_EXIT(et); 725 out: 726 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 727 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 728 /* 729 * If the implicit bind in the connect call fails, restore 730 * the flags we modified. 731 */ 732 if (error != 0 && inp->inp_lport == 0) { 733 inp->inp_vflag = vflagsav; 734 inp->inp_inc.inc_flags = incflagsav; 735 } 736 737 TCPDEBUG2(PRU_CONNECT); 738 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 739 INP_WUNLOCK(inp); 740 return (error); 741 } 742 #endif /* INET6 */ 743 744 /* 745 * Initiate disconnect from peer. 746 * If connection never passed embryonic stage, just drop; 747 * else if don't need to let data drain, then can just drop anyways, 748 * else have to begin TCP shutdown process: mark socket disconnecting, 749 * drain unread data, state switch to reflect user close, and 750 * send segment (e.g. FIN) to peer. Socket will be really disconnected 751 * when peer sends FIN and acks ours. 752 * 753 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 754 */ 755 static int 756 tcp_usr_disconnect(struct socket *so) 757 { 758 struct inpcb *inp; 759 struct tcpcb *tp = NULL; 760 struct epoch_tracker et; 761 int error = 0; 762 763 TCPDEBUG0; 764 NET_EPOCH_ENTER(et); 765 inp = sotoinpcb(so); 766 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 767 INP_WLOCK(inp); 768 if (inp->inp_flags & INP_TIMEWAIT) 769 goto out; 770 if (inp->inp_flags & INP_DROPPED) { 771 error = ECONNRESET; 772 goto out; 773 } 774 tp = intotcpcb(inp); 775 TCPDEBUG1(); 776 tcp_disconnect(tp); 777 out: 778 TCPDEBUG2(PRU_DISCONNECT); 779 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 780 INP_WUNLOCK(inp); 781 NET_EPOCH_EXIT(et); 782 return (error); 783 } 784 785 #ifdef INET 786 /* 787 * Accept a connection. Essentially all the work is done at higher levels; 788 * just return the address of the peer, storing through addr. 789 */ 790 static int 791 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 792 { 793 int error = 0; 794 struct inpcb *inp = NULL; 795 struct tcpcb *tp = NULL; 796 struct in_addr addr; 797 in_port_t port = 0; 798 TCPDEBUG0; 799 800 if (so->so_state & SS_ISDISCONNECTED) 801 return (ECONNABORTED); 802 803 inp = sotoinpcb(so); 804 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 805 INP_WLOCK(inp); 806 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 807 error = ECONNABORTED; 808 goto out; 809 } 810 tp = intotcpcb(inp); 811 TCPDEBUG1(); 812 813 /* 814 * We inline in_getpeeraddr and COMMON_END here, so that we can 815 * copy the data of interest and defer the malloc until after we 816 * release the lock. 817 */ 818 port = inp->inp_fport; 819 addr = inp->inp_faddr; 820 821 out: 822 TCPDEBUG2(PRU_ACCEPT); 823 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 824 INP_WUNLOCK(inp); 825 if (error == 0) 826 *nam = in_sockaddr(port, &addr); 827 return error; 828 } 829 #endif /* INET */ 830 831 #ifdef INET6 832 static int 833 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 834 { 835 struct inpcb *inp = NULL; 836 int error = 0; 837 struct tcpcb *tp = NULL; 838 struct in_addr addr; 839 struct in6_addr addr6; 840 struct epoch_tracker et; 841 in_port_t port = 0; 842 int v4 = 0; 843 TCPDEBUG0; 844 845 if (so->so_state & SS_ISDISCONNECTED) 846 return (ECONNABORTED); 847 848 inp = sotoinpcb(so); 849 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 850 NET_EPOCH_ENTER(et); 851 INP_WLOCK(inp); 852 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 853 error = ECONNABORTED; 854 goto out; 855 } 856 tp = intotcpcb(inp); 857 TCPDEBUG1(); 858 859 /* 860 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 861 * copy the data of interest and defer the malloc until after we 862 * release the lock. 863 */ 864 if (inp->inp_vflag & INP_IPV4) { 865 v4 = 1; 866 port = inp->inp_fport; 867 addr = inp->inp_faddr; 868 } else { 869 port = inp->inp_fport; 870 addr6 = inp->in6p_faddr; 871 } 872 873 out: 874 TCPDEBUG2(PRU_ACCEPT); 875 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 876 INP_WUNLOCK(inp); 877 NET_EPOCH_EXIT(et); 878 if (error == 0) { 879 if (v4) 880 *nam = in6_v4mapsin6_sockaddr(port, &addr); 881 else 882 *nam = in6_sockaddr(port, &addr6); 883 } 884 return error; 885 } 886 #endif /* INET6 */ 887 888 /* 889 * Mark the connection as being incapable of further output. 890 */ 891 static int 892 tcp_usr_shutdown(struct socket *so) 893 { 894 int error = 0; 895 struct inpcb *inp; 896 struct tcpcb *tp = NULL; 897 struct epoch_tracker et; 898 899 TCPDEBUG0; 900 inp = sotoinpcb(so); 901 KASSERT(inp != NULL, ("inp == NULL")); 902 INP_WLOCK(inp); 903 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 904 INP_WUNLOCK(inp); 905 return (ECONNRESET); 906 } 907 tp = intotcpcb(inp); 908 NET_EPOCH_ENTER(et); 909 TCPDEBUG1(); 910 socantsendmore(so); 911 tcp_usrclosed(tp); 912 if (!(inp->inp_flags & INP_DROPPED)) 913 error = tcp_output_nodrop(tp); 914 TCPDEBUG2(PRU_SHUTDOWN); 915 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 916 error = tcp_unlock_or_drop(tp, error); 917 NET_EPOCH_EXIT(et); 918 919 return (error); 920 } 921 922 /* 923 * After a receive, possibly send window update to peer. 924 */ 925 static int 926 tcp_usr_rcvd(struct socket *so, int flags) 927 { 928 struct epoch_tracker et; 929 struct inpcb *inp; 930 struct tcpcb *tp = NULL; 931 int outrv = 0, error = 0; 932 933 TCPDEBUG0; 934 inp = sotoinpcb(so); 935 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 936 INP_WLOCK(inp); 937 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 938 INP_WUNLOCK(inp); 939 return (ECONNRESET); 940 } 941 tp = intotcpcb(inp); 942 NET_EPOCH_ENTER(et); 943 TCPDEBUG1(); 944 /* 945 * For passively-created TFO connections, don't attempt a window 946 * update while still in SYN_RECEIVED as this may trigger an early 947 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 948 * application response data, or failing that, when the DELACK timer 949 * expires. 950 */ 951 if (IS_FASTOPEN(tp->t_flags) && 952 (tp->t_state == TCPS_SYN_RECEIVED)) 953 goto out; 954 #ifdef TCP_OFFLOAD 955 if (tp->t_flags & TF_TOE) 956 tcp_offload_rcvd(tp); 957 else 958 #endif 959 outrv = tcp_output_nodrop(tp); 960 out: 961 TCPDEBUG2(PRU_RCVD); 962 TCP_PROBE2(debug__user, tp, PRU_RCVD); 963 (void) tcp_unlock_or_drop(tp, outrv); 964 NET_EPOCH_EXIT(et); 965 return (error); 966 } 967 968 /* 969 * Do a send by putting data in output queue and updating urgent 970 * marker if URG set. Possibly send more data. Unlike the other 971 * pru_*() routines, the mbuf chains are our responsibility. We 972 * must either enqueue them or free them. The other pru_* routines 973 * generally are caller-frees. 974 */ 975 static int 976 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 977 struct sockaddr *nam, struct mbuf *control, struct thread *td) 978 { 979 struct epoch_tracker et; 980 int error = 0; 981 struct inpcb *inp; 982 struct tcpcb *tp = NULL; 983 #ifdef INET 984 #ifdef INET6 985 struct sockaddr_in sin; 986 #endif 987 struct sockaddr_in *sinp; 988 #endif 989 #ifdef INET6 990 int isipv6; 991 #endif 992 u_int8_t incflagsav; 993 u_char vflagsav; 994 bool restoreflags; 995 TCPDEBUG0; 996 997 if (control != NULL) { 998 /* TCP doesn't do control messages (rights, creds, etc) */ 999 if (control->m_len) { 1000 m_freem(control); 1001 return (EINVAL); 1002 } 1003 m_freem(control); /* empty control, just free it */ 1004 } 1005 1006 inp = sotoinpcb(so); 1007 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 1008 INP_WLOCK(inp); 1009 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1010 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1011 m_freem(m); 1012 INP_WUNLOCK(inp); 1013 return (ECONNRESET); 1014 } 1015 1016 vflagsav = inp->inp_vflag; 1017 incflagsav = inp->inp_inc.inc_flags; 1018 restoreflags = false; 1019 tp = intotcpcb(inp); 1020 1021 NET_EPOCH_ENTER(et); 1022 if ((flags & PRUS_OOB) != 0 && 1023 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 1024 goto out; 1025 1026 TCPDEBUG1(); 1027 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 1028 if (tp->t_state == TCPS_LISTEN) { 1029 error = EINVAL; 1030 goto out; 1031 } 1032 switch (nam->sa_family) { 1033 #ifdef INET 1034 case AF_INET: 1035 sinp = (struct sockaddr_in *)nam; 1036 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 1037 error = EINVAL; 1038 goto out; 1039 } 1040 if ((inp->inp_vflag & INP_IPV6) != 0) { 1041 error = EAFNOSUPPORT; 1042 goto out; 1043 } 1044 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 1045 error = EAFNOSUPPORT; 1046 goto out; 1047 } 1048 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 1049 error = EACCES; 1050 goto out; 1051 } 1052 if ((error = prison_remote_ip4(td->td_ucred, 1053 &sinp->sin_addr))) 1054 goto out; 1055 #ifdef INET6 1056 isipv6 = 0; 1057 #endif 1058 break; 1059 #endif /* INET */ 1060 #ifdef INET6 1061 case AF_INET6: 1062 { 1063 struct sockaddr_in6 *sin6; 1064 1065 sin6 = (struct sockaddr_in6 *)nam; 1066 if (sin6->sin6_len != sizeof(*sin6)) { 1067 error = EINVAL; 1068 goto out; 1069 } 1070 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1071 error = EAFNOSUPPORT; 1072 goto out; 1073 } 1074 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1075 error = EAFNOSUPPORT; 1076 goto out; 1077 } 1078 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1079 #ifdef INET 1080 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1081 error = EINVAL; 1082 goto out; 1083 } 1084 if ((inp->inp_vflag & INP_IPV4) == 0) { 1085 error = EAFNOSUPPORT; 1086 goto out; 1087 } 1088 restoreflags = true; 1089 inp->inp_vflag &= ~INP_IPV6; 1090 sinp = &sin; 1091 in6_sin6_2_sin(sinp, sin6); 1092 if (IN_MULTICAST( 1093 ntohl(sinp->sin_addr.s_addr))) { 1094 error = EAFNOSUPPORT; 1095 goto out; 1096 } 1097 if ((error = prison_remote_ip4(td->td_ucred, 1098 &sinp->sin_addr))) 1099 goto out; 1100 isipv6 = 0; 1101 #else /* !INET */ 1102 error = EAFNOSUPPORT; 1103 goto out; 1104 #endif /* INET */ 1105 } else { 1106 if ((inp->inp_vflag & INP_IPV6) == 0) { 1107 error = EAFNOSUPPORT; 1108 goto out; 1109 } 1110 restoreflags = true; 1111 inp->inp_vflag &= ~INP_IPV4; 1112 inp->inp_inc.inc_flags |= INC_ISIPV6; 1113 if ((error = prison_remote_ip6(td->td_ucred, 1114 &sin6->sin6_addr))) 1115 goto out; 1116 isipv6 = 1; 1117 } 1118 break; 1119 } 1120 #endif /* INET6 */ 1121 default: 1122 error = EAFNOSUPPORT; 1123 goto out; 1124 } 1125 } 1126 if (!(flags & PRUS_OOB)) { 1127 sbappendstream(&so->so_snd, m, flags); 1128 m = NULL; 1129 if (nam && tp->t_state < TCPS_SYN_SENT) { 1130 KASSERT(tp->t_state == TCPS_CLOSED, 1131 ("%s: tp %p is listening", __func__, tp)); 1132 1133 /* 1134 * Do implied connect if not yet connected, 1135 * initialize window to default value, and 1136 * initialize maxseg using peer's cached MSS. 1137 */ 1138 #ifdef INET6 1139 if (isipv6) 1140 error = tcp6_connect(tp, nam, td); 1141 #endif /* INET6 */ 1142 #if defined(INET6) && defined(INET) 1143 else 1144 #endif 1145 #ifdef INET 1146 error = tcp_connect(tp, 1147 (struct sockaddr *)sinp, td); 1148 #endif 1149 /* 1150 * The bind operation in tcp_connect succeeded. We 1151 * no longer want to restore the flags if later 1152 * operations fail. 1153 */ 1154 if (error == 0 || inp->inp_lport != 0) 1155 restoreflags = false; 1156 1157 if (error) { 1158 /* m is freed if PRUS_NOTREADY is unset. */ 1159 sbflush(&so->so_snd); 1160 goto out; 1161 } 1162 if (IS_FASTOPEN(tp->t_flags)) 1163 tcp_fastopen_connect(tp); 1164 else { 1165 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1166 tcp_mss(tp, -1); 1167 } 1168 } 1169 if (flags & PRUS_EOF) { 1170 /* 1171 * Close the send side of the connection after 1172 * the data is sent. 1173 */ 1174 socantsendmore(so); 1175 tcp_usrclosed(tp); 1176 } 1177 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1178 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1179 (tp->t_fbyte_out == 0) && 1180 (so->so_snd.sb_ccc > 0)) { 1181 tp->t_fbyte_out = ticks; 1182 if (tp->t_fbyte_out == 0) 1183 tp->t_fbyte_out = 1; 1184 if (tp->t_fbyte_out && tp->t_fbyte_in) 1185 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1186 } 1187 if (!(inp->inp_flags & INP_DROPPED) && 1188 !(flags & PRUS_NOTREADY)) { 1189 if (flags & PRUS_MORETOCOME) 1190 tp->t_flags |= TF_MORETOCOME; 1191 error = tcp_output_nodrop(tp); 1192 if (flags & PRUS_MORETOCOME) 1193 tp->t_flags &= ~TF_MORETOCOME; 1194 } 1195 } else { 1196 /* 1197 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1198 */ 1199 SOCKBUF_LOCK(&so->so_snd); 1200 if (sbspace(&so->so_snd) < -512) { 1201 SOCKBUF_UNLOCK(&so->so_snd); 1202 error = ENOBUFS; 1203 goto out; 1204 } 1205 /* 1206 * According to RFC961 (Assigned Protocols), 1207 * the urgent pointer points to the last octet 1208 * of urgent data. We continue, however, 1209 * to consider it to indicate the first octet 1210 * of data past the urgent section. 1211 * Otherwise, snd_up should be one lower. 1212 */ 1213 sbappendstream_locked(&so->so_snd, m, flags); 1214 SOCKBUF_UNLOCK(&so->so_snd); 1215 m = NULL; 1216 if (nam && tp->t_state < TCPS_SYN_SENT) { 1217 /* 1218 * Do implied connect if not yet connected, 1219 * initialize window to default value, and 1220 * initialize maxseg using peer's cached MSS. 1221 */ 1222 1223 /* 1224 * Not going to contemplate SYN|URG 1225 */ 1226 if (IS_FASTOPEN(tp->t_flags)) 1227 tp->t_flags &= ~TF_FASTOPEN; 1228 #ifdef INET6 1229 if (isipv6) 1230 error = tcp6_connect(tp, nam, td); 1231 #endif /* INET6 */ 1232 #if defined(INET6) && defined(INET) 1233 else 1234 #endif 1235 #ifdef INET 1236 error = tcp_connect(tp, 1237 (struct sockaddr *)sinp, td); 1238 #endif 1239 /* 1240 * The bind operation in tcp_connect succeeded. We 1241 * no longer want to restore the flags if later 1242 * operations fail. 1243 */ 1244 if (error == 0 || inp->inp_lport != 0) 1245 restoreflags = false; 1246 1247 if (error != 0) { 1248 /* m is freed if PRUS_NOTREADY is unset. */ 1249 sbflush(&so->so_snd); 1250 goto out; 1251 } 1252 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1253 tcp_mss(tp, -1); 1254 } 1255 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1256 if ((flags & PRUS_NOTREADY) == 0) { 1257 tp->t_flags |= TF_FORCEDATA; 1258 error = tcp_output_nodrop(tp); 1259 tp->t_flags &= ~TF_FORCEDATA; 1260 } 1261 } 1262 TCP_LOG_EVENT(tp, NULL, 1263 &inp->inp_socket->so_rcv, 1264 &inp->inp_socket->so_snd, 1265 TCP_LOG_USERSEND, error, 1266 0, NULL, false); 1267 1268 out: 1269 /* 1270 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1271 * responsible for freeing memory. 1272 */ 1273 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1274 m_freem(m); 1275 1276 /* 1277 * If the request was unsuccessful and we changed flags, 1278 * restore the original flags. 1279 */ 1280 if (error != 0 && restoreflags) { 1281 inp->inp_vflag = vflagsav; 1282 inp->inp_inc.inc_flags = incflagsav; 1283 } 1284 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1285 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1286 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1287 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1288 error = tcp_unlock_or_drop(tp, error); 1289 NET_EPOCH_EXIT(et); 1290 return (error); 1291 } 1292 1293 static int 1294 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1295 { 1296 struct epoch_tracker et; 1297 struct inpcb *inp; 1298 struct tcpcb *tp; 1299 int error; 1300 1301 inp = sotoinpcb(so); 1302 INP_WLOCK(inp); 1303 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1304 INP_WUNLOCK(inp); 1305 mb_free_notready(m, count); 1306 return (ECONNRESET); 1307 } 1308 tp = intotcpcb(inp); 1309 1310 SOCKBUF_LOCK(&so->so_snd); 1311 error = sbready(&so->so_snd, m, count); 1312 SOCKBUF_UNLOCK(&so->so_snd); 1313 if (error) { 1314 INP_WUNLOCK(inp); 1315 return (error); 1316 } 1317 NET_EPOCH_ENTER(et); 1318 error = tcp_output_unlock(tp); 1319 NET_EPOCH_EXIT(et); 1320 1321 return (error); 1322 } 1323 1324 /* 1325 * Abort the TCP. Drop the connection abruptly. 1326 */ 1327 static void 1328 tcp_usr_abort(struct socket *so) 1329 { 1330 struct inpcb *inp; 1331 struct tcpcb *tp = NULL; 1332 struct epoch_tracker et; 1333 TCPDEBUG0; 1334 1335 inp = sotoinpcb(so); 1336 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1337 1338 NET_EPOCH_ENTER(et); 1339 INP_WLOCK(inp); 1340 KASSERT(inp->inp_socket != NULL, 1341 ("tcp_usr_abort: inp_socket == NULL")); 1342 1343 /* 1344 * If we still have full TCP state, and we're not dropped, drop. 1345 */ 1346 if (!(inp->inp_flags & INP_TIMEWAIT) && 1347 !(inp->inp_flags & INP_DROPPED)) { 1348 tp = intotcpcb(inp); 1349 TCPDEBUG1(); 1350 tp = tcp_drop(tp, ECONNABORTED); 1351 if (tp == NULL) 1352 goto dropped; 1353 TCPDEBUG2(PRU_ABORT); 1354 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1355 } 1356 if (!(inp->inp_flags & INP_DROPPED)) { 1357 SOCK_LOCK(so); 1358 so->so_state |= SS_PROTOREF; 1359 SOCK_UNLOCK(so); 1360 inp->inp_flags |= INP_SOCKREF; 1361 } 1362 INP_WUNLOCK(inp); 1363 dropped: 1364 NET_EPOCH_EXIT(et); 1365 } 1366 1367 /* 1368 * TCP socket is closed. Start friendly disconnect. 1369 */ 1370 static void 1371 tcp_usr_close(struct socket *so) 1372 { 1373 struct inpcb *inp; 1374 struct tcpcb *tp = NULL; 1375 struct epoch_tracker et; 1376 TCPDEBUG0; 1377 1378 inp = sotoinpcb(so); 1379 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1380 1381 NET_EPOCH_ENTER(et); 1382 INP_WLOCK(inp); 1383 KASSERT(inp->inp_socket != NULL, 1384 ("tcp_usr_close: inp_socket == NULL")); 1385 1386 /* 1387 * If we still have full TCP state, and we're not dropped, initiate 1388 * a disconnect. 1389 */ 1390 if (!(inp->inp_flags & INP_TIMEWAIT) && 1391 !(inp->inp_flags & INP_DROPPED)) { 1392 tp = intotcpcb(inp); 1393 TCPDEBUG1(); 1394 tcp_disconnect(tp); 1395 TCPDEBUG2(PRU_CLOSE); 1396 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1397 } 1398 if (!(inp->inp_flags & INP_DROPPED)) { 1399 SOCK_LOCK(so); 1400 so->so_state |= SS_PROTOREF; 1401 SOCK_UNLOCK(so); 1402 inp->inp_flags |= INP_SOCKREF; 1403 } 1404 INP_WUNLOCK(inp); 1405 NET_EPOCH_EXIT(et); 1406 } 1407 1408 static int 1409 tcp_pru_options_support(struct tcpcb *tp, int flags) 1410 { 1411 /* 1412 * If the specific TCP stack has a pru_options 1413 * specified then it does not always support 1414 * all the PRU_XX options and we must ask it. 1415 * If the function is not specified then all 1416 * of the PRU_XX options are supported. 1417 */ 1418 int ret = 0; 1419 1420 if (tp->t_fb->tfb_pru_options) { 1421 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1422 } 1423 return (ret); 1424 } 1425 1426 /* 1427 * Receive out-of-band data. 1428 */ 1429 static int 1430 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1431 { 1432 int error = 0; 1433 struct inpcb *inp; 1434 struct tcpcb *tp = NULL; 1435 1436 TCPDEBUG0; 1437 inp = sotoinpcb(so); 1438 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1439 INP_WLOCK(inp); 1440 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1441 error = ECONNRESET; 1442 goto out; 1443 } 1444 tp = intotcpcb(inp); 1445 error = tcp_pru_options_support(tp, PRUS_OOB); 1446 if (error) { 1447 goto out; 1448 } 1449 TCPDEBUG1(); 1450 if ((so->so_oobmark == 0 && 1451 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1452 so->so_options & SO_OOBINLINE || 1453 tp->t_oobflags & TCPOOB_HADDATA) { 1454 error = EINVAL; 1455 goto out; 1456 } 1457 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1458 error = EWOULDBLOCK; 1459 goto out; 1460 } 1461 m->m_len = 1; 1462 *mtod(m, caddr_t) = tp->t_iobc; 1463 if ((flags & MSG_PEEK) == 0) 1464 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1465 1466 out: 1467 TCPDEBUG2(PRU_RCVOOB); 1468 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1469 INP_WUNLOCK(inp); 1470 return (error); 1471 } 1472 1473 #ifdef INET 1474 struct pr_usrreqs tcp_usrreqs = { 1475 .pru_abort = tcp_usr_abort, 1476 .pru_accept = tcp_usr_accept, 1477 .pru_attach = tcp_usr_attach, 1478 .pru_bind = tcp_usr_bind, 1479 .pru_connect = tcp_usr_connect, 1480 .pru_control = in_control, 1481 .pru_detach = tcp_usr_detach, 1482 .pru_disconnect = tcp_usr_disconnect, 1483 .pru_listen = tcp_usr_listen, 1484 .pru_peeraddr = in_getpeeraddr, 1485 .pru_rcvd = tcp_usr_rcvd, 1486 .pru_rcvoob = tcp_usr_rcvoob, 1487 .pru_send = tcp_usr_send, 1488 .pru_ready = tcp_usr_ready, 1489 .pru_shutdown = tcp_usr_shutdown, 1490 .pru_sockaddr = in_getsockaddr, 1491 .pru_sosetlabel = in_pcbsosetlabel, 1492 .pru_close = tcp_usr_close, 1493 }; 1494 #endif /* INET */ 1495 1496 #ifdef INET6 1497 struct pr_usrreqs tcp6_usrreqs = { 1498 .pru_abort = tcp_usr_abort, 1499 .pru_accept = tcp6_usr_accept, 1500 .pru_attach = tcp_usr_attach, 1501 .pru_bind = tcp6_usr_bind, 1502 .pru_connect = tcp6_usr_connect, 1503 .pru_control = in6_control, 1504 .pru_detach = tcp_usr_detach, 1505 .pru_disconnect = tcp_usr_disconnect, 1506 .pru_listen = tcp6_usr_listen, 1507 .pru_peeraddr = in6_mapped_peeraddr, 1508 .pru_rcvd = tcp_usr_rcvd, 1509 .pru_rcvoob = tcp_usr_rcvoob, 1510 .pru_send = tcp_usr_send, 1511 .pru_ready = tcp_usr_ready, 1512 .pru_shutdown = tcp_usr_shutdown, 1513 .pru_sockaddr = in6_mapped_sockaddr, 1514 .pru_sosetlabel = in_pcbsosetlabel, 1515 .pru_close = tcp_usr_close, 1516 }; 1517 #endif /* INET6 */ 1518 1519 #ifdef INET 1520 /* 1521 * Common subroutine to open a TCP connection to remote host specified 1522 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1523 * port number if needed. Call in_pcbconnect_setup to do the routing and 1524 * to choose a local host address (interface). If there is an existing 1525 * incarnation of the same connection in TIME-WAIT state and if the remote 1526 * host was sending CC options and if the connection duration was < MSL, then 1527 * truncate the previous TIME-WAIT state and proceed. 1528 * Initialize connection parameters and enter SYN-SENT state. 1529 */ 1530 static int 1531 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1532 { 1533 struct inpcb *inp = tp->t_inpcb, *oinp; 1534 struct socket *so = inp->inp_socket; 1535 struct in_addr laddr; 1536 u_short lport; 1537 int error; 1538 1539 NET_EPOCH_ASSERT(); 1540 INP_WLOCK_ASSERT(inp); 1541 INP_HASH_WLOCK(&V_tcbinfo); 1542 1543 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1544 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1545 if (error) 1546 goto out; 1547 } 1548 1549 /* 1550 * Cannot simply call in_pcbconnect, because there might be an 1551 * earlier incarnation of this same connection still in 1552 * TIME_WAIT state, creating an ADDRINUSE error. 1553 */ 1554 laddr = inp->inp_laddr; 1555 lport = inp->inp_lport; 1556 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1557 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1558 if (error && oinp == NULL) 1559 goto out; 1560 if (oinp) { 1561 error = EADDRINUSE; 1562 goto out; 1563 } 1564 /* Handle initial bind if it hadn't been done in advance. */ 1565 if (inp->inp_lport == 0) { 1566 inp->inp_lport = lport; 1567 if (in_pcbinshash(inp) != 0) { 1568 inp->inp_lport = 0; 1569 error = EAGAIN; 1570 goto out; 1571 } 1572 } 1573 inp->inp_laddr = laddr; 1574 in_pcbrehash(inp); 1575 INP_HASH_WUNLOCK(&V_tcbinfo); 1576 1577 /* 1578 * Compute window scaling to request: 1579 * Scale to fit into sweet spot. See tcp_syncache.c. 1580 * XXX: This should move to tcp_output(). 1581 */ 1582 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1583 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1584 tp->request_r_scale++; 1585 1586 soisconnecting(so); 1587 TCPSTAT_INC(tcps_connattempt); 1588 tcp_state_change(tp, TCPS_SYN_SENT); 1589 tp->iss = tcp_new_isn(&inp->inp_inc); 1590 if (tp->t_flags & TF_REQ_TSTMP) 1591 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1592 tcp_sendseqinit(tp); 1593 1594 return 0; 1595 1596 out: 1597 INP_HASH_WUNLOCK(&V_tcbinfo); 1598 return (error); 1599 } 1600 #endif /* INET */ 1601 1602 #ifdef INET6 1603 static int 1604 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1605 { 1606 struct inpcb *inp = tp->t_inpcb; 1607 int error; 1608 1609 INP_WLOCK_ASSERT(inp); 1610 INP_HASH_WLOCK(&V_tcbinfo); 1611 1612 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1613 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1614 if (error) 1615 goto out; 1616 } 1617 error = in6_pcbconnect(inp, nam, td->td_ucred); 1618 if (error != 0) 1619 goto out; 1620 INP_HASH_WUNLOCK(&V_tcbinfo); 1621 1622 /* Compute window scaling to request. */ 1623 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1624 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1625 tp->request_r_scale++; 1626 1627 soisconnecting(inp->inp_socket); 1628 TCPSTAT_INC(tcps_connattempt); 1629 tcp_state_change(tp, TCPS_SYN_SENT); 1630 tp->iss = tcp_new_isn(&inp->inp_inc); 1631 if (tp->t_flags & TF_REQ_TSTMP) 1632 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1633 tcp_sendseqinit(tp); 1634 1635 return 0; 1636 1637 out: 1638 INP_HASH_WUNLOCK(&V_tcbinfo); 1639 return error; 1640 } 1641 #endif /* INET6 */ 1642 1643 /* 1644 * Export TCP internal state information via a struct tcp_info, based on the 1645 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1646 * (TCP state machine, etc). We export all information using FreeBSD-native 1647 * constants -- for example, the numeric values for tcpi_state will differ 1648 * from Linux. 1649 */ 1650 static void 1651 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1652 { 1653 1654 INP_WLOCK_ASSERT(tp->t_inpcb); 1655 bzero(ti, sizeof(*ti)); 1656 1657 ti->tcpi_state = tp->t_state; 1658 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1659 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1660 if (tp->t_flags & TF_SACK_PERMIT) 1661 ti->tcpi_options |= TCPI_OPT_SACK; 1662 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1663 ti->tcpi_options |= TCPI_OPT_WSCALE; 1664 ti->tcpi_snd_wscale = tp->snd_scale; 1665 ti->tcpi_rcv_wscale = tp->rcv_scale; 1666 } 1667 if (tp->t_flags2 & TF2_ECN_PERMIT) 1668 ti->tcpi_options |= TCPI_OPT_ECN; 1669 1670 ti->tcpi_rto = tp->t_rxtcur * tick; 1671 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1672 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1673 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1674 1675 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1676 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1677 1678 /* 1679 * FreeBSD-specific extension fields for tcp_info. 1680 */ 1681 ti->tcpi_rcv_space = tp->rcv_wnd; 1682 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1683 ti->tcpi_snd_wnd = tp->snd_wnd; 1684 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1685 ti->tcpi_snd_nxt = tp->snd_nxt; 1686 ti->tcpi_snd_mss = tp->t_maxseg; 1687 ti->tcpi_rcv_mss = tp->t_maxseg; 1688 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1689 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1690 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1691 #ifdef TCP_OFFLOAD 1692 if (tp->t_flags & TF_TOE) { 1693 ti->tcpi_options |= TCPI_OPT_TOE; 1694 tcp_offload_tcp_info(tp, ti); 1695 } 1696 #endif 1697 } 1698 1699 /* 1700 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1701 * socket option arguments. When it re-acquires the lock after the copy, it 1702 * has to revalidate that the connection is still valid for the socket 1703 * option. 1704 */ 1705 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1706 INP_WLOCK(inp); \ 1707 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1708 INP_WUNLOCK(inp); \ 1709 cleanup; \ 1710 return (ECONNRESET); \ 1711 } \ 1712 tp = intotcpcb(inp); \ 1713 } while(0) 1714 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1715 1716 static int 1717 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1718 { 1719 struct tcpcb *tp = intotcpcb(inp); 1720 int error = 0; 1721 1722 MPASS(sopt->sopt_dir == SOPT_SET); 1723 INP_WLOCK_ASSERT(inp); 1724 1725 if (sopt->sopt_level != IPPROTO_TCP) { 1726 INP_WUNLOCK(inp); 1727 #ifdef INET6 1728 if (inp->inp_vflag & INP_IPV6PROTO) 1729 error = ip6_ctloutput(inp->inp_socket, sopt); 1730 #endif 1731 #if defined(INET6) && defined(INET) 1732 else 1733 #endif 1734 #ifdef INET 1735 error = ip_ctloutput(inp->inp_socket, sopt); 1736 #endif 1737 /* 1738 * When an IP-level socket option affects TCP, pass control 1739 * down to stack tfb_tcp_ctloutput, otherwise return what 1740 * IP level returned. 1741 */ 1742 switch (sopt->sopt_level) { 1743 #ifdef INET6 1744 case IPPROTO_IPV6: 1745 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1746 return (error); 1747 switch (sopt->sopt_name) { 1748 case IPV6_TCLASS: 1749 /* Notify tcp stacks that care (e.g. RACK). */ 1750 break; 1751 case IPV6_USE_MIN_MTU: 1752 /* Update t_maxseg accordingly. */ 1753 break; 1754 default: 1755 return (error); 1756 } 1757 break; 1758 #endif 1759 #ifdef INET 1760 case IPPROTO_IP: 1761 switch (sopt->sopt_name) { 1762 case IP_TOS: 1763 inp->inp_ip_tos &= ~IPTOS_ECN_MASK; 1764 break; 1765 case IP_TTL: 1766 /* Notify tcp stacks that care (e.g. RACK). */ 1767 break; 1768 default: 1769 return (error); 1770 } 1771 break; 1772 #endif 1773 default: 1774 return (error); 1775 } 1776 INP_WLOCK(inp); 1777 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1778 INP_WUNLOCK(inp); 1779 return (ECONNRESET); 1780 } 1781 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1782 /* 1783 * Protect the TCP option TCP_FUNCTION_BLK so 1784 * that a sub-function can *never* overwrite this. 1785 */ 1786 struct tcp_function_set fsn; 1787 struct tcp_function_block *blk; 1788 1789 INP_WUNLOCK(inp); 1790 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1791 if (error) 1792 return (error); 1793 1794 INP_WLOCK(inp); 1795 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1796 INP_WUNLOCK(inp); 1797 return (ECONNRESET); 1798 } 1799 tp = intotcpcb(inp); 1800 1801 blk = find_and_ref_tcp_functions(&fsn); 1802 if (blk == NULL) { 1803 INP_WUNLOCK(inp); 1804 return (ENOENT); 1805 } 1806 if (tp->t_fb == blk) { 1807 /* You already have this */ 1808 refcount_release(&blk->tfb_refcnt); 1809 INP_WUNLOCK(inp); 1810 return (0); 1811 } 1812 if (tp->t_state != TCPS_CLOSED) { 1813 /* 1814 * The user has advanced the state 1815 * past the initial point, we may not 1816 * be able to switch. 1817 */ 1818 if (blk->tfb_tcp_handoff_ok != NULL) { 1819 /* 1820 * Does the stack provide a 1821 * query mechanism, if so it may 1822 * still be possible? 1823 */ 1824 error = (*blk->tfb_tcp_handoff_ok)(tp); 1825 } else 1826 error = EINVAL; 1827 if (error) { 1828 refcount_release(&blk->tfb_refcnt); 1829 INP_WUNLOCK(inp); 1830 return(error); 1831 } 1832 } 1833 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1834 refcount_release(&blk->tfb_refcnt); 1835 INP_WUNLOCK(inp); 1836 return (ENOENT); 1837 } 1838 /* 1839 * Release the old refcnt, the 1840 * lookup acquired a ref on the 1841 * new one already. 1842 */ 1843 if (tp->t_fb->tfb_tcp_fb_fini) { 1844 struct epoch_tracker et; 1845 /* 1846 * Tell the stack to cleanup with 0 i.e. 1847 * the tcb is not going away. 1848 */ 1849 NET_EPOCH_ENTER(et); 1850 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1851 NET_EPOCH_EXIT(et); 1852 } 1853 #ifdef TCPHPTS 1854 /* Assure that we are not on any hpts */ 1855 tcp_hpts_remove(tp->t_inpcb); 1856 #endif 1857 if (blk->tfb_tcp_fb_init) { 1858 error = (*blk->tfb_tcp_fb_init)(tp); 1859 if (error) { 1860 refcount_release(&blk->tfb_refcnt); 1861 if (tp->t_fb->tfb_tcp_fb_init) { 1862 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1863 /* Fall back failed, drop the connection */ 1864 INP_WUNLOCK(inp); 1865 soabort(inp->inp_socket); 1866 return(error); 1867 } 1868 } 1869 goto err_out; 1870 } 1871 } 1872 refcount_release(&tp->t_fb->tfb_refcnt); 1873 tp->t_fb = blk; 1874 #ifdef TCP_OFFLOAD 1875 if (tp->t_flags & TF_TOE) { 1876 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1877 sopt->sopt_name); 1878 } 1879 #endif 1880 err_out: 1881 INP_WUNLOCK(inp); 1882 return (error); 1883 } 1884 1885 tp = intotcpcb(inp); 1886 1887 /* Pass in the INP locked, callee must unlock it. */ 1888 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt)); 1889 } 1890 1891 static int 1892 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1893 { 1894 int error = 0; 1895 struct tcpcb *tp; 1896 1897 MPASS(sopt->sopt_dir == SOPT_GET); 1898 INP_WLOCK_ASSERT(inp); 1899 1900 if (sopt->sopt_level != IPPROTO_TCP) { 1901 INP_WUNLOCK(inp); 1902 #ifdef INET6 1903 if (inp->inp_vflag & INP_IPV6PROTO) 1904 error = ip6_ctloutput(inp->inp_socket, sopt); 1905 #endif /* INET6 */ 1906 #if defined(INET6) && defined(INET) 1907 else 1908 #endif 1909 #ifdef INET 1910 error = ip_ctloutput(inp->inp_socket, sopt); 1911 #endif 1912 return (error); 1913 } 1914 tp = intotcpcb(inp); 1915 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1916 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1917 struct tcp_function_set fsn; 1918 1919 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1920 memset(&fsn, 0, sizeof(fsn)); 1921 find_tcp_function_alias(tp->t_fb, &fsn); 1922 } else { 1923 strncpy(fsn.function_set_name, 1924 tp->t_fb->tfb_tcp_block_name, 1925 TCP_FUNCTION_NAME_LEN_MAX); 1926 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1927 } 1928 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1929 INP_WUNLOCK(inp); 1930 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1931 return (error); 1932 } 1933 1934 /* Pass in the INP locked, callee must unlock it. */ 1935 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt)); 1936 } 1937 1938 int 1939 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1940 { 1941 struct inpcb *inp; 1942 1943 inp = sotoinpcb(so); 1944 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1945 1946 INP_WLOCK(inp); 1947 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1948 INP_WUNLOCK(inp); 1949 return (ECONNRESET); 1950 } 1951 if (sopt->sopt_dir == SOPT_SET) 1952 return (tcp_ctloutput_set(inp, sopt)); 1953 else if (sopt->sopt_dir == SOPT_GET) 1954 return (tcp_ctloutput_get(inp, sopt)); 1955 else 1956 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1957 } 1958 1959 /* 1960 * If this assert becomes untrue, we need to change the size of the buf 1961 * variable in tcp_default_ctloutput(). 1962 */ 1963 #ifdef CTASSERT 1964 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1965 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1966 #endif 1967 1968 #ifdef KERN_TLS 1969 static int 1970 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1971 { 1972 struct tls_enable_v0 tls_v0; 1973 int error; 1974 1975 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1976 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1977 sizeof(tls_v0)); 1978 if (error) 1979 return (error); 1980 memset(tls, 0, sizeof(*tls)); 1981 tls->cipher_key = tls_v0.cipher_key; 1982 tls->iv = tls_v0.iv; 1983 tls->auth_key = tls_v0.auth_key; 1984 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1985 tls->cipher_key_len = tls_v0.cipher_key_len; 1986 tls->iv_len = tls_v0.iv_len; 1987 tls->auth_algorithm = tls_v0.auth_algorithm; 1988 tls->auth_key_len = tls_v0.auth_key_len; 1989 tls->flags = tls_v0.flags; 1990 tls->tls_vmajor = tls_v0.tls_vmajor; 1991 tls->tls_vminor = tls_v0.tls_vminor; 1992 return (0); 1993 } 1994 1995 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1996 } 1997 #endif 1998 1999 extern struct cc_algo newreno_cc_algo; 2000 2001 static int 2002 tcp_congestion(struct inpcb *inp, struct sockopt *sopt) 2003 { 2004 struct cc_algo *algo; 2005 void *ptr = NULL; 2006 struct tcpcb *tp; 2007 struct cc_var cc_mem; 2008 char buf[TCP_CA_NAME_MAX]; 2009 size_t mem_sz; 2010 int error; 2011 2012 INP_WUNLOCK(inp); 2013 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 2014 if (error) 2015 return(error); 2016 buf[sopt->sopt_valsize] = '\0'; 2017 CC_LIST_RLOCK(); 2018 STAILQ_FOREACH(algo, &cc_list, entries) 2019 if (strncmp(buf, algo->name, 2020 TCP_CA_NAME_MAX) == 0) { 2021 if (algo->flags & CC_MODULE_BEING_REMOVED) { 2022 /* We can't "see" modules being unloaded */ 2023 continue; 2024 } 2025 break; 2026 } 2027 if (algo == NULL) { 2028 CC_LIST_RUNLOCK(); 2029 return(ESRCH); 2030 } 2031 do_over: 2032 if (algo->cb_init != NULL) { 2033 /* We can now pre-get the memory for the CC */ 2034 mem_sz = (*algo->cc_data_sz)(); 2035 if (mem_sz == 0) { 2036 goto no_mem_needed; 2037 } 2038 CC_LIST_RUNLOCK(); 2039 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 2040 CC_LIST_RLOCK(); 2041 STAILQ_FOREACH(algo, &cc_list, entries) 2042 if (strncmp(buf, algo->name, 2043 TCP_CA_NAME_MAX) == 0) 2044 break; 2045 if (algo == NULL) { 2046 if (ptr) 2047 free(ptr, M_CC_MEM); 2048 CC_LIST_RUNLOCK(); 2049 return(ESRCH); 2050 } 2051 } else { 2052 no_mem_needed: 2053 mem_sz = 0; 2054 ptr = NULL; 2055 } 2056 /* 2057 * Make sure its all clean and zero and also get 2058 * back the inplock. 2059 */ 2060 memset(&cc_mem, 0, sizeof(cc_mem)); 2061 if (mem_sz != (*algo->cc_data_sz)()) { 2062 if (ptr) 2063 free(ptr, M_CC_MEM); 2064 goto do_over; 2065 } 2066 INP_WLOCK(inp); 2067 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 2068 INP_WUNLOCK(inp); 2069 CC_LIST_RUNLOCK(); 2070 free(ptr, M_CC_MEM); 2071 return (ECONNRESET); 2072 } 2073 tp = intotcpcb(inp); 2074 if (ptr != NULL) 2075 memset(ptr, 0, mem_sz); 2076 CC_LIST_RUNLOCK(); 2077 cc_mem.ccvc.tcp = tp; 2078 /* 2079 * We once again hold a write lock over the tcb so it's 2080 * safe to do these things without ordering concerns. 2081 * Note here we init into stack memory. 2082 */ 2083 if (algo->cb_init != NULL) 2084 error = algo->cb_init(&cc_mem, ptr); 2085 else 2086 error = 0; 2087 /* 2088 * The CC algorithms, when given their memory 2089 * should not fail we could in theory have a 2090 * KASSERT here. 2091 */ 2092 if (error == 0) { 2093 /* 2094 * Touchdown, lets go ahead and move the 2095 * connection to the new CC module by 2096 * copying in the cc_mem after we call 2097 * the old ones cleanup (if any). 2098 */ 2099 if (CC_ALGO(tp)->cb_destroy != NULL) 2100 CC_ALGO(tp)->cb_destroy(tp->ccv); 2101 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2102 tp->cc_algo = algo; 2103 /* Ok now are we where we have gotten past any conn_init? */ 2104 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2105 /* Yep run the connection init for the new CC */ 2106 CC_ALGO(tp)->conn_init(tp->ccv); 2107 } 2108 } else if (ptr) 2109 free(ptr, M_CC_MEM); 2110 INP_WUNLOCK(inp); 2111 return (error); 2112 } 2113 2114 int 2115 tcp_default_ctloutput(struct inpcb *inp, struct sockopt *sopt) 2116 { 2117 struct tcpcb *tp; 2118 int error, opt, optval; 2119 u_int ui; 2120 struct tcp_info ti; 2121 #ifdef KERN_TLS 2122 struct tls_enable tls; 2123 #endif 2124 char *pbuf, buf[TCP_LOG_ID_LEN]; 2125 #ifdef STATS 2126 struct statsblob *sbp; 2127 #endif 2128 size_t len; 2129 2130 INP_WLOCK_ASSERT(inp); 2131 2132 tp = intotcpcb(inp); 2133 switch (sopt->sopt_level) { 2134 #ifdef INET6 2135 case IPPROTO_IPV6: 2136 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2137 switch (sopt->sopt_name) { 2138 case IPV6_USE_MIN_MTU: 2139 tcp6_use_min_mtu(tp); 2140 /* FALLTHROUGH */ 2141 } 2142 INP_WUNLOCK(inp); 2143 return (0); 2144 #endif 2145 #ifdef INET 2146 case IPPROTO_IP: 2147 INP_WUNLOCK(inp); 2148 return (0); 2149 #endif 2150 } 2151 2152 /* 2153 * For TCP_CCALGOOPT forward the control to CC module, for both 2154 * SOPT_SET and SOPT_GET. 2155 */ 2156 switch (sopt->sopt_name) { 2157 case TCP_CCALGOOPT: 2158 INP_WUNLOCK(inp); 2159 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2160 return (EINVAL); 2161 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2162 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2163 sopt->sopt_valsize); 2164 if (error) { 2165 free(pbuf, M_TEMP); 2166 return (error); 2167 } 2168 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2169 if (CC_ALGO(tp)->ctl_output != NULL) 2170 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 2171 else 2172 error = ENOENT; 2173 INP_WUNLOCK(inp); 2174 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2175 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2176 free(pbuf, M_TEMP); 2177 return (error); 2178 } 2179 2180 switch (sopt->sopt_dir) { 2181 case SOPT_SET: 2182 switch (sopt->sopt_name) { 2183 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2184 case TCP_MD5SIG: 2185 if (!TCPMD5_ENABLED()) { 2186 INP_WUNLOCK(inp); 2187 return (ENOPROTOOPT); 2188 } 2189 error = TCPMD5_PCBCTL(inp, sopt); 2190 if (error) 2191 return (error); 2192 goto unlock_and_done; 2193 #endif /* IPSEC */ 2194 2195 case TCP_NODELAY: 2196 case TCP_NOOPT: 2197 case TCP_LRD: 2198 INP_WUNLOCK(inp); 2199 error = sooptcopyin(sopt, &optval, sizeof optval, 2200 sizeof optval); 2201 if (error) 2202 return (error); 2203 2204 INP_WLOCK_RECHECK(inp); 2205 switch (sopt->sopt_name) { 2206 case TCP_NODELAY: 2207 opt = TF_NODELAY; 2208 break; 2209 case TCP_NOOPT: 2210 opt = TF_NOOPT; 2211 break; 2212 case TCP_LRD: 2213 opt = TF_LRD; 2214 break; 2215 default: 2216 opt = 0; /* dead code to fool gcc */ 2217 break; 2218 } 2219 2220 if (optval) 2221 tp->t_flags |= opt; 2222 else 2223 tp->t_flags &= ~opt; 2224 unlock_and_done: 2225 #ifdef TCP_OFFLOAD 2226 if (tp->t_flags & TF_TOE) { 2227 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2228 sopt->sopt_name); 2229 } 2230 #endif 2231 INP_WUNLOCK(inp); 2232 break; 2233 2234 case TCP_NOPUSH: 2235 INP_WUNLOCK(inp); 2236 error = sooptcopyin(sopt, &optval, sizeof optval, 2237 sizeof optval); 2238 if (error) 2239 return (error); 2240 2241 INP_WLOCK_RECHECK(inp); 2242 if (optval) 2243 tp->t_flags |= TF_NOPUSH; 2244 else if (tp->t_flags & TF_NOPUSH) { 2245 tp->t_flags &= ~TF_NOPUSH; 2246 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2247 struct epoch_tracker et; 2248 2249 NET_EPOCH_ENTER(et); 2250 error = tcp_output_nodrop(tp); 2251 NET_EPOCH_EXIT(et); 2252 } 2253 } 2254 goto unlock_and_done; 2255 2256 case TCP_REMOTE_UDP_ENCAPS_PORT: 2257 INP_WUNLOCK(inp); 2258 error = sooptcopyin(sopt, &optval, sizeof optval, 2259 sizeof optval); 2260 if (error) 2261 return (error); 2262 if ((optval < TCP_TUNNELING_PORT_MIN) || 2263 (optval > TCP_TUNNELING_PORT_MAX)) { 2264 /* Its got to be in range */ 2265 return (EINVAL); 2266 } 2267 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2268 /* You have to have enabled a UDP tunneling port first */ 2269 return (EINVAL); 2270 } 2271 INP_WLOCK_RECHECK(inp); 2272 if (tp->t_state != TCPS_CLOSED) { 2273 /* You can't change after you are connected */ 2274 error = EINVAL; 2275 } else { 2276 /* Ok we are all good set the port */ 2277 tp->t_port = htons(optval); 2278 } 2279 goto unlock_and_done; 2280 2281 case TCP_MAXSEG: 2282 INP_WUNLOCK(inp); 2283 error = sooptcopyin(sopt, &optval, sizeof optval, 2284 sizeof optval); 2285 if (error) 2286 return (error); 2287 2288 INP_WLOCK_RECHECK(inp); 2289 if (optval > 0 && optval <= tp->t_maxseg && 2290 optval + 40 >= V_tcp_minmss) 2291 tp->t_maxseg = optval; 2292 else 2293 error = EINVAL; 2294 goto unlock_and_done; 2295 2296 case TCP_INFO: 2297 INP_WUNLOCK(inp); 2298 error = EINVAL; 2299 break; 2300 2301 case TCP_STATS: 2302 INP_WUNLOCK(inp); 2303 #ifdef STATS 2304 error = sooptcopyin(sopt, &optval, sizeof optval, 2305 sizeof optval); 2306 if (error) 2307 return (error); 2308 2309 if (optval > 0) 2310 sbp = stats_blob_alloc( 2311 V_tcp_perconn_stats_dflt_tpl, 0); 2312 else 2313 sbp = NULL; 2314 2315 INP_WLOCK_RECHECK(inp); 2316 if ((tp->t_stats != NULL && sbp == NULL) || 2317 (tp->t_stats == NULL && sbp != NULL)) { 2318 struct statsblob *t = tp->t_stats; 2319 tp->t_stats = sbp; 2320 sbp = t; 2321 } 2322 INP_WUNLOCK(inp); 2323 2324 stats_blob_destroy(sbp); 2325 #else 2326 return (EOPNOTSUPP); 2327 #endif /* !STATS */ 2328 break; 2329 2330 case TCP_CONGESTION: 2331 error = tcp_congestion(inp, sopt); 2332 break; 2333 2334 case TCP_REUSPORT_LB_NUMA: 2335 INP_WUNLOCK(inp); 2336 error = sooptcopyin(sopt, &optval, sizeof(optval), 2337 sizeof(optval)); 2338 INP_WLOCK_RECHECK(inp); 2339 if (!error) 2340 error = in_pcblbgroup_numa(inp, optval); 2341 INP_WUNLOCK(inp); 2342 break; 2343 2344 #ifdef KERN_TLS 2345 case TCP_TXTLS_ENABLE: 2346 INP_WUNLOCK(inp); 2347 error = copyin_tls_enable(sopt, &tls); 2348 if (error) 2349 break; 2350 error = ktls_enable_tx(inp->inp_socket, &tls); 2351 break; 2352 case TCP_TXTLS_MODE: 2353 INP_WUNLOCK(inp); 2354 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2355 if (error) 2356 return (error); 2357 2358 INP_WLOCK_RECHECK(inp); 2359 error = ktls_set_tx_mode(inp->inp_socket, ui); 2360 INP_WUNLOCK(inp); 2361 break; 2362 case TCP_RXTLS_ENABLE: 2363 INP_WUNLOCK(inp); 2364 error = sooptcopyin(sopt, &tls, sizeof(tls), 2365 sizeof(tls)); 2366 if (error) 2367 break; 2368 error = ktls_enable_rx(inp->inp_socket, &tls); 2369 break; 2370 #endif 2371 2372 case TCP_KEEPIDLE: 2373 case TCP_KEEPINTVL: 2374 case TCP_KEEPINIT: 2375 INP_WUNLOCK(inp); 2376 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2377 if (error) 2378 return (error); 2379 2380 if (ui > (UINT_MAX / hz)) { 2381 error = EINVAL; 2382 break; 2383 } 2384 ui *= hz; 2385 2386 INP_WLOCK_RECHECK(inp); 2387 switch (sopt->sopt_name) { 2388 case TCP_KEEPIDLE: 2389 tp->t_keepidle = ui; 2390 /* 2391 * XXX: better check current remaining 2392 * timeout and "merge" it with new value. 2393 */ 2394 if ((tp->t_state > TCPS_LISTEN) && 2395 (tp->t_state <= TCPS_CLOSING)) 2396 tcp_timer_activate(tp, TT_KEEP, 2397 TP_KEEPIDLE(tp)); 2398 break; 2399 case TCP_KEEPINTVL: 2400 tp->t_keepintvl = ui; 2401 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2402 (TP_MAXIDLE(tp) > 0)) 2403 tcp_timer_activate(tp, TT_2MSL, 2404 TP_MAXIDLE(tp)); 2405 break; 2406 case TCP_KEEPINIT: 2407 tp->t_keepinit = ui; 2408 if (tp->t_state == TCPS_SYN_RECEIVED || 2409 tp->t_state == TCPS_SYN_SENT) 2410 tcp_timer_activate(tp, TT_KEEP, 2411 TP_KEEPINIT(tp)); 2412 break; 2413 } 2414 goto unlock_and_done; 2415 2416 case TCP_KEEPCNT: 2417 INP_WUNLOCK(inp); 2418 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2419 if (error) 2420 return (error); 2421 2422 INP_WLOCK_RECHECK(inp); 2423 tp->t_keepcnt = ui; 2424 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2425 (TP_MAXIDLE(tp) > 0)) 2426 tcp_timer_activate(tp, TT_2MSL, 2427 TP_MAXIDLE(tp)); 2428 goto unlock_and_done; 2429 2430 #ifdef TCPPCAP 2431 case TCP_PCAP_OUT: 2432 case TCP_PCAP_IN: 2433 INP_WUNLOCK(inp); 2434 error = sooptcopyin(sopt, &optval, sizeof optval, 2435 sizeof optval); 2436 if (error) 2437 return (error); 2438 2439 INP_WLOCK_RECHECK(inp); 2440 if (optval >= 0) 2441 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2442 &(tp->t_outpkts) : &(tp->t_inpkts), 2443 optval); 2444 else 2445 error = EINVAL; 2446 goto unlock_and_done; 2447 #endif 2448 2449 case TCP_FASTOPEN: { 2450 struct tcp_fastopen tfo_optval; 2451 2452 INP_WUNLOCK(inp); 2453 if (!V_tcp_fastopen_client_enable && 2454 !V_tcp_fastopen_server_enable) 2455 return (EPERM); 2456 2457 error = sooptcopyin(sopt, &tfo_optval, 2458 sizeof(tfo_optval), sizeof(int)); 2459 if (error) 2460 return (error); 2461 2462 INP_WLOCK_RECHECK(inp); 2463 if ((tp->t_state != TCPS_CLOSED) && 2464 (tp->t_state != TCPS_LISTEN)) { 2465 error = EINVAL; 2466 goto unlock_and_done; 2467 } 2468 if (tfo_optval.enable) { 2469 if (tp->t_state == TCPS_LISTEN) { 2470 if (!V_tcp_fastopen_server_enable) { 2471 error = EPERM; 2472 goto unlock_and_done; 2473 } 2474 2475 if (tp->t_tfo_pending == NULL) 2476 tp->t_tfo_pending = 2477 tcp_fastopen_alloc_counter(); 2478 } else { 2479 /* 2480 * If a pre-shared key was provided, 2481 * stash it in the client cookie 2482 * field of the tcpcb for use during 2483 * connect. 2484 */ 2485 if (sopt->sopt_valsize == 2486 sizeof(tfo_optval)) { 2487 memcpy(tp->t_tfo_cookie.client, 2488 tfo_optval.psk, 2489 TCP_FASTOPEN_PSK_LEN); 2490 tp->t_tfo_client_cookie_len = 2491 TCP_FASTOPEN_PSK_LEN; 2492 } 2493 } 2494 tp->t_flags |= TF_FASTOPEN; 2495 } else 2496 tp->t_flags &= ~TF_FASTOPEN; 2497 goto unlock_and_done; 2498 } 2499 2500 #ifdef TCP_BLACKBOX 2501 case TCP_LOG: 2502 INP_WUNLOCK(inp); 2503 error = sooptcopyin(sopt, &optval, sizeof optval, 2504 sizeof optval); 2505 if (error) 2506 return (error); 2507 2508 INP_WLOCK_RECHECK(inp); 2509 error = tcp_log_state_change(tp, optval); 2510 goto unlock_and_done; 2511 2512 case TCP_LOGBUF: 2513 INP_WUNLOCK(inp); 2514 error = EINVAL; 2515 break; 2516 2517 case TCP_LOGID: 2518 INP_WUNLOCK(inp); 2519 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2520 if (error) 2521 break; 2522 buf[sopt->sopt_valsize] = '\0'; 2523 INP_WLOCK_RECHECK(inp); 2524 error = tcp_log_set_id(tp, buf); 2525 /* tcp_log_set_id() unlocks the INP. */ 2526 break; 2527 2528 case TCP_LOGDUMP: 2529 case TCP_LOGDUMPID: 2530 INP_WUNLOCK(inp); 2531 error = 2532 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2533 if (error) 2534 break; 2535 buf[sopt->sopt_valsize] = '\0'; 2536 INP_WLOCK_RECHECK(inp); 2537 if (sopt->sopt_name == TCP_LOGDUMP) { 2538 error = tcp_log_dump_tp_logbuf(tp, buf, 2539 M_WAITOK, true); 2540 INP_WUNLOCK(inp); 2541 } else { 2542 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2543 /* 2544 * tcp_log_dump_tp_bucket_logbufs() drops the 2545 * INP lock. 2546 */ 2547 } 2548 break; 2549 #endif 2550 2551 default: 2552 INP_WUNLOCK(inp); 2553 error = ENOPROTOOPT; 2554 break; 2555 } 2556 break; 2557 2558 case SOPT_GET: 2559 tp = intotcpcb(inp); 2560 switch (sopt->sopt_name) { 2561 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2562 case TCP_MD5SIG: 2563 if (!TCPMD5_ENABLED()) { 2564 INP_WUNLOCK(inp); 2565 return (ENOPROTOOPT); 2566 } 2567 error = TCPMD5_PCBCTL(inp, sopt); 2568 break; 2569 #endif 2570 2571 case TCP_NODELAY: 2572 optval = tp->t_flags & TF_NODELAY; 2573 INP_WUNLOCK(inp); 2574 error = sooptcopyout(sopt, &optval, sizeof optval); 2575 break; 2576 case TCP_MAXSEG: 2577 optval = tp->t_maxseg; 2578 INP_WUNLOCK(inp); 2579 error = sooptcopyout(sopt, &optval, sizeof optval); 2580 break; 2581 case TCP_REMOTE_UDP_ENCAPS_PORT: 2582 optval = ntohs(tp->t_port); 2583 INP_WUNLOCK(inp); 2584 error = sooptcopyout(sopt, &optval, sizeof optval); 2585 break; 2586 case TCP_NOOPT: 2587 optval = tp->t_flags & TF_NOOPT; 2588 INP_WUNLOCK(inp); 2589 error = sooptcopyout(sopt, &optval, sizeof optval); 2590 break; 2591 case TCP_NOPUSH: 2592 optval = tp->t_flags & TF_NOPUSH; 2593 INP_WUNLOCK(inp); 2594 error = sooptcopyout(sopt, &optval, sizeof optval); 2595 break; 2596 case TCP_INFO: 2597 tcp_fill_info(tp, &ti); 2598 INP_WUNLOCK(inp); 2599 error = sooptcopyout(sopt, &ti, sizeof ti); 2600 break; 2601 case TCP_STATS: 2602 { 2603 #ifdef STATS 2604 int nheld; 2605 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2606 2607 error = 0; 2608 socklen_t outsbsz = sopt->sopt_valsize; 2609 if (tp->t_stats == NULL) 2610 error = ENOENT; 2611 else if (outsbsz >= tp->t_stats->cursz) 2612 outsbsz = tp->t_stats->cursz; 2613 else if (outsbsz >= sizeof(struct statsblob)) 2614 outsbsz = sizeof(struct statsblob); 2615 else 2616 error = EINVAL; 2617 INP_WUNLOCK(inp); 2618 if (error) 2619 break; 2620 2621 sbp = sopt->sopt_val; 2622 nheld = atop(round_page(((vm_offset_t)sbp) + 2623 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2624 vm_page_t ma[nheld]; 2625 if (vm_fault_quick_hold_pages( 2626 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2627 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2628 nheld) < 0) { 2629 error = EFAULT; 2630 break; 2631 } 2632 2633 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2634 SIZEOF_MEMBER(struct statsblob, flags)))) 2635 goto unhold; 2636 2637 INP_WLOCK_RECHECK(inp); 2638 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2639 sbflags | SB_CLONE_USRDSTNOFAULT); 2640 INP_WUNLOCK(inp); 2641 sopt->sopt_valsize = outsbsz; 2642 unhold: 2643 vm_page_unhold_pages(ma, nheld); 2644 #else 2645 INP_WUNLOCK(inp); 2646 error = EOPNOTSUPP; 2647 #endif /* !STATS */ 2648 break; 2649 } 2650 case TCP_CONGESTION: 2651 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2652 INP_WUNLOCK(inp); 2653 error = sooptcopyout(sopt, buf, len + 1); 2654 break; 2655 case TCP_KEEPIDLE: 2656 case TCP_KEEPINTVL: 2657 case TCP_KEEPINIT: 2658 case TCP_KEEPCNT: 2659 switch (sopt->sopt_name) { 2660 case TCP_KEEPIDLE: 2661 ui = TP_KEEPIDLE(tp) / hz; 2662 break; 2663 case TCP_KEEPINTVL: 2664 ui = TP_KEEPINTVL(tp) / hz; 2665 break; 2666 case TCP_KEEPINIT: 2667 ui = TP_KEEPINIT(tp) / hz; 2668 break; 2669 case TCP_KEEPCNT: 2670 ui = TP_KEEPCNT(tp); 2671 break; 2672 } 2673 INP_WUNLOCK(inp); 2674 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2675 break; 2676 #ifdef TCPPCAP 2677 case TCP_PCAP_OUT: 2678 case TCP_PCAP_IN: 2679 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2680 &(tp->t_outpkts) : &(tp->t_inpkts)); 2681 INP_WUNLOCK(inp); 2682 error = sooptcopyout(sopt, &optval, sizeof optval); 2683 break; 2684 #endif 2685 case TCP_FASTOPEN: 2686 optval = tp->t_flags & TF_FASTOPEN; 2687 INP_WUNLOCK(inp); 2688 error = sooptcopyout(sopt, &optval, sizeof optval); 2689 break; 2690 #ifdef TCP_BLACKBOX 2691 case TCP_LOG: 2692 optval = tp->t_logstate; 2693 INP_WUNLOCK(inp); 2694 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2695 break; 2696 case TCP_LOGBUF: 2697 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2698 error = tcp_log_getlogbuf(sopt, tp); 2699 break; 2700 case TCP_LOGID: 2701 len = tcp_log_get_id(tp, buf); 2702 INP_WUNLOCK(inp); 2703 error = sooptcopyout(sopt, buf, len + 1); 2704 break; 2705 case TCP_LOGDUMP: 2706 case TCP_LOGDUMPID: 2707 INP_WUNLOCK(inp); 2708 error = EINVAL; 2709 break; 2710 #endif 2711 #ifdef KERN_TLS 2712 case TCP_TXTLS_MODE: 2713 error = ktls_get_tx_mode(inp->inp_socket, &optval); 2714 INP_WUNLOCK(inp); 2715 if (error == 0) 2716 error = sooptcopyout(sopt, &optval, 2717 sizeof(optval)); 2718 break; 2719 case TCP_RXTLS_MODE: 2720 error = ktls_get_rx_mode(inp->inp_socket, &optval); 2721 INP_WUNLOCK(inp); 2722 if (error == 0) 2723 error = sooptcopyout(sopt, &optval, 2724 sizeof(optval)); 2725 break; 2726 #endif 2727 case TCP_LRD: 2728 optval = tp->t_flags & TF_LRD; 2729 INP_WUNLOCK(inp); 2730 error = sooptcopyout(sopt, &optval, sizeof optval); 2731 break; 2732 default: 2733 INP_WUNLOCK(inp); 2734 error = ENOPROTOOPT; 2735 break; 2736 } 2737 break; 2738 } 2739 return (error); 2740 } 2741 #undef INP_WLOCK_RECHECK 2742 #undef INP_WLOCK_RECHECK_CLEANUP 2743 2744 /* 2745 * Initiate (or continue) disconnect. 2746 * If embryonic state, just send reset (once). 2747 * If in ``let data drain'' option and linger null, just drop. 2748 * Otherwise (hard), mark socket disconnecting and drop 2749 * current input data; switch states based on user close, and 2750 * send segment to peer (with FIN). 2751 */ 2752 static void 2753 tcp_disconnect(struct tcpcb *tp) 2754 { 2755 struct inpcb *inp = tp->t_inpcb; 2756 struct socket *so = inp->inp_socket; 2757 2758 NET_EPOCH_ASSERT(); 2759 INP_WLOCK_ASSERT(inp); 2760 2761 /* 2762 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2763 * socket is still open. 2764 */ 2765 if (tp->t_state < TCPS_ESTABLISHED && 2766 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2767 tp = tcp_close(tp); 2768 KASSERT(tp != NULL, 2769 ("tcp_disconnect: tcp_close() returned NULL")); 2770 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2771 tp = tcp_drop(tp, 0); 2772 KASSERT(tp != NULL, 2773 ("tcp_disconnect: tcp_drop() returned NULL")); 2774 } else { 2775 soisdisconnecting(so); 2776 sbflush(&so->so_rcv); 2777 tcp_usrclosed(tp); 2778 if (!(inp->inp_flags & INP_DROPPED)) 2779 /* Ignore stack's drop request, we already at it. */ 2780 (void)tcp_output_nodrop(tp); 2781 } 2782 } 2783 2784 /* 2785 * User issued close, and wish to trail through shutdown states: 2786 * if never received SYN, just forget it. If got a SYN from peer, 2787 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2788 * If already got a FIN from peer, then almost done; go to LAST_ACK 2789 * state. In all other cases, have already sent FIN to peer (e.g. 2790 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2791 * for peer to send FIN or not respond to keep-alives, etc. 2792 * We can let the user exit from the close as soon as the FIN is acked. 2793 */ 2794 static void 2795 tcp_usrclosed(struct tcpcb *tp) 2796 { 2797 2798 NET_EPOCH_ASSERT(); 2799 INP_WLOCK_ASSERT(tp->t_inpcb); 2800 2801 switch (tp->t_state) { 2802 case TCPS_LISTEN: 2803 #ifdef TCP_OFFLOAD 2804 tcp_offload_listen_stop(tp); 2805 #endif 2806 tcp_state_change(tp, TCPS_CLOSED); 2807 /* FALLTHROUGH */ 2808 case TCPS_CLOSED: 2809 tp = tcp_close(tp); 2810 /* 2811 * tcp_close() should never return NULL here as the socket is 2812 * still open. 2813 */ 2814 KASSERT(tp != NULL, 2815 ("tcp_usrclosed: tcp_close() returned NULL")); 2816 break; 2817 2818 case TCPS_SYN_SENT: 2819 case TCPS_SYN_RECEIVED: 2820 tp->t_flags |= TF_NEEDFIN; 2821 break; 2822 2823 case TCPS_ESTABLISHED: 2824 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2825 break; 2826 2827 case TCPS_CLOSE_WAIT: 2828 tcp_state_change(tp, TCPS_LAST_ACK); 2829 break; 2830 } 2831 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2832 soisdisconnected(tp->t_inpcb->inp_socket); 2833 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2834 if (tp->t_state == TCPS_FIN_WAIT_2) { 2835 int timeout; 2836 2837 timeout = (tcp_fast_finwait2_recycle) ? 2838 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2839 tcp_timer_activate(tp, TT_2MSL, timeout); 2840 } 2841 } 2842 } 2843 2844 #ifdef DDB 2845 static void 2846 db_print_indent(int indent) 2847 { 2848 int i; 2849 2850 for (i = 0; i < indent; i++) 2851 db_printf(" "); 2852 } 2853 2854 static void 2855 db_print_tstate(int t_state) 2856 { 2857 2858 switch (t_state) { 2859 case TCPS_CLOSED: 2860 db_printf("TCPS_CLOSED"); 2861 return; 2862 2863 case TCPS_LISTEN: 2864 db_printf("TCPS_LISTEN"); 2865 return; 2866 2867 case TCPS_SYN_SENT: 2868 db_printf("TCPS_SYN_SENT"); 2869 return; 2870 2871 case TCPS_SYN_RECEIVED: 2872 db_printf("TCPS_SYN_RECEIVED"); 2873 return; 2874 2875 case TCPS_ESTABLISHED: 2876 db_printf("TCPS_ESTABLISHED"); 2877 return; 2878 2879 case TCPS_CLOSE_WAIT: 2880 db_printf("TCPS_CLOSE_WAIT"); 2881 return; 2882 2883 case TCPS_FIN_WAIT_1: 2884 db_printf("TCPS_FIN_WAIT_1"); 2885 return; 2886 2887 case TCPS_CLOSING: 2888 db_printf("TCPS_CLOSING"); 2889 return; 2890 2891 case TCPS_LAST_ACK: 2892 db_printf("TCPS_LAST_ACK"); 2893 return; 2894 2895 case TCPS_FIN_WAIT_2: 2896 db_printf("TCPS_FIN_WAIT_2"); 2897 return; 2898 2899 case TCPS_TIME_WAIT: 2900 db_printf("TCPS_TIME_WAIT"); 2901 return; 2902 2903 default: 2904 db_printf("unknown"); 2905 return; 2906 } 2907 } 2908 2909 static void 2910 db_print_tflags(u_int t_flags) 2911 { 2912 int comma; 2913 2914 comma = 0; 2915 if (t_flags & TF_ACKNOW) { 2916 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2917 comma = 1; 2918 } 2919 if (t_flags & TF_DELACK) { 2920 db_printf("%sTF_DELACK", comma ? ", " : ""); 2921 comma = 1; 2922 } 2923 if (t_flags & TF_NODELAY) { 2924 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2925 comma = 1; 2926 } 2927 if (t_flags & TF_NOOPT) { 2928 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2929 comma = 1; 2930 } 2931 if (t_flags & TF_SENTFIN) { 2932 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2933 comma = 1; 2934 } 2935 if (t_flags & TF_REQ_SCALE) { 2936 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2937 comma = 1; 2938 } 2939 if (t_flags & TF_RCVD_SCALE) { 2940 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2941 comma = 1; 2942 } 2943 if (t_flags & TF_REQ_TSTMP) { 2944 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2945 comma = 1; 2946 } 2947 if (t_flags & TF_RCVD_TSTMP) { 2948 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2949 comma = 1; 2950 } 2951 if (t_flags & TF_SACK_PERMIT) { 2952 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2953 comma = 1; 2954 } 2955 if (t_flags & TF_NEEDSYN) { 2956 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2957 comma = 1; 2958 } 2959 if (t_flags & TF_NEEDFIN) { 2960 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2961 comma = 1; 2962 } 2963 if (t_flags & TF_NOPUSH) { 2964 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2965 comma = 1; 2966 } 2967 if (t_flags & TF_MORETOCOME) { 2968 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2969 comma = 1; 2970 } 2971 if (t_flags & TF_LQ_OVERFLOW) { 2972 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2973 comma = 1; 2974 } 2975 if (t_flags & TF_LASTIDLE) { 2976 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2977 comma = 1; 2978 } 2979 if (t_flags & TF_RXWIN0SENT) { 2980 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2981 comma = 1; 2982 } 2983 if (t_flags & TF_FASTRECOVERY) { 2984 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2985 comma = 1; 2986 } 2987 if (t_flags & TF_CONGRECOVERY) { 2988 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2989 comma = 1; 2990 } 2991 if (t_flags & TF_WASFRECOVERY) { 2992 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2993 comma = 1; 2994 } 2995 if (t_flags & TF_SIGNATURE) { 2996 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2997 comma = 1; 2998 } 2999 if (t_flags & TF_FORCEDATA) { 3000 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 3001 comma = 1; 3002 } 3003 if (t_flags & TF_TSO) { 3004 db_printf("%sTF_TSO", comma ? ", " : ""); 3005 comma = 1; 3006 } 3007 if (t_flags & TF_FASTOPEN) { 3008 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 3009 comma = 1; 3010 } 3011 } 3012 3013 static void 3014 db_print_tflags2(u_int t_flags2) 3015 { 3016 int comma; 3017 3018 comma = 0; 3019 if (t_flags2 & TF2_ECN_PERMIT) { 3020 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 3021 comma = 1; 3022 } 3023 } 3024 3025 static void 3026 db_print_toobflags(char t_oobflags) 3027 { 3028 int comma; 3029 3030 comma = 0; 3031 if (t_oobflags & TCPOOB_HAVEDATA) { 3032 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3033 comma = 1; 3034 } 3035 if (t_oobflags & TCPOOB_HADDATA) { 3036 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3037 comma = 1; 3038 } 3039 } 3040 3041 static void 3042 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3043 { 3044 3045 db_print_indent(indent); 3046 db_printf("%s at %p\n", name, tp); 3047 3048 indent += 2; 3049 3050 db_print_indent(indent); 3051 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3052 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3053 3054 db_print_indent(indent); 3055 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 3056 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 3057 3058 db_print_indent(indent); 3059 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 3060 &tp->t_timers->tt_delack, tp->t_inpcb); 3061 3062 db_print_indent(indent); 3063 db_printf("t_state: %d (", tp->t_state); 3064 db_print_tstate(tp->t_state); 3065 db_printf(")\n"); 3066 3067 db_print_indent(indent); 3068 db_printf("t_flags: 0x%x (", tp->t_flags); 3069 db_print_tflags(tp->t_flags); 3070 db_printf(")\n"); 3071 3072 db_print_indent(indent); 3073 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3074 db_print_tflags2(tp->t_flags2); 3075 db_printf(")\n"); 3076 3077 db_print_indent(indent); 3078 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 3079 tp->snd_una, tp->snd_max, tp->snd_nxt); 3080 3081 db_print_indent(indent); 3082 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3083 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3084 3085 db_print_indent(indent); 3086 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3087 tp->iss, tp->irs, tp->rcv_nxt); 3088 3089 db_print_indent(indent); 3090 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3091 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3092 3093 db_print_indent(indent); 3094 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3095 tp->snd_wnd, tp->snd_cwnd); 3096 3097 db_print_indent(indent); 3098 db_printf("snd_ssthresh: %u snd_recover: " 3099 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3100 3101 db_print_indent(indent); 3102 db_printf("t_rcvtime: %u t_startime: %u\n", 3103 tp->t_rcvtime, tp->t_starttime); 3104 3105 db_print_indent(indent); 3106 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3107 tp->t_rtttime, tp->t_rtseq); 3108 3109 db_print_indent(indent); 3110 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3111 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3112 3113 db_print_indent(indent); 3114 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 3115 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 3116 tp->t_rttbest); 3117 3118 db_print_indent(indent); 3119 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 3120 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3121 3122 db_print_indent(indent); 3123 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3124 db_print_toobflags(tp->t_oobflags); 3125 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3126 3127 db_print_indent(indent); 3128 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3129 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3130 3131 db_print_indent(indent); 3132 db_printf("ts_recent: %u ts_recent_age: %u\n", 3133 tp->ts_recent, tp->ts_recent_age); 3134 3135 db_print_indent(indent); 3136 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3137 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3138 3139 db_print_indent(indent); 3140 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3141 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3142 tp->snd_recover_prev, tp->t_badrxtwin); 3143 3144 db_print_indent(indent); 3145 db_printf("snd_numholes: %d snd_holes first: %p\n", 3146 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3147 3148 db_print_indent(indent); 3149 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3150 tp->snd_fack, tp->rcv_numsacks); 3151 3152 /* Skip sackblks, sackhint. */ 3153 3154 db_print_indent(indent); 3155 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3156 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3157 } 3158 3159 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3160 { 3161 struct tcpcb *tp; 3162 3163 if (!have_addr) { 3164 db_printf("usage: show tcpcb <addr>\n"); 3165 return; 3166 } 3167 tp = (struct tcpcb *)addr; 3168 3169 db_print_tcpcb(tp, "tcpcb", 0); 3170 } 3171 #endif 3172