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_var.h> 86 #ifdef INET6 87 #include <netinet/ip6.h> 88 #include <netinet6/in6_pcb.h> 89 #include <netinet6/ip6_var.h> 90 #include <netinet6/scope6_var.h> 91 #endif 92 #include <netinet/tcp.h> 93 #include <netinet/tcp_fsm.h> 94 #include <netinet/tcp_seq.h> 95 #include <netinet/tcp_timer.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/tcp_log_buf.h> 98 #include <netinet/tcpip.h> 99 #include <netinet/cc/cc.h> 100 #include <netinet/tcp_fastopen.h> 101 #include <netinet/tcp_hpts.h> 102 #ifdef TCPPCAP 103 #include <netinet/tcp_pcap.h> 104 #endif 105 #ifdef TCPDEBUG 106 #include <netinet/tcp_debug.h> 107 #endif 108 #ifdef TCP_OFFLOAD 109 #include <netinet/tcp_offload.h> 110 #endif 111 #include <netipsec/ipsec_support.h> 112 113 #include <vm/vm.h> 114 #include <vm/vm_param.h> 115 #include <vm/pmap.h> 116 #include <vm/vm_extern.h> 117 #include <vm/vm_map.h> 118 #include <vm/vm_page.h> 119 120 /* 121 * TCP protocol interface to socket abstraction. 122 */ 123 #ifdef INET 124 static int tcp_connect(struct tcpcb *, struct sockaddr *, 125 struct thread *td); 126 #endif /* INET */ 127 #ifdef INET6 128 static int tcp6_connect(struct tcpcb *, struct sockaddr *, 129 struct thread *td); 130 #endif /* INET6 */ 131 static void tcp_disconnect(struct tcpcb *); 132 static void tcp_usrclosed(struct tcpcb *); 133 static void tcp_fill_info(struct tcpcb *, struct tcp_info *); 134 135 static int tcp_pru_options_support(struct tcpcb *tp, int flags); 136 137 #ifdef TCPDEBUG 138 #define TCPDEBUG0 int ostate = 0 139 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 140 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 141 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 142 #else 143 #define TCPDEBUG0 144 #define TCPDEBUG1() 145 #define TCPDEBUG2(req) 146 #endif 147 148 /* 149 * tcp_require_unique port requires a globally-unique source port for each 150 * outgoing connection. The default is to require the 4-tuple to be unique. 151 */ 152 VNET_DEFINE(int, tcp_require_unique_port) = 0; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port, 154 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0, 155 "Require globally-unique ephemeral port for outgoing connections"); 156 #define V_tcp_require_unique_port VNET(tcp_require_unique_port) 157 158 /* 159 * TCP attaches to socket via pru_attach(), reserving space, 160 * and an internet control block. 161 */ 162 static int 163 tcp_usr_attach(struct socket *so, int proto, struct thread *td) 164 { 165 struct inpcb *inp; 166 struct tcpcb *tp = NULL; 167 int error; 168 TCPDEBUG0; 169 170 inp = sotoinpcb(so); 171 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL")); 172 TCPDEBUG1(); 173 174 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 175 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace); 176 if (error) 177 goto out; 178 } 179 180 so->so_rcv.sb_flags |= SB_AUTOSIZE; 181 so->so_snd.sb_flags |= SB_AUTOSIZE; 182 error = in_pcballoc(so, &V_tcbinfo); 183 if (error) 184 goto out; 185 inp = sotoinpcb(so); 186 #ifdef INET6 187 if (inp->inp_vflag & INP_IPV6PROTO) { 188 inp->inp_vflag |= INP_IPV6; 189 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 190 inp->inp_vflag |= INP_IPV4; 191 inp->in6p_hops = -1; /* use kernel default */ 192 } 193 else 194 #endif 195 inp->inp_vflag |= INP_IPV4; 196 tp = tcp_newtcpcb(inp); 197 if (tp == NULL) { 198 error = ENOBUFS; 199 in_pcbdetach(inp); 200 in_pcbfree(inp); 201 goto out; 202 } 203 tp->t_state = TCPS_CLOSED; 204 INP_WUNLOCK(inp); 205 TCPSTATES_INC(TCPS_CLOSED); 206 out: 207 TCPDEBUG2(PRU_ATTACH); 208 TCP_PROBE2(debug__user, tp, PRU_ATTACH); 209 return (error); 210 } 211 212 /* 213 * tcp_usr_detach is called when the socket layer loses its final reference 214 * to the socket, be it a file descriptor reference, a reference from TCP, 215 * etc. At this point, there is only one case in which we will keep around 216 * inpcb state: time wait. 217 */ 218 static void 219 tcp_usr_detach(struct socket *so) 220 { 221 struct inpcb *inp; 222 struct tcpcb *tp; 223 224 inp = sotoinpcb(so); 225 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 226 INP_WLOCK(inp); 227 KASSERT(so->so_pcb == inp && inp->inp_socket == so, 228 ("%s: socket %p inp %p mismatch", __func__, so, inp)); 229 230 tp = intotcpcb(inp); 231 232 if (inp->inp_flags & INP_TIMEWAIT) { 233 /* 234 * There are two cases to handle: one in which the time wait 235 * state is being discarded (INP_DROPPED), and one in which 236 * this connection will remain in timewait. In the former, 237 * it is time to discard all state (except tcptw, which has 238 * already been discarded by the timewait close code, which 239 * should be further up the call stack somewhere). In the 240 * latter case, we detach from the socket, but leave the pcb 241 * present until timewait ends. 242 * 243 * XXXRW: Would it be cleaner to free the tcptw here? 244 * 245 * Astute question indeed, from twtcp perspective there are 246 * four cases to consider: 247 * 248 * #1 tcp_usr_detach is called at tcptw creation time by 249 * tcp_twstart, then do not discard the newly created tcptw 250 * and leave inpcb present until timewait ends 251 * #2 tcp_usr_detach is called at tcptw creation time by 252 * tcp_twstart, but connection is local and tw will be 253 * discarded immediately 254 * #3 tcp_usr_detach is called at timewait end (or reuse) by 255 * tcp_twclose, then the tcptw has already been discarded 256 * (or reused) and inpcb is freed here 257 * #4 tcp_usr_detach is called() after timewait ends (or reuse) 258 * (e.g. by soclose), then tcptw has already been discarded 259 * (or reused) and inpcb is freed here 260 * 261 * In all three cases the tcptw should not be freed here. 262 */ 263 if (inp->inp_flags & INP_DROPPED) { 264 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && " 265 "INP_DROPPED && tp != NULL")); 266 in_pcbdetach(inp); 267 in_pcbfree(inp); 268 } else { 269 in_pcbdetach(inp); 270 INP_WUNLOCK(inp); 271 } 272 } else { 273 /* 274 * If the connection is not in timewait, we consider two 275 * two conditions: one in which no further processing is 276 * necessary (dropped || embryonic), and one in which TCP is 277 * not yet done, but no longer requires the socket, so the 278 * pcb will persist for the time being. 279 * 280 * XXXRW: Does the second case still occur? 281 */ 282 if (inp->inp_flags & INP_DROPPED || 283 tp->t_state < TCPS_SYN_SENT) { 284 tcp_discardcb(tp); 285 in_pcbdetach(inp); 286 in_pcbfree(inp); 287 } else { 288 in_pcbdetach(inp); 289 INP_WUNLOCK(inp); 290 } 291 } 292 } 293 294 #ifdef INET 295 /* 296 * Give the socket an address. 297 */ 298 static int 299 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 300 { 301 int error = 0; 302 struct inpcb *inp; 303 struct tcpcb *tp = NULL; 304 struct sockaddr_in *sinp; 305 306 sinp = (struct sockaddr_in *)nam; 307 if (nam->sa_family != AF_INET) { 308 /* 309 * Preserve compatibility with old programs. 310 */ 311 if (nam->sa_family != AF_UNSPEC || 312 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 313 sinp->sin_addr.s_addr != INADDR_ANY) 314 return (EAFNOSUPPORT); 315 nam->sa_family = AF_INET; 316 } 317 if (nam->sa_len != sizeof(*sinp)) 318 return (EINVAL); 319 320 /* 321 * Must check for multicast addresses and disallow binding 322 * to them. 323 */ 324 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 325 return (EAFNOSUPPORT); 326 327 TCPDEBUG0; 328 inp = sotoinpcb(so); 329 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL")); 330 INP_WLOCK(inp); 331 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 332 error = EINVAL; 333 goto out; 334 } 335 tp = intotcpcb(inp); 336 TCPDEBUG1(); 337 INP_HASH_WLOCK(&V_tcbinfo); 338 error = in_pcbbind(inp, nam, td->td_ucred); 339 INP_HASH_WUNLOCK(&V_tcbinfo); 340 out: 341 TCPDEBUG2(PRU_BIND); 342 TCP_PROBE2(debug__user, tp, PRU_BIND); 343 INP_WUNLOCK(inp); 344 345 return (error); 346 } 347 #endif /* INET */ 348 349 #ifdef INET6 350 static int 351 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 352 { 353 int error = 0; 354 struct inpcb *inp; 355 struct tcpcb *tp = NULL; 356 struct sockaddr_in6 *sin6; 357 u_char vflagsav; 358 359 sin6 = (struct sockaddr_in6 *)nam; 360 if (nam->sa_family != AF_INET6) 361 return (EAFNOSUPPORT); 362 if (nam->sa_len != sizeof(*sin6)) 363 return (EINVAL); 364 365 /* 366 * Must check for multicast addresses and disallow binding 367 * to them. 368 */ 369 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 370 return (EAFNOSUPPORT); 371 372 TCPDEBUG0; 373 inp = sotoinpcb(so); 374 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL")); 375 INP_WLOCK(inp); 376 vflagsav = inp->inp_vflag; 377 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 378 error = EINVAL; 379 goto out; 380 } 381 tp = intotcpcb(inp); 382 TCPDEBUG1(); 383 INP_HASH_WLOCK(&V_tcbinfo); 384 inp->inp_vflag &= ~INP_IPV4; 385 inp->inp_vflag |= INP_IPV6; 386 #ifdef INET 387 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 388 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 389 inp->inp_vflag |= INP_IPV4; 390 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 391 struct sockaddr_in sin; 392 393 in6_sin6_2_sin(&sin, sin6); 394 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 395 error = EAFNOSUPPORT; 396 INP_HASH_WUNLOCK(&V_tcbinfo); 397 goto out; 398 } 399 inp->inp_vflag |= INP_IPV4; 400 inp->inp_vflag &= ~INP_IPV6; 401 error = in_pcbbind(inp, (struct sockaddr *)&sin, 402 td->td_ucred); 403 INP_HASH_WUNLOCK(&V_tcbinfo); 404 goto out; 405 } 406 } 407 #endif 408 error = in6_pcbbind(inp, nam, td->td_ucred); 409 INP_HASH_WUNLOCK(&V_tcbinfo); 410 out: 411 if (error != 0) 412 inp->inp_vflag = vflagsav; 413 TCPDEBUG2(PRU_BIND); 414 TCP_PROBE2(debug__user, tp, PRU_BIND); 415 INP_WUNLOCK(inp); 416 return (error); 417 } 418 #endif /* INET6 */ 419 420 #ifdef INET 421 /* 422 * Prepare to accept connections. 423 */ 424 static int 425 tcp_usr_listen(struct socket *so, int backlog, struct thread *td) 426 { 427 int error = 0; 428 struct inpcb *inp; 429 struct tcpcb *tp = NULL; 430 431 TCPDEBUG0; 432 inp = sotoinpcb(so); 433 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL")); 434 INP_WLOCK(inp); 435 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 436 error = EINVAL; 437 goto out; 438 } 439 tp = intotcpcb(inp); 440 TCPDEBUG1(); 441 SOCK_LOCK(so); 442 error = solisten_proto_check(so); 443 if (error != 0) { 444 SOCK_UNLOCK(so); 445 goto out; 446 } 447 if (inp->inp_lport == 0) { 448 INP_HASH_WLOCK(&V_tcbinfo); 449 error = in_pcbbind(inp, NULL, td->td_ucred); 450 INP_HASH_WUNLOCK(&V_tcbinfo); 451 } 452 if (error == 0) { 453 tcp_state_change(tp, TCPS_LISTEN); 454 solisten_proto(so, backlog); 455 #ifdef TCP_OFFLOAD 456 if ((so->so_options & SO_NO_OFFLOAD) == 0) 457 tcp_offload_listen_start(tp); 458 #endif 459 } else { 460 solisten_proto_abort(so); 461 } 462 SOCK_UNLOCK(so); 463 464 if (IS_FASTOPEN(tp->t_flags)) 465 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 466 467 out: 468 TCPDEBUG2(PRU_LISTEN); 469 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 470 INP_WUNLOCK(inp); 471 return (error); 472 } 473 #endif /* INET */ 474 475 #ifdef INET6 476 static int 477 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td) 478 { 479 int error = 0; 480 struct inpcb *inp; 481 struct tcpcb *tp = NULL; 482 u_char vflagsav; 483 484 TCPDEBUG0; 485 inp = sotoinpcb(so); 486 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL")); 487 INP_WLOCK(inp); 488 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 489 error = EINVAL; 490 goto out; 491 } 492 vflagsav = inp->inp_vflag; 493 tp = intotcpcb(inp); 494 TCPDEBUG1(); 495 SOCK_LOCK(so); 496 error = solisten_proto_check(so); 497 if (error != 0) { 498 SOCK_UNLOCK(so); 499 goto out; 500 } 501 INP_HASH_WLOCK(&V_tcbinfo); 502 if (inp->inp_lport == 0) { 503 inp->inp_vflag &= ~INP_IPV4; 504 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) 505 inp->inp_vflag |= INP_IPV4; 506 error = in6_pcbbind(inp, NULL, td->td_ucred); 507 } 508 INP_HASH_WUNLOCK(&V_tcbinfo); 509 if (error == 0) { 510 tcp_state_change(tp, TCPS_LISTEN); 511 solisten_proto(so, backlog); 512 #ifdef TCP_OFFLOAD 513 if ((so->so_options & SO_NO_OFFLOAD) == 0) 514 tcp_offload_listen_start(tp); 515 #endif 516 } else { 517 solisten_proto_abort(so); 518 } 519 SOCK_UNLOCK(so); 520 521 if (IS_FASTOPEN(tp->t_flags)) 522 tp->t_tfo_pending = tcp_fastopen_alloc_counter(); 523 524 if (error != 0) 525 inp->inp_vflag = vflagsav; 526 527 out: 528 TCPDEBUG2(PRU_LISTEN); 529 TCP_PROBE2(debug__user, tp, PRU_LISTEN); 530 INP_WUNLOCK(inp); 531 return (error); 532 } 533 #endif /* INET6 */ 534 535 #ifdef INET 536 /* 537 * Initiate connection to peer. 538 * Create a template for use in transmissions on this connection. 539 * Enter SYN_SENT state, and mark socket as connecting. 540 * Start keep-alive timer, and seed output sequence space. 541 * Send initial segment on connection. 542 */ 543 static int 544 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 545 { 546 struct epoch_tracker et; 547 int error = 0; 548 struct inpcb *inp; 549 struct tcpcb *tp = NULL; 550 struct sockaddr_in *sinp; 551 552 sinp = (struct sockaddr_in *)nam; 553 if (nam->sa_family != AF_INET) 554 return (EAFNOSUPPORT); 555 if (nam->sa_len != sizeof (*sinp)) 556 return (EINVAL); 557 558 /* 559 * Must disallow TCP ``connections'' to multicast addresses. 560 */ 561 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 562 return (EAFNOSUPPORT); 563 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) 564 return (EACCES); 565 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0) 566 return (error); 567 568 TCPDEBUG0; 569 inp = sotoinpcb(so); 570 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL")); 571 INP_WLOCK(inp); 572 if (inp->inp_flags & INP_TIMEWAIT) { 573 error = EADDRINUSE; 574 goto out; 575 } 576 if (inp->inp_flags & INP_DROPPED) { 577 error = ECONNREFUSED; 578 goto out; 579 } 580 if (SOLISTENING(so)) { 581 error = EOPNOTSUPP; 582 goto out; 583 } 584 tp = intotcpcb(inp); 585 TCPDEBUG1(); 586 NET_EPOCH_ENTER(et); 587 if ((error = tcp_connect(tp, nam, td)) != 0) 588 goto out_in_epoch; 589 #ifdef TCP_OFFLOAD 590 if (registered_toedevs > 0 && 591 (so->so_options & SO_NO_OFFLOAD) == 0 && 592 (error = tcp_offload_connect(so, nam)) == 0) 593 goto out_in_epoch; 594 #endif 595 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 596 error = tcp_output(tp); 597 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 598 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 599 out_in_epoch: 600 NET_EPOCH_EXIT(et); 601 out: 602 TCPDEBUG2(PRU_CONNECT); 603 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 604 INP_WUNLOCK(inp); 605 return (error); 606 } 607 #endif /* INET */ 608 609 #ifdef INET6 610 static int 611 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 612 { 613 struct epoch_tracker et; 614 int error = 0; 615 struct inpcb *inp; 616 struct tcpcb *tp = NULL; 617 struct sockaddr_in6 *sin6; 618 u_int8_t incflagsav; 619 u_char vflagsav; 620 621 TCPDEBUG0; 622 623 sin6 = (struct sockaddr_in6 *)nam; 624 if (nam->sa_family != AF_INET6) 625 return (EAFNOSUPPORT); 626 if (nam->sa_len != sizeof (*sin6)) 627 return (EINVAL); 628 629 /* 630 * Must disallow TCP ``connections'' to multicast addresses. 631 */ 632 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 633 return (EAFNOSUPPORT); 634 635 inp = sotoinpcb(so); 636 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 637 INP_WLOCK(inp); 638 vflagsav = inp->inp_vflag; 639 incflagsav = inp->inp_inc.inc_flags; 640 if (inp->inp_flags & INP_TIMEWAIT) { 641 error = EADDRINUSE; 642 goto out; 643 } 644 if (inp->inp_flags & INP_DROPPED) { 645 error = ECONNREFUSED; 646 goto out; 647 } 648 if (SOLISTENING(so)) { 649 error = EINVAL; 650 goto out; 651 } 652 tp = intotcpcb(inp); 653 TCPDEBUG1(); 654 #ifdef INET 655 /* 656 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 657 * therefore probably require the hash lock, which isn't held here. 658 * Is this a significant problem? 659 */ 660 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 661 struct sockaddr_in sin; 662 663 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 664 error = EINVAL; 665 goto out; 666 } 667 if ((inp->inp_vflag & INP_IPV4) == 0) { 668 error = EAFNOSUPPORT; 669 goto out; 670 } 671 672 in6_sin6_2_sin(&sin, sin6); 673 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 674 error = EAFNOSUPPORT; 675 goto out; 676 } 677 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 678 error = EACCES; 679 goto out; 680 } 681 if ((error = prison_remote_ip4(td->td_ucred, 682 &sin.sin_addr)) != 0) 683 goto out; 684 inp->inp_vflag |= INP_IPV4; 685 inp->inp_vflag &= ~INP_IPV6; 686 NET_EPOCH_ENTER(et); 687 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 688 goto out_in_epoch; 689 #ifdef TCP_OFFLOAD 690 if (registered_toedevs > 0 && 691 (so->so_options & SO_NO_OFFLOAD) == 0 && 692 (error = tcp_offload_connect(so, nam)) == 0) 693 goto out_in_epoch; 694 #endif 695 error = tcp_output(tp); 696 goto out_in_epoch; 697 } else { 698 if ((inp->inp_vflag & INP_IPV6) == 0) { 699 error = EAFNOSUPPORT; 700 goto out; 701 } 702 } 703 #endif 704 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 705 goto out; 706 inp->inp_vflag &= ~INP_IPV4; 707 inp->inp_vflag |= INP_IPV6; 708 inp->inp_inc.inc_flags |= INC_ISIPV6; 709 if ((error = tcp6_connect(tp, nam, td)) != 0) 710 goto out; 711 #ifdef TCP_OFFLOAD 712 if (registered_toedevs > 0 && 713 (so->so_options & SO_NO_OFFLOAD) == 0 && 714 (error = tcp_offload_connect(so, nam)) == 0) 715 goto out; 716 #endif 717 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 718 NET_EPOCH_ENTER(et); 719 error = tcp_output(tp); 720 #ifdef INET 721 out_in_epoch: 722 #endif 723 NET_EPOCH_EXIT(et); 724 out: 725 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()" 726 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error)); 727 /* 728 * If the implicit bind in the connect call fails, restore 729 * the flags we modified. 730 */ 731 if (error != 0 && inp->inp_lport == 0) { 732 inp->inp_vflag = vflagsav; 733 inp->inp_inc.inc_flags = incflagsav; 734 } 735 736 TCPDEBUG2(PRU_CONNECT); 737 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 738 INP_WUNLOCK(inp); 739 return (error); 740 } 741 #endif /* INET6 */ 742 743 /* 744 * Initiate disconnect from peer. 745 * If connection never passed embryonic stage, just drop; 746 * else if don't need to let data drain, then can just drop anyways, 747 * else have to begin TCP shutdown process: mark socket disconnecting, 748 * drain unread data, state switch to reflect user close, and 749 * send segment (e.g. FIN) to peer. Socket will be really disconnected 750 * when peer sends FIN and acks ours. 751 * 752 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 753 */ 754 static int 755 tcp_usr_disconnect(struct socket *so) 756 { 757 struct inpcb *inp; 758 struct tcpcb *tp = NULL; 759 struct epoch_tracker et; 760 int error = 0; 761 762 TCPDEBUG0; 763 NET_EPOCH_ENTER(et); 764 inp = sotoinpcb(so); 765 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 766 INP_WLOCK(inp); 767 if (inp->inp_flags & INP_TIMEWAIT) 768 goto out; 769 if (inp->inp_flags & INP_DROPPED) { 770 error = ECONNRESET; 771 goto out; 772 } 773 tp = intotcpcb(inp); 774 TCPDEBUG1(); 775 tcp_disconnect(tp); 776 out: 777 TCPDEBUG2(PRU_DISCONNECT); 778 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 779 INP_WUNLOCK(inp); 780 NET_EPOCH_EXIT(et); 781 return (error); 782 } 783 784 #ifdef INET 785 /* 786 * Accept a connection. Essentially all the work is done at higher levels; 787 * just return the address of the peer, storing through addr. 788 */ 789 static int 790 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 791 { 792 int error = 0; 793 struct inpcb *inp = NULL; 794 struct tcpcb *tp = NULL; 795 struct in_addr addr; 796 in_port_t port = 0; 797 TCPDEBUG0; 798 799 if (so->so_state & SS_ISDISCONNECTED) 800 return (ECONNABORTED); 801 802 inp = sotoinpcb(so); 803 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 804 INP_WLOCK(inp); 805 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 806 error = ECONNABORTED; 807 goto out; 808 } 809 tp = intotcpcb(inp); 810 TCPDEBUG1(); 811 812 /* 813 * We inline in_getpeeraddr and COMMON_END here, so that we can 814 * copy the data of interest and defer the malloc until after we 815 * release the lock. 816 */ 817 port = inp->inp_fport; 818 addr = inp->inp_faddr; 819 820 out: 821 TCPDEBUG2(PRU_ACCEPT); 822 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 823 INP_WUNLOCK(inp); 824 if (error == 0) 825 *nam = in_sockaddr(port, &addr); 826 return error; 827 } 828 #endif /* INET */ 829 830 #ifdef INET6 831 static int 832 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 833 { 834 struct inpcb *inp = NULL; 835 int error = 0; 836 struct tcpcb *tp = NULL; 837 struct in_addr addr; 838 struct in6_addr addr6; 839 struct epoch_tracker et; 840 in_port_t port = 0; 841 int v4 = 0; 842 TCPDEBUG0; 843 844 if (so->so_state & SS_ISDISCONNECTED) 845 return (ECONNABORTED); 846 847 inp = sotoinpcb(so); 848 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 849 NET_EPOCH_ENTER(et); 850 INP_WLOCK(inp); 851 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 852 error = ECONNABORTED; 853 goto out; 854 } 855 tp = intotcpcb(inp); 856 TCPDEBUG1(); 857 858 /* 859 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 860 * copy the data of interest and defer the malloc until after we 861 * release the lock. 862 */ 863 if (inp->inp_vflag & INP_IPV4) { 864 v4 = 1; 865 port = inp->inp_fport; 866 addr = inp->inp_faddr; 867 } else { 868 port = inp->inp_fport; 869 addr6 = inp->in6p_faddr; 870 } 871 872 out: 873 TCPDEBUG2(PRU_ACCEPT); 874 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 875 INP_WUNLOCK(inp); 876 NET_EPOCH_EXIT(et); 877 if (error == 0) { 878 if (v4) 879 *nam = in6_v4mapsin6_sockaddr(port, &addr); 880 else 881 *nam = in6_sockaddr(port, &addr6); 882 } 883 return error; 884 } 885 #endif /* INET6 */ 886 887 /* 888 * Mark the connection as being incapable of further output. 889 */ 890 static int 891 tcp_usr_shutdown(struct socket *so) 892 { 893 int error = 0; 894 struct inpcb *inp; 895 struct tcpcb *tp = NULL; 896 struct epoch_tracker et; 897 898 TCPDEBUG0; 899 inp = sotoinpcb(so); 900 KASSERT(inp != NULL, ("inp == NULL")); 901 INP_WLOCK(inp); 902 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 903 INP_WUNLOCK(inp); 904 return (ECONNRESET); 905 } 906 tp = intotcpcb(inp); 907 NET_EPOCH_ENTER(et); 908 TCPDEBUG1(); 909 socantsendmore(so); 910 tcp_usrclosed(tp); 911 if (!(inp->inp_flags & INP_DROPPED)) 912 error = tcp_output_nodrop(tp); 913 TCPDEBUG2(PRU_SHUTDOWN); 914 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 915 error = tcp_unlock_or_drop(tp, error); 916 NET_EPOCH_EXIT(et); 917 918 return (error); 919 } 920 921 /* 922 * After a receive, possibly send window update to peer. 923 */ 924 static int 925 tcp_usr_rcvd(struct socket *so, int flags) 926 { 927 struct epoch_tracker et; 928 struct inpcb *inp; 929 struct tcpcb *tp = NULL; 930 int outrv = 0, error = 0; 931 932 TCPDEBUG0; 933 inp = sotoinpcb(so); 934 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 935 INP_WLOCK(inp); 936 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 937 INP_WUNLOCK(inp); 938 return (ECONNRESET); 939 } 940 tp = intotcpcb(inp); 941 NET_EPOCH_ENTER(et); 942 TCPDEBUG1(); 943 /* 944 * For passively-created TFO connections, don't attempt a window 945 * update while still in SYN_RECEIVED as this may trigger an early 946 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 947 * application response data, or failing that, when the DELACK timer 948 * expires. 949 */ 950 if (IS_FASTOPEN(tp->t_flags) && 951 (tp->t_state == TCPS_SYN_RECEIVED)) 952 goto out; 953 #ifdef TCP_OFFLOAD 954 if (tp->t_flags & TF_TOE) 955 tcp_offload_rcvd(tp); 956 else 957 #endif 958 outrv = tcp_output_nodrop(tp); 959 out: 960 TCPDEBUG2(PRU_RCVD); 961 TCP_PROBE2(debug__user, tp, PRU_RCVD); 962 (void) tcp_unlock_or_drop(tp, outrv); 963 NET_EPOCH_EXIT(et); 964 return (error); 965 } 966 967 /* 968 * Do a send by putting data in output queue and updating urgent 969 * marker if URG set. Possibly send more data. Unlike the other 970 * pru_*() routines, the mbuf chains are our responsibility. We 971 * must either enqueue them or free them. The other pru_* routines 972 * generally are caller-frees. 973 */ 974 static int 975 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 976 struct sockaddr *nam, struct mbuf *control, struct thread *td) 977 { 978 struct epoch_tracker et; 979 int error = 0; 980 struct inpcb *inp; 981 struct tcpcb *tp = NULL; 982 #ifdef INET 983 #ifdef INET6 984 struct sockaddr_in sin; 985 #endif 986 struct sockaddr_in *sinp; 987 #endif 988 #ifdef INET6 989 int isipv6; 990 #endif 991 u_int8_t incflagsav; 992 u_char vflagsav; 993 bool restoreflags; 994 TCPDEBUG0; 995 996 if (control != NULL) { 997 /* TCP doesn't do control messages (rights, creds, etc) */ 998 if (control->m_len) { 999 m_freem(control); 1000 return (EINVAL); 1001 } 1002 m_freem(control); /* empty control, just free it */ 1003 } 1004 1005 inp = sotoinpcb(so); 1006 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 1007 INP_WLOCK(inp); 1008 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1009 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1010 m_freem(m); 1011 INP_WUNLOCK(inp); 1012 return (ECONNRESET); 1013 } 1014 1015 vflagsav = inp->inp_vflag; 1016 incflagsav = inp->inp_inc.inc_flags; 1017 restoreflags = false; 1018 tp = intotcpcb(inp); 1019 1020 NET_EPOCH_ENTER(et); 1021 if ((flags & PRUS_OOB) != 0 && 1022 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) 1023 goto out; 1024 1025 TCPDEBUG1(); 1026 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) { 1027 if (tp->t_state == TCPS_LISTEN) { 1028 error = EINVAL; 1029 goto out; 1030 } 1031 switch (nam->sa_family) { 1032 #ifdef INET 1033 case AF_INET: 1034 sinp = (struct sockaddr_in *)nam; 1035 if (sinp->sin_len != sizeof(struct sockaddr_in)) { 1036 error = EINVAL; 1037 goto out; 1038 } 1039 if ((inp->inp_vflag & INP_IPV6) != 0) { 1040 error = EAFNOSUPPORT; 1041 goto out; 1042 } 1043 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 1044 error = EAFNOSUPPORT; 1045 goto out; 1046 } 1047 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) { 1048 error = EACCES; 1049 goto out; 1050 } 1051 if ((error = prison_remote_ip4(td->td_ucred, 1052 &sinp->sin_addr))) 1053 goto out; 1054 #ifdef INET6 1055 isipv6 = 0; 1056 #endif 1057 break; 1058 #endif /* INET */ 1059 #ifdef INET6 1060 case AF_INET6: 1061 { 1062 struct sockaddr_in6 *sin6; 1063 1064 sin6 = (struct sockaddr_in6 *)nam; 1065 if (sin6->sin6_len != sizeof(*sin6)) { 1066 error = EINVAL; 1067 goto out; 1068 } 1069 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) { 1070 error = EAFNOSUPPORT; 1071 goto out; 1072 } 1073 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 1074 error = EAFNOSUPPORT; 1075 goto out; 1076 } 1077 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 1078 #ifdef INET 1079 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 1080 error = EINVAL; 1081 goto out; 1082 } 1083 if ((inp->inp_vflag & INP_IPV4) == 0) { 1084 error = EAFNOSUPPORT; 1085 goto out; 1086 } 1087 restoreflags = true; 1088 inp->inp_vflag &= ~INP_IPV6; 1089 sinp = &sin; 1090 in6_sin6_2_sin(sinp, sin6); 1091 if (IN_MULTICAST( 1092 ntohl(sinp->sin_addr.s_addr))) { 1093 error = EAFNOSUPPORT; 1094 goto out; 1095 } 1096 if ((error = prison_remote_ip4(td->td_ucred, 1097 &sinp->sin_addr))) 1098 goto out; 1099 isipv6 = 0; 1100 #else /* !INET */ 1101 error = EAFNOSUPPORT; 1102 goto out; 1103 #endif /* INET */ 1104 } else { 1105 if ((inp->inp_vflag & INP_IPV6) == 0) { 1106 error = EAFNOSUPPORT; 1107 goto out; 1108 } 1109 restoreflags = true; 1110 inp->inp_vflag &= ~INP_IPV4; 1111 inp->inp_inc.inc_flags |= INC_ISIPV6; 1112 if ((error = prison_remote_ip6(td->td_ucred, 1113 &sin6->sin6_addr))) 1114 goto out; 1115 isipv6 = 1; 1116 } 1117 break; 1118 } 1119 #endif /* INET6 */ 1120 default: 1121 error = EAFNOSUPPORT; 1122 goto out; 1123 } 1124 } 1125 if (!(flags & PRUS_OOB)) { 1126 sbappendstream(&so->so_snd, m, flags); 1127 m = NULL; 1128 if (nam && tp->t_state < TCPS_SYN_SENT) { 1129 KASSERT(tp->t_state == TCPS_CLOSED, 1130 ("%s: tp %p is listening", __func__, tp)); 1131 1132 /* 1133 * Do implied connect if not yet connected, 1134 * initialize window to default value, and 1135 * initialize maxseg using peer's cached MSS. 1136 */ 1137 #ifdef INET6 1138 if (isipv6) 1139 error = tcp6_connect(tp, nam, td); 1140 #endif /* INET6 */ 1141 #if defined(INET6) && defined(INET) 1142 else 1143 #endif 1144 #ifdef INET 1145 error = tcp_connect(tp, 1146 (struct sockaddr *)sinp, td); 1147 #endif 1148 /* 1149 * The bind operation in tcp_connect succeeded. We 1150 * no longer want to restore the flags if later 1151 * operations fail. 1152 */ 1153 if (error == 0 || inp->inp_lport != 0) 1154 restoreflags = false; 1155 1156 if (error) { 1157 /* m is freed if PRUS_NOTREADY is unset. */ 1158 sbflush(&so->so_snd); 1159 goto out; 1160 } 1161 if (IS_FASTOPEN(tp->t_flags)) 1162 tcp_fastopen_connect(tp); 1163 else { 1164 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1165 tcp_mss(tp, -1); 1166 } 1167 } 1168 if (flags & PRUS_EOF) { 1169 /* 1170 * Close the send side of the connection after 1171 * the data is sent. 1172 */ 1173 socantsendmore(so); 1174 tcp_usrclosed(tp); 1175 } 1176 if (TCPS_HAVEESTABLISHED(tp->t_state) && 1177 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1178 (tp->t_fbyte_out == 0) && 1179 (so->so_snd.sb_ccc > 0)) { 1180 tp->t_fbyte_out = ticks; 1181 if (tp->t_fbyte_out == 0) 1182 tp->t_fbyte_out = 1; 1183 if (tp->t_fbyte_out && tp->t_fbyte_in) 1184 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1185 } 1186 if (!(inp->inp_flags & INP_DROPPED) && 1187 !(flags & PRUS_NOTREADY)) { 1188 if (flags & PRUS_MORETOCOME) 1189 tp->t_flags |= TF_MORETOCOME; 1190 error = tcp_output_nodrop(tp); 1191 if (flags & PRUS_MORETOCOME) 1192 tp->t_flags &= ~TF_MORETOCOME; 1193 } 1194 } else { 1195 /* 1196 * XXXRW: PRUS_EOF not implemented with PRUS_OOB? 1197 */ 1198 SOCKBUF_LOCK(&so->so_snd); 1199 if (sbspace(&so->so_snd) < -512) { 1200 SOCKBUF_UNLOCK(&so->so_snd); 1201 error = ENOBUFS; 1202 goto out; 1203 } 1204 /* 1205 * According to RFC961 (Assigned Protocols), 1206 * the urgent pointer points to the last octet 1207 * of urgent data. We continue, however, 1208 * to consider it to indicate the first octet 1209 * of data past the urgent section. 1210 * Otherwise, snd_up should be one lower. 1211 */ 1212 sbappendstream_locked(&so->so_snd, m, flags); 1213 SOCKBUF_UNLOCK(&so->so_snd); 1214 m = NULL; 1215 if (nam && tp->t_state < TCPS_SYN_SENT) { 1216 /* 1217 * Do implied connect if not yet connected, 1218 * initialize window to default value, and 1219 * initialize maxseg using peer's cached MSS. 1220 */ 1221 1222 /* 1223 * Not going to contemplate SYN|URG 1224 */ 1225 if (IS_FASTOPEN(tp->t_flags)) 1226 tp->t_flags &= ~TF_FASTOPEN; 1227 #ifdef INET6 1228 if (isipv6) 1229 error = tcp6_connect(tp, nam, td); 1230 #endif /* INET6 */ 1231 #if defined(INET6) && defined(INET) 1232 else 1233 #endif 1234 #ifdef INET 1235 error = tcp_connect(tp, 1236 (struct sockaddr *)sinp, td); 1237 #endif 1238 /* 1239 * The bind operation in tcp_connect succeeded. We 1240 * no longer want to restore the flags if later 1241 * operations fail. 1242 */ 1243 if (error == 0 || inp->inp_lport != 0) 1244 restoreflags = false; 1245 1246 if (error != 0) { 1247 /* m is freed if PRUS_NOTREADY is unset. */ 1248 sbflush(&so->so_snd); 1249 goto out; 1250 } 1251 tp->snd_wnd = TTCP_CLIENT_SND_WND; 1252 tcp_mss(tp, -1); 1253 } 1254 tp->snd_up = tp->snd_una + sbavail(&so->so_snd); 1255 if ((flags & PRUS_NOTREADY) == 0) { 1256 tp->t_flags |= TF_FORCEDATA; 1257 error = tcp_output_nodrop(tp); 1258 tp->t_flags &= ~TF_FORCEDATA; 1259 } 1260 } 1261 TCP_LOG_EVENT(tp, NULL, 1262 &inp->inp_socket->so_rcv, 1263 &inp->inp_socket->so_snd, 1264 TCP_LOG_USERSEND, error, 1265 0, NULL, false); 1266 1267 out: 1268 /* 1269 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is 1270 * responsible for freeing memory. 1271 */ 1272 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1273 m_freem(m); 1274 1275 /* 1276 * If the request was unsuccessful and we changed flags, 1277 * restore the original flags. 1278 */ 1279 if (error != 0 && restoreflags) { 1280 inp->inp_vflag = vflagsav; 1281 inp->inp_inc.inc_flags = incflagsav; 1282 } 1283 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB : 1284 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1285 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB : 1286 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND)); 1287 error = tcp_unlock_or_drop(tp, error); 1288 NET_EPOCH_EXIT(et); 1289 return (error); 1290 } 1291 1292 static int 1293 tcp_usr_ready(struct socket *so, struct mbuf *m, int count) 1294 { 1295 struct epoch_tracker et; 1296 struct inpcb *inp; 1297 struct tcpcb *tp; 1298 int error; 1299 1300 inp = sotoinpcb(so); 1301 INP_WLOCK(inp); 1302 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1303 INP_WUNLOCK(inp); 1304 mb_free_notready(m, count); 1305 return (ECONNRESET); 1306 } 1307 tp = intotcpcb(inp); 1308 1309 SOCKBUF_LOCK(&so->so_snd); 1310 error = sbready(&so->so_snd, m, count); 1311 SOCKBUF_UNLOCK(&so->so_snd); 1312 if (error) { 1313 INP_WUNLOCK(inp); 1314 return (error); 1315 } 1316 NET_EPOCH_ENTER(et); 1317 error = tcp_output_unlock(tp); 1318 NET_EPOCH_EXIT(et); 1319 1320 return (error); 1321 } 1322 1323 /* 1324 * Abort the TCP. Drop the connection abruptly. 1325 */ 1326 static void 1327 tcp_usr_abort(struct socket *so) 1328 { 1329 struct inpcb *inp; 1330 struct tcpcb *tp = NULL; 1331 struct epoch_tracker et; 1332 TCPDEBUG0; 1333 1334 inp = sotoinpcb(so); 1335 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL")); 1336 1337 NET_EPOCH_ENTER(et); 1338 INP_WLOCK(inp); 1339 KASSERT(inp->inp_socket != NULL, 1340 ("tcp_usr_abort: inp_socket == NULL")); 1341 1342 /* 1343 * If we still have full TCP state, and we're not dropped, drop. 1344 */ 1345 if (!(inp->inp_flags & INP_TIMEWAIT) && 1346 !(inp->inp_flags & INP_DROPPED)) { 1347 tp = intotcpcb(inp); 1348 TCPDEBUG1(); 1349 tp = tcp_drop(tp, ECONNABORTED); 1350 if (tp == NULL) 1351 goto dropped; 1352 TCPDEBUG2(PRU_ABORT); 1353 TCP_PROBE2(debug__user, tp, PRU_ABORT); 1354 } 1355 if (!(inp->inp_flags & INP_DROPPED)) { 1356 SOCK_LOCK(so); 1357 so->so_state |= SS_PROTOREF; 1358 SOCK_UNLOCK(so); 1359 inp->inp_flags |= INP_SOCKREF; 1360 } 1361 INP_WUNLOCK(inp); 1362 dropped: 1363 NET_EPOCH_EXIT(et); 1364 } 1365 1366 /* 1367 * TCP socket is closed. Start friendly disconnect. 1368 */ 1369 static void 1370 tcp_usr_close(struct socket *so) 1371 { 1372 struct inpcb *inp; 1373 struct tcpcb *tp = NULL; 1374 struct epoch_tracker et; 1375 TCPDEBUG0; 1376 1377 inp = sotoinpcb(so); 1378 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL")); 1379 1380 NET_EPOCH_ENTER(et); 1381 INP_WLOCK(inp); 1382 KASSERT(inp->inp_socket != NULL, 1383 ("tcp_usr_close: inp_socket == NULL")); 1384 1385 /* 1386 * If we still have full TCP state, and we're not dropped, initiate 1387 * a disconnect. 1388 */ 1389 if (!(inp->inp_flags & INP_TIMEWAIT) && 1390 !(inp->inp_flags & INP_DROPPED)) { 1391 tp = intotcpcb(inp); 1392 TCPDEBUG1(); 1393 tcp_disconnect(tp); 1394 TCPDEBUG2(PRU_CLOSE); 1395 TCP_PROBE2(debug__user, tp, PRU_CLOSE); 1396 } 1397 if (!(inp->inp_flags & INP_DROPPED)) { 1398 SOCK_LOCK(so); 1399 so->so_state |= SS_PROTOREF; 1400 SOCK_UNLOCK(so); 1401 inp->inp_flags |= INP_SOCKREF; 1402 } 1403 INP_WUNLOCK(inp); 1404 NET_EPOCH_EXIT(et); 1405 } 1406 1407 static int 1408 tcp_pru_options_support(struct tcpcb *tp, int flags) 1409 { 1410 /* 1411 * If the specific TCP stack has a pru_options 1412 * specified then it does not always support 1413 * all the PRU_XX options and we must ask it. 1414 * If the function is not specified then all 1415 * of the PRU_XX options are supported. 1416 */ 1417 int ret = 0; 1418 1419 if (tp->t_fb->tfb_pru_options) { 1420 ret = (*tp->t_fb->tfb_pru_options)(tp, flags); 1421 } 1422 return (ret); 1423 } 1424 1425 /* 1426 * Receive out-of-band data. 1427 */ 1428 static int 1429 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags) 1430 { 1431 int error = 0; 1432 struct inpcb *inp; 1433 struct tcpcb *tp = NULL; 1434 1435 TCPDEBUG0; 1436 inp = sotoinpcb(so); 1437 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL")); 1438 INP_WLOCK(inp); 1439 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1440 error = ECONNRESET; 1441 goto out; 1442 } 1443 tp = intotcpcb(inp); 1444 error = tcp_pru_options_support(tp, PRUS_OOB); 1445 if (error) { 1446 goto out; 1447 } 1448 TCPDEBUG1(); 1449 if ((so->so_oobmark == 0 && 1450 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) || 1451 so->so_options & SO_OOBINLINE || 1452 tp->t_oobflags & TCPOOB_HADDATA) { 1453 error = EINVAL; 1454 goto out; 1455 } 1456 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 1457 error = EWOULDBLOCK; 1458 goto out; 1459 } 1460 m->m_len = 1; 1461 *mtod(m, caddr_t) = tp->t_iobc; 1462 if ((flags & MSG_PEEK) == 0) 1463 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1464 1465 out: 1466 TCPDEBUG2(PRU_RCVOOB); 1467 TCP_PROBE2(debug__user, tp, PRU_RCVOOB); 1468 INP_WUNLOCK(inp); 1469 return (error); 1470 } 1471 1472 #ifdef INET 1473 struct pr_usrreqs tcp_usrreqs = { 1474 .pru_abort = tcp_usr_abort, 1475 .pru_accept = tcp_usr_accept, 1476 .pru_attach = tcp_usr_attach, 1477 .pru_bind = tcp_usr_bind, 1478 .pru_connect = tcp_usr_connect, 1479 .pru_control = in_control, 1480 .pru_detach = tcp_usr_detach, 1481 .pru_disconnect = tcp_usr_disconnect, 1482 .pru_listen = tcp_usr_listen, 1483 .pru_peeraddr = in_getpeeraddr, 1484 .pru_rcvd = tcp_usr_rcvd, 1485 .pru_rcvoob = tcp_usr_rcvoob, 1486 .pru_send = tcp_usr_send, 1487 .pru_ready = tcp_usr_ready, 1488 .pru_shutdown = tcp_usr_shutdown, 1489 .pru_sockaddr = in_getsockaddr, 1490 .pru_sosetlabel = in_pcbsosetlabel, 1491 .pru_close = tcp_usr_close, 1492 }; 1493 #endif /* INET */ 1494 1495 #ifdef INET6 1496 struct pr_usrreqs tcp6_usrreqs = { 1497 .pru_abort = tcp_usr_abort, 1498 .pru_accept = tcp6_usr_accept, 1499 .pru_attach = tcp_usr_attach, 1500 .pru_bind = tcp6_usr_bind, 1501 .pru_connect = tcp6_usr_connect, 1502 .pru_control = in6_control, 1503 .pru_detach = tcp_usr_detach, 1504 .pru_disconnect = tcp_usr_disconnect, 1505 .pru_listen = tcp6_usr_listen, 1506 .pru_peeraddr = in6_mapped_peeraddr, 1507 .pru_rcvd = tcp_usr_rcvd, 1508 .pru_rcvoob = tcp_usr_rcvoob, 1509 .pru_send = tcp_usr_send, 1510 .pru_ready = tcp_usr_ready, 1511 .pru_shutdown = tcp_usr_shutdown, 1512 .pru_sockaddr = in6_mapped_sockaddr, 1513 .pru_sosetlabel = in_pcbsosetlabel, 1514 .pru_close = tcp_usr_close, 1515 }; 1516 #endif /* INET6 */ 1517 1518 #ifdef INET 1519 /* 1520 * Common subroutine to open a TCP connection to remote host specified 1521 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1522 * port number if needed. Call in_pcbconnect_setup to do the routing and 1523 * to choose a local host address (interface). If there is an existing 1524 * incarnation of the same connection in TIME-WAIT state and if the remote 1525 * host was sending CC options and if the connection duration was < MSL, then 1526 * truncate the previous TIME-WAIT state and proceed. 1527 * Initialize connection parameters and enter SYN-SENT state. 1528 */ 1529 static int 1530 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1531 { 1532 struct inpcb *inp = tp->t_inpcb, *oinp; 1533 struct socket *so = inp->inp_socket; 1534 struct in_addr laddr; 1535 u_short lport; 1536 int error; 1537 1538 NET_EPOCH_ASSERT(); 1539 INP_WLOCK_ASSERT(inp); 1540 INP_HASH_WLOCK(&V_tcbinfo); 1541 1542 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1543 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1544 if (error) 1545 goto out; 1546 } 1547 1548 /* 1549 * Cannot simply call in_pcbconnect, because there might be an 1550 * earlier incarnation of this same connection still in 1551 * TIME_WAIT state, creating an ADDRINUSE error. 1552 */ 1553 laddr = inp->inp_laddr; 1554 lport = inp->inp_lport; 1555 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport, 1556 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred); 1557 if (error && oinp == NULL) 1558 goto out; 1559 if (oinp) { 1560 error = EADDRINUSE; 1561 goto out; 1562 } 1563 /* Handle initial bind if it hadn't been done in advance. */ 1564 if (inp->inp_lport == 0) { 1565 inp->inp_lport = lport; 1566 if (in_pcbinshash(inp) != 0) { 1567 inp->inp_lport = 0; 1568 error = EAGAIN; 1569 goto out; 1570 } 1571 } 1572 inp->inp_laddr = laddr; 1573 in_pcbrehash(inp); 1574 INP_HASH_WUNLOCK(&V_tcbinfo); 1575 1576 /* 1577 * Compute window scaling to request: 1578 * Scale to fit into sweet spot. See tcp_syncache.c. 1579 * XXX: This should move to tcp_output(). 1580 */ 1581 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1582 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1583 tp->request_r_scale++; 1584 1585 soisconnecting(so); 1586 TCPSTAT_INC(tcps_connattempt); 1587 tcp_state_change(tp, TCPS_SYN_SENT); 1588 tp->iss = tcp_new_isn(&inp->inp_inc); 1589 if (tp->t_flags & TF_REQ_TSTMP) 1590 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1591 tcp_sendseqinit(tp); 1592 1593 return 0; 1594 1595 out: 1596 INP_HASH_WUNLOCK(&V_tcbinfo); 1597 return (error); 1598 } 1599 #endif /* INET */ 1600 1601 #ifdef INET6 1602 static int 1603 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td) 1604 { 1605 struct inpcb *inp = tp->t_inpcb; 1606 int error; 1607 1608 INP_WLOCK_ASSERT(inp); 1609 INP_HASH_WLOCK(&V_tcbinfo); 1610 1611 if (V_tcp_require_unique_port && inp->inp_lport == 0) { 1612 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred); 1613 if (error) 1614 goto out; 1615 } 1616 error = in6_pcbconnect(inp, nam, td->td_ucred); 1617 if (error != 0) 1618 goto out; 1619 INP_HASH_WUNLOCK(&V_tcbinfo); 1620 1621 /* Compute window scaling to request. */ 1622 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1623 (TCP_MAXWIN << tp->request_r_scale) < sb_max) 1624 tp->request_r_scale++; 1625 1626 soisconnecting(inp->inp_socket); 1627 TCPSTAT_INC(tcps_connattempt); 1628 tcp_state_change(tp, TCPS_SYN_SENT); 1629 tp->iss = tcp_new_isn(&inp->inp_inc); 1630 if (tp->t_flags & TF_REQ_TSTMP) 1631 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc); 1632 tcp_sendseqinit(tp); 1633 1634 return 0; 1635 1636 out: 1637 INP_HASH_WUNLOCK(&V_tcbinfo); 1638 return error; 1639 } 1640 #endif /* INET6 */ 1641 1642 /* 1643 * Export TCP internal state information via a struct tcp_info, based on the 1644 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently 1645 * (TCP state machine, etc). We export all information using FreeBSD-native 1646 * constants -- for example, the numeric values for tcpi_state will differ 1647 * from Linux. 1648 */ 1649 static void 1650 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) 1651 { 1652 1653 INP_WLOCK_ASSERT(tp->t_inpcb); 1654 bzero(ti, sizeof(*ti)); 1655 1656 ti->tcpi_state = tp->t_state; 1657 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) 1658 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; 1659 if (tp->t_flags & TF_SACK_PERMIT) 1660 ti->tcpi_options |= TCPI_OPT_SACK; 1661 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { 1662 ti->tcpi_options |= TCPI_OPT_WSCALE; 1663 ti->tcpi_snd_wscale = tp->snd_scale; 1664 ti->tcpi_rcv_wscale = tp->rcv_scale; 1665 } 1666 if (tp->t_flags2 & TF2_ECN_PERMIT) 1667 ti->tcpi_options |= TCPI_OPT_ECN; 1668 1669 ti->tcpi_rto = tp->t_rxtcur * tick; 1670 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick; 1671 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 1672 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT; 1673 1674 ti->tcpi_snd_ssthresh = tp->snd_ssthresh; 1675 ti->tcpi_snd_cwnd = tp->snd_cwnd; 1676 1677 /* 1678 * FreeBSD-specific extension fields for tcp_info. 1679 */ 1680 ti->tcpi_rcv_space = tp->rcv_wnd; 1681 ti->tcpi_rcv_nxt = tp->rcv_nxt; 1682 ti->tcpi_snd_wnd = tp->snd_wnd; 1683 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */ 1684 ti->tcpi_snd_nxt = tp->snd_nxt; 1685 ti->tcpi_snd_mss = tp->t_maxseg; 1686 ti->tcpi_rcv_mss = tp->t_maxseg; 1687 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack; 1688 ti->tcpi_rcv_ooopack = tp->t_rcvoopack; 1689 ti->tcpi_snd_zerowin = tp->t_sndzerowin; 1690 #ifdef TCP_OFFLOAD 1691 if (tp->t_flags & TF_TOE) { 1692 ti->tcpi_options |= TCPI_OPT_TOE; 1693 tcp_offload_tcp_info(tp, ti); 1694 } 1695 #endif 1696 } 1697 1698 /* 1699 * tcp_ctloutput() must drop the inpcb lock before performing copyin on 1700 * socket option arguments. When it re-acquires the lock after the copy, it 1701 * has to revalidate that the connection is still valid for the socket 1702 * option. 1703 */ 1704 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \ 1705 INP_WLOCK(inp); \ 1706 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ 1707 INP_WUNLOCK(inp); \ 1708 cleanup; \ 1709 return (ECONNRESET); \ 1710 } \ 1711 tp = intotcpcb(inp); \ 1712 } while(0) 1713 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */) 1714 1715 static int 1716 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt) 1717 { 1718 struct tcpcb *tp = intotcpcb(inp); 1719 int error = 0; 1720 1721 MPASS(sopt->sopt_dir == SOPT_SET); 1722 1723 if (sopt->sopt_level != IPPROTO_TCP) { 1724 #ifdef INET6 1725 if (inp->inp_vflag & INP_IPV6PROTO) 1726 error = ip6_ctloutput(inp->inp_socket, sopt); 1727 #endif 1728 #if defined(INET6) && defined(INET) 1729 else 1730 #endif 1731 #ifdef INET 1732 error = ip_ctloutput(inp->inp_socket, sopt); 1733 #endif 1734 /* 1735 * When an IP-level socket option affects TCP, pass control 1736 * down to stack tfb_tcp_ctloutput, otherwise return what 1737 * IP level returned. 1738 */ 1739 switch (sopt->sopt_level) { 1740 #ifdef INET6 1741 case IPPROTO_IPV6: 1742 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) 1743 return (error); 1744 switch (sopt->sopt_name) { 1745 case IPV6_TCLASS: 1746 /* Notify tcp stacks that care (e.g. RACK). */ 1747 break; 1748 case IPV6_USE_MIN_MTU: 1749 /* Update t_maxseg accordingly. */ 1750 break; 1751 default: 1752 return (error); 1753 } 1754 break; 1755 #endif 1756 #ifdef INET 1757 case IPPROTO_IP: 1758 switch (sopt->sopt_name) { 1759 case IP_TOS: 1760 case IP_TTL: 1761 /* Notify tcp stacks that care (e.g. RACK). */ 1762 break; 1763 default: 1764 return (error); 1765 } 1766 break; 1767 #endif 1768 default: 1769 return (error); 1770 } 1771 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) { 1772 /* 1773 * Protect the TCP option TCP_FUNCTION_BLK so 1774 * that a sub-function can *never* overwrite this. 1775 */ 1776 struct tcp_function_set fsn; 1777 struct tcp_function_block *blk; 1778 1779 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn); 1780 if (error) 1781 return (error); 1782 1783 INP_WLOCK(inp); 1784 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1785 INP_WUNLOCK(inp); 1786 return (ECONNRESET); 1787 } 1788 tp = intotcpcb(inp); 1789 1790 blk = find_and_ref_tcp_functions(&fsn); 1791 if (blk == NULL) { 1792 INP_WUNLOCK(inp); 1793 return (ENOENT); 1794 } 1795 if (tp->t_fb == blk) { 1796 /* You already have this */ 1797 refcount_release(&blk->tfb_refcnt); 1798 INP_WUNLOCK(inp); 1799 return (0); 1800 } 1801 if (tp->t_state != TCPS_CLOSED) { 1802 /* 1803 * The user has advanced the state 1804 * past the initial point, we may not 1805 * be able to switch. 1806 */ 1807 if (blk->tfb_tcp_handoff_ok != NULL) { 1808 /* 1809 * Does the stack provide a 1810 * query mechanism, if so it may 1811 * still be possible? 1812 */ 1813 error = (*blk->tfb_tcp_handoff_ok)(tp); 1814 } else 1815 error = EINVAL; 1816 if (error) { 1817 refcount_release(&blk->tfb_refcnt); 1818 INP_WUNLOCK(inp); 1819 return(error); 1820 } 1821 } 1822 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1823 refcount_release(&blk->tfb_refcnt); 1824 INP_WUNLOCK(inp); 1825 return (ENOENT); 1826 } 1827 /* 1828 * Release the old refcnt, the 1829 * lookup acquired a ref on the 1830 * new one already. 1831 */ 1832 if (tp->t_fb->tfb_tcp_fb_fini) { 1833 struct epoch_tracker et; 1834 /* 1835 * Tell the stack to cleanup with 0 i.e. 1836 * the tcb is not going away. 1837 */ 1838 NET_EPOCH_ENTER(et); 1839 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 1840 NET_EPOCH_EXIT(et); 1841 } 1842 #ifdef TCPHPTS 1843 /* Assure that we are not on any hpts */ 1844 tcp_hpts_remove(tp->t_inpcb); 1845 #endif 1846 if (blk->tfb_tcp_fb_init) { 1847 error = (*blk->tfb_tcp_fb_init)(tp); 1848 if (error) { 1849 refcount_release(&blk->tfb_refcnt); 1850 if (tp->t_fb->tfb_tcp_fb_init) { 1851 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1852 /* Fall back failed, drop the connection */ 1853 INP_WUNLOCK(inp); 1854 soabort(inp->inp_socket); 1855 return(error); 1856 } 1857 } 1858 goto err_out; 1859 } 1860 } 1861 refcount_release(&tp->t_fb->tfb_refcnt); 1862 tp->t_fb = blk; 1863 #ifdef TCP_OFFLOAD 1864 if (tp->t_flags & TF_TOE) { 1865 tcp_offload_ctloutput(tp, sopt->sopt_dir, 1866 sopt->sopt_name); 1867 } 1868 #endif 1869 err_out: 1870 INP_WUNLOCK(inp); 1871 return (error); 1872 } 1873 1874 INP_WLOCK(inp); 1875 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1876 INP_WUNLOCK(inp); 1877 return (ECONNRESET); 1878 } 1879 tp = intotcpcb(inp); 1880 1881 /* Pass in the INP locked, caller must unlock it. */ 1882 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp)); 1883 } 1884 1885 static int 1886 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt) 1887 { 1888 int error = 0; 1889 struct tcpcb *tp; 1890 1891 MPASS(sopt->sopt_dir == SOPT_GET); 1892 1893 if (sopt->sopt_level != IPPROTO_TCP) { 1894 #ifdef INET6 1895 if (inp->inp_vflag & INP_IPV6PROTO) 1896 error = ip6_ctloutput(inp->inp_socket, sopt); 1897 #endif /* INET6 */ 1898 #if defined(INET6) && defined(INET) 1899 else 1900 #endif 1901 #ifdef INET 1902 error = ip_ctloutput(inp->inp_socket, sopt); 1903 #endif 1904 return (error); 1905 } 1906 INP_WLOCK(inp); 1907 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1908 INP_WUNLOCK(inp); 1909 return (ECONNRESET); 1910 } 1911 tp = intotcpcb(inp); 1912 if (((sopt->sopt_name == TCP_FUNCTION_BLK) || 1913 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) { 1914 struct tcp_function_set fsn; 1915 1916 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) { 1917 memset(&fsn, 0, sizeof(fsn)); 1918 find_tcp_function_alias(tp->t_fb, &fsn); 1919 } else { 1920 strncpy(fsn.function_set_name, 1921 tp->t_fb->tfb_tcp_block_name, 1922 TCP_FUNCTION_NAME_LEN_MAX); 1923 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 1924 } 1925 fsn.pcbcnt = tp->t_fb->tfb_refcnt; 1926 INP_WUNLOCK(inp); 1927 error = sooptcopyout(sopt, &fsn, sizeof fsn); 1928 return (error); 1929 } 1930 1931 /* Pass in the INP locked, caller must unlock it. */ 1932 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp)); 1933 } 1934 1935 int 1936 tcp_ctloutput(struct socket *so, struct sockopt *sopt) 1937 { 1938 struct inpcb *inp; 1939 1940 inp = sotoinpcb(so); 1941 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1942 1943 if (sopt->sopt_dir == SOPT_SET) 1944 return (tcp_ctloutput_set(inp, sopt)); 1945 else if (sopt->sopt_dir == SOPT_GET) 1946 return (tcp_ctloutput_get(inp, sopt)); 1947 else 1948 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1949 } 1950 1951 /* 1952 * If this assert becomes untrue, we need to change the size of the buf 1953 * variable in tcp_default_ctloutput(). 1954 */ 1955 #ifdef CTASSERT 1956 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1957 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1958 #endif 1959 1960 #ifdef KERN_TLS 1961 static int 1962 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1963 { 1964 struct tls_enable_v0 tls_v0; 1965 int error; 1966 1967 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1968 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1969 sizeof(tls_v0)); 1970 if (error) 1971 return (error); 1972 memset(tls, 0, sizeof(*tls)); 1973 tls->cipher_key = tls_v0.cipher_key; 1974 tls->iv = tls_v0.iv; 1975 tls->auth_key = tls_v0.auth_key; 1976 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1977 tls->cipher_key_len = tls_v0.cipher_key_len; 1978 tls->iv_len = tls_v0.iv_len; 1979 tls->auth_algorithm = tls_v0.auth_algorithm; 1980 tls->auth_key_len = tls_v0.auth_key_len; 1981 tls->flags = tls_v0.flags; 1982 tls->tls_vmajor = tls_v0.tls_vmajor; 1983 tls->tls_vminor = tls_v0.tls_vminor; 1984 return (0); 1985 } 1986 1987 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1988 } 1989 #endif 1990 1991 extern struct cc_algo newreno_cc_algo; 1992 1993 static int 1994 tcp_congestion(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 1995 { 1996 struct cc_algo *algo; 1997 void *ptr = NULL; 1998 struct cc_var cc_mem; 1999 char buf[TCP_CA_NAME_MAX]; 2000 size_t mem_sz; 2001 int error; 2002 2003 INP_WUNLOCK(inp); 2004 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 2005 if (error) 2006 return(error); 2007 buf[sopt->sopt_valsize] = '\0'; 2008 CC_LIST_RLOCK(); 2009 STAILQ_FOREACH(algo, &cc_list, entries) 2010 if (strncmp(buf, algo->name, 2011 TCP_CA_NAME_MAX) == 0) { 2012 if (algo->flags & CC_MODULE_BEING_REMOVED) { 2013 /* We can't "see" modules being unloaded */ 2014 continue; 2015 } 2016 break; 2017 } 2018 if (algo == NULL) { 2019 CC_LIST_RUNLOCK(); 2020 return(ESRCH); 2021 } 2022 do_over: 2023 if (algo->cb_init != NULL) { 2024 /* We can now pre-get the memory for the CC */ 2025 mem_sz = (*algo->cc_data_sz)(); 2026 if (mem_sz == 0) { 2027 goto no_mem_needed; 2028 } 2029 CC_LIST_RUNLOCK(); 2030 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 2031 CC_LIST_RLOCK(); 2032 STAILQ_FOREACH(algo, &cc_list, entries) 2033 if (strncmp(buf, algo->name, 2034 TCP_CA_NAME_MAX) == 0) 2035 break; 2036 if (algo == NULL) { 2037 if (ptr) 2038 free(ptr, M_CC_MEM); 2039 CC_LIST_RUNLOCK(); 2040 return(ESRCH); 2041 } 2042 } else { 2043 no_mem_needed: 2044 mem_sz = 0; 2045 ptr = NULL; 2046 } 2047 /* 2048 * Make sure its all clean and zero and also get 2049 * back the inplock. 2050 */ 2051 memset(&cc_mem, 0, sizeof(cc_mem)); 2052 if (mem_sz != (*algo->cc_data_sz)()) { 2053 if (ptr) 2054 free(ptr, M_CC_MEM); 2055 goto do_over; 2056 } 2057 INP_WLOCK(inp); 2058 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 2059 INP_WUNLOCK(inp); 2060 CC_LIST_RUNLOCK(); 2061 free(ptr, M_CC_MEM); 2062 return (ECONNRESET); 2063 } 2064 tp = intotcpcb(inp); 2065 if (ptr != NULL) 2066 memset(ptr, 0, mem_sz); 2067 CC_LIST_RUNLOCK(); 2068 cc_mem.ccvc.tcp = tp; 2069 /* 2070 * We once again hold a write lock over the tcb so it's 2071 * safe to do these things without ordering concerns. 2072 * Note here we init into stack memory. 2073 */ 2074 if (algo->cb_init != NULL) 2075 error = algo->cb_init(&cc_mem, ptr); 2076 else 2077 error = 0; 2078 /* 2079 * The CC algorithms, when given their memory 2080 * should not fail we could in theory have a 2081 * KASSERT here. 2082 */ 2083 if (error == 0) { 2084 /* 2085 * Touchdown, lets go ahead and move the 2086 * connection to the new CC module by 2087 * copying in the cc_mem after we call 2088 * the old ones cleanup (if any). 2089 */ 2090 if (CC_ALGO(tp)->cb_destroy != NULL) 2091 CC_ALGO(tp)->cb_destroy(tp->ccv); 2092 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2093 tp->cc_algo = algo; 2094 /* Ok now are we where we have gotten past any conn_init? */ 2095 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2096 /* Yep run the connection init for the new CC */ 2097 CC_ALGO(tp)->conn_init(tp->ccv); 2098 } 2099 } else if (ptr) 2100 free(ptr, M_CC_MEM); 2101 INP_WUNLOCK(inp); 2102 return (error); 2103 } 2104 2105 int 2106 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 2107 { 2108 int error, opt, optval; 2109 u_int ui; 2110 struct tcp_info ti; 2111 #ifdef KERN_TLS 2112 struct tls_enable tls; 2113 #endif 2114 char *pbuf, buf[TCP_LOG_ID_LEN]; 2115 #ifdef STATS 2116 struct statsblob *sbp; 2117 #endif 2118 size_t len; 2119 2120 INP_WLOCK_ASSERT(inp); 2121 2122 switch (sopt->sopt_level) { 2123 #ifdef INET6 2124 case IPPROTO_IPV6: 2125 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2126 switch (sopt->sopt_name) { 2127 case IPV6_USE_MIN_MTU: 2128 tcp6_use_min_mtu(tp); 2129 /* FALLTHROUGH */ 2130 } 2131 INP_WUNLOCK(inp); 2132 return (0); 2133 #endif 2134 #ifdef INET 2135 case IPPROTO_IP: 2136 INP_WUNLOCK(inp); 2137 return (0); 2138 #endif 2139 } 2140 2141 /* 2142 * For TCP_CCALGOOPT forward the control to CC module, for both 2143 * SOPT_SET and SOPT_GET. 2144 */ 2145 switch (sopt->sopt_name) { 2146 case TCP_CCALGOOPT: 2147 INP_WUNLOCK(inp); 2148 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2149 return (EINVAL); 2150 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2151 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2152 sopt->sopt_valsize); 2153 if (error) { 2154 free(pbuf, M_TEMP); 2155 return (error); 2156 } 2157 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2158 if (CC_ALGO(tp)->ctl_output != NULL) 2159 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 2160 else 2161 error = ENOENT; 2162 INP_WUNLOCK(inp); 2163 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2164 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2165 free(pbuf, M_TEMP); 2166 return (error); 2167 } 2168 2169 switch (sopt->sopt_dir) { 2170 case SOPT_SET: 2171 switch (sopt->sopt_name) { 2172 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2173 case TCP_MD5SIG: 2174 if (!TCPMD5_ENABLED()) { 2175 INP_WUNLOCK(inp); 2176 return (ENOPROTOOPT); 2177 } 2178 error = TCPMD5_PCBCTL(inp, sopt); 2179 if (error) 2180 return (error); 2181 goto unlock_and_done; 2182 #endif /* IPSEC */ 2183 2184 case TCP_NODELAY: 2185 case TCP_NOOPT: 2186 case TCP_LRD: 2187 INP_WUNLOCK(inp); 2188 error = sooptcopyin(sopt, &optval, sizeof optval, 2189 sizeof optval); 2190 if (error) 2191 return (error); 2192 2193 INP_WLOCK_RECHECK(inp); 2194 switch (sopt->sopt_name) { 2195 case TCP_NODELAY: 2196 opt = TF_NODELAY; 2197 break; 2198 case TCP_NOOPT: 2199 opt = TF_NOOPT; 2200 break; 2201 case TCP_LRD: 2202 opt = TF_LRD; 2203 break; 2204 default: 2205 opt = 0; /* dead code to fool gcc */ 2206 break; 2207 } 2208 2209 if (optval) 2210 tp->t_flags |= opt; 2211 else 2212 tp->t_flags &= ~opt; 2213 unlock_and_done: 2214 #ifdef TCP_OFFLOAD 2215 if (tp->t_flags & TF_TOE) { 2216 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2217 sopt->sopt_name); 2218 } 2219 #endif 2220 INP_WUNLOCK(inp); 2221 break; 2222 2223 case TCP_NOPUSH: 2224 INP_WUNLOCK(inp); 2225 error = sooptcopyin(sopt, &optval, sizeof optval, 2226 sizeof optval); 2227 if (error) 2228 return (error); 2229 2230 INP_WLOCK_RECHECK(inp); 2231 if (optval) 2232 tp->t_flags |= TF_NOPUSH; 2233 else if (tp->t_flags & TF_NOPUSH) { 2234 tp->t_flags &= ~TF_NOPUSH; 2235 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2236 struct epoch_tracker et; 2237 2238 NET_EPOCH_ENTER(et); 2239 error = tcp_output_nodrop(tp); 2240 NET_EPOCH_EXIT(et); 2241 } 2242 } 2243 goto unlock_and_done; 2244 2245 case TCP_REMOTE_UDP_ENCAPS_PORT: 2246 INP_WUNLOCK(inp); 2247 error = sooptcopyin(sopt, &optval, sizeof optval, 2248 sizeof optval); 2249 if (error) 2250 return (error); 2251 if ((optval < TCP_TUNNELING_PORT_MIN) || 2252 (optval > TCP_TUNNELING_PORT_MAX)) { 2253 /* Its got to be in range */ 2254 return (EINVAL); 2255 } 2256 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2257 /* You have to have enabled a UDP tunneling port first */ 2258 return (EINVAL); 2259 } 2260 INP_WLOCK_RECHECK(inp); 2261 if (tp->t_state != TCPS_CLOSED) { 2262 /* You can't change after you are connected */ 2263 error = EINVAL; 2264 } else { 2265 /* Ok we are all good set the port */ 2266 tp->t_port = htons(optval); 2267 } 2268 goto unlock_and_done; 2269 2270 case TCP_MAXSEG: 2271 INP_WUNLOCK(inp); 2272 error = sooptcopyin(sopt, &optval, sizeof optval, 2273 sizeof optval); 2274 if (error) 2275 return (error); 2276 2277 INP_WLOCK_RECHECK(inp); 2278 if (optval > 0 && optval <= tp->t_maxseg && 2279 optval + 40 >= V_tcp_minmss) 2280 tp->t_maxseg = optval; 2281 else 2282 error = EINVAL; 2283 goto unlock_and_done; 2284 2285 case TCP_INFO: 2286 INP_WUNLOCK(inp); 2287 error = EINVAL; 2288 break; 2289 2290 case TCP_STATS: 2291 INP_WUNLOCK(inp); 2292 #ifdef STATS 2293 error = sooptcopyin(sopt, &optval, sizeof optval, 2294 sizeof optval); 2295 if (error) 2296 return (error); 2297 2298 if (optval > 0) 2299 sbp = stats_blob_alloc( 2300 V_tcp_perconn_stats_dflt_tpl, 0); 2301 else 2302 sbp = NULL; 2303 2304 INP_WLOCK_RECHECK(inp); 2305 if ((tp->t_stats != NULL && sbp == NULL) || 2306 (tp->t_stats == NULL && sbp != NULL)) { 2307 struct statsblob *t = tp->t_stats; 2308 tp->t_stats = sbp; 2309 sbp = t; 2310 } 2311 INP_WUNLOCK(inp); 2312 2313 stats_blob_destroy(sbp); 2314 #else 2315 return (EOPNOTSUPP); 2316 #endif /* !STATS */ 2317 break; 2318 2319 case TCP_CONGESTION: 2320 error = tcp_congestion(so, sopt, inp, tp); 2321 break; 2322 2323 case TCP_REUSPORT_LB_NUMA: 2324 INP_WUNLOCK(inp); 2325 error = sooptcopyin(sopt, &optval, sizeof(optval), 2326 sizeof(optval)); 2327 INP_WLOCK_RECHECK(inp); 2328 if (!error) 2329 error = in_pcblbgroup_numa(inp, optval); 2330 INP_WUNLOCK(inp); 2331 break; 2332 2333 #ifdef KERN_TLS 2334 case TCP_TXTLS_ENABLE: 2335 INP_WUNLOCK(inp); 2336 error = copyin_tls_enable(sopt, &tls); 2337 if (error) 2338 break; 2339 error = ktls_enable_tx(so, &tls); 2340 break; 2341 case TCP_TXTLS_MODE: 2342 INP_WUNLOCK(inp); 2343 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2344 if (error) 2345 return (error); 2346 2347 INP_WLOCK_RECHECK(inp); 2348 error = ktls_set_tx_mode(so, ui); 2349 INP_WUNLOCK(inp); 2350 break; 2351 case TCP_RXTLS_ENABLE: 2352 INP_WUNLOCK(inp); 2353 error = sooptcopyin(sopt, &tls, sizeof(tls), 2354 sizeof(tls)); 2355 if (error) 2356 break; 2357 error = ktls_enable_rx(so, &tls); 2358 break; 2359 #endif 2360 2361 case TCP_KEEPIDLE: 2362 case TCP_KEEPINTVL: 2363 case TCP_KEEPINIT: 2364 INP_WUNLOCK(inp); 2365 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2366 if (error) 2367 return (error); 2368 2369 if (ui > (UINT_MAX / hz)) { 2370 error = EINVAL; 2371 break; 2372 } 2373 ui *= hz; 2374 2375 INP_WLOCK_RECHECK(inp); 2376 switch (sopt->sopt_name) { 2377 case TCP_KEEPIDLE: 2378 tp->t_keepidle = ui; 2379 /* 2380 * XXX: better check current remaining 2381 * timeout and "merge" it with new value. 2382 */ 2383 if ((tp->t_state > TCPS_LISTEN) && 2384 (tp->t_state <= TCPS_CLOSING)) 2385 tcp_timer_activate(tp, TT_KEEP, 2386 TP_KEEPIDLE(tp)); 2387 break; 2388 case TCP_KEEPINTVL: 2389 tp->t_keepintvl = ui; 2390 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2391 (TP_MAXIDLE(tp) > 0)) 2392 tcp_timer_activate(tp, TT_2MSL, 2393 TP_MAXIDLE(tp)); 2394 break; 2395 case TCP_KEEPINIT: 2396 tp->t_keepinit = ui; 2397 if (tp->t_state == TCPS_SYN_RECEIVED || 2398 tp->t_state == TCPS_SYN_SENT) 2399 tcp_timer_activate(tp, TT_KEEP, 2400 TP_KEEPINIT(tp)); 2401 break; 2402 } 2403 goto unlock_and_done; 2404 2405 case TCP_KEEPCNT: 2406 INP_WUNLOCK(inp); 2407 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2408 if (error) 2409 return (error); 2410 2411 INP_WLOCK_RECHECK(inp); 2412 tp->t_keepcnt = ui; 2413 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2414 (TP_MAXIDLE(tp) > 0)) 2415 tcp_timer_activate(tp, TT_2MSL, 2416 TP_MAXIDLE(tp)); 2417 goto unlock_and_done; 2418 2419 #ifdef TCPPCAP 2420 case TCP_PCAP_OUT: 2421 case TCP_PCAP_IN: 2422 INP_WUNLOCK(inp); 2423 error = sooptcopyin(sopt, &optval, sizeof optval, 2424 sizeof optval); 2425 if (error) 2426 return (error); 2427 2428 INP_WLOCK_RECHECK(inp); 2429 if (optval >= 0) 2430 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2431 &(tp->t_outpkts) : &(tp->t_inpkts), 2432 optval); 2433 else 2434 error = EINVAL; 2435 goto unlock_and_done; 2436 #endif 2437 2438 case TCP_FASTOPEN: { 2439 struct tcp_fastopen tfo_optval; 2440 2441 INP_WUNLOCK(inp); 2442 if (!V_tcp_fastopen_client_enable && 2443 !V_tcp_fastopen_server_enable) 2444 return (EPERM); 2445 2446 error = sooptcopyin(sopt, &tfo_optval, 2447 sizeof(tfo_optval), sizeof(int)); 2448 if (error) 2449 return (error); 2450 2451 INP_WLOCK_RECHECK(inp); 2452 if ((tp->t_state != TCPS_CLOSED) && 2453 (tp->t_state != TCPS_LISTEN)) { 2454 error = EINVAL; 2455 goto unlock_and_done; 2456 } 2457 if (tfo_optval.enable) { 2458 if (tp->t_state == TCPS_LISTEN) { 2459 if (!V_tcp_fastopen_server_enable) { 2460 error = EPERM; 2461 goto unlock_and_done; 2462 } 2463 2464 if (tp->t_tfo_pending == NULL) 2465 tp->t_tfo_pending = 2466 tcp_fastopen_alloc_counter(); 2467 } else { 2468 /* 2469 * If a pre-shared key was provided, 2470 * stash it in the client cookie 2471 * field of the tcpcb for use during 2472 * connect. 2473 */ 2474 if (sopt->sopt_valsize == 2475 sizeof(tfo_optval)) { 2476 memcpy(tp->t_tfo_cookie.client, 2477 tfo_optval.psk, 2478 TCP_FASTOPEN_PSK_LEN); 2479 tp->t_tfo_client_cookie_len = 2480 TCP_FASTOPEN_PSK_LEN; 2481 } 2482 } 2483 tp->t_flags |= TF_FASTOPEN; 2484 } else 2485 tp->t_flags &= ~TF_FASTOPEN; 2486 goto unlock_and_done; 2487 } 2488 2489 #ifdef TCP_BLACKBOX 2490 case TCP_LOG: 2491 INP_WUNLOCK(inp); 2492 error = sooptcopyin(sopt, &optval, sizeof optval, 2493 sizeof optval); 2494 if (error) 2495 return (error); 2496 2497 INP_WLOCK_RECHECK(inp); 2498 error = tcp_log_state_change(tp, optval); 2499 goto unlock_and_done; 2500 2501 case TCP_LOGBUF: 2502 INP_WUNLOCK(inp); 2503 error = EINVAL; 2504 break; 2505 2506 case TCP_LOGID: 2507 INP_WUNLOCK(inp); 2508 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2509 if (error) 2510 break; 2511 buf[sopt->sopt_valsize] = '\0'; 2512 INP_WLOCK_RECHECK(inp); 2513 error = tcp_log_set_id(tp, buf); 2514 /* tcp_log_set_id() unlocks the INP. */ 2515 break; 2516 2517 case TCP_LOGDUMP: 2518 case TCP_LOGDUMPID: 2519 INP_WUNLOCK(inp); 2520 error = 2521 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2522 if (error) 2523 break; 2524 buf[sopt->sopt_valsize] = '\0'; 2525 INP_WLOCK_RECHECK(inp); 2526 if (sopt->sopt_name == TCP_LOGDUMP) { 2527 error = tcp_log_dump_tp_logbuf(tp, buf, 2528 M_WAITOK, true); 2529 INP_WUNLOCK(inp); 2530 } else { 2531 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2532 /* 2533 * tcp_log_dump_tp_bucket_logbufs() drops the 2534 * INP lock. 2535 */ 2536 } 2537 break; 2538 #endif 2539 2540 default: 2541 INP_WUNLOCK(inp); 2542 error = ENOPROTOOPT; 2543 break; 2544 } 2545 break; 2546 2547 case SOPT_GET: 2548 tp = intotcpcb(inp); 2549 switch (sopt->sopt_name) { 2550 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2551 case TCP_MD5SIG: 2552 if (!TCPMD5_ENABLED()) { 2553 INP_WUNLOCK(inp); 2554 return (ENOPROTOOPT); 2555 } 2556 error = TCPMD5_PCBCTL(inp, sopt); 2557 break; 2558 #endif 2559 2560 case TCP_NODELAY: 2561 optval = tp->t_flags & TF_NODELAY; 2562 INP_WUNLOCK(inp); 2563 error = sooptcopyout(sopt, &optval, sizeof optval); 2564 break; 2565 case TCP_MAXSEG: 2566 optval = tp->t_maxseg; 2567 INP_WUNLOCK(inp); 2568 error = sooptcopyout(sopt, &optval, sizeof optval); 2569 break; 2570 case TCP_REMOTE_UDP_ENCAPS_PORT: 2571 optval = ntohs(tp->t_port); 2572 INP_WUNLOCK(inp); 2573 error = sooptcopyout(sopt, &optval, sizeof optval); 2574 break; 2575 case TCP_NOOPT: 2576 optval = tp->t_flags & TF_NOOPT; 2577 INP_WUNLOCK(inp); 2578 error = sooptcopyout(sopt, &optval, sizeof optval); 2579 break; 2580 case TCP_NOPUSH: 2581 optval = tp->t_flags & TF_NOPUSH; 2582 INP_WUNLOCK(inp); 2583 error = sooptcopyout(sopt, &optval, sizeof optval); 2584 break; 2585 case TCP_INFO: 2586 tcp_fill_info(tp, &ti); 2587 INP_WUNLOCK(inp); 2588 error = sooptcopyout(sopt, &ti, sizeof ti); 2589 break; 2590 case TCP_STATS: 2591 { 2592 #ifdef STATS 2593 int nheld; 2594 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2595 2596 error = 0; 2597 socklen_t outsbsz = sopt->sopt_valsize; 2598 if (tp->t_stats == NULL) 2599 error = ENOENT; 2600 else if (outsbsz >= tp->t_stats->cursz) 2601 outsbsz = tp->t_stats->cursz; 2602 else if (outsbsz >= sizeof(struct statsblob)) 2603 outsbsz = sizeof(struct statsblob); 2604 else 2605 error = EINVAL; 2606 INP_WUNLOCK(inp); 2607 if (error) 2608 break; 2609 2610 sbp = sopt->sopt_val; 2611 nheld = atop(round_page(((vm_offset_t)sbp) + 2612 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2613 vm_page_t ma[nheld]; 2614 if (vm_fault_quick_hold_pages( 2615 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2616 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2617 nheld) < 0) { 2618 error = EFAULT; 2619 break; 2620 } 2621 2622 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2623 SIZEOF_MEMBER(struct statsblob, flags)))) 2624 goto unhold; 2625 2626 INP_WLOCK_RECHECK(inp); 2627 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2628 sbflags | SB_CLONE_USRDSTNOFAULT); 2629 INP_WUNLOCK(inp); 2630 sopt->sopt_valsize = outsbsz; 2631 unhold: 2632 vm_page_unhold_pages(ma, nheld); 2633 #else 2634 INP_WUNLOCK(inp); 2635 error = EOPNOTSUPP; 2636 #endif /* !STATS */ 2637 break; 2638 } 2639 case TCP_CONGESTION: 2640 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2641 INP_WUNLOCK(inp); 2642 error = sooptcopyout(sopt, buf, len + 1); 2643 break; 2644 case TCP_KEEPIDLE: 2645 case TCP_KEEPINTVL: 2646 case TCP_KEEPINIT: 2647 case TCP_KEEPCNT: 2648 switch (sopt->sopt_name) { 2649 case TCP_KEEPIDLE: 2650 ui = TP_KEEPIDLE(tp) / hz; 2651 break; 2652 case TCP_KEEPINTVL: 2653 ui = TP_KEEPINTVL(tp) / hz; 2654 break; 2655 case TCP_KEEPINIT: 2656 ui = TP_KEEPINIT(tp) / hz; 2657 break; 2658 case TCP_KEEPCNT: 2659 ui = TP_KEEPCNT(tp); 2660 break; 2661 } 2662 INP_WUNLOCK(inp); 2663 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2664 break; 2665 #ifdef TCPPCAP 2666 case TCP_PCAP_OUT: 2667 case TCP_PCAP_IN: 2668 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2669 &(tp->t_outpkts) : &(tp->t_inpkts)); 2670 INP_WUNLOCK(inp); 2671 error = sooptcopyout(sopt, &optval, sizeof optval); 2672 break; 2673 #endif 2674 case TCP_FASTOPEN: 2675 optval = tp->t_flags & TF_FASTOPEN; 2676 INP_WUNLOCK(inp); 2677 error = sooptcopyout(sopt, &optval, sizeof optval); 2678 break; 2679 #ifdef TCP_BLACKBOX 2680 case TCP_LOG: 2681 optval = tp->t_logstate; 2682 INP_WUNLOCK(inp); 2683 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2684 break; 2685 case TCP_LOGBUF: 2686 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2687 error = tcp_log_getlogbuf(sopt, tp); 2688 break; 2689 case TCP_LOGID: 2690 len = tcp_log_get_id(tp, buf); 2691 INP_WUNLOCK(inp); 2692 error = sooptcopyout(sopt, buf, len + 1); 2693 break; 2694 case TCP_LOGDUMP: 2695 case TCP_LOGDUMPID: 2696 INP_WUNLOCK(inp); 2697 error = EINVAL; 2698 break; 2699 #endif 2700 #ifdef KERN_TLS 2701 case TCP_TXTLS_MODE: 2702 error = ktls_get_tx_mode(so, &optval); 2703 INP_WUNLOCK(inp); 2704 if (error == 0) 2705 error = sooptcopyout(sopt, &optval, 2706 sizeof(optval)); 2707 break; 2708 case TCP_RXTLS_MODE: 2709 error = ktls_get_rx_mode(so, &optval); 2710 INP_WUNLOCK(inp); 2711 if (error == 0) 2712 error = sooptcopyout(sopt, &optval, 2713 sizeof(optval)); 2714 break; 2715 #endif 2716 case TCP_LRD: 2717 optval = tp->t_flags & TF_LRD; 2718 INP_WUNLOCK(inp); 2719 error = sooptcopyout(sopt, &optval, sizeof optval); 2720 break; 2721 default: 2722 INP_WUNLOCK(inp); 2723 error = ENOPROTOOPT; 2724 break; 2725 } 2726 break; 2727 } 2728 return (error); 2729 } 2730 #undef INP_WLOCK_RECHECK 2731 #undef INP_WLOCK_RECHECK_CLEANUP 2732 2733 /* 2734 * Initiate (or continue) disconnect. 2735 * If embryonic state, just send reset (once). 2736 * If in ``let data drain'' option and linger null, just drop. 2737 * Otherwise (hard), mark socket disconnecting and drop 2738 * current input data; switch states based on user close, and 2739 * send segment to peer (with FIN). 2740 */ 2741 static void 2742 tcp_disconnect(struct tcpcb *tp) 2743 { 2744 struct inpcb *inp = tp->t_inpcb; 2745 struct socket *so = inp->inp_socket; 2746 2747 NET_EPOCH_ASSERT(); 2748 INP_WLOCK_ASSERT(inp); 2749 2750 /* 2751 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2752 * socket is still open. 2753 */ 2754 if (tp->t_state < TCPS_ESTABLISHED && 2755 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2756 tp = tcp_close(tp); 2757 KASSERT(tp != NULL, 2758 ("tcp_disconnect: tcp_close() returned NULL")); 2759 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2760 tp = tcp_drop(tp, 0); 2761 KASSERT(tp != NULL, 2762 ("tcp_disconnect: tcp_drop() returned NULL")); 2763 } else { 2764 soisdisconnecting(so); 2765 sbflush(&so->so_rcv); 2766 tcp_usrclosed(tp); 2767 if (!(inp->inp_flags & INP_DROPPED)) 2768 /* Ignore stack's drop request, we already at it. */ 2769 (void)tcp_output_nodrop(tp); 2770 } 2771 } 2772 2773 /* 2774 * User issued close, and wish to trail through shutdown states: 2775 * if never received SYN, just forget it. If got a SYN from peer, 2776 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2777 * If already got a FIN from peer, then almost done; go to LAST_ACK 2778 * state. In all other cases, have already sent FIN to peer (e.g. 2779 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2780 * for peer to send FIN or not respond to keep-alives, etc. 2781 * We can let the user exit from the close as soon as the FIN is acked. 2782 */ 2783 static void 2784 tcp_usrclosed(struct tcpcb *tp) 2785 { 2786 2787 NET_EPOCH_ASSERT(); 2788 INP_WLOCK_ASSERT(tp->t_inpcb); 2789 2790 switch (tp->t_state) { 2791 case TCPS_LISTEN: 2792 #ifdef TCP_OFFLOAD 2793 tcp_offload_listen_stop(tp); 2794 #endif 2795 tcp_state_change(tp, TCPS_CLOSED); 2796 /* FALLTHROUGH */ 2797 case TCPS_CLOSED: 2798 tp = tcp_close(tp); 2799 /* 2800 * tcp_close() should never return NULL here as the socket is 2801 * still open. 2802 */ 2803 KASSERT(tp != NULL, 2804 ("tcp_usrclosed: tcp_close() returned NULL")); 2805 break; 2806 2807 case TCPS_SYN_SENT: 2808 case TCPS_SYN_RECEIVED: 2809 tp->t_flags |= TF_NEEDFIN; 2810 break; 2811 2812 case TCPS_ESTABLISHED: 2813 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2814 break; 2815 2816 case TCPS_CLOSE_WAIT: 2817 tcp_state_change(tp, TCPS_LAST_ACK); 2818 break; 2819 } 2820 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2821 soisdisconnected(tp->t_inpcb->inp_socket); 2822 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2823 if (tp->t_state == TCPS_FIN_WAIT_2) { 2824 int timeout; 2825 2826 timeout = (tcp_fast_finwait2_recycle) ? 2827 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2828 tcp_timer_activate(tp, TT_2MSL, timeout); 2829 } 2830 } 2831 } 2832 2833 #ifdef DDB 2834 static void 2835 db_print_indent(int indent) 2836 { 2837 int i; 2838 2839 for (i = 0; i < indent; i++) 2840 db_printf(" "); 2841 } 2842 2843 static void 2844 db_print_tstate(int t_state) 2845 { 2846 2847 switch (t_state) { 2848 case TCPS_CLOSED: 2849 db_printf("TCPS_CLOSED"); 2850 return; 2851 2852 case TCPS_LISTEN: 2853 db_printf("TCPS_LISTEN"); 2854 return; 2855 2856 case TCPS_SYN_SENT: 2857 db_printf("TCPS_SYN_SENT"); 2858 return; 2859 2860 case TCPS_SYN_RECEIVED: 2861 db_printf("TCPS_SYN_RECEIVED"); 2862 return; 2863 2864 case TCPS_ESTABLISHED: 2865 db_printf("TCPS_ESTABLISHED"); 2866 return; 2867 2868 case TCPS_CLOSE_WAIT: 2869 db_printf("TCPS_CLOSE_WAIT"); 2870 return; 2871 2872 case TCPS_FIN_WAIT_1: 2873 db_printf("TCPS_FIN_WAIT_1"); 2874 return; 2875 2876 case TCPS_CLOSING: 2877 db_printf("TCPS_CLOSING"); 2878 return; 2879 2880 case TCPS_LAST_ACK: 2881 db_printf("TCPS_LAST_ACK"); 2882 return; 2883 2884 case TCPS_FIN_WAIT_2: 2885 db_printf("TCPS_FIN_WAIT_2"); 2886 return; 2887 2888 case TCPS_TIME_WAIT: 2889 db_printf("TCPS_TIME_WAIT"); 2890 return; 2891 2892 default: 2893 db_printf("unknown"); 2894 return; 2895 } 2896 } 2897 2898 static void 2899 db_print_tflags(u_int t_flags) 2900 { 2901 int comma; 2902 2903 comma = 0; 2904 if (t_flags & TF_ACKNOW) { 2905 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2906 comma = 1; 2907 } 2908 if (t_flags & TF_DELACK) { 2909 db_printf("%sTF_DELACK", comma ? ", " : ""); 2910 comma = 1; 2911 } 2912 if (t_flags & TF_NODELAY) { 2913 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2914 comma = 1; 2915 } 2916 if (t_flags & TF_NOOPT) { 2917 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2918 comma = 1; 2919 } 2920 if (t_flags & TF_SENTFIN) { 2921 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2922 comma = 1; 2923 } 2924 if (t_flags & TF_REQ_SCALE) { 2925 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2926 comma = 1; 2927 } 2928 if (t_flags & TF_RCVD_SCALE) { 2929 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2930 comma = 1; 2931 } 2932 if (t_flags & TF_REQ_TSTMP) { 2933 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2934 comma = 1; 2935 } 2936 if (t_flags & TF_RCVD_TSTMP) { 2937 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2938 comma = 1; 2939 } 2940 if (t_flags & TF_SACK_PERMIT) { 2941 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2942 comma = 1; 2943 } 2944 if (t_flags & TF_NEEDSYN) { 2945 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2946 comma = 1; 2947 } 2948 if (t_flags & TF_NEEDFIN) { 2949 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2950 comma = 1; 2951 } 2952 if (t_flags & TF_NOPUSH) { 2953 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2954 comma = 1; 2955 } 2956 if (t_flags & TF_MORETOCOME) { 2957 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2958 comma = 1; 2959 } 2960 if (t_flags & TF_LQ_OVERFLOW) { 2961 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2962 comma = 1; 2963 } 2964 if (t_flags & TF_LASTIDLE) { 2965 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2966 comma = 1; 2967 } 2968 if (t_flags & TF_RXWIN0SENT) { 2969 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2970 comma = 1; 2971 } 2972 if (t_flags & TF_FASTRECOVERY) { 2973 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2974 comma = 1; 2975 } 2976 if (t_flags & TF_CONGRECOVERY) { 2977 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2978 comma = 1; 2979 } 2980 if (t_flags & TF_WASFRECOVERY) { 2981 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2982 comma = 1; 2983 } 2984 if (t_flags & TF_SIGNATURE) { 2985 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2986 comma = 1; 2987 } 2988 if (t_flags & TF_FORCEDATA) { 2989 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2990 comma = 1; 2991 } 2992 if (t_flags & TF_TSO) { 2993 db_printf("%sTF_TSO", comma ? ", " : ""); 2994 comma = 1; 2995 } 2996 if (t_flags & TF_FASTOPEN) { 2997 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2998 comma = 1; 2999 } 3000 } 3001 3002 static void 3003 db_print_tflags2(u_int t_flags2) 3004 { 3005 int comma; 3006 3007 comma = 0; 3008 if (t_flags2 & TF2_ECN_PERMIT) { 3009 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 3010 comma = 1; 3011 } 3012 } 3013 3014 static void 3015 db_print_toobflags(char t_oobflags) 3016 { 3017 int comma; 3018 3019 comma = 0; 3020 if (t_oobflags & TCPOOB_HAVEDATA) { 3021 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3022 comma = 1; 3023 } 3024 if (t_oobflags & TCPOOB_HADDATA) { 3025 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3026 comma = 1; 3027 } 3028 } 3029 3030 static void 3031 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3032 { 3033 3034 db_print_indent(indent); 3035 db_printf("%s at %p\n", name, tp); 3036 3037 indent += 2; 3038 3039 db_print_indent(indent); 3040 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3041 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3042 3043 db_print_indent(indent); 3044 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 3045 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 3046 3047 db_print_indent(indent); 3048 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 3049 &tp->t_timers->tt_delack, tp->t_inpcb); 3050 3051 db_print_indent(indent); 3052 db_printf("t_state: %d (", tp->t_state); 3053 db_print_tstate(tp->t_state); 3054 db_printf(")\n"); 3055 3056 db_print_indent(indent); 3057 db_printf("t_flags: 0x%x (", tp->t_flags); 3058 db_print_tflags(tp->t_flags); 3059 db_printf(")\n"); 3060 3061 db_print_indent(indent); 3062 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3063 db_print_tflags2(tp->t_flags2); 3064 db_printf(")\n"); 3065 3066 db_print_indent(indent); 3067 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 3068 tp->snd_una, tp->snd_max, tp->snd_nxt); 3069 3070 db_print_indent(indent); 3071 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3072 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3073 3074 db_print_indent(indent); 3075 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3076 tp->iss, tp->irs, tp->rcv_nxt); 3077 3078 db_print_indent(indent); 3079 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3080 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3081 3082 db_print_indent(indent); 3083 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3084 tp->snd_wnd, tp->snd_cwnd); 3085 3086 db_print_indent(indent); 3087 db_printf("snd_ssthresh: %u snd_recover: " 3088 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3089 3090 db_print_indent(indent); 3091 db_printf("t_rcvtime: %u t_startime: %u\n", 3092 tp->t_rcvtime, tp->t_starttime); 3093 3094 db_print_indent(indent); 3095 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3096 tp->t_rtttime, tp->t_rtseq); 3097 3098 db_print_indent(indent); 3099 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3100 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3101 3102 db_print_indent(indent); 3103 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 3104 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 3105 tp->t_rttbest); 3106 3107 db_print_indent(indent); 3108 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 3109 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3110 3111 db_print_indent(indent); 3112 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3113 db_print_toobflags(tp->t_oobflags); 3114 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3115 3116 db_print_indent(indent); 3117 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3118 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3119 3120 db_print_indent(indent); 3121 db_printf("ts_recent: %u ts_recent_age: %u\n", 3122 tp->ts_recent, tp->ts_recent_age); 3123 3124 db_print_indent(indent); 3125 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3126 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3127 3128 db_print_indent(indent); 3129 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3130 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3131 tp->snd_recover_prev, tp->t_badrxtwin); 3132 3133 db_print_indent(indent); 3134 db_printf("snd_numholes: %d snd_holes first: %p\n", 3135 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3136 3137 db_print_indent(indent); 3138 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3139 tp->snd_fack, tp->rcv_numsacks); 3140 3141 /* Skip sackblks, sackhint. */ 3142 3143 db_print_indent(indent); 3144 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3145 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3146 } 3147 3148 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3149 { 3150 struct tcpcb *tp; 3151 3152 if (!have_addr) { 3153 db_printf("usage: show tcpcb <addr>\n"); 3154 return; 3155 } 3156 tp = (struct tcpcb *)addr; 3157 3158 db_print_tcpcb(tp, "tcpcb", 0); 3159 } 3160 #endif 3161