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 = tp->t_fb->tfb_tcp_output(tp); 597 out_in_epoch: 598 NET_EPOCH_EXIT(et); 599 out: 600 TCPDEBUG2(PRU_CONNECT); 601 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 602 INP_WUNLOCK(inp); 603 return (error); 604 } 605 #endif /* INET */ 606 607 #ifdef INET6 608 static int 609 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 610 { 611 struct epoch_tracker et; 612 int error = 0; 613 struct inpcb *inp; 614 struct tcpcb *tp = NULL; 615 struct sockaddr_in6 *sin6; 616 u_int8_t incflagsav; 617 u_char vflagsav; 618 619 TCPDEBUG0; 620 621 sin6 = (struct sockaddr_in6 *)nam; 622 if (nam->sa_family != AF_INET6) 623 return (EAFNOSUPPORT); 624 if (nam->sa_len != sizeof (*sin6)) 625 return (EINVAL); 626 627 /* 628 * Must disallow TCP ``connections'' to multicast addresses. 629 */ 630 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 631 return (EAFNOSUPPORT); 632 633 inp = sotoinpcb(so); 634 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL")); 635 INP_WLOCK(inp); 636 vflagsav = inp->inp_vflag; 637 incflagsav = inp->inp_inc.inc_flags; 638 if (inp->inp_flags & INP_TIMEWAIT) { 639 error = EADDRINUSE; 640 goto out; 641 } 642 if (inp->inp_flags & INP_DROPPED) { 643 error = ECONNREFUSED; 644 goto out; 645 } 646 if (SOLISTENING(so)) { 647 error = EINVAL; 648 goto out; 649 } 650 tp = intotcpcb(inp); 651 TCPDEBUG1(); 652 #ifdef INET 653 /* 654 * XXXRW: Some confusion: V4/V6 flags relate to binding, and 655 * therefore probably require the hash lock, which isn't held here. 656 * Is this a significant problem? 657 */ 658 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { 659 struct sockaddr_in sin; 660 661 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 662 error = EINVAL; 663 goto out; 664 } 665 if ((inp->inp_vflag & INP_IPV4) == 0) { 666 error = EAFNOSUPPORT; 667 goto out; 668 } 669 670 in6_sin6_2_sin(&sin, sin6); 671 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { 672 error = EAFNOSUPPORT; 673 goto out; 674 } 675 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) { 676 error = EACCES; 677 goto out; 678 } 679 if ((error = prison_remote_ip4(td->td_ucred, 680 &sin.sin_addr)) != 0) 681 goto out; 682 inp->inp_vflag |= INP_IPV4; 683 inp->inp_vflag &= ~INP_IPV6; 684 NET_EPOCH_ENTER(et); 685 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0) 686 goto out_in_epoch; 687 #ifdef TCP_OFFLOAD 688 if (registered_toedevs > 0 && 689 (so->so_options & SO_NO_OFFLOAD) == 0 && 690 (error = tcp_offload_connect(so, nam)) == 0) 691 goto out_in_epoch; 692 #endif 693 error = tp->t_fb->tfb_tcp_output(tp); 694 goto out_in_epoch; 695 } else { 696 if ((inp->inp_vflag & INP_IPV6) == 0) { 697 error = EAFNOSUPPORT; 698 goto out; 699 } 700 } 701 #endif 702 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0) 703 goto out; 704 inp->inp_vflag &= ~INP_IPV4; 705 inp->inp_vflag |= INP_IPV6; 706 inp->inp_inc.inc_flags |= INC_ISIPV6; 707 if ((error = tcp6_connect(tp, nam, td)) != 0) 708 goto out; 709 #ifdef TCP_OFFLOAD 710 if (registered_toedevs > 0 && 711 (so->so_options & SO_NO_OFFLOAD) == 0 && 712 (error = tcp_offload_connect(so, nam)) == 0) 713 goto out; 714 #endif 715 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 716 NET_EPOCH_ENTER(et); 717 error = tp->t_fb->tfb_tcp_output(tp); 718 #ifdef INET 719 out_in_epoch: 720 #endif 721 NET_EPOCH_EXIT(et); 722 out: 723 /* 724 * If the implicit bind in the connect call fails, restore 725 * the flags we modified. 726 */ 727 if (error != 0 && inp->inp_lport == 0) { 728 inp->inp_vflag = vflagsav; 729 inp->inp_inc.inc_flags = incflagsav; 730 } 731 732 TCPDEBUG2(PRU_CONNECT); 733 TCP_PROBE2(debug__user, tp, PRU_CONNECT); 734 INP_WUNLOCK(inp); 735 return (error); 736 } 737 #endif /* INET6 */ 738 739 /* 740 * Initiate disconnect from peer. 741 * If connection never passed embryonic stage, just drop; 742 * else if don't need to let data drain, then can just drop anyways, 743 * else have to begin TCP shutdown process: mark socket disconnecting, 744 * drain unread data, state switch to reflect user close, and 745 * send segment (e.g. FIN) to peer. Socket will be really disconnected 746 * when peer sends FIN and acks ours. 747 * 748 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 749 */ 750 static int 751 tcp_usr_disconnect(struct socket *so) 752 { 753 struct inpcb *inp; 754 struct tcpcb *tp = NULL; 755 struct epoch_tracker et; 756 int error = 0; 757 758 TCPDEBUG0; 759 NET_EPOCH_ENTER(et); 760 inp = sotoinpcb(so); 761 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL")); 762 INP_WLOCK(inp); 763 if (inp->inp_flags & INP_TIMEWAIT) 764 goto out; 765 if (inp->inp_flags & INP_DROPPED) { 766 error = ECONNRESET; 767 goto out; 768 } 769 tp = intotcpcb(inp); 770 TCPDEBUG1(); 771 tcp_disconnect(tp); 772 out: 773 TCPDEBUG2(PRU_DISCONNECT); 774 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT); 775 INP_WUNLOCK(inp); 776 NET_EPOCH_EXIT(et); 777 return (error); 778 } 779 780 #ifdef INET 781 /* 782 * Accept a connection. Essentially all the work is done at higher levels; 783 * just return the address of the peer, storing through addr. 784 */ 785 static int 786 tcp_usr_accept(struct socket *so, struct sockaddr **nam) 787 { 788 int error = 0; 789 struct inpcb *inp = NULL; 790 struct tcpcb *tp = NULL; 791 struct in_addr addr; 792 in_port_t port = 0; 793 TCPDEBUG0; 794 795 if (so->so_state & SS_ISDISCONNECTED) 796 return (ECONNABORTED); 797 798 inp = sotoinpcb(so); 799 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL")); 800 INP_WLOCK(inp); 801 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 802 error = ECONNABORTED; 803 goto out; 804 } 805 tp = intotcpcb(inp); 806 TCPDEBUG1(); 807 808 /* 809 * We inline in_getpeeraddr and COMMON_END here, so that we can 810 * copy the data of interest and defer the malloc until after we 811 * release the lock. 812 */ 813 port = inp->inp_fport; 814 addr = inp->inp_faddr; 815 816 out: 817 TCPDEBUG2(PRU_ACCEPT); 818 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 819 INP_WUNLOCK(inp); 820 if (error == 0) 821 *nam = in_sockaddr(port, &addr); 822 return error; 823 } 824 #endif /* INET */ 825 826 #ifdef INET6 827 static int 828 tcp6_usr_accept(struct socket *so, struct sockaddr **nam) 829 { 830 struct inpcb *inp = NULL; 831 int error = 0; 832 struct tcpcb *tp = NULL; 833 struct in_addr addr; 834 struct in6_addr addr6; 835 struct epoch_tracker et; 836 in_port_t port = 0; 837 int v4 = 0; 838 TCPDEBUG0; 839 840 if (so->so_state & SS_ISDISCONNECTED) 841 return (ECONNABORTED); 842 843 inp = sotoinpcb(so); 844 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL")); 845 NET_EPOCH_ENTER(et); 846 INP_WLOCK(inp); 847 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 848 error = ECONNABORTED; 849 goto out; 850 } 851 tp = intotcpcb(inp); 852 TCPDEBUG1(); 853 854 /* 855 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can 856 * copy the data of interest and defer the malloc until after we 857 * release the lock. 858 */ 859 if (inp->inp_vflag & INP_IPV4) { 860 v4 = 1; 861 port = inp->inp_fport; 862 addr = inp->inp_faddr; 863 } else { 864 port = inp->inp_fport; 865 addr6 = inp->in6p_faddr; 866 } 867 868 out: 869 TCPDEBUG2(PRU_ACCEPT); 870 TCP_PROBE2(debug__user, tp, PRU_ACCEPT); 871 INP_WUNLOCK(inp); 872 NET_EPOCH_EXIT(et); 873 if (error == 0) { 874 if (v4) 875 *nam = in6_v4mapsin6_sockaddr(port, &addr); 876 else 877 *nam = in6_sockaddr(port, &addr6); 878 } 879 return error; 880 } 881 #endif /* INET6 */ 882 883 /* 884 * Mark the connection as being incapable of further output. 885 */ 886 static int 887 tcp_usr_shutdown(struct socket *so) 888 { 889 int error = 0; 890 struct inpcb *inp; 891 struct tcpcb *tp = NULL; 892 struct epoch_tracker et; 893 894 TCPDEBUG0; 895 NET_EPOCH_ENTER(et); 896 inp = sotoinpcb(so); 897 KASSERT(inp != NULL, ("inp == NULL")); 898 INP_WLOCK(inp); 899 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 900 error = ECONNRESET; 901 goto out; 902 } 903 tp = intotcpcb(inp); 904 TCPDEBUG1(); 905 socantsendmore(so); 906 tcp_usrclosed(tp); 907 if (!(inp->inp_flags & INP_DROPPED)) 908 error = tp->t_fb->tfb_tcp_output(tp); 909 910 out: 911 TCPDEBUG2(PRU_SHUTDOWN); 912 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN); 913 INP_WUNLOCK(inp); 914 NET_EPOCH_EXIT(et); 915 916 return (error); 917 } 918 919 /* 920 * After a receive, possibly send window update to peer. 921 */ 922 static int 923 tcp_usr_rcvd(struct socket *so, int flags) 924 { 925 struct epoch_tracker et; 926 struct inpcb *inp; 927 struct tcpcb *tp = NULL; 928 int error = 0; 929 930 TCPDEBUG0; 931 inp = sotoinpcb(so); 932 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL")); 933 INP_WLOCK(inp); 934 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 935 error = ECONNRESET; 936 goto out; 937 } 938 tp = intotcpcb(inp); 939 TCPDEBUG1(); 940 /* 941 * For passively-created TFO connections, don't attempt a window 942 * update while still in SYN_RECEIVED as this may trigger an early 943 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with 944 * application response data, or failing that, when the DELACK timer 945 * expires. 946 */ 947 if (IS_FASTOPEN(tp->t_flags) && 948 (tp->t_state == TCPS_SYN_RECEIVED)) 949 goto out; 950 NET_EPOCH_ENTER(et); 951 #ifdef TCP_OFFLOAD 952 if (tp->t_flags & TF_TOE) 953 tcp_offload_rcvd(tp); 954 else 955 #endif 956 tp->t_fb->tfb_tcp_output(tp); 957 NET_EPOCH_EXIT(et); 958 out: 959 TCPDEBUG2(PRU_RCVD); 960 TCP_PROBE2(debug__user, tp, PRU_RCVD); 961 INP_WUNLOCK(inp); 962 return (error); 963 } 964 965 /* 966 * Do a send by putting data in output queue and updating urgent 967 * marker if URG set. Possibly send more data. Unlike the other 968 * pru_*() routines, the mbuf chains are our responsibility. We 969 * must either enqueue them or free them. The other pru_* routines 970 * generally are caller-frees. 971 */ 972 static int 973 tcp_usr_send(struct socket *so, int flags, struct mbuf *m, 974 struct sockaddr *nam, struct mbuf *control, struct thread *td) 975 { 976 struct epoch_tracker et; 977 int error = 0; 978 struct inpcb *inp; 979 struct tcpcb *tp = NULL; 980 #ifdef INET 981 #ifdef INET6 982 struct sockaddr_in sin; 983 #endif 984 struct sockaddr_in *sinp; 985 #endif 986 #ifdef INET6 987 int isipv6; 988 #endif 989 u_int8_t incflagsav; 990 u_char vflagsav; 991 bool restoreflags; 992 TCPDEBUG0; 993 994 /* 995 * We require the pcbinfo "read lock" if we will close the socket 996 * as part of this call. 997 */ 998 NET_EPOCH_ENTER(et); 999 inp = sotoinpcb(so); 1000 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL")); 1001 INP_WLOCK(inp); 1002 vflagsav = inp->inp_vflag; 1003 incflagsav = inp->inp_inc.inc_flags; 1004 restoreflags = false; 1005 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 1006 if (control) 1007 m_freem(control); 1008 error = ECONNRESET; 1009 goto out; 1010 } 1011 if (control != NULL) { 1012 /* TCP doesn't do control messages (rights, creds, etc) */ 1013 if (control->m_len) { 1014 m_freem(control); 1015 error = EINVAL; 1016 goto out; 1017 } 1018 m_freem(control); /* empty control, just free it */ 1019 control = NULL; 1020 } 1021 tp = intotcpcb(inp); 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 = tp->t_fb->tfb_tcp_output(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 = tp->t_fb->tfb_tcp_output(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 INP_WUNLOCK(inp); 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 == 0) { 1314 NET_EPOCH_ENTER(et); 1315 error = tp->t_fb->tfb_tcp_output(tp); 1316 NET_EPOCH_EXIT(et); 1317 } 1318 INP_WUNLOCK(inp); 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, HPTS_REMOVE_ALL); 1845 #endif 1846 if (blk->tfb_tcp_fb_init) { 1847 error = (*blk->tfb_tcp_fb_init)(tp); 1848 if (error) { 1849 refcount_release(&blk->tfb_refcnt); 1850 if (tp->t_fb->tfb_tcp_fb_init) { 1851 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) { 1852 /* Fall back failed, drop the connection */ 1853 INP_WUNLOCK(inp); 1854 soabort(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 int error; 1939 struct inpcb *inp; 1940 1941 error = 0; 1942 inp = sotoinpcb(so); 1943 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL")); 1944 1945 if (sopt->sopt_dir == SOPT_SET) 1946 return (tcp_ctloutput_set(inp, sopt)); 1947 else if (sopt->sopt_dir == SOPT_GET) 1948 return (tcp_ctloutput_get(inp, sopt)); 1949 else 1950 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir); 1951 } 1952 1953 /* 1954 * If this assert becomes untrue, we need to change the size of the buf 1955 * variable in tcp_default_ctloutput(). 1956 */ 1957 #ifdef CTASSERT 1958 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN); 1959 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN); 1960 #endif 1961 1962 #ifdef KERN_TLS 1963 static int 1964 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls) 1965 { 1966 struct tls_enable_v0 tls_v0; 1967 int error; 1968 1969 if (sopt->sopt_valsize == sizeof(tls_v0)) { 1970 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0), 1971 sizeof(tls_v0)); 1972 if (error) 1973 return (error); 1974 memset(tls, 0, sizeof(*tls)); 1975 tls->cipher_key = tls_v0.cipher_key; 1976 tls->iv = tls_v0.iv; 1977 tls->auth_key = tls_v0.auth_key; 1978 tls->cipher_algorithm = tls_v0.cipher_algorithm; 1979 tls->cipher_key_len = tls_v0.cipher_key_len; 1980 tls->iv_len = tls_v0.iv_len; 1981 tls->auth_algorithm = tls_v0.auth_algorithm; 1982 tls->auth_key_len = tls_v0.auth_key_len; 1983 tls->flags = tls_v0.flags; 1984 tls->tls_vmajor = tls_v0.tls_vmajor; 1985 tls->tls_vminor = tls_v0.tls_vminor; 1986 return (0); 1987 } 1988 1989 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls))); 1990 } 1991 #endif 1992 1993 extern struct cc_algo newreno_cc_algo; 1994 1995 static int 1996 tcp_congestion(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 1997 { 1998 struct cc_algo *algo; 1999 void *ptr = NULL; 2000 struct cc_var cc_mem; 2001 char buf[TCP_CA_NAME_MAX]; 2002 size_t mem_sz; 2003 int error; 2004 2005 INP_WUNLOCK(inp); 2006 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1); 2007 if (error) 2008 return(error); 2009 buf[sopt->sopt_valsize] = '\0'; 2010 CC_LIST_RLOCK(); 2011 STAILQ_FOREACH(algo, &cc_list, entries) 2012 if (strncmp(buf, algo->name, 2013 TCP_CA_NAME_MAX) == 0) { 2014 if (algo->flags & CC_MODULE_BEING_REMOVED) { 2015 /* We can't "see" modules being unloaded */ 2016 continue; 2017 } 2018 break; 2019 } 2020 if (algo == NULL) { 2021 CC_LIST_RUNLOCK(); 2022 return(ESRCH); 2023 } 2024 do_over: 2025 if (algo->cb_init != NULL) { 2026 /* We can now pre-get the memory for the CC */ 2027 mem_sz = (*algo->cc_data_sz)(); 2028 if (mem_sz == 0) { 2029 goto no_mem_needed; 2030 } 2031 CC_LIST_RUNLOCK(); 2032 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK); 2033 CC_LIST_RLOCK(); 2034 STAILQ_FOREACH(algo, &cc_list, entries) 2035 if (strncmp(buf, algo->name, 2036 TCP_CA_NAME_MAX) == 0) 2037 break; 2038 if (algo == NULL) { 2039 if (ptr) 2040 free(ptr, M_CC_MEM); 2041 CC_LIST_RUNLOCK(); 2042 return(ESRCH); 2043 } 2044 } else { 2045 no_mem_needed: 2046 mem_sz = 0; 2047 ptr = NULL; 2048 } 2049 /* 2050 * Make sure its all clean and zero and also get 2051 * back the inplock. 2052 */ 2053 memset(&cc_mem, 0, sizeof(cc_mem)); 2054 if (mem_sz != (*algo->cc_data_sz)()) { 2055 if (ptr) 2056 free(ptr, M_CC_MEM); 2057 goto do_over; 2058 } 2059 INP_WLOCK(inp); 2060 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 2061 INP_WUNLOCK(inp); 2062 CC_LIST_RUNLOCK(); 2063 free(ptr, M_CC_MEM); 2064 return (ECONNRESET); 2065 } 2066 tp = intotcpcb(inp); 2067 if (ptr != NULL) 2068 memset(ptr, 0, mem_sz); 2069 CC_LIST_RUNLOCK(); 2070 cc_mem.ccvc.tcp = tp; 2071 /* 2072 * We once again hold a write lock over the tcb so it's 2073 * safe to do these things without ordering concerns. 2074 * Note here we init into stack memory. 2075 */ 2076 if (algo->cb_init != NULL) 2077 error = algo->cb_init(&cc_mem, ptr); 2078 else 2079 error = 0; 2080 /* 2081 * The CC algorithms, when given their memory 2082 * should not fail we could in theory have a 2083 * KASSERT here. 2084 */ 2085 if (error == 0) { 2086 /* 2087 * Touchdown, lets go ahead and move the 2088 * connection to the new CC module by 2089 * copying in the cc_mem after we call 2090 * the old ones cleanup (if any). 2091 */ 2092 if (CC_ALGO(tp)->cb_destroy != NULL) 2093 CC_ALGO(tp)->cb_destroy(tp->ccv); 2094 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2095 tp->cc_algo = algo; 2096 /* Ok now are we where we have gotten past any conn_init? */ 2097 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) { 2098 /* Yep run the connection init for the new CC */ 2099 CC_ALGO(tp)->conn_init(tp->ccv); 2100 } 2101 } else if (ptr) 2102 free(ptr, M_CC_MEM); 2103 INP_WUNLOCK(inp); 2104 return (error); 2105 } 2106 2107 int 2108 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) 2109 { 2110 int error, opt, optval; 2111 u_int ui; 2112 struct tcp_info ti; 2113 #ifdef KERN_TLS 2114 struct tls_enable tls; 2115 #endif 2116 char *pbuf, buf[TCP_LOG_ID_LEN]; 2117 #ifdef STATS 2118 struct statsblob *sbp; 2119 #endif 2120 size_t len; 2121 2122 INP_WLOCK_ASSERT(inp); 2123 2124 switch (sopt->sopt_level) { 2125 #ifdef INET6 2126 case IPPROTO_IPV6: 2127 MPASS(inp->inp_vflag & INP_IPV6PROTO); 2128 switch (sopt->sopt_name) { 2129 case IPV6_USE_MIN_MTU: 2130 tcp6_use_min_mtu(tp); 2131 /* FALLTHROUGH */ 2132 } 2133 INP_WUNLOCK(inp); 2134 return (0); 2135 #endif 2136 #ifdef INET 2137 case IPPROTO_IP: 2138 INP_WUNLOCK(inp); 2139 return (0); 2140 #endif 2141 } 2142 2143 /* 2144 * For TCP_CCALGOOPT forward the control to CC module, for both 2145 * SOPT_SET and SOPT_GET. 2146 */ 2147 switch (sopt->sopt_name) { 2148 case TCP_CCALGOOPT: 2149 INP_WUNLOCK(inp); 2150 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT) 2151 return (EINVAL); 2152 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO); 2153 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize, 2154 sopt->sopt_valsize); 2155 if (error) { 2156 free(pbuf, M_TEMP); 2157 return (error); 2158 } 2159 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP)); 2160 if (CC_ALGO(tp)->ctl_output != NULL) 2161 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf); 2162 else 2163 error = ENOENT; 2164 INP_WUNLOCK(inp); 2165 if (error == 0 && sopt->sopt_dir == SOPT_GET) 2166 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize); 2167 free(pbuf, M_TEMP); 2168 return (error); 2169 } 2170 2171 switch (sopt->sopt_dir) { 2172 case SOPT_SET: 2173 switch (sopt->sopt_name) { 2174 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2175 case TCP_MD5SIG: 2176 if (!TCPMD5_ENABLED()) { 2177 INP_WUNLOCK(inp); 2178 return (ENOPROTOOPT); 2179 } 2180 error = TCPMD5_PCBCTL(inp, sopt); 2181 if (error) 2182 return (error); 2183 goto unlock_and_done; 2184 #endif /* IPSEC */ 2185 2186 case TCP_NODELAY: 2187 case TCP_NOOPT: 2188 case TCP_LRD: 2189 INP_WUNLOCK(inp); 2190 error = sooptcopyin(sopt, &optval, sizeof optval, 2191 sizeof optval); 2192 if (error) 2193 return (error); 2194 2195 INP_WLOCK_RECHECK(inp); 2196 switch (sopt->sopt_name) { 2197 case TCP_NODELAY: 2198 opt = TF_NODELAY; 2199 break; 2200 case TCP_NOOPT: 2201 opt = TF_NOOPT; 2202 break; 2203 case TCP_LRD: 2204 opt = TF_LRD; 2205 break; 2206 default: 2207 opt = 0; /* dead code to fool gcc */ 2208 break; 2209 } 2210 2211 if (optval) 2212 tp->t_flags |= opt; 2213 else 2214 tp->t_flags &= ~opt; 2215 unlock_and_done: 2216 #ifdef TCP_OFFLOAD 2217 if (tp->t_flags & TF_TOE) { 2218 tcp_offload_ctloutput(tp, sopt->sopt_dir, 2219 sopt->sopt_name); 2220 } 2221 #endif 2222 INP_WUNLOCK(inp); 2223 break; 2224 2225 case TCP_NOPUSH: 2226 INP_WUNLOCK(inp); 2227 error = sooptcopyin(sopt, &optval, sizeof optval, 2228 sizeof optval); 2229 if (error) 2230 return (error); 2231 2232 INP_WLOCK_RECHECK(inp); 2233 if (optval) 2234 tp->t_flags |= TF_NOPUSH; 2235 else if (tp->t_flags & TF_NOPUSH) { 2236 tp->t_flags &= ~TF_NOPUSH; 2237 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2238 struct epoch_tracker et; 2239 2240 NET_EPOCH_ENTER(et); 2241 error = tp->t_fb->tfb_tcp_output(tp); 2242 NET_EPOCH_EXIT(et); 2243 } 2244 } 2245 goto unlock_and_done; 2246 2247 case TCP_REMOTE_UDP_ENCAPS_PORT: 2248 INP_WUNLOCK(inp); 2249 error = sooptcopyin(sopt, &optval, sizeof optval, 2250 sizeof optval); 2251 if (error) 2252 return (error); 2253 if ((optval < TCP_TUNNELING_PORT_MIN) || 2254 (optval > TCP_TUNNELING_PORT_MAX)) { 2255 /* Its got to be in range */ 2256 return (EINVAL); 2257 } 2258 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) { 2259 /* You have to have enabled a UDP tunneling port first */ 2260 return (EINVAL); 2261 } 2262 INP_WLOCK_RECHECK(inp); 2263 if (tp->t_state != TCPS_CLOSED) { 2264 /* You can't change after you are connected */ 2265 error = EINVAL; 2266 } else { 2267 /* Ok we are all good set the port */ 2268 tp->t_port = htons(optval); 2269 } 2270 goto unlock_and_done; 2271 2272 case TCP_MAXSEG: 2273 INP_WUNLOCK(inp); 2274 error = sooptcopyin(sopt, &optval, sizeof optval, 2275 sizeof optval); 2276 if (error) 2277 return (error); 2278 2279 INP_WLOCK_RECHECK(inp); 2280 if (optval > 0 && optval <= tp->t_maxseg && 2281 optval + 40 >= V_tcp_minmss) 2282 tp->t_maxseg = optval; 2283 else 2284 error = EINVAL; 2285 goto unlock_and_done; 2286 2287 case TCP_INFO: 2288 INP_WUNLOCK(inp); 2289 error = EINVAL; 2290 break; 2291 2292 case TCP_STATS: 2293 INP_WUNLOCK(inp); 2294 #ifdef STATS 2295 error = sooptcopyin(sopt, &optval, sizeof optval, 2296 sizeof optval); 2297 if (error) 2298 return (error); 2299 2300 if (optval > 0) 2301 sbp = stats_blob_alloc( 2302 V_tcp_perconn_stats_dflt_tpl, 0); 2303 else 2304 sbp = NULL; 2305 2306 INP_WLOCK_RECHECK(inp); 2307 if ((tp->t_stats != NULL && sbp == NULL) || 2308 (tp->t_stats == NULL && sbp != NULL)) { 2309 struct statsblob *t = tp->t_stats; 2310 tp->t_stats = sbp; 2311 sbp = t; 2312 } 2313 INP_WUNLOCK(inp); 2314 2315 stats_blob_destroy(sbp); 2316 #else 2317 return (EOPNOTSUPP); 2318 #endif /* !STATS */ 2319 break; 2320 2321 case TCP_CONGESTION: 2322 error = tcp_congestion(so, sopt, inp, tp); 2323 break; 2324 2325 case TCP_REUSPORT_LB_NUMA: 2326 INP_WUNLOCK(inp); 2327 error = sooptcopyin(sopt, &optval, sizeof(optval), 2328 sizeof(optval)); 2329 INP_WLOCK_RECHECK(inp); 2330 if (!error) 2331 error = in_pcblbgroup_numa(inp, optval); 2332 INP_WUNLOCK(inp); 2333 break; 2334 2335 #ifdef KERN_TLS 2336 case TCP_TXTLS_ENABLE: 2337 INP_WUNLOCK(inp); 2338 error = copyin_tls_enable(sopt, &tls); 2339 if (error) 2340 break; 2341 error = ktls_enable_tx(so, &tls); 2342 break; 2343 case TCP_TXTLS_MODE: 2344 INP_WUNLOCK(inp); 2345 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2346 if (error) 2347 return (error); 2348 2349 INP_WLOCK_RECHECK(inp); 2350 error = ktls_set_tx_mode(so, ui); 2351 INP_WUNLOCK(inp); 2352 break; 2353 case TCP_RXTLS_ENABLE: 2354 INP_WUNLOCK(inp); 2355 error = sooptcopyin(sopt, &tls, sizeof(tls), 2356 sizeof(tls)); 2357 if (error) 2358 break; 2359 error = ktls_enable_rx(so, &tls); 2360 break; 2361 #endif 2362 2363 case TCP_KEEPIDLE: 2364 case TCP_KEEPINTVL: 2365 case TCP_KEEPINIT: 2366 INP_WUNLOCK(inp); 2367 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2368 if (error) 2369 return (error); 2370 2371 if (ui > (UINT_MAX / hz)) { 2372 error = EINVAL; 2373 break; 2374 } 2375 ui *= hz; 2376 2377 INP_WLOCK_RECHECK(inp); 2378 switch (sopt->sopt_name) { 2379 case TCP_KEEPIDLE: 2380 tp->t_keepidle = ui; 2381 /* 2382 * XXX: better check current remaining 2383 * timeout and "merge" it with new value. 2384 */ 2385 if ((tp->t_state > TCPS_LISTEN) && 2386 (tp->t_state <= TCPS_CLOSING)) 2387 tcp_timer_activate(tp, TT_KEEP, 2388 TP_KEEPIDLE(tp)); 2389 break; 2390 case TCP_KEEPINTVL: 2391 tp->t_keepintvl = ui; 2392 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2393 (TP_MAXIDLE(tp) > 0)) 2394 tcp_timer_activate(tp, TT_2MSL, 2395 TP_MAXIDLE(tp)); 2396 break; 2397 case TCP_KEEPINIT: 2398 tp->t_keepinit = ui; 2399 if (tp->t_state == TCPS_SYN_RECEIVED || 2400 tp->t_state == TCPS_SYN_SENT) 2401 tcp_timer_activate(tp, TT_KEEP, 2402 TP_KEEPINIT(tp)); 2403 break; 2404 } 2405 goto unlock_and_done; 2406 2407 case TCP_KEEPCNT: 2408 INP_WUNLOCK(inp); 2409 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui)); 2410 if (error) 2411 return (error); 2412 2413 INP_WLOCK_RECHECK(inp); 2414 tp->t_keepcnt = ui; 2415 if ((tp->t_state == TCPS_FIN_WAIT_2) && 2416 (TP_MAXIDLE(tp) > 0)) 2417 tcp_timer_activate(tp, TT_2MSL, 2418 TP_MAXIDLE(tp)); 2419 goto unlock_and_done; 2420 2421 #ifdef TCPPCAP 2422 case TCP_PCAP_OUT: 2423 case TCP_PCAP_IN: 2424 INP_WUNLOCK(inp); 2425 error = sooptcopyin(sopt, &optval, sizeof optval, 2426 sizeof optval); 2427 if (error) 2428 return (error); 2429 2430 INP_WLOCK_RECHECK(inp); 2431 if (optval >= 0) 2432 tcp_pcap_set_sock_max(TCP_PCAP_OUT ? 2433 &(tp->t_outpkts) : &(tp->t_inpkts), 2434 optval); 2435 else 2436 error = EINVAL; 2437 goto unlock_and_done; 2438 #endif 2439 2440 case TCP_FASTOPEN: { 2441 struct tcp_fastopen tfo_optval; 2442 2443 INP_WUNLOCK(inp); 2444 if (!V_tcp_fastopen_client_enable && 2445 !V_tcp_fastopen_server_enable) 2446 return (EPERM); 2447 2448 error = sooptcopyin(sopt, &tfo_optval, 2449 sizeof(tfo_optval), sizeof(int)); 2450 if (error) 2451 return (error); 2452 2453 INP_WLOCK_RECHECK(inp); 2454 if ((tp->t_state != TCPS_CLOSED) && 2455 (tp->t_state != TCPS_LISTEN)) { 2456 error = EINVAL; 2457 goto unlock_and_done; 2458 } 2459 if (tfo_optval.enable) { 2460 if (tp->t_state == TCPS_LISTEN) { 2461 if (!V_tcp_fastopen_server_enable) { 2462 error = EPERM; 2463 goto unlock_and_done; 2464 } 2465 2466 if (tp->t_tfo_pending == NULL) 2467 tp->t_tfo_pending = 2468 tcp_fastopen_alloc_counter(); 2469 } else { 2470 /* 2471 * If a pre-shared key was provided, 2472 * stash it in the client cookie 2473 * field of the tcpcb for use during 2474 * connect. 2475 */ 2476 if (sopt->sopt_valsize == 2477 sizeof(tfo_optval)) { 2478 memcpy(tp->t_tfo_cookie.client, 2479 tfo_optval.psk, 2480 TCP_FASTOPEN_PSK_LEN); 2481 tp->t_tfo_client_cookie_len = 2482 TCP_FASTOPEN_PSK_LEN; 2483 } 2484 } 2485 tp->t_flags |= TF_FASTOPEN; 2486 } else 2487 tp->t_flags &= ~TF_FASTOPEN; 2488 goto unlock_and_done; 2489 } 2490 2491 #ifdef TCP_BLACKBOX 2492 case TCP_LOG: 2493 INP_WUNLOCK(inp); 2494 error = sooptcopyin(sopt, &optval, sizeof optval, 2495 sizeof optval); 2496 if (error) 2497 return (error); 2498 2499 INP_WLOCK_RECHECK(inp); 2500 error = tcp_log_state_change(tp, optval); 2501 goto unlock_and_done; 2502 2503 case TCP_LOGBUF: 2504 INP_WUNLOCK(inp); 2505 error = EINVAL; 2506 break; 2507 2508 case TCP_LOGID: 2509 INP_WUNLOCK(inp); 2510 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0); 2511 if (error) 2512 break; 2513 buf[sopt->sopt_valsize] = '\0'; 2514 INP_WLOCK_RECHECK(inp); 2515 error = tcp_log_set_id(tp, buf); 2516 /* tcp_log_set_id() unlocks the INP. */ 2517 break; 2518 2519 case TCP_LOGDUMP: 2520 case TCP_LOGDUMPID: 2521 INP_WUNLOCK(inp); 2522 error = 2523 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0); 2524 if (error) 2525 break; 2526 buf[sopt->sopt_valsize] = '\0'; 2527 INP_WLOCK_RECHECK(inp); 2528 if (sopt->sopt_name == TCP_LOGDUMP) { 2529 error = tcp_log_dump_tp_logbuf(tp, buf, 2530 M_WAITOK, true); 2531 INP_WUNLOCK(inp); 2532 } else { 2533 tcp_log_dump_tp_bucket_logbufs(tp, buf); 2534 /* 2535 * tcp_log_dump_tp_bucket_logbufs() drops the 2536 * INP lock. 2537 */ 2538 } 2539 break; 2540 #endif 2541 2542 default: 2543 INP_WUNLOCK(inp); 2544 error = ENOPROTOOPT; 2545 break; 2546 } 2547 break; 2548 2549 case SOPT_GET: 2550 tp = intotcpcb(inp); 2551 switch (sopt->sopt_name) { 2552 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2553 case TCP_MD5SIG: 2554 if (!TCPMD5_ENABLED()) { 2555 INP_WUNLOCK(inp); 2556 return (ENOPROTOOPT); 2557 } 2558 error = TCPMD5_PCBCTL(inp, sopt); 2559 break; 2560 #endif 2561 2562 case TCP_NODELAY: 2563 optval = tp->t_flags & TF_NODELAY; 2564 INP_WUNLOCK(inp); 2565 error = sooptcopyout(sopt, &optval, sizeof optval); 2566 break; 2567 case TCP_MAXSEG: 2568 optval = tp->t_maxseg; 2569 INP_WUNLOCK(inp); 2570 error = sooptcopyout(sopt, &optval, sizeof optval); 2571 break; 2572 case TCP_REMOTE_UDP_ENCAPS_PORT: 2573 optval = ntohs(tp->t_port); 2574 INP_WUNLOCK(inp); 2575 error = sooptcopyout(sopt, &optval, sizeof optval); 2576 break; 2577 case TCP_NOOPT: 2578 optval = tp->t_flags & TF_NOOPT; 2579 INP_WUNLOCK(inp); 2580 error = sooptcopyout(sopt, &optval, sizeof optval); 2581 break; 2582 case TCP_NOPUSH: 2583 optval = tp->t_flags & TF_NOPUSH; 2584 INP_WUNLOCK(inp); 2585 error = sooptcopyout(sopt, &optval, sizeof optval); 2586 break; 2587 case TCP_INFO: 2588 tcp_fill_info(tp, &ti); 2589 INP_WUNLOCK(inp); 2590 error = sooptcopyout(sopt, &ti, sizeof ti); 2591 break; 2592 case TCP_STATS: 2593 { 2594 #ifdef STATS 2595 int nheld; 2596 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0; 2597 2598 error = 0; 2599 socklen_t outsbsz = sopt->sopt_valsize; 2600 if (tp->t_stats == NULL) 2601 error = ENOENT; 2602 else if (outsbsz >= tp->t_stats->cursz) 2603 outsbsz = tp->t_stats->cursz; 2604 else if (outsbsz >= sizeof(struct statsblob)) 2605 outsbsz = sizeof(struct statsblob); 2606 else 2607 error = EINVAL; 2608 INP_WUNLOCK(inp); 2609 if (error) 2610 break; 2611 2612 sbp = sopt->sopt_val; 2613 nheld = atop(round_page(((vm_offset_t)sbp) + 2614 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp)); 2615 vm_page_t ma[nheld]; 2616 if (vm_fault_quick_hold_pages( 2617 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp, 2618 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma, 2619 nheld) < 0) { 2620 error = EFAULT; 2621 break; 2622 } 2623 2624 if ((error = copyin_nofault(&(sbp->flags), &sbflags, 2625 SIZEOF_MEMBER(struct statsblob, flags)))) 2626 goto unhold; 2627 2628 INP_WLOCK_RECHECK(inp); 2629 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats, 2630 sbflags | SB_CLONE_USRDSTNOFAULT); 2631 INP_WUNLOCK(inp); 2632 sopt->sopt_valsize = outsbsz; 2633 unhold: 2634 vm_page_unhold_pages(ma, nheld); 2635 #else 2636 INP_WUNLOCK(inp); 2637 error = EOPNOTSUPP; 2638 #endif /* !STATS */ 2639 break; 2640 } 2641 case TCP_CONGESTION: 2642 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX); 2643 INP_WUNLOCK(inp); 2644 error = sooptcopyout(sopt, buf, len + 1); 2645 break; 2646 case TCP_KEEPIDLE: 2647 case TCP_KEEPINTVL: 2648 case TCP_KEEPINIT: 2649 case TCP_KEEPCNT: 2650 switch (sopt->sopt_name) { 2651 case TCP_KEEPIDLE: 2652 ui = TP_KEEPIDLE(tp) / hz; 2653 break; 2654 case TCP_KEEPINTVL: 2655 ui = TP_KEEPINTVL(tp) / hz; 2656 break; 2657 case TCP_KEEPINIT: 2658 ui = TP_KEEPINIT(tp) / hz; 2659 break; 2660 case TCP_KEEPCNT: 2661 ui = TP_KEEPCNT(tp); 2662 break; 2663 } 2664 INP_WUNLOCK(inp); 2665 error = sooptcopyout(sopt, &ui, sizeof(ui)); 2666 break; 2667 #ifdef TCPPCAP 2668 case TCP_PCAP_OUT: 2669 case TCP_PCAP_IN: 2670 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ? 2671 &(tp->t_outpkts) : &(tp->t_inpkts)); 2672 INP_WUNLOCK(inp); 2673 error = sooptcopyout(sopt, &optval, sizeof optval); 2674 break; 2675 #endif 2676 case TCP_FASTOPEN: 2677 optval = tp->t_flags & TF_FASTOPEN; 2678 INP_WUNLOCK(inp); 2679 error = sooptcopyout(sopt, &optval, sizeof optval); 2680 break; 2681 #ifdef TCP_BLACKBOX 2682 case TCP_LOG: 2683 optval = tp->t_logstate; 2684 INP_WUNLOCK(inp); 2685 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2686 break; 2687 case TCP_LOGBUF: 2688 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */ 2689 error = tcp_log_getlogbuf(sopt, tp); 2690 break; 2691 case TCP_LOGID: 2692 len = tcp_log_get_id(tp, buf); 2693 INP_WUNLOCK(inp); 2694 error = sooptcopyout(sopt, buf, len + 1); 2695 break; 2696 case TCP_LOGDUMP: 2697 case TCP_LOGDUMPID: 2698 INP_WUNLOCK(inp); 2699 error = EINVAL; 2700 break; 2701 #endif 2702 #ifdef KERN_TLS 2703 case TCP_TXTLS_MODE: 2704 error = ktls_get_tx_mode(so, &optval); 2705 INP_WUNLOCK(inp); 2706 if (error == 0) 2707 error = sooptcopyout(sopt, &optval, 2708 sizeof(optval)); 2709 break; 2710 case TCP_RXTLS_MODE: 2711 error = ktls_get_rx_mode(so, &optval); 2712 INP_WUNLOCK(inp); 2713 if (error == 0) 2714 error = sooptcopyout(sopt, &optval, 2715 sizeof(optval)); 2716 break; 2717 #endif 2718 case TCP_LRD: 2719 optval = tp->t_flags & TF_LRD; 2720 INP_WUNLOCK(inp); 2721 error = sooptcopyout(sopt, &optval, sizeof optval); 2722 break; 2723 default: 2724 INP_WUNLOCK(inp); 2725 error = ENOPROTOOPT; 2726 break; 2727 } 2728 break; 2729 } 2730 return (error); 2731 } 2732 #undef INP_WLOCK_RECHECK 2733 #undef INP_WLOCK_RECHECK_CLEANUP 2734 2735 /* 2736 * Initiate (or continue) disconnect. 2737 * If embryonic state, just send reset (once). 2738 * If in ``let data drain'' option and linger null, just drop. 2739 * Otherwise (hard), mark socket disconnecting and drop 2740 * current input data; switch states based on user close, and 2741 * send segment to peer (with FIN). 2742 */ 2743 static void 2744 tcp_disconnect(struct tcpcb *tp) 2745 { 2746 struct inpcb *inp = tp->t_inpcb; 2747 struct socket *so = inp->inp_socket; 2748 2749 NET_EPOCH_ASSERT(); 2750 INP_WLOCK_ASSERT(inp); 2751 2752 /* 2753 * Neither tcp_close() nor tcp_drop() should return NULL, as the 2754 * socket is still open. 2755 */ 2756 if (tp->t_state < TCPS_ESTABLISHED && 2757 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) { 2758 tp = tcp_close(tp); 2759 KASSERT(tp != NULL, 2760 ("tcp_disconnect: tcp_close() returned NULL")); 2761 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 2762 tp = tcp_drop(tp, 0); 2763 KASSERT(tp != NULL, 2764 ("tcp_disconnect: tcp_drop() returned NULL")); 2765 } else { 2766 soisdisconnecting(so); 2767 sbflush(&so->so_rcv); 2768 tcp_usrclosed(tp); 2769 if (!(inp->inp_flags & INP_DROPPED)) 2770 tp->t_fb->tfb_tcp_output(tp); 2771 } 2772 } 2773 2774 /* 2775 * User issued close, and wish to trail through shutdown states: 2776 * if never received SYN, just forget it. If got a SYN from peer, 2777 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 2778 * If already got a FIN from peer, then almost done; go to LAST_ACK 2779 * state. In all other cases, have already sent FIN to peer (e.g. 2780 * after PRU_SHUTDOWN), and just have to play tedious game waiting 2781 * for peer to send FIN or not respond to keep-alives, etc. 2782 * We can let the user exit from the close as soon as the FIN is acked. 2783 */ 2784 static void 2785 tcp_usrclosed(struct tcpcb *tp) 2786 { 2787 2788 NET_EPOCH_ASSERT(); 2789 INP_WLOCK_ASSERT(tp->t_inpcb); 2790 2791 switch (tp->t_state) { 2792 case TCPS_LISTEN: 2793 #ifdef TCP_OFFLOAD 2794 tcp_offload_listen_stop(tp); 2795 #endif 2796 tcp_state_change(tp, TCPS_CLOSED); 2797 /* FALLTHROUGH */ 2798 case TCPS_CLOSED: 2799 tp = tcp_close(tp); 2800 /* 2801 * tcp_close() should never return NULL here as the socket is 2802 * still open. 2803 */ 2804 KASSERT(tp != NULL, 2805 ("tcp_usrclosed: tcp_close() returned NULL")); 2806 break; 2807 2808 case TCPS_SYN_SENT: 2809 case TCPS_SYN_RECEIVED: 2810 tp->t_flags |= TF_NEEDFIN; 2811 break; 2812 2813 case TCPS_ESTABLISHED: 2814 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2815 break; 2816 2817 case TCPS_CLOSE_WAIT: 2818 tcp_state_change(tp, TCPS_LAST_ACK); 2819 break; 2820 } 2821 if (tp->t_state >= TCPS_FIN_WAIT_2) { 2822 soisdisconnected(tp->t_inpcb->inp_socket); 2823 /* Prevent the connection hanging in FIN_WAIT_2 forever. */ 2824 if (tp->t_state == TCPS_FIN_WAIT_2) { 2825 int timeout; 2826 2827 timeout = (tcp_fast_finwait2_recycle) ? 2828 tcp_finwait2_timeout : TP_MAXIDLE(tp); 2829 tcp_timer_activate(tp, TT_2MSL, timeout); 2830 } 2831 } 2832 } 2833 2834 #ifdef DDB 2835 static void 2836 db_print_indent(int indent) 2837 { 2838 int i; 2839 2840 for (i = 0; i < indent; i++) 2841 db_printf(" "); 2842 } 2843 2844 static void 2845 db_print_tstate(int t_state) 2846 { 2847 2848 switch (t_state) { 2849 case TCPS_CLOSED: 2850 db_printf("TCPS_CLOSED"); 2851 return; 2852 2853 case TCPS_LISTEN: 2854 db_printf("TCPS_LISTEN"); 2855 return; 2856 2857 case TCPS_SYN_SENT: 2858 db_printf("TCPS_SYN_SENT"); 2859 return; 2860 2861 case TCPS_SYN_RECEIVED: 2862 db_printf("TCPS_SYN_RECEIVED"); 2863 return; 2864 2865 case TCPS_ESTABLISHED: 2866 db_printf("TCPS_ESTABLISHED"); 2867 return; 2868 2869 case TCPS_CLOSE_WAIT: 2870 db_printf("TCPS_CLOSE_WAIT"); 2871 return; 2872 2873 case TCPS_FIN_WAIT_1: 2874 db_printf("TCPS_FIN_WAIT_1"); 2875 return; 2876 2877 case TCPS_CLOSING: 2878 db_printf("TCPS_CLOSING"); 2879 return; 2880 2881 case TCPS_LAST_ACK: 2882 db_printf("TCPS_LAST_ACK"); 2883 return; 2884 2885 case TCPS_FIN_WAIT_2: 2886 db_printf("TCPS_FIN_WAIT_2"); 2887 return; 2888 2889 case TCPS_TIME_WAIT: 2890 db_printf("TCPS_TIME_WAIT"); 2891 return; 2892 2893 default: 2894 db_printf("unknown"); 2895 return; 2896 } 2897 } 2898 2899 static void 2900 db_print_tflags(u_int t_flags) 2901 { 2902 int comma; 2903 2904 comma = 0; 2905 if (t_flags & TF_ACKNOW) { 2906 db_printf("%sTF_ACKNOW", comma ? ", " : ""); 2907 comma = 1; 2908 } 2909 if (t_flags & TF_DELACK) { 2910 db_printf("%sTF_DELACK", comma ? ", " : ""); 2911 comma = 1; 2912 } 2913 if (t_flags & TF_NODELAY) { 2914 db_printf("%sTF_NODELAY", comma ? ", " : ""); 2915 comma = 1; 2916 } 2917 if (t_flags & TF_NOOPT) { 2918 db_printf("%sTF_NOOPT", comma ? ", " : ""); 2919 comma = 1; 2920 } 2921 if (t_flags & TF_SENTFIN) { 2922 db_printf("%sTF_SENTFIN", comma ? ", " : ""); 2923 comma = 1; 2924 } 2925 if (t_flags & TF_REQ_SCALE) { 2926 db_printf("%sTF_REQ_SCALE", comma ? ", " : ""); 2927 comma = 1; 2928 } 2929 if (t_flags & TF_RCVD_SCALE) { 2930 db_printf("%sTF_RECVD_SCALE", comma ? ", " : ""); 2931 comma = 1; 2932 } 2933 if (t_flags & TF_REQ_TSTMP) { 2934 db_printf("%sTF_REQ_TSTMP", comma ? ", " : ""); 2935 comma = 1; 2936 } 2937 if (t_flags & TF_RCVD_TSTMP) { 2938 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : ""); 2939 comma = 1; 2940 } 2941 if (t_flags & TF_SACK_PERMIT) { 2942 db_printf("%sTF_SACK_PERMIT", comma ? ", " : ""); 2943 comma = 1; 2944 } 2945 if (t_flags & TF_NEEDSYN) { 2946 db_printf("%sTF_NEEDSYN", comma ? ", " : ""); 2947 comma = 1; 2948 } 2949 if (t_flags & TF_NEEDFIN) { 2950 db_printf("%sTF_NEEDFIN", comma ? ", " : ""); 2951 comma = 1; 2952 } 2953 if (t_flags & TF_NOPUSH) { 2954 db_printf("%sTF_NOPUSH", comma ? ", " : ""); 2955 comma = 1; 2956 } 2957 if (t_flags & TF_MORETOCOME) { 2958 db_printf("%sTF_MORETOCOME", comma ? ", " : ""); 2959 comma = 1; 2960 } 2961 if (t_flags & TF_LQ_OVERFLOW) { 2962 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : ""); 2963 comma = 1; 2964 } 2965 if (t_flags & TF_LASTIDLE) { 2966 db_printf("%sTF_LASTIDLE", comma ? ", " : ""); 2967 comma = 1; 2968 } 2969 if (t_flags & TF_RXWIN0SENT) { 2970 db_printf("%sTF_RXWIN0SENT", comma ? ", " : ""); 2971 comma = 1; 2972 } 2973 if (t_flags & TF_FASTRECOVERY) { 2974 db_printf("%sTF_FASTRECOVERY", comma ? ", " : ""); 2975 comma = 1; 2976 } 2977 if (t_flags & TF_CONGRECOVERY) { 2978 db_printf("%sTF_CONGRECOVERY", comma ? ", " : ""); 2979 comma = 1; 2980 } 2981 if (t_flags & TF_WASFRECOVERY) { 2982 db_printf("%sTF_WASFRECOVERY", comma ? ", " : ""); 2983 comma = 1; 2984 } 2985 if (t_flags & TF_SIGNATURE) { 2986 db_printf("%sTF_SIGNATURE", comma ? ", " : ""); 2987 comma = 1; 2988 } 2989 if (t_flags & TF_FORCEDATA) { 2990 db_printf("%sTF_FORCEDATA", comma ? ", " : ""); 2991 comma = 1; 2992 } 2993 if (t_flags & TF_TSO) { 2994 db_printf("%sTF_TSO", comma ? ", " : ""); 2995 comma = 1; 2996 } 2997 if (t_flags & TF_FASTOPEN) { 2998 db_printf("%sTF_FASTOPEN", comma ? ", " : ""); 2999 comma = 1; 3000 } 3001 } 3002 3003 static void 3004 db_print_tflags2(u_int t_flags2) 3005 { 3006 int comma; 3007 3008 comma = 0; 3009 if (t_flags2 & TF2_ECN_PERMIT) { 3010 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : ""); 3011 comma = 1; 3012 } 3013 } 3014 3015 static void 3016 db_print_toobflags(char t_oobflags) 3017 { 3018 int comma; 3019 3020 comma = 0; 3021 if (t_oobflags & TCPOOB_HAVEDATA) { 3022 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : ""); 3023 comma = 1; 3024 } 3025 if (t_oobflags & TCPOOB_HADDATA) { 3026 db_printf("%sTCPOOB_HADDATA", comma ? ", " : ""); 3027 comma = 1; 3028 } 3029 } 3030 3031 static void 3032 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent) 3033 { 3034 3035 db_print_indent(indent); 3036 db_printf("%s at %p\n", name, tp); 3037 3038 indent += 2; 3039 3040 db_print_indent(indent); 3041 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n", 3042 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks); 3043 3044 db_print_indent(indent); 3045 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n", 3046 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep); 3047 3048 db_print_indent(indent); 3049 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl, 3050 &tp->t_timers->tt_delack, tp->t_inpcb); 3051 3052 db_print_indent(indent); 3053 db_printf("t_state: %d (", tp->t_state); 3054 db_print_tstate(tp->t_state); 3055 db_printf(")\n"); 3056 3057 db_print_indent(indent); 3058 db_printf("t_flags: 0x%x (", tp->t_flags); 3059 db_print_tflags(tp->t_flags); 3060 db_printf(")\n"); 3061 3062 db_print_indent(indent); 3063 db_printf("t_flags2: 0x%x (", tp->t_flags2); 3064 db_print_tflags2(tp->t_flags2); 3065 db_printf(")\n"); 3066 3067 db_print_indent(indent); 3068 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n", 3069 tp->snd_una, tp->snd_max, tp->snd_nxt); 3070 3071 db_print_indent(indent); 3072 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n", 3073 tp->snd_up, tp->snd_wl1, tp->snd_wl2); 3074 3075 db_print_indent(indent); 3076 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n", 3077 tp->iss, tp->irs, tp->rcv_nxt); 3078 3079 db_print_indent(indent); 3080 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n", 3081 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up); 3082 3083 db_print_indent(indent); 3084 db_printf("snd_wnd: %u snd_cwnd: %u\n", 3085 tp->snd_wnd, tp->snd_cwnd); 3086 3087 db_print_indent(indent); 3088 db_printf("snd_ssthresh: %u snd_recover: " 3089 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover); 3090 3091 db_print_indent(indent); 3092 db_printf("t_rcvtime: %u t_startime: %u\n", 3093 tp->t_rcvtime, tp->t_starttime); 3094 3095 db_print_indent(indent); 3096 db_printf("t_rttime: %u t_rtsq: 0x%08x\n", 3097 tp->t_rtttime, tp->t_rtseq); 3098 3099 db_print_indent(indent); 3100 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n", 3101 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt); 3102 3103 db_print_indent(indent); 3104 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u " 3105 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin, 3106 tp->t_rttbest); 3107 3108 db_print_indent(indent); 3109 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n", 3110 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror); 3111 3112 db_print_indent(indent); 3113 db_printf("t_oobflags: 0x%x (", tp->t_oobflags); 3114 db_print_toobflags(tp->t_oobflags); 3115 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc); 3116 3117 db_print_indent(indent); 3118 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n", 3119 tp->snd_scale, tp->rcv_scale, tp->request_r_scale); 3120 3121 db_print_indent(indent); 3122 db_printf("ts_recent: %u ts_recent_age: %u\n", 3123 tp->ts_recent, tp->ts_recent_age); 3124 3125 db_print_indent(indent); 3126 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: " 3127 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev); 3128 3129 db_print_indent(indent); 3130 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x " 3131 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev, 3132 tp->snd_recover_prev, tp->t_badrxtwin); 3133 3134 db_print_indent(indent); 3135 db_printf("snd_numholes: %d snd_holes first: %p\n", 3136 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes)); 3137 3138 db_print_indent(indent); 3139 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n", 3140 tp->snd_fack, tp->rcv_numsacks); 3141 3142 /* Skip sackblks, sackhint. */ 3143 3144 db_print_indent(indent); 3145 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n", 3146 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt); 3147 } 3148 3149 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb) 3150 { 3151 struct tcpcb *tp; 3152 3153 if (!have_addr) { 3154 db_printf("usage: show tcpcb <addr>\n"); 3155 return; 3156 } 3157 tp = (struct tcpcb *)addr; 3158 3159 db_print_tcpcb(tp, "tcpcb", 0); 3160 } 3161 #endif 3162