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