1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 67 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $ 68 */ 69 70 #include "opt_ipsec.h" 71 #include "opt_inet.h" 72 #include "opt_inet6.h" 73 #include "opt_tcpdebug.h" 74 75 #include <sys/param.h> 76 #include <sys/systm.h> 77 #include <sys/kernel.h> 78 #include <sys/malloc.h> 79 #include <sys/sysctl.h> 80 #include <sys/globaldata.h> 81 #include <sys/thread.h> 82 83 #include <sys/mbuf.h> 84 #ifdef INET6 85 #include <sys/domain.h> 86 #endif /* INET6 */ 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/protosw.h> 90 91 #include <sys/thread2.h> 92 #include <sys/msgport2.h> 93 #include <sys/socketvar2.h> 94 95 #include <net/if.h> 96 #include <net/netisr.h> 97 #include <net/route.h> 98 99 #include <net/netmsg2.h> 100 101 #include <netinet/in.h> 102 #include <netinet/in_systm.h> 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #endif 106 #include <netinet/in_pcb.h> 107 #ifdef INET6 108 #include <netinet6/in6_pcb.h> 109 #endif 110 #include <netinet/in_var.h> 111 #include <netinet/ip_var.h> 112 #ifdef INET6 113 #include <netinet6/ip6_var.h> 114 #include <netinet6/tcp6_var.h> 115 #endif 116 #include <netinet/tcp.h> 117 #include <netinet/tcp_fsm.h> 118 #include <netinet/tcp_seq.h> 119 #include <netinet/tcp_timer.h> 120 #include <netinet/tcp_timer2.h> 121 #include <netinet/tcp_var.h> 122 #include <netinet/tcpip.h> 123 #ifdef TCPDEBUG 124 #include <netinet/tcp_debug.h> 125 #endif 126 127 #ifdef IPSEC 128 #include <netinet6/ipsec.h> 129 #endif /*IPSEC*/ 130 131 /* 132 * TCP protocol interface to socket abstraction. 133 */ 134 extern char *tcpstates[]; /* XXX ??? */ 135 136 static int tcp_attach (struct socket *, struct pru_attach_info *); 137 static void tcp_connect (netmsg_t msg); 138 #ifdef INET6 139 static void tcp6_connect (netmsg_t msg); 140 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags, 141 struct mbuf **mp, 142 struct sockaddr_in6 *sin6, 143 struct in6_addr *addr6); 144 #endif /* INET6 */ 145 static struct tcpcb * 146 tcp_disconnect (struct tcpcb *); 147 static struct tcpcb * 148 tcp_usrclosed (struct tcpcb *); 149 150 #ifdef TCPDEBUG 151 #define TCPDEBUG0 int ostate = 0 152 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 153 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 154 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 155 #else 156 #define TCPDEBUG0 157 #define TCPDEBUG1() 158 #define TCPDEBUG2(req) 159 #endif 160 161 static int tcp_lport_extension = 1; 162 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW, 164 &tcp_lport_extension, 0, ""); 165 166 /* 167 * TCP attaches to socket via pru_attach(), reserving space, 168 * and an internet control block. This is likely occuring on 169 * cpu0 and may have to move later when we bind/connect. 170 */ 171 static void 172 tcp_usr_attach(netmsg_t msg) 173 { 174 struct socket *so = msg->base.nm_so; 175 struct pru_attach_info *ai = msg->attach.nm_ai; 176 int error; 177 struct inpcb *inp; 178 struct tcpcb *tp = 0; 179 TCPDEBUG0; 180 181 soreference(so); 182 inp = so->so_pcb; 183 TCPDEBUG1(); 184 if (inp) { 185 error = EISCONN; 186 goto out; 187 } 188 189 error = tcp_attach(so, ai); 190 if (error) 191 goto out; 192 193 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 194 so->so_linger = TCP_LINGERTIME; 195 tp = sototcpcb(so); 196 out: 197 sofree(so); /* from ref above */ 198 TCPDEBUG2(PRU_ATTACH); 199 lwkt_replymsg(&msg->lmsg, error); 200 } 201 202 /* 203 * pru_detach() detaches the TCP protocol from the socket. 204 * If the protocol state is non-embryonic, then can't 205 * do this directly: have to initiate a pru_disconnect(), 206 * which may finish later; embryonic TCB's can just 207 * be discarded here. 208 */ 209 static void 210 tcp_usr_detach(netmsg_t msg) 211 { 212 struct socket *so = msg->base.nm_so; 213 int error = 0; 214 struct inpcb *inp; 215 struct tcpcb *tp; 216 TCPDEBUG0; 217 218 inp = so->so_pcb; 219 220 /* 221 * If the inp is already detached it may have been due to an async 222 * close. Just return as if no error occured. 223 * 224 * It's possible for the tcpcb (tp) to disconnect from the inp due 225 * to tcp_drop()->tcp_close() being called. This may occur *after* 226 * the detach message has been queued so we may find a NULL tp here. 227 */ 228 if (inp) { 229 if ((tp = intotcpcb(inp)) != NULL) { 230 TCPDEBUG1(); 231 tp = tcp_disconnect(tp); 232 TCPDEBUG2(PRU_DETACH); 233 } 234 } 235 lwkt_replymsg(&msg->lmsg, error); 236 } 237 238 /* 239 * NOTE: ignore_error is non-zero for certain disconnection races 240 * which we want to silently allow, otherwise close() may return 241 * an unexpected error. 242 * 243 * NOTE: The variables (msg) and (tp) are assumed. 244 */ 245 #define COMMON_START(so, inp, ignore_error) \ 246 TCPDEBUG0; \ 247 \ 248 inp = so->so_pcb; \ 249 do { \ 250 if (inp == NULL) { \ 251 error = ignore_error ? 0 : EINVAL; \ 252 tp = NULL; \ 253 goto out; \ 254 } \ 255 tp = intotcpcb(inp); \ 256 TCPDEBUG1(); \ 257 } while(0) 258 259 #define COMMON_END1(req, noreply) \ 260 out: do { \ 261 TCPDEBUG2(req); \ 262 if (!(noreply)) \ 263 lwkt_replymsg(&msg->lmsg, error); \ 264 return; \ 265 } while(0) 266 267 #define COMMON_END(req) COMMON_END1((req), 0) 268 269 /* 270 * Give the socket an address. 271 */ 272 static void 273 tcp_usr_bind(netmsg_t msg) 274 { 275 struct socket *so = msg->bind.base.nm_so; 276 struct sockaddr *nam = msg->bind.nm_nam; 277 struct thread *td = msg->bind.nm_td; 278 int error = 0; 279 struct inpcb *inp; 280 struct tcpcb *tp; 281 struct sockaddr_in *sinp; 282 283 COMMON_START(so, inp, 0); 284 285 /* 286 * Must check for multicast addresses and disallow binding 287 * to them. 288 */ 289 sinp = (struct sockaddr_in *)nam; 290 if (sinp->sin_family == AF_INET && 291 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 292 error = EAFNOSUPPORT; 293 goto out; 294 } 295 error = in_pcbbind(inp, nam, td); 296 if (error) 297 goto out; 298 COMMON_END(PRU_BIND); 299 300 } 301 302 #ifdef INET6 303 304 static void 305 tcp6_usr_bind(netmsg_t msg) 306 { 307 struct socket *so = msg->bind.base.nm_so; 308 struct sockaddr *nam = msg->bind.nm_nam; 309 struct thread *td = msg->bind.nm_td; 310 int error = 0; 311 struct inpcb *inp; 312 struct tcpcb *tp; 313 struct sockaddr_in6 *sin6p; 314 315 COMMON_START(so, inp, 0); 316 317 /* 318 * Must check for multicast addresses and disallow binding 319 * to them. 320 */ 321 sin6p = (struct sockaddr_in6 *)nam; 322 if (sin6p->sin6_family == AF_INET6 && 323 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 324 error = EAFNOSUPPORT; 325 goto out; 326 } 327 inp->inp_vflag &= ~INP_IPV4; 328 inp->inp_vflag |= INP_IPV6; 329 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 330 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 331 inp->inp_vflag |= INP_IPV4; 332 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 333 struct sockaddr_in sin; 334 335 in6_sin6_2_sin(&sin, sin6p); 336 inp->inp_vflag |= INP_IPV4; 337 inp->inp_vflag &= ~INP_IPV6; 338 error = in_pcbbind(inp, (struct sockaddr *)&sin, td); 339 goto out; 340 } 341 } 342 error = in6_pcbbind(inp, nam, td); 343 if (error) 344 goto out; 345 COMMON_END(PRU_BIND); 346 } 347 #endif /* INET6 */ 348 349 #ifdef SMP 350 351 struct netmsg_inswildcard { 352 struct netmsg_base base; 353 struct inpcb *nm_inp; 354 }; 355 356 static void 357 in_pcbinswildcardhash_handler(netmsg_t msg) 358 { 359 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg; 360 int cpu = mycpuid, nextcpu; 361 362 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]); 363 364 nextcpu = cpu + 1; 365 if (nextcpu < ncpus2) 366 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg); 367 else 368 lwkt_replymsg(&nm->base.lmsg, 0); 369 } 370 371 #endif 372 373 /* 374 * Prepare to accept connections. 375 */ 376 static void 377 tcp_usr_listen(netmsg_t msg) 378 { 379 struct socket *so = msg->listen.base.nm_so; 380 struct thread *td = msg->listen.nm_td; 381 int error = 0; 382 struct inpcb *inp; 383 struct tcpcb *tp; 384 #ifdef SMP 385 struct netmsg_inswildcard nm; 386 #endif 387 388 COMMON_START(so, inp, 0); 389 390 if (tp->t_flags & TF_LISTEN) 391 goto out; 392 393 if (inp->inp_lport == 0) { 394 error = in_pcbbind(inp, NULL, td); 395 if (error) 396 goto out; 397 } 398 399 tp->t_state = TCPS_LISTEN; 400 tp->t_flags |= TF_LISTEN; 401 tp->tt_msg = NULL; /* Catch any invalid timer usage */ 402 403 #ifdef SMP 404 if (ncpus > 1) { 405 /* 406 * We have to set the flag because we can't have other cpus 407 * messing with our inp's flags. 408 */ 409 KASSERT(!(inp->inp_flags & INP_CONNECTED), 410 ("already on connhash\n")); 411 KASSERT(!(inp->inp_flags & INP_WILDCARD), 412 ("already on wildcardhash\n")); 413 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP), 414 ("already on MP wildcardhash\n")); 415 inp->inp_flags |= INP_WILDCARD_MP; 416 417 KKASSERT(so->so_port == cpu_portfn(0)); 418 KKASSERT(&curthread->td_msgport == cpu_portfn(0)); 419 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 420 421 netmsg_init(&nm.base, NULL, &curthread->td_msgport, 422 MSGF_PRIORITY, in_pcbinswildcardhash_handler); 423 nm.nm_inp = inp; 424 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0); 425 } 426 #endif 427 in_pcbinswildcardhash(inp); 428 COMMON_END(PRU_LISTEN); 429 } 430 431 #ifdef INET6 432 433 static void 434 tcp6_usr_listen(netmsg_t msg) 435 { 436 struct socket *so = msg->listen.base.nm_so; 437 struct thread *td = msg->listen.nm_td; 438 int error = 0; 439 struct inpcb *inp; 440 struct tcpcb *tp; 441 #ifdef SMP 442 struct netmsg_inswildcard nm; 443 #endif 444 445 COMMON_START(so, inp, 0); 446 447 if (tp->t_flags & TF_LISTEN) 448 goto out; 449 450 if (inp->inp_lport == 0) { 451 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY)) 452 inp->inp_vflag |= INP_IPV4; 453 else 454 inp->inp_vflag &= ~INP_IPV4; 455 error = in6_pcbbind(inp, NULL, td); 456 if (error) 457 goto out; 458 } 459 460 tp->t_state = TCPS_LISTEN; 461 tp->t_flags |= TF_LISTEN; 462 tp->tt_msg = NULL; /* Catch any invalid timer usage */ 463 464 #ifdef SMP 465 if (ncpus > 1) { 466 /* 467 * We have to set the flag because we can't have other cpus 468 * messing with our inp's flags. 469 */ 470 KASSERT(!(inp->inp_flags & INP_CONNECTED), 471 ("already on connhash\n")); 472 KASSERT(!(inp->inp_flags & INP_WILDCARD), 473 ("already on wildcardhash\n")); 474 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP), 475 ("already on MP wildcardhash\n")); 476 inp->inp_flags |= INP_WILDCARD_MP; 477 478 KKASSERT(so->so_port == cpu_portfn(0)); 479 KKASSERT(&curthread->td_msgport == cpu_portfn(0)); 480 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 481 482 netmsg_init(&nm.base, NULL, &curthread->td_msgport, 483 MSGF_PRIORITY, in_pcbinswildcardhash_handler); 484 nm.nm_inp = inp; 485 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0); 486 } 487 #endif 488 in_pcbinswildcardhash(inp); 489 COMMON_END(PRU_LISTEN); 490 } 491 #endif /* INET6 */ 492 493 /* 494 * Initiate connection to peer. 495 * Create a template for use in transmissions on this connection. 496 * Enter SYN_SENT state, and mark socket as connecting. 497 * Start keep-alive timer, and seed output sequence space. 498 * Send initial segment on connection. 499 */ 500 static void 501 tcp_usr_connect(netmsg_t msg) 502 { 503 struct socket *so = msg->connect.base.nm_so; 504 struct sockaddr *nam = msg->connect.nm_nam; 505 struct thread *td = msg->connect.nm_td; 506 int error = 0; 507 struct inpcb *inp; 508 struct tcpcb *tp; 509 struct sockaddr_in *sinp; 510 511 COMMON_START(so, inp, 0); 512 513 /* 514 * Must disallow TCP ``connections'' to multicast addresses. 515 */ 516 sinp = (struct sockaddr_in *)nam; 517 if (sinp->sin_family == AF_INET 518 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 519 error = EAFNOSUPPORT; 520 goto out; 521 } 522 523 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) { 524 error = EAFNOSUPPORT; /* IPv6 only jail */ 525 goto out; 526 } 527 528 tcp_connect(msg); 529 /* msg is invalid now */ 530 return; 531 out: 532 if (msg->connect.nm_m) { 533 m_freem(msg->connect.nm_m); 534 msg->connect.nm_m = NULL; 535 } 536 lwkt_replymsg(&msg->lmsg, error); 537 } 538 539 #ifdef INET6 540 541 static void 542 tcp6_usr_connect(netmsg_t msg) 543 { 544 struct socket *so = msg->connect.base.nm_so; 545 struct sockaddr *nam = msg->connect.nm_nam; 546 struct thread *td = msg->connect.nm_td; 547 int error = 0; 548 struct inpcb *inp; 549 struct tcpcb *tp; 550 struct sockaddr_in6 *sin6p; 551 552 COMMON_START(so, inp, 0); 553 554 /* 555 * Must disallow TCP ``connections'' to multicast addresses. 556 */ 557 sin6p = (struct sockaddr_in6 *)nam; 558 if (sin6p->sin6_family == AF_INET6 559 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 560 error = EAFNOSUPPORT; 561 goto out; 562 } 563 564 if (!prison_remote_ip(td, nam)) { 565 error = EAFNOSUPPORT; /* IPv4 only jail */ 566 goto out; 567 } 568 569 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 570 struct sockaddr_in *sinp; 571 572 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 573 error = EINVAL; 574 goto out; 575 } 576 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT); 577 in6_sin6_2_sin(sinp, sin6p); 578 inp->inp_vflag |= INP_IPV4; 579 inp->inp_vflag &= ~INP_IPV6; 580 msg->connect.nm_nam = (struct sockaddr *)sinp; 581 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC; 582 tcp_connect(msg); 583 /* msg is invalid now */ 584 return; 585 } 586 inp->inp_vflag &= ~INP_IPV4; 587 inp->inp_vflag |= INP_IPV6; 588 inp->inp_inc.inc_isipv6 = 1; 589 590 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK; 591 tcp6_connect(msg); 592 /* msg is invalid now */ 593 return; 594 out: 595 if (msg->connect.nm_m) { 596 m_freem(msg->connect.nm_m); 597 msg->connect.nm_m = NULL; 598 } 599 lwkt_replymsg(&msg->lmsg, error); 600 } 601 602 #endif /* INET6 */ 603 604 /* 605 * Initiate disconnect from peer. 606 * If connection never passed embryonic stage, just drop; 607 * else if don't need to let data drain, then can just drop anyways, 608 * else have to begin TCP shutdown process: mark socket disconnecting, 609 * drain unread data, state switch to reflect user close, and 610 * send segment (e.g. FIN) to peer. Socket will be really disconnected 611 * when peer sends FIN and acks ours. 612 * 613 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 614 */ 615 static void 616 tcp_usr_disconnect(netmsg_t msg) 617 { 618 struct socket *so = msg->disconnect.base.nm_so; 619 int error = 0; 620 struct inpcb *inp; 621 struct tcpcb *tp; 622 623 COMMON_START(so, inp, 1); 624 tp = tcp_disconnect(tp); 625 COMMON_END(PRU_DISCONNECT); 626 } 627 628 /* 629 * Accept a connection. Essentially all the work is 630 * done at higher levels; just return the address 631 * of the peer, storing through addr. 632 */ 633 static void 634 tcp_usr_accept(netmsg_t msg) 635 { 636 struct socket *so = msg->accept.base.nm_so; 637 struct sockaddr **nam = msg->accept.nm_nam; 638 int error = 0; 639 struct inpcb *inp; 640 struct tcpcb *tp = NULL; 641 TCPDEBUG0; 642 643 inp = so->so_pcb; 644 if (so->so_state & SS_ISDISCONNECTED) { 645 error = ECONNABORTED; 646 goto out; 647 } 648 if (inp == 0) { 649 error = EINVAL; 650 goto out; 651 } 652 653 tp = intotcpcb(inp); 654 TCPDEBUG1(); 655 in_setpeeraddr(so, nam); 656 COMMON_END(PRU_ACCEPT); 657 } 658 659 #ifdef INET6 660 static void 661 tcp6_usr_accept(netmsg_t msg) 662 { 663 struct socket *so = msg->accept.base.nm_so; 664 struct sockaddr **nam = msg->accept.nm_nam; 665 int error = 0; 666 struct inpcb *inp; 667 struct tcpcb *tp = NULL; 668 TCPDEBUG0; 669 670 inp = so->so_pcb; 671 672 if (so->so_state & SS_ISDISCONNECTED) { 673 error = ECONNABORTED; 674 goto out; 675 } 676 if (inp == 0) { 677 error = EINVAL; 678 goto out; 679 } 680 tp = intotcpcb(inp); 681 TCPDEBUG1(); 682 in6_mapped_peeraddr(so, nam); 683 COMMON_END(PRU_ACCEPT); 684 } 685 #endif /* INET6 */ 686 /* 687 * Mark the connection as being incapable of further output. 688 */ 689 static void 690 tcp_usr_shutdown(netmsg_t msg) 691 { 692 struct socket *so = msg->shutdown.base.nm_so; 693 int error = 0; 694 struct inpcb *inp; 695 struct tcpcb *tp; 696 697 COMMON_START(so, inp, 0); 698 socantsendmore(so); 699 tp = tcp_usrclosed(tp); 700 if (tp) 701 error = tcp_output(tp); 702 COMMON_END(PRU_SHUTDOWN); 703 } 704 705 /* 706 * After a receive, possibly send window update to peer. 707 */ 708 static void 709 tcp_usr_rcvd(netmsg_t msg) 710 { 711 struct socket *so = msg->rcvd.base.nm_so; 712 int error = 0; 713 struct inpcb *inp; 714 struct tcpcb *tp; 715 716 COMMON_START(so, inp, 0); 717 tcp_output(tp); 718 COMMON_END(PRU_RCVD); 719 } 720 721 /* 722 * Do a send by putting data in output queue and updating urgent 723 * marker if URG set. Possibly send more data. Unlike the other 724 * pru_*() routines, the mbuf chains are our responsibility. We 725 * must either enqueue them or free them. The other pru_* routines 726 * generally are caller-frees. 727 */ 728 static void 729 tcp_usr_send(netmsg_t msg) 730 { 731 struct socket *so = msg->send.base.nm_so; 732 int flags = msg->send.nm_flags; 733 struct mbuf *m = msg->send.nm_m; 734 int error = 0; 735 struct inpcb *inp; 736 struct tcpcb *tp; 737 TCPDEBUG0; 738 739 KKASSERT(msg->send.nm_control == NULL); 740 KKASSERT(msg->send.nm_addr == NULL); 741 KKASSERT((flags & PRUS_FREEADDR) == 0); 742 743 inp = so->so_pcb; 744 745 if (inp == NULL) { 746 /* 747 * OOPS! we lost a race, the TCP session got reset after 748 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a 749 * network interrupt in the non-critical section of sosend(). 750 */ 751 m_freem(m); 752 error = ECONNRESET; /* XXX EPIPE? */ 753 tp = NULL; 754 TCPDEBUG1(); 755 goto out; 756 } 757 tp = intotcpcb(inp); 758 TCPDEBUG1(); 759 760 /* 761 * Don't let too much OOB data build up 762 */ 763 if (flags & PRUS_OOB) { 764 if (ssb_space(&so->so_snd) < -512) { 765 m_freem(m); 766 error = ENOBUFS; 767 goto out; 768 } 769 } 770 771 /* 772 * Pump the data into the socket. 773 */ 774 if (m) 775 ssb_appendstream(&so->so_snd, m); 776 if (flags & PRUS_OOB) { 777 /* 778 * According to RFC961 (Assigned Protocols), 779 * the urgent pointer points to the last octet 780 * of urgent data. We continue, however, 781 * to consider it to indicate the first octet 782 * of data past the urgent section. 783 * Otherwise, snd_up should be one lower. 784 */ 785 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 786 tp->t_flags |= TF_FORCE; 787 error = tcp_output(tp); 788 tp->t_flags &= ~TF_FORCE; 789 } else { 790 if (flags & PRUS_EOF) { 791 /* 792 * Close the send side of the connection after 793 * the data is sent. 794 */ 795 socantsendmore(so); 796 tp = tcp_usrclosed(tp); 797 } 798 if (tp != NULL) { 799 if (flags & PRUS_MORETOCOME) 800 tp->t_flags |= TF_MORETOCOME; 801 error = tcp_output(tp); 802 if (flags & PRUS_MORETOCOME) 803 tp->t_flags &= ~TF_MORETOCOME; 804 } 805 } 806 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB : 807 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), 808 (flags & PRUS_NOREPLY)); 809 } 810 811 /* 812 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort() 813 * will sofree() it when we return. 814 */ 815 static void 816 tcp_usr_abort(netmsg_t msg) 817 { 818 struct socket *so = msg->abort.base.nm_so; 819 int error = 0; 820 struct inpcb *inp; 821 struct tcpcb *tp; 822 823 COMMON_START(so, inp, 1); 824 tp = tcp_drop(tp, ECONNABORTED); 825 COMMON_END(PRU_ABORT); 826 } 827 828 /* 829 * Receive out-of-band data. 830 */ 831 static void 832 tcp_usr_rcvoob(netmsg_t msg) 833 { 834 struct socket *so = msg->rcvoob.base.nm_so; 835 struct mbuf *m = msg->rcvoob.nm_m; 836 int flags = msg->rcvoob.nm_flags; 837 int error = 0; 838 struct inpcb *inp; 839 struct tcpcb *tp; 840 841 COMMON_START(so, inp, 0); 842 if ((so->so_oobmark == 0 && 843 (so->so_state & SS_RCVATMARK) == 0) || 844 so->so_options & SO_OOBINLINE || 845 tp->t_oobflags & TCPOOB_HADDATA) { 846 error = EINVAL; 847 goto out; 848 } 849 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 850 error = EWOULDBLOCK; 851 goto out; 852 } 853 m->m_len = 1; 854 *mtod(m, caddr_t) = tp->t_iobc; 855 if ((flags & MSG_PEEK) == 0) 856 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 857 COMMON_END(PRU_RCVOOB); 858 } 859 860 static void 861 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr) 862 { 863 in_savefaddr(so, faddr); 864 } 865 866 #ifdef INET6 867 static void 868 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr) 869 { 870 in6_mapped_savefaddr(so, faddr); 871 } 872 #endif 873 874 /* xxx - should be const */ 875 struct pr_usrreqs tcp_usrreqs = { 876 .pru_abort = tcp_usr_abort, 877 .pru_accept = tcp_usr_accept, 878 .pru_attach = tcp_usr_attach, 879 .pru_bind = tcp_usr_bind, 880 .pru_connect = tcp_usr_connect, 881 .pru_connect2 = pr_generic_notsupp, 882 .pru_control = in_control_dispatch, 883 .pru_detach = tcp_usr_detach, 884 .pru_disconnect = tcp_usr_disconnect, 885 .pru_listen = tcp_usr_listen, 886 .pru_peeraddr = in_setpeeraddr_dispatch, 887 .pru_rcvd = tcp_usr_rcvd, 888 .pru_rcvoob = tcp_usr_rcvoob, 889 .pru_send = tcp_usr_send, 890 .pru_sense = pru_sense_null, 891 .pru_shutdown = tcp_usr_shutdown, 892 .pru_sockaddr = in_setsockaddr_dispatch, 893 .pru_sosend = sosendtcp, 894 .pru_soreceive = soreceive, 895 .pru_savefaddr = tcp_usr_savefaddr 896 }; 897 898 #ifdef INET6 899 struct pr_usrreqs tcp6_usrreqs = { 900 .pru_abort = tcp_usr_abort, 901 .pru_accept = tcp6_usr_accept, 902 .pru_attach = tcp_usr_attach, 903 .pru_bind = tcp6_usr_bind, 904 .pru_connect = tcp6_usr_connect, 905 .pru_connect2 = pr_generic_notsupp, 906 .pru_control = in6_control_dispatch, 907 .pru_detach = tcp_usr_detach, 908 .pru_disconnect = tcp_usr_disconnect, 909 .pru_listen = tcp6_usr_listen, 910 .pru_peeraddr = in6_mapped_peeraddr_dispatch, 911 .pru_rcvd = tcp_usr_rcvd, 912 .pru_rcvoob = tcp_usr_rcvoob, 913 .pru_send = tcp_usr_send, 914 .pru_sense = pru_sense_null, 915 .pru_shutdown = tcp_usr_shutdown, 916 .pru_sockaddr = in6_mapped_sockaddr_dispatch, 917 .pru_sosend = sosendtcp, 918 .pru_soreceive = soreceive, 919 .pru_savefaddr = tcp6_usr_savefaddr 920 }; 921 #endif /* INET6 */ 922 923 static int 924 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m, 925 struct sockaddr_in *sin, struct sockaddr_in *if_sin) 926 { 927 struct inpcb *inp = tp->t_inpcb, *oinp; 928 struct socket *so = inp->inp_socket; 929 struct route *ro = &inp->inp_route; 930 931 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid], 932 sin->sin_addr, sin->sin_port, 933 (inp->inp_laddr.s_addr != INADDR_ANY ? 934 inp->inp_laddr : if_sin->sin_addr), 935 inp->inp_lport, 0, NULL); 936 if (oinp != NULL) { 937 m_freem(m); 938 return (EADDRINUSE); 939 } 940 if (inp->inp_laddr.s_addr == INADDR_ANY) 941 inp->inp_laddr = if_sin->sin_addr; 942 inp->inp_faddr = sin->sin_addr; 943 inp->inp_fport = sin->sin_port; 944 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid]; 945 in_pcbinsconnhash(inp); 946 947 /* 948 * We are now on the inpcb's owner CPU, if the cached route was 949 * freed because the rtentry's owner CPU is not the current CPU 950 * (e.g. in tcp_connect()), then we try to reallocate it here with 951 * the hope that a rtentry may be cloned from a RTF_PRCLONING 952 * rtentry. 953 */ 954 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 955 ro->ro_rt == NULL) { 956 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 957 ro->ro_dst.sa_family = AF_INET; 958 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 959 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = 960 sin->sin_addr; 961 rtalloc(ro); 962 } 963 964 /* 965 * Now that no more errors can occur, change the protocol processing 966 * port to the current thread (which is the correct thread). 967 * 968 * Create TCP timer message now; we are on the tcpcb's owner 969 * CPU/thread. 970 */ 971 tcp_create_timermsg(tp, &curthread->td_msgport); 972 973 /* 974 * Compute window scaling to request. Use a larger scaling then 975 * needed for the initial receive buffer in case the receive buffer 976 * gets expanded. 977 */ 978 if (tp->request_r_scale < TCP_MIN_WINSHIFT) 979 tp->request_r_scale = TCP_MIN_WINSHIFT; 980 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 981 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat 982 ) { 983 tp->request_r_scale++; 984 } 985 986 soisconnecting(so); 987 tcpstat.tcps_connattempt++; 988 tp->t_state = TCPS_SYN_SENT; 989 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 990 tp->iss = tcp_new_isn(tp); 991 tcp_sendseqinit(tp); 992 if (m) { 993 ssb_appendstream(&so->so_snd, m); 994 m = NULL; 995 if (flags & PRUS_OOB) 996 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 997 } 998 999 /* 1000 * Close the send side of the connection after 1001 * the data is sent if flagged. 1002 */ 1003 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) { 1004 socantsendmore(so); 1005 tp = tcp_usrclosed(tp); 1006 } 1007 return (tcp_output(tp)); 1008 } 1009 1010 /* 1011 * Common subroutine to open a TCP connection to remote host specified 1012 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1013 * port number if needed. Call in_pcbladdr to do the routing and to choose 1014 * a local host address (interface). 1015 * Initialize connection parameters and enter SYN-SENT state. 1016 */ 1017 static void 1018 tcp_connect(netmsg_t msg) 1019 { 1020 struct socket *so = msg->connect.base.nm_so; 1021 struct sockaddr *nam = msg->connect.nm_nam; 1022 struct thread *td = msg->connect.nm_td; 1023 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1024 struct sockaddr_in *if_sin; 1025 struct inpcb *inp; 1026 struct tcpcb *tp; 1027 int error, calc_laddr = 1; 1028 #ifdef SMP 1029 lwkt_port_t port; 1030 #endif 1031 1032 COMMON_START(so, inp, 0); 1033 1034 /* 1035 * Reconnect our pcb if we have to 1036 */ 1037 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) { 1038 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT; 1039 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1040 } 1041 1042 /* 1043 * Bind if we have to 1044 */ 1045 if (inp->inp_lport == 0) { 1046 if (tcp_lport_extension) { 1047 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY); 1048 1049 error = in_pcbladdr(inp, nam, &if_sin, td); 1050 if (error) 1051 goto out; 1052 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr; 1053 1054 error = in_pcbconn_bind(inp, nam, td); 1055 if (error) 1056 goto out; 1057 1058 calc_laddr = 0; 1059 } else { 1060 error = in_pcbbind(inp, NULL, td); 1061 if (error) 1062 goto out; 1063 } 1064 } 1065 1066 if (calc_laddr) { 1067 /* 1068 * Calculate the correct protocol processing thread. The 1069 * connect operation must run there. Set the forwarding 1070 * port before we forward the message or it will get bounced 1071 * right back to us. 1072 */ 1073 error = in_pcbladdr(inp, nam, &if_sin, td); 1074 if (error) 1075 goto out; 1076 } 1077 KKASSERT(inp->inp_socket == so); 1078 1079 #ifdef SMP 1080 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port, 1081 (inp->inp_laddr.s_addr ? 1082 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr), 1083 inp->inp_lport); 1084 1085 if (port != &curthread->td_msgport) { 1086 struct route *ro = &inp->inp_route; 1087 1088 /* 1089 * in_pcbladdr() may have allocated a route entry for us 1090 * on the current CPU, but we need a route entry on the 1091 * inpcb's owner CPU, so free it here. 1092 */ 1093 if (ro->ro_rt != NULL) 1094 RTFREE(ro->ro_rt); 1095 bzero(ro, sizeof(*ro)); 1096 1097 /* 1098 * We are moving the protocol processing port the socket 1099 * is on, we have to unlink here and re-link on the 1100 * target cpu. 1101 */ 1102 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1103 sosetport(so, port); 1104 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT; 1105 msg->connect.base.nm_dispatch = tcp_connect; 1106 1107 lwkt_forwardmsg(port, &msg->connect.base.lmsg); 1108 /* msg invalid now */ 1109 return; 1110 } 1111 #else 1112 KKASSERT(so->so_port == &curthread->td_msgport); 1113 #endif 1114 error = tcp_connect_oncpu(tp, msg->connect.nm_flags, 1115 msg->connect.nm_m, sin, if_sin); 1116 msg->connect.nm_m = NULL; 1117 out: 1118 if (msg->connect.nm_m) { 1119 m_freem(msg->connect.nm_m); 1120 msg->connect.nm_m = NULL; 1121 } 1122 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) { 1123 kfree(msg->connect.nm_nam, M_LWKTMSG); 1124 msg->connect.nm_nam = NULL; 1125 } 1126 lwkt_replymsg(&msg->connect.base.lmsg, error); 1127 /* msg invalid now */ 1128 } 1129 1130 #ifdef INET6 1131 1132 static void 1133 tcp6_connect(netmsg_t msg) 1134 { 1135 struct tcpcb *tp; 1136 struct socket *so = msg->connect.base.nm_so; 1137 struct sockaddr *nam = msg->connect.nm_nam; 1138 struct thread *td = msg->connect.nm_td; 1139 struct inpcb *inp; 1140 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1141 struct in6_addr *addr6; 1142 #ifdef SMP 1143 lwkt_port_t port; 1144 #endif 1145 int error; 1146 1147 COMMON_START(so, inp, 0); 1148 1149 /* 1150 * Reconnect our pcb if we have to 1151 */ 1152 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) { 1153 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT; 1154 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1155 } 1156 1157 /* 1158 * Bind if we have to 1159 */ 1160 if (inp->inp_lport == 0) { 1161 error = in6_pcbbind(inp, NULL, td); 1162 if (error) 1163 goto out; 1164 } 1165 1166 /* 1167 * Cannot simply call in_pcbconnect, because there might be an 1168 * earlier incarnation of this same connection still in 1169 * TIME_WAIT state, creating an ADDRINUSE error. 1170 */ 1171 error = in6_pcbladdr(inp, nam, &addr6, td); 1172 if (error) 1173 goto out; 1174 1175 #ifdef SMP 1176 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */ 1177 1178 if (port != &curthread->td_msgport) { 1179 struct route *ro = &inp->inp_route; 1180 1181 /* 1182 * in_pcbladdr() may have allocated a route entry for us 1183 * on the current CPU, but we need a route entry on the 1184 * inpcb's owner CPU, so free it here. 1185 */ 1186 if (ro->ro_rt != NULL) 1187 RTFREE(ro->ro_rt); 1188 bzero(ro, sizeof(*ro)); 1189 1190 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1191 sosetport(so, port); 1192 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT; 1193 msg->connect.base.nm_dispatch = tcp6_connect; 1194 1195 lwkt_forwardmsg(port, &msg->connect.base.lmsg); 1196 /* msg invalid now */ 1197 return; 1198 } 1199 #endif 1200 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags, 1201 &msg->connect.nm_m, sin6, addr6); 1202 /* nm_m may still be intact */ 1203 out: 1204 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) { 1205 tcp_connect(msg); 1206 /* msg invalid now */ 1207 } else { 1208 if (msg->connect.nm_m) { 1209 m_freem(msg->connect.nm_m); 1210 msg->connect.nm_m = NULL; 1211 } 1212 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) { 1213 kfree(msg->connect.nm_nam, M_LWKTMSG); 1214 msg->connect.nm_nam = NULL; 1215 } 1216 lwkt_replymsg(&msg->connect.base.lmsg, error); 1217 /* msg invalid now */ 1218 } 1219 } 1220 1221 static int 1222 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp, 1223 struct sockaddr_in6 *sin6, struct in6_addr *addr6) 1224 { 1225 struct mbuf *m = *mp; 1226 struct inpcb *inp = tp->t_inpcb; 1227 struct socket *so = inp->inp_socket; 1228 struct inpcb *oinp; 1229 1230 /* 1231 * Cannot simply call in_pcbconnect, because there might be an 1232 * earlier incarnation of this same connection still in 1233 * TIME_WAIT state, creating an ADDRINUSE error. 1234 */ 1235 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo, 1236 &sin6->sin6_addr, sin6->sin6_port, 1237 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ? 1238 addr6 : &inp->in6p_laddr), 1239 inp->inp_lport, 0, NULL); 1240 if (oinp) 1241 return (EADDRINUSE); 1242 1243 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1244 inp->in6p_laddr = *addr6; 1245 inp->in6p_faddr = sin6->sin6_addr; 1246 inp->inp_fport = sin6->sin6_port; 1247 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0) 1248 inp->in6p_flowinfo = sin6->sin6_flowinfo; 1249 in_pcbinsconnhash(inp); 1250 1251 /* 1252 * Now that no more errors can occur, change the protocol processing 1253 * port to the current thread (which is the correct thread). 1254 * 1255 * Create TCP timer message now; we are on the tcpcb's owner 1256 * CPU/thread. 1257 */ 1258 tcp_create_timermsg(tp, &curthread->td_msgport); 1259 1260 /* Compute window scaling to request. */ 1261 if (tp->request_r_scale < TCP_MIN_WINSHIFT) 1262 tp->request_r_scale = TCP_MIN_WINSHIFT; 1263 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1264 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) { 1265 tp->request_r_scale++; 1266 } 1267 1268 soisconnecting(so); 1269 tcpstat.tcps_connattempt++; 1270 tp->t_state = TCPS_SYN_SENT; 1271 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 1272 tp->iss = tcp_new_isn(tp); 1273 tcp_sendseqinit(tp); 1274 if (m) { 1275 ssb_appendstream(&so->so_snd, m); 1276 *mp = NULL; 1277 if (flags & PRUS_OOB) 1278 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 1279 } 1280 1281 /* 1282 * Close the send side of the connection after 1283 * the data is sent if flagged. 1284 */ 1285 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) { 1286 socantsendmore(so); 1287 tp = tcp_usrclosed(tp); 1288 } 1289 return (tcp_output(tp)); 1290 } 1291 1292 #endif /* INET6 */ 1293 1294 /* 1295 * The new sockopt interface makes it possible for us to block in the 1296 * copyin/out step (if we take a page fault). Taking a page fault while 1297 * in a critical section is probably a Bad Thing. (Since sockets and pcbs 1298 * both now use TSM, there probably isn't any need for this function to 1299 * run in a critical section any more. This needs more examination.) 1300 */ 1301 void 1302 tcp_ctloutput(netmsg_t msg) 1303 { 1304 struct socket *so = msg->base.nm_so; 1305 struct sockopt *sopt = msg->ctloutput.nm_sopt; 1306 int error, opt, optval, opthz; 1307 struct inpcb *inp; 1308 struct tcpcb *tp; 1309 1310 error = 0; 1311 inp = so->so_pcb; 1312 if (inp == NULL) { 1313 error = ECONNRESET; 1314 goto done; 1315 } 1316 1317 if (sopt->sopt_level != IPPROTO_TCP) { 1318 #ifdef INET6 1319 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1320 ip6_ctloutput_dispatch(msg); 1321 else 1322 #endif /* INET6 */ 1323 ip_ctloutput(msg); 1324 /* msg invalid now */ 1325 return; 1326 } 1327 tp = intotcpcb(inp); 1328 1329 switch (sopt->sopt_dir) { 1330 case SOPT_SET: 1331 error = soopt_to_kbuf(sopt, &optval, sizeof optval, 1332 sizeof optval); 1333 if (error) 1334 break; 1335 switch (sopt->sopt_name) { 1336 case TCP_FASTKEEP: 1337 if (optval > 0) { 1338 if ((tp->t_flags & TF_FASTKEEP) == 0) { 1339 tp->t_flags |= TF_FASTKEEP; 1340 tcp_timer_keep_activity(tp, 0); 1341 } 1342 } else { 1343 tp->t_flags &= ~TF_FASTKEEP; 1344 } 1345 break; 1346 #ifdef TCP_SIGNATURE 1347 case TCP_SIGNATURE_ENABLE: 1348 if (optval > 0) 1349 tp->t_flags |= TF_SIGNATURE; 1350 else 1351 tp->t_flags &= ~TF_SIGNATURE; 1352 break; 1353 #endif /* TCP_SIGNATURE */ 1354 case TCP_NODELAY: 1355 case TCP_NOOPT: 1356 switch (sopt->sopt_name) { 1357 case TCP_NODELAY: 1358 opt = TF_NODELAY; 1359 break; 1360 case TCP_NOOPT: 1361 opt = TF_NOOPT; 1362 break; 1363 default: 1364 opt = 0; /* dead code to fool gcc */ 1365 break; 1366 } 1367 1368 if (optval) 1369 tp->t_flags |= opt; 1370 else 1371 tp->t_flags &= ~opt; 1372 break; 1373 1374 case TCP_NOPUSH: 1375 if (optval) 1376 tp->t_flags |= TF_NOPUSH; 1377 else { 1378 tp->t_flags &= ~TF_NOPUSH; 1379 error = tcp_output(tp); 1380 } 1381 break; 1382 1383 case TCP_MAXSEG: 1384 /* 1385 * Must be between 0 and maxseg. If the requested 1386 * maxseg is too small to satisfy the desired minmss, 1387 * pump it up (silently so sysctl modifications of 1388 * minmss do not create unexpected program failures). 1389 * Handle degenerate cases. 1390 */ 1391 if (optval > 0 && optval <= tp->t_maxseg) { 1392 if (optval + 40 < tcp_minmss) { 1393 optval = tcp_minmss - 40; 1394 if (optval < 0) 1395 optval = 1; 1396 } 1397 tp->t_maxseg = optval; 1398 } else { 1399 error = EINVAL; 1400 } 1401 break; 1402 1403 case TCP_KEEPINIT: 1404 opthz = ((int64_t)optval * hz) / 1000; 1405 if (opthz >= 1) 1406 tp->t_keepinit = opthz; 1407 else 1408 error = EINVAL; 1409 break; 1410 1411 case TCP_KEEPIDLE: 1412 opthz = ((int64_t)optval * hz) / 1000; 1413 if (opthz >= 1) 1414 tp->t_keepidle = opthz; 1415 else 1416 error = EINVAL; 1417 break; 1418 1419 case TCP_KEEPINTVL: 1420 opthz = ((int64_t)optval * hz) / 1000; 1421 if (opthz >= 1) { 1422 tp->t_keepintvl = opthz; 1423 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 1424 } else { 1425 error = EINVAL; 1426 } 1427 break; 1428 1429 case TCP_KEEPCNT: 1430 if (optval > 0) { 1431 tp->t_keepcnt = optval; 1432 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 1433 } else { 1434 error = EINVAL; 1435 } 1436 break; 1437 1438 default: 1439 error = ENOPROTOOPT; 1440 break; 1441 } 1442 break; 1443 1444 case SOPT_GET: 1445 switch (sopt->sopt_name) { 1446 #ifdef TCP_SIGNATURE 1447 case TCP_SIGNATURE_ENABLE: 1448 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1449 break; 1450 #endif /* TCP_SIGNATURE */ 1451 case TCP_NODELAY: 1452 optval = tp->t_flags & TF_NODELAY; 1453 break; 1454 case TCP_MAXSEG: 1455 optval = tp->t_maxseg; 1456 break; 1457 case TCP_NOOPT: 1458 optval = tp->t_flags & TF_NOOPT; 1459 break; 1460 case TCP_NOPUSH: 1461 optval = tp->t_flags & TF_NOPUSH; 1462 break; 1463 case TCP_KEEPINIT: 1464 optval = ((int64_t)tp->t_keepinit * 1000) / hz; 1465 break; 1466 case TCP_KEEPIDLE: 1467 optval = ((int64_t)tp->t_keepidle * 1000) / hz; 1468 break; 1469 case TCP_KEEPINTVL: 1470 optval = ((int64_t)tp->t_keepintvl * 1000) / hz; 1471 break; 1472 case TCP_KEEPCNT: 1473 optval = tp->t_keepcnt; 1474 break; 1475 default: 1476 error = ENOPROTOOPT; 1477 break; 1478 } 1479 if (error == 0) 1480 soopt_from_kbuf(sopt, &optval, sizeof optval); 1481 break; 1482 } 1483 done: 1484 lwkt_replymsg(&msg->lmsg, error); 1485 } 1486 1487 /* 1488 * tcp_sendspace and tcp_recvspace are the default send and receive window 1489 * sizes, respectively. These are obsolescent (this information should 1490 * be set by the route). 1491 * 1492 * Use a default that does not require tcp window scaling to be turned 1493 * on. Individual programs or the administrator can increase the default. 1494 */ 1495 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1496 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1497 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1498 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1499 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1500 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1501 1502 /* 1503 * Attach TCP protocol to socket, allocating internet protocol control 1504 * block, tcp control block, bufer space, and entering LISTEN state 1505 * if to accept connections. 1506 */ 1507 static int 1508 tcp_attach(struct socket *so, struct pru_attach_info *ai) 1509 { 1510 struct tcpcb *tp; 1511 struct inpcb *inp; 1512 int error; 1513 int cpu; 1514 #ifdef INET6 1515 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1516 #endif 1517 1518 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 1519 lwkt_gettoken(&so->so_rcv.ssb_token); 1520 error = soreserve(so, tcp_sendspace, tcp_recvspace, 1521 ai->sb_rlimit); 1522 lwkt_reltoken(&so->so_rcv.ssb_token); 1523 if (error) 1524 return (error); 1525 } 1526 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE); 1527 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE); 1528 cpu = mycpu->gd_cpuid; 1529 1530 /* 1531 * Set the default port for protocol processing. This will likely 1532 * change when we connect. 1533 */ 1534 error = in_pcballoc(so, &tcbinfo[cpu]); 1535 if (error) 1536 return (error); 1537 inp = so->so_pcb; 1538 #ifdef INET6 1539 if (isipv6) { 1540 inp->inp_vflag |= INP_IPV6; 1541 inp->in6p_hops = -1; /* use kernel default */ 1542 } 1543 else 1544 #endif 1545 inp->inp_vflag |= INP_IPV4; 1546 tp = tcp_newtcpcb(inp); 1547 if (tp == NULL) { 1548 /* 1549 * Make sure the socket is destroyed by the pcbdetach. 1550 */ 1551 soreference(so); 1552 #ifdef INET6 1553 if (isipv6) 1554 in6_pcbdetach(inp); 1555 else 1556 #endif 1557 in_pcbdetach(inp); 1558 sofree(so); /* from ref above */ 1559 return (ENOBUFS); 1560 } 1561 tp->t_state = TCPS_CLOSED; 1562 return (0); 1563 } 1564 1565 /* 1566 * Initiate (or continue) disconnect. 1567 * If embryonic state, just send reset (once). 1568 * If in ``let data drain'' option and linger null, just drop. 1569 * Otherwise (hard), mark socket disconnecting and drop 1570 * current input data; switch states based on user close, and 1571 * send segment to peer (with FIN). 1572 */ 1573 static struct tcpcb * 1574 tcp_disconnect(struct tcpcb *tp) 1575 { 1576 struct socket *so = tp->t_inpcb->inp_socket; 1577 1578 if (tp->t_state < TCPS_ESTABLISHED) { 1579 tp = tcp_close(tp); 1580 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1581 tp = tcp_drop(tp, 0); 1582 } else { 1583 lwkt_gettoken(&so->so_rcv.ssb_token); 1584 soisdisconnecting(so); 1585 sbflush(&so->so_rcv.sb); 1586 tp = tcp_usrclosed(tp); 1587 if (tp) 1588 tcp_output(tp); 1589 lwkt_reltoken(&so->so_rcv.ssb_token); 1590 } 1591 return (tp); 1592 } 1593 1594 /* 1595 * User issued close, and wish to trail through shutdown states: 1596 * if never received SYN, just forget it. If got a SYN from peer, 1597 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1598 * If already got a FIN from peer, then almost done; go to LAST_ACK 1599 * state. In all other cases, have already sent FIN to peer (e.g. 1600 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1601 * for peer to send FIN or not respond to keep-alives, etc. 1602 * We can let the user exit from the close as soon as the FIN is acked. 1603 */ 1604 static struct tcpcb * 1605 tcp_usrclosed(struct tcpcb *tp) 1606 { 1607 1608 switch (tp->t_state) { 1609 1610 case TCPS_CLOSED: 1611 case TCPS_LISTEN: 1612 tp->t_state = TCPS_CLOSED; 1613 tp = tcp_close(tp); 1614 break; 1615 1616 case TCPS_SYN_SENT: 1617 case TCPS_SYN_RECEIVED: 1618 tp->t_flags |= TF_NEEDFIN; 1619 break; 1620 1621 case TCPS_ESTABLISHED: 1622 tp->t_state = TCPS_FIN_WAIT_1; 1623 break; 1624 1625 case TCPS_CLOSE_WAIT: 1626 tp->t_state = TCPS_LAST_ACK; 1627 break; 1628 } 1629 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1630 soisdisconnected(tp->t_inpcb->inp_socket); 1631 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1632 if (tp->t_state == TCPS_FIN_WAIT_2) { 1633 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle, 1634 tcp_timer_2msl); 1635 } 1636 } 1637 return (tp); 1638 } 1639