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. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94 63 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $ 64 */ 65 66 #include "opt_ipsec.h" 67 #include "opt_inet.h" 68 #include "opt_inet6.h" 69 #include "opt_tcpdebug.h" 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/kernel.h> 74 #include <sys/malloc.h> 75 #include <sys/sysctl.h> 76 #include <sys/globaldata.h> 77 #include <sys/thread.h> 78 79 #include <sys/mbuf.h> 80 #ifdef INET6 81 #include <sys/domain.h> 82 #endif /* INET6 */ 83 #include <sys/socket.h> 84 #include <sys/socketvar.h> 85 #include <sys/socketops.h> 86 #include <sys/protosw.h> 87 88 #include <sys/thread2.h> 89 #include <sys/msgport2.h> 90 #include <sys/socketvar2.h> 91 92 #include <net/if.h> 93 #include <net/netisr.h> 94 #include <net/route.h> 95 96 #include <net/netmsg2.h> 97 #include <net/netisr2.h> 98 99 #include <netinet/in.h> 100 #include <netinet/in_systm.h> 101 #ifdef INET6 102 #include <netinet/ip6.h> 103 #endif 104 #include <netinet/in_pcb.h> 105 #ifdef INET6 106 #include <netinet6/in6_pcb.h> 107 #endif 108 #include <netinet/in_var.h> 109 #include <netinet/ip_var.h> 110 #ifdef INET6 111 #include <netinet6/ip6_var.h> 112 #include <netinet6/tcp6_var.h> 113 #endif 114 #include <netinet/tcp.h> 115 #include <netinet/tcp_fsm.h> 116 #include <netinet/tcp_seq.h> 117 #include <netinet/tcp_timer.h> 118 #include <netinet/tcp_timer2.h> 119 #include <netinet/tcp_var.h> 120 #include <netinet/tcpip.h> 121 #ifdef TCPDEBUG 122 #include <netinet/tcp_debug.h> 123 #endif 124 125 #ifdef IPSEC 126 #include <netinet6/ipsec.h> 127 #endif /*IPSEC*/ 128 129 /* 130 * TCP protocol interface to socket abstraction. 131 */ 132 extern char *tcpstates[]; /* XXX ??? */ 133 134 static int tcp_attach (struct socket *, struct pru_attach_info *); 135 static void tcp_connect (netmsg_t msg); 136 #ifdef INET6 137 static void tcp6_connect (netmsg_t msg); 138 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags, 139 struct mbuf **mp, 140 struct sockaddr_in6 *sin6, 141 struct in6_addr *addr6); 142 #endif /* INET6 */ 143 static struct tcpcb * 144 tcp_disconnect (struct tcpcb *); 145 static struct tcpcb * 146 tcp_usrclosed (struct tcpcb *); 147 148 #ifdef TCPDEBUG 149 #define TCPDEBUG0 int ostate = 0 150 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 151 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \ 152 tcp_trace(TA_USER, ostate, tp, 0, 0, req) 153 #else 154 #define TCPDEBUG0 155 #define TCPDEBUG1() 156 #define TCPDEBUG2(req) 157 #endif 158 159 static int tcp_lport_extension = 1; 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW, 161 &tcp_lport_extension, 0, ""); 162 163 /* 164 * For some ill optimized programs, which try to use TCP_NOPUSH 165 * to improve performance, will have small amount of data sits 166 * in the sending buffer. These small amount of data will _not_ 167 * be pushed into the network until more data are written into 168 * the socket or the socket write side is shutdown. 169 */ 170 static int tcp_disable_nopush = 1; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW, 172 &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect"); 173 174 /* 175 * TCP attaches to socket via pru_attach(), reserving space, 176 * and an internet control block. This is likely occuring on 177 * cpu0 and may have to move later when we bind/connect. 178 */ 179 static void 180 tcp_usr_attach(netmsg_t msg) 181 { 182 struct socket *so = msg->base.nm_so; 183 struct pru_attach_info *ai = msg->attach.nm_ai; 184 int error; 185 struct inpcb *inp; 186 struct tcpcb *tp = NULL; 187 TCPDEBUG0; 188 189 soreference(so); 190 inp = so->so_pcb; 191 TCPDEBUG1(); 192 if (inp) { 193 error = EISCONN; 194 goto out; 195 } 196 197 error = tcp_attach(so, ai); 198 if (error) 199 goto out; 200 201 if ((so->so_options & SO_LINGER) && so->so_linger == 0) 202 so->so_linger = TCP_LINGERTIME; 203 tp = sototcpcb(so); 204 out: 205 sofree(so); /* from ref above */ 206 TCPDEBUG2(PRU_ATTACH); 207 lwkt_replymsg(&msg->lmsg, error); 208 } 209 210 /* 211 * pru_detach() detaches the TCP protocol from the socket. 212 * If the protocol state is non-embryonic, then can't 213 * do this directly: have to initiate a pru_disconnect(), 214 * which may finish later; embryonic TCB's can just 215 * be discarded here. 216 */ 217 static void 218 tcp_usr_detach(netmsg_t msg) 219 { 220 struct socket *so = msg->base.nm_so; 221 int error = 0; 222 struct inpcb *inp; 223 struct tcpcb *tp; 224 TCPDEBUG0; 225 226 inp = so->so_pcb; 227 228 /* 229 * If the inp is already detached it may have been due to an async 230 * close. Just return as if no error occured. 231 * 232 * It's possible for the tcpcb (tp) to disconnect from the inp due 233 * to tcp_drop()->tcp_close() being called. This may occur *after* 234 * the detach message has been queued so we may find a NULL tp here. 235 */ 236 if (inp) { 237 if ((tp = intotcpcb(inp)) != NULL) { 238 TCPDEBUG1(); 239 tp = tcp_disconnect(tp); 240 TCPDEBUG2(PRU_DETACH); 241 } 242 } 243 lwkt_replymsg(&msg->lmsg, error); 244 } 245 246 /* 247 * NOTE: ignore_error is non-zero for certain disconnection races 248 * which we want to silently allow, otherwise close() may return 249 * an unexpected error. 250 * 251 * NOTE: The variables (msg) and (tp) are assumed. 252 */ 253 #define COMMON_START(so, inp, ignore_error) \ 254 TCPDEBUG0; \ 255 \ 256 inp = so->so_pcb; \ 257 do { \ 258 if (inp == NULL) { \ 259 error = ignore_error ? 0 : EINVAL; \ 260 tp = NULL; \ 261 goto out; \ 262 } \ 263 tp = intotcpcb(inp); \ 264 TCPDEBUG1(); \ 265 } while(0) 266 267 #define COMMON_END1(req, noreply) \ 268 out: do { \ 269 TCPDEBUG2(req); \ 270 if (!(noreply)) \ 271 lwkt_replymsg(&msg->lmsg, error); \ 272 return; \ 273 } while(0) 274 275 #define COMMON_END(req) COMMON_END1((req), 0) 276 277 /* 278 * Give the socket an address. 279 */ 280 static void 281 tcp_usr_bind(netmsg_t msg) 282 { 283 struct socket *so = msg->bind.base.nm_so; 284 struct sockaddr *nam = msg->bind.nm_nam; 285 struct thread *td = msg->bind.nm_td; 286 int error = 0; 287 struct inpcb *inp; 288 struct tcpcb *tp; 289 struct sockaddr_in *sinp; 290 291 COMMON_START(so, inp, 0); 292 293 /* 294 * Must check for multicast addresses and disallow binding 295 * to them. 296 */ 297 sinp = (struct sockaddr_in *)nam; 298 if (sinp->sin_family == AF_INET && 299 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 300 error = EAFNOSUPPORT; 301 goto out; 302 } 303 error = in_pcbbind(inp, nam, td); 304 if (error) 305 goto out; 306 COMMON_END(PRU_BIND); 307 308 } 309 310 #ifdef INET6 311 312 static void 313 tcp6_usr_bind(netmsg_t msg) 314 { 315 struct socket *so = msg->bind.base.nm_so; 316 struct sockaddr *nam = msg->bind.nm_nam; 317 struct thread *td = msg->bind.nm_td; 318 int error = 0; 319 struct inpcb *inp; 320 struct tcpcb *tp; 321 struct sockaddr_in6 *sin6p; 322 323 COMMON_START(so, inp, 0); 324 325 /* 326 * Must check for multicast addresses and disallow binding 327 * to them. 328 */ 329 sin6p = (struct sockaddr_in6 *)nam; 330 if (sin6p->sin6_family == AF_INET6 && 331 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 332 error = EAFNOSUPPORT; 333 goto out; 334 } 335 inp->inp_vflag &= ~INP_IPV4; 336 inp->inp_vflag |= INP_IPV6; 337 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) { 338 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr)) 339 inp->inp_vflag |= INP_IPV4; 340 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 341 struct sockaddr_in sin; 342 343 in6_sin6_2_sin(&sin, sin6p); 344 inp->inp_vflag |= INP_IPV4; 345 inp->inp_vflag &= ~INP_IPV6; 346 error = in_pcbbind(inp, (struct sockaddr *)&sin, td); 347 goto out; 348 } 349 } 350 error = in6_pcbbind(inp, nam, td); 351 if (error) 352 goto out; 353 COMMON_END(PRU_BIND); 354 } 355 #endif /* INET6 */ 356 357 struct netmsg_inswildcard { 358 struct netmsg_base base; 359 struct inpcb *nm_inp; 360 }; 361 362 static void 363 in_pcbinswildcardhash_handler(netmsg_t msg) 364 { 365 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg; 366 int cpu = mycpuid, nextcpu; 367 368 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]); 369 370 nextcpu = cpu + 1; 371 if (nextcpu < ncpus2) 372 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg); 373 else 374 lwkt_replymsg(&nm->base.lmsg, 0); 375 } 376 377 /* 378 * Prepare to accept connections. 379 */ 380 static void 381 tcp_usr_listen(netmsg_t msg) 382 { 383 struct socket *so = msg->listen.base.nm_so; 384 struct thread *td = msg->listen.nm_td; 385 int error = 0; 386 struct inpcb *inp; 387 struct tcpcb *tp; 388 struct netmsg_inswildcard nm; 389 lwkt_port_t port0 = netisr_cpuport(0); 390 391 COMMON_START(so, inp, 0); 392 393 if (&curthread->td_msgport != port0) { 394 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0, 395 ("already asked to relink")); 396 397 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]); 398 sosetport(so, port0); 399 msg->listen.nm_flags |= PRUL_RELINK; 400 401 lwkt_forwardmsg(port0, &msg->listen.base.lmsg); 402 /* msg invalid now */ 403 return; 404 } 405 KASSERT(so->so_port == port0, ("so_port is not netisr0")); 406 407 if (msg->listen.nm_flags & PRUL_RELINK) { 408 msg->listen.nm_flags &= ~PRUL_RELINK; 409 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]); 410 } 411 KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0")); 412 413 if (tp->t_flags & TF_LISTEN) 414 goto out; 415 416 if (inp->inp_lport == 0) { 417 error = in_pcbbind(inp, NULL, td); 418 if (error) 419 goto out; 420 } 421 422 tp->t_state = TCPS_LISTEN; 423 tp->t_flags |= TF_LISTEN; 424 tp->tt_msg = NULL; /* Catch any invalid timer usage */ 425 426 if (ncpus > 1) { 427 /* 428 * We have to set the flag because we can't have other cpus 429 * messing with our inp's flags. 430 */ 431 KASSERT(!(inp->inp_flags & INP_CONNECTED), 432 ("already on connhash")); 433 KASSERT(!(inp->inp_flags & INP_WILDCARD), 434 ("already on wildcardhash")); 435 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP), 436 ("already on MP wildcardhash")); 437 inp->inp_flags |= INP_WILDCARD_MP; 438 439 netmsg_init(&nm.base, NULL, &curthread->td_msgport, 440 MSGF_PRIORITY, in_pcbinswildcardhash_handler); 441 nm.nm_inp = inp; 442 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0); 443 } 444 in_pcbinswildcardhash(inp); 445 COMMON_END(PRU_LISTEN); 446 } 447 448 #ifdef INET6 449 450 static void 451 tcp6_usr_listen(netmsg_t msg) 452 { 453 struct socket *so = msg->listen.base.nm_so; 454 struct thread *td = msg->listen.nm_td; 455 int error = 0; 456 struct inpcb *inp; 457 struct tcpcb *tp; 458 struct netmsg_inswildcard nm; 459 460 COMMON_START(so, inp, 0); 461 462 if (tp->t_flags & TF_LISTEN) 463 goto out; 464 465 if (inp->inp_lport == 0) { 466 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY)) 467 inp->inp_vflag |= INP_IPV4; 468 else 469 inp->inp_vflag &= ~INP_IPV4; 470 error = in6_pcbbind(inp, NULL, td); 471 if (error) 472 goto out; 473 } 474 475 tp->t_state = TCPS_LISTEN; 476 tp->t_flags |= TF_LISTEN; 477 tp->tt_msg = NULL; /* Catch any invalid timer usage */ 478 479 if (ncpus > 1) { 480 /* 481 * We have to set the flag because we can't have other cpus 482 * messing with our inp's flags. 483 */ 484 KASSERT(!(inp->inp_flags & INP_CONNECTED), 485 ("already on connhash")); 486 KASSERT(!(inp->inp_flags & INP_WILDCARD), 487 ("already on wildcardhash")); 488 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP), 489 ("already on MP wildcardhash")); 490 inp->inp_flags |= INP_WILDCARD_MP; 491 492 KKASSERT(so->so_port == netisr_cpuport(0)); 493 KKASSERT(&curthread->td_msgport == netisr_cpuport(0)); 494 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 495 496 netmsg_init(&nm.base, NULL, &curthread->td_msgport, 497 MSGF_PRIORITY, in_pcbinswildcardhash_handler); 498 nm.nm_inp = inp; 499 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0); 500 } 501 in_pcbinswildcardhash(inp); 502 COMMON_END(PRU_LISTEN); 503 } 504 #endif /* INET6 */ 505 506 /* 507 * Initiate connection to peer. 508 * Create a template for use in transmissions on this connection. 509 * Enter SYN_SENT state, and mark socket as connecting. 510 * Start keep-alive timer, and seed output sequence space. 511 * Send initial segment on connection. 512 */ 513 static void 514 tcp_usr_connect(netmsg_t msg) 515 { 516 struct socket *so = msg->connect.base.nm_so; 517 struct sockaddr *nam = msg->connect.nm_nam; 518 struct thread *td = msg->connect.nm_td; 519 int error = 0; 520 struct inpcb *inp; 521 struct tcpcb *tp; 522 struct sockaddr_in *sinp; 523 524 COMMON_START(so, inp, 0); 525 526 /* 527 * Must disallow TCP ``connections'' to multicast addresses. 528 */ 529 sinp = (struct sockaddr_in *)nam; 530 if (sinp->sin_family == AF_INET 531 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { 532 error = EAFNOSUPPORT; 533 goto out; 534 } 535 536 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) { 537 error = EAFNOSUPPORT; /* IPv6 only jail */ 538 goto out; 539 } 540 541 tcp_connect(msg); 542 /* msg is invalid now */ 543 return; 544 out: 545 if (msg->connect.nm_m) { 546 m_freem(msg->connect.nm_m); 547 msg->connect.nm_m = NULL; 548 } 549 if (msg->connect.nm_flags & PRUC_HELDTD) 550 lwkt_rele(td); 551 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) { 552 so->so_error = error; 553 soisdisconnected(so); 554 } 555 lwkt_replymsg(&msg->lmsg, error); 556 } 557 558 #ifdef INET6 559 560 static void 561 tcp6_usr_connect(netmsg_t msg) 562 { 563 struct socket *so = msg->connect.base.nm_so; 564 struct sockaddr *nam = msg->connect.nm_nam; 565 struct thread *td = msg->connect.nm_td; 566 int error = 0; 567 struct inpcb *inp; 568 struct tcpcb *tp; 569 struct sockaddr_in6 *sin6p; 570 571 COMMON_START(so, inp, 0); 572 573 /* 574 * Must disallow TCP ``connections'' to multicast addresses. 575 */ 576 sin6p = (struct sockaddr_in6 *)nam; 577 if (sin6p->sin6_family == AF_INET6 578 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { 579 error = EAFNOSUPPORT; 580 goto out; 581 } 582 583 if (!prison_remote_ip(td, nam)) { 584 error = EAFNOSUPPORT; /* IPv4 only jail */ 585 goto out; 586 } 587 588 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) { 589 struct sockaddr_in *sinp; 590 591 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) { 592 error = EINVAL; 593 goto out; 594 } 595 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT); 596 in6_sin6_2_sin(sinp, sin6p); 597 inp->inp_vflag |= INP_IPV4; 598 inp->inp_vflag &= ~INP_IPV6; 599 msg->connect.nm_nam = (struct sockaddr *)sinp; 600 msg->connect.nm_flags |= PRUC_NAMALLOC; 601 tcp_connect(msg); 602 /* msg is invalid now */ 603 return; 604 } 605 inp->inp_vflag &= ~INP_IPV4; 606 inp->inp_vflag |= INP_IPV6; 607 inp->inp_inc.inc_isipv6 = 1; 608 609 msg->connect.nm_flags |= PRUC_FALLBACK; 610 tcp6_connect(msg); 611 /* msg is invalid now */ 612 return; 613 out: 614 if (msg->connect.nm_m) { 615 m_freem(msg->connect.nm_m); 616 msg->connect.nm_m = NULL; 617 } 618 lwkt_replymsg(&msg->lmsg, error); 619 } 620 621 #endif /* INET6 */ 622 623 /* 624 * Initiate disconnect from peer. 625 * If connection never passed embryonic stage, just drop; 626 * else if don't need to let data drain, then can just drop anyways, 627 * else have to begin TCP shutdown process: mark socket disconnecting, 628 * drain unread data, state switch to reflect user close, and 629 * send segment (e.g. FIN) to peer. Socket will be really disconnected 630 * when peer sends FIN and acks ours. 631 * 632 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB. 633 */ 634 static void 635 tcp_usr_disconnect(netmsg_t msg) 636 { 637 struct socket *so = msg->disconnect.base.nm_so; 638 int error = 0; 639 struct inpcb *inp; 640 struct tcpcb *tp; 641 642 COMMON_START(so, inp, 1); 643 tp = tcp_disconnect(tp); 644 COMMON_END(PRU_DISCONNECT); 645 } 646 647 /* 648 * Accept a connection. Essentially all the work is 649 * done at higher levels; just return the address 650 * of the peer, storing through addr. 651 */ 652 static void 653 tcp_usr_accept(netmsg_t msg) 654 { 655 struct socket *so = msg->accept.base.nm_so; 656 struct sockaddr **nam = msg->accept.nm_nam; 657 int error = 0; 658 struct inpcb *inp; 659 struct tcpcb *tp = NULL; 660 TCPDEBUG0; 661 662 inp = so->so_pcb; 663 if (so->so_state & SS_ISDISCONNECTED) { 664 error = ECONNABORTED; 665 goto out; 666 } 667 if (inp == 0) { 668 error = EINVAL; 669 goto out; 670 } 671 672 tp = intotcpcb(inp); 673 TCPDEBUG1(); 674 in_setpeeraddr(so, nam); 675 COMMON_END(PRU_ACCEPT); 676 } 677 678 #ifdef INET6 679 static void 680 tcp6_usr_accept(netmsg_t msg) 681 { 682 struct socket *so = msg->accept.base.nm_so; 683 struct sockaddr **nam = msg->accept.nm_nam; 684 int error = 0; 685 struct inpcb *inp; 686 struct tcpcb *tp = NULL; 687 TCPDEBUG0; 688 689 inp = so->so_pcb; 690 691 if (so->so_state & SS_ISDISCONNECTED) { 692 error = ECONNABORTED; 693 goto out; 694 } 695 if (inp == 0) { 696 error = EINVAL; 697 goto out; 698 } 699 tp = intotcpcb(inp); 700 TCPDEBUG1(); 701 in6_mapped_peeraddr(so, nam); 702 COMMON_END(PRU_ACCEPT); 703 } 704 #endif /* INET6 */ 705 /* 706 * Mark the connection as being incapable of further output. 707 */ 708 static void 709 tcp_usr_shutdown(netmsg_t msg) 710 { 711 struct socket *so = msg->shutdown.base.nm_so; 712 int error = 0; 713 struct inpcb *inp; 714 struct tcpcb *tp; 715 716 COMMON_START(so, inp, 0); 717 socantsendmore(so); 718 tp = tcp_usrclosed(tp); 719 if (tp) 720 error = tcp_output(tp); 721 COMMON_END(PRU_SHUTDOWN); 722 } 723 724 /* 725 * After a receive, possibly send window update to peer. 726 */ 727 static void 728 tcp_usr_rcvd(netmsg_t msg) 729 { 730 struct socket *so = msg->rcvd.base.nm_so; 731 int error = 0, noreply = 0; 732 struct inpcb *inp; 733 struct tcpcb *tp; 734 735 COMMON_START(so, inp, 0); 736 737 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) { 738 noreply = 1; 739 so_async_rcvd_reply(so); 740 } 741 tcp_output(tp); 742 743 COMMON_END1(PRU_RCVD, noreply); 744 } 745 746 /* 747 * Do a send by putting data in output queue and updating urgent 748 * marker if URG set. Possibly send more data. Unlike the other 749 * pru_*() routines, the mbuf chains are our responsibility. We 750 * must either enqueue them or free them. The other pru_* routines 751 * generally are caller-frees. 752 */ 753 static void 754 tcp_usr_send(netmsg_t msg) 755 { 756 struct socket *so = msg->send.base.nm_so; 757 int flags = msg->send.nm_flags; 758 struct mbuf *m = msg->send.nm_m; 759 int error = 0; 760 struct inpcb *inp; 761 struct tcpcb *tp; 762 TCPDEBUG0; 763 764 KKASSERT(msg->send.nm_control == NULL); 765 KKASSERT(msg->send.nm_addr == NULL); 766 KKASSERT((flags & PRUS_FREEADDR) == 0); 767 768 inp = so->so_pcb; 769 770 if (inp == NULL) { 771 /* 772 * OOPS! we lost a race, the TCP session got reset after 773 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a 774 * network interrupt in the non-critical section of sosend(). 775 */ 776 m_freem(m); 777 error = ECONNRESET; /* XXX EPIPE? */ 778 tp = NULL; 779 TCPDEBUG1(); 780 goto out; 781 } 782 tp = intotcpcb(inp); 783 TCPDEBUG1(); 784 785 #ifdef foo 786 /* 787 * This is no longer necessary, since: 788 * - sosendtcp() has already checked it for us 789 * - It does not work with asynchronized send 790 */ 791 792 /* 793 * Don't let too much OOB data build up 794 */ 795 if (flags & PRUS_OOB) { 796 if (ssb_space(&so->so_snd) < -512) { 797 m_freem(m); 798 error = ENOBUFS; 799 goto out; 800 } 801 } 802 #endif 803 804 /* 805 * Pump the data into the socket. 806 */ 807 if (m) { 808 ssb_appendstream(&so->so_snd, m); 809 sowwakeup(so); 810 } 811 if (flags & PRUS_OOB) { 812 /* 813 * According to RFC961 (Assigned Protocols), 814 * the urgent pointer points to the last octet 815 * of urgent data. We continue, however, 816 * to consider it to indicate the first octet 817 * of data past the urgent section. 818 * Otherwise, snd_up should be one lower. 819 */ 820 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 821 tp->t_flags |= TF_FORCE; 822 error = tcp_output(tp); 823 tp->t_flags &= ~TF_FORCE; 824 } else { 825 if (flags & PRUS_EOF) { 826 /* 827 * Close the send side of the connection after 828 * the data is sent. 829 */ 830 socantsendmore(so); 831 tp = tcp_usrclosed(tp); 832 } 833 if (tp != NULL && !tcp_output_pending(tp)) { 834 if (flags & PRUS_MORETOCOME) 835 tp->t_flags |= TF_MORETOCOME; 836 error = tcp_output_fair(tp); 837 if (flags & PRUS_MORETOCOME) 838 tp->t_flags &= ~TF_MORETOCOME; 839 } 840 } 841 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB : 842 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), 843 (flags & PRUS_NOREPLY)); 844 } 845 846 /* 847 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort() 848 * will sofree() it when we return. 849 */ 850 static void 851 tcp_usr_abort(netmsg_t msg) 852 { 853 struct socket *so = msg->abort.base.nm_so; 854 int error = 0; 855 struct inpcb *inp; 856 struct tcpcb *tp; 857 858 COMMON_START(so, inp, 1); 859 tp = tcp_drop(tp, ECONNABORTED); 860 COMMON_END(PRU_ABORT); 861 } 862 863 /* 864 * Receive out-of-band data. 865 */ 866 static void 867 tcp_usr_rcvoob(netmsg_t msg) 868 { 869 struct socket *so = msg->rcvoob.base.nm_so; 870 struct mbuf *m = msg->rcvoob.nm_m; 871 int flags = msg->rcvoob.nm_flags; 872 int error = 0; 873 struct inpcb *inp; 874 struct tcpcb *tp; 875 876 COMMON_START(so, inp, 0); 877 if ((so->so_oobmark == 0 && 878 (so->so_state & SS_RCVATMARK) == 0) || 879 so->so_options & SO_OOBINLINE || 880 tp->t_oobflags & TCPOOB_HADDATA) { 881 error = EINVAL; 882 goto out; 883 } 884 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) { 885 error = EWOULDBLOCK; 886 goto out; 887 } 888 m->m_len = 1; 889 *mtod(m, caddr_t) = tp->t_iobc; 890 if ((flags & MSG_PEEK) == 0) 891 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA); 892 COMMON_END(PRU_RCVOOB); 893 } 894 895 static void 896 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr) 897 { 898 in_savefaddr(so, faddr); 899 } 900 901 #ifdef INET6 902 static void 903 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr) 904 { 905 in6_mapped_savefaddr(so, faddr); 906 } 907 #endif 908 909 static int 910 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam, 911 struct thread *td __unused) 912 { 913 const struct sockaddr_in *sinp; 914 915 sinp = (const struct sockaddr_in *)nam; 916 if (sinp->sin_family == AF_INET && 917 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) 918 return EAFNOSUPPORT; 919 920 soisconnecting(so); 921 return 0; 922 } 923 924 /* xxx - should be const */ 925 struct pr_usrreqs tcp_usrreqs = { 926 .pru_abort = tcp_usr_abort, 927 .pru_accept = tcp_usr_accept, 928 .pru_attach = tcp_usr_attach, 929 .pru_bind = tcp_usr_bind, 930 .pru_connect = tcp_usr_connect, 931 .pru_connect2 = pr_generic_notsupp, 932 .pru_control = in_control_dispatch, 933 .pru_detach = tcp_usr_detach, 934 .pru_disconnect = tcp_usr_disconnect, 935 .pru_listen = tcp_usr_listen, 936 .pru_peeraddr = in_setpeeraddr_dispatch, 937 .pru_rcvd = tcp_usr_rcvd, 938 .pru_rcvoob = tcp_usr_rcvoob, 939 .pru_send = tcp_usr_send, 940 .pru_sense = pru_sense_null, 941 .pru_shutdown = tcp_usr_shutdown, 942 .pru_sockaddr = in_setsockaddr_dispatch, 943 .pru_sosend = sosendtcp, 944 .pru_soreceive = sorecvtcp, 945 .pru_savefaddr = tcp_usr_savefaddr, 946 .pru_preconnect = tcp_usr_preconnect 947 }; 948 949 #ifdef INET6 950 struct pr_usrreqs tcp6_usrreqs = { 951 .pru_abort = tcp_usr_abort, 952 .pru_accept = tcp6_usr_accept, 953 .pru_attach = tcp_usr_attach, 954 .pru_bind = tcp6_usr_bind, 955 .pru_connect = tcp6_usr_connect, 956 .pru_connect2 = pr_generic_notsupp, 957 .pru_control = in6_control_dispatch, 958 .pru_detach = tcp_usr_detach, 959 .pru_disconnect = tcp_usr_disconnect, 960 .pru_listen = tcp6_usr_listen, 961 .pru_peeraddr = in6_mapped_peeraddr_dispatch, 962 .pru_rcvd = tcp_usr_rcvd, 963 .pru_rcvoob = tcp_usr_rcvoob, 964 .pru_send = tcp_usr_send, 965 .pru_sense = pru_sense_null, 966 .pru_shutdown = tcp_usr_shutdown, 967 .pru_sockaddr = in6_mapped_sockaddr_dispatch, 968 .pru_sosend = sosendtcp, 969 .pru_soreceive = sorecvtcp, 970 .pru_savefaddr = tcp6_usr_savefaddr 971 }; 972 #endif /* INET6 */ 973 974 static int 975 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m, 976 struct sockaddr_in *sin, struct sockaddr_in *if_sin) 977 { 978 struct inpcb *inp = tp->t_inpcb, *oinp; 979 struct socket *so = inp->inp_socket; 980 struct route *ro = &inp->inp_route; 981 982 KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid], 983 ("pcbinfo mismatch")); 984 985 oinp = in_pcblookup_hash(inp->inp_pcbinfo, 986 sin->sin_addr, sin->sin_port, 987 (inp->inp_laddr.s_addr != INADDR_ANY ? 988 inp->inp_laddr : if_sin->sin_addr), 989 inp->inp_lport, 0, NULL); 990 if (oinp != NULL) { 991 m_freem(m); 992 return (EADDRINUSE); 993 } 994 if (inp->inp_laddr.s_addr == INADDR_ANY) 995 inp->inp_laddr = if_sin->sin_addr; 996 inp->inp_faddr = sin->sin_addr; 997 inp->inp_fport = sin->sin_port; 998 in_pcbinsconnhash(inp); 999 1000 /* 1001 * We are now on the inpcb's owner CPU, if the cached route was 1002 * freed because the rtentry's owner CPU is not the current CPU 1003 * (e.g. in tcp_connect()), then we try to reallocate it here with 1004 * the hope that a rtentry may be cloned from a RTF_PRCLONING 1005 * rtentry. 1006 */ 1007 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 1008 ro->ro_rt == NULL) { 1009 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 1010 ro->ro_dst.sa_family = AF_INET; 1011 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1012 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = 1013 sin->sin_addr; 1014 rtalloc(ro); 1015 } 1016 1017 /* 1018 * Now that no more errors can occur, change the protocol processing 1019 * port to the current thread (which is the correct thread). 1020 * 1021 * Create TCP timer message now; we are on the tcpcb's owner 1022 * CPU/thread. 1023 */ 1024 tcp_create_timermsg(tp, &curthread->td_msgport); 1025 1026 /* 1027 * Compute window scaling to request. Use a larger scaling then 1028 * needed for the initial receive buffer in case the receive buffer 1029 * gets expanded. 1030 */ 1031 if (tp->request_r_scale < TCP_MIN_WINSHIFT) 1032 tp->request_r_scale = TCP_MIN_WINSHIFT; 1033 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1034 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat 1035 ) { 1036 tp->request_r_scale++; 1037 } 1038 1039 soisconnecting(so); 1040 tcpstat.tcps_connattempt++; 1041 tp->t_state = TCPS_SYN_SENT; 1042 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 1043 tp->iss = tcp_new_isn(tp); 1044 tcp_sendseqinit(tp); 1045 if (m) { 1046 ssb_appendstream(&so->so_snd, m); 1047 m = NULL; 1048 if (flags & PRUS_OOB) 1049 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 1050 } 1051 1052 /* 1053 * Close the send side of the connection after 1054 * the data is sent if flagged. 1055 */ 1056 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) { 1057 socantsendmore(so); 1058 tp = tcp_usrclosed(tp); 1059 } 1060 return (tcp_output(tp)); 1061 } 1062 1063 /* 1064 * Common subroutine to open a TCP connection to remote host specified 1065 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local 1066 * port number if needed. Call in_pcbladdr to do the routing and to choose 1067 * a local host address (interface). 1068 * Initialize connection parameters and enter SYN-SENT state. 1069 */ 1070 static void 1071 tcp_connect(netmsg_t msg) 1072 { 1073 struct socket *so = msg->connect.base.nm_so; 1074 struct sockaddr *nam = msg->connect.nm_nam; 1075 struct thread *td = msg->connect.nm_td; 1076 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1077 struct sockaddr_in *if_sin = NULL; 1078 struct inpcb *inp; 1079 struct tcpcb *tp; 1080 int error; 1081 lwkt_port_t port; 1082 1083 COMMON_START(so, inp, 0); 1084 1085 /* 1086 * Reconnect our pcb if we have to 1087 */ 1088 if (msg->connect.nm_flags & PRUC_RECONNECT) { 1089 msg->connect.nm_flags &= ~PRUC_RECONNECT; 1090 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1091 } 1092 1093 /* 1094 * Bind if we have to 1095 */ 1096 if (inp->inp_lport == 0) { 1097 if (tcp_lport_extension) { 1098 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY); 1099 1100 error = in_pcbladdr(inp, nam, &if_sin, td); 1101 if (error) 1102 goto out; 1103 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr; 1104 1105 error = in_pcbbind_remote(inp, nam, td); 1106 if (error) 1107 goto out; 1108 1109 msg->connect.nm_flags |= PRUC_HASLADDR; 1110 } else { 1111 error = in_pcbbind(inp, NULL, td); 1112 if (error) 1113 goto out; 1114 } 1115 } 1116 1117 if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) { 1118 /* 1119 * Calculate the correct protocol processing thread. The 1120 * connect operation must run there. Set the forwarding 1121 * port before we forward the message or it will get bounced 1122 * right back to us. 1123 */ 1124 error = in_pcbladdr(inp, nam, &if_sin, td); 1125 if (error) 1126 goto out; 1127 } 1128 KKASSERT(inp->inp_socket == so); 1129 1130 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port, 1131 (inp->inp_laddr.s_addr != INADDR_ANY ? 1132 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr), 1133 inp->inp_lport); 1134 1135 if (port != &curthread->td_msgport) { 1136 struct route *ro = &inp->inp_route; 1137 1138 /* 1139 * in_pcbladdr() may have allocated a route entry for us 1140 * on the current CPU, but we need a route entry on the 1141 * inpcb's owner CPU, so free it here. 1142 */ 1143 if (ro->ro_rt != NULL) 1144 RTFREE(ro->ro_rt); 1145 bzero(ro, sizeof(*ro)); 1146 1147 /* 1148 * We are moving the protocol processing port the socket 1149 * is on, we have to unlink here and re-link on the 1150 * target cpu. 1151 */ 1152 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1153 sosetport(so, port); 1154 msg->connect.nm_flags |= PRUC_RECONNECT; 1155 msg->connect.base.nm_dispatch = tcp_connect; 1156 1157 lwkt_forwardmsg(port, &msg->connect.base.lmsg); 1158 /* msg invalid now */ 1159 return; 1160 } else if (msg->connect.nm_flags & PRUC_HELDTD) { 1161 /* 1162 * The original thread is no longer needed; release it. 1163 */ 1164 lwkt_rele(td); 1165 msg->connect.nm_flags &= ~PRUC_HELDTD; 1166 } 1167 error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags, 1168 msg->connect.nm_m, sin, if_sin); 1169 msg->connect.nm_m = NULL; 1170 out: 1171 if (msg->connect.nm_m) { 1172 m_freem(msg->connect.nm_m); 1173 msg->connect.nm_m = NULL; 1174 } 1175 if (msg->connect.nm_flags & PRUC_NAMALLOC) { 1176 kfree(msg->connect.nm_nam, M_LWKTMSG); 1177 msg->connect.nm_nam = NULL; 1178 } 1179 if (msg->connect.nm_flags & PRUC_HELDTD) 1180 lwkt_rele(td); 1181 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) { 1182 so->so_error = error; 1183 soisdisconnected(so); 1184 } 1185 lwkt_replymsg(&msg->connect.base.lmsg, error); 1186 /* msg invalid now */ 1187 } 1188 1189 #ifdef INET6 1190 1191 static void 1192 tcp6_connect(netmsg_t msg) 1193 { 1194 struct tcpcb *tp; 1195 struct socket *so = msg->connect.base.nm_so; 1196 struct sockaddr *nam = msg->connect.nm_nam; 1197 struct thread *td = msg->connect.nm_td; 1198 struct inpcb *inp; 1199 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 1200 struct in6_addr *addr6; 1201 lwkt_port_t port; 1202 int error; 1203 1204 COMMON_START(so, inp, 0); 1205 1206 /* 1207 * Reconnect our pcb if we have to 1208 */ 1209 if (msg->connect.nm_flags & PRUC_RECONNECT) { 1210 msg->connect.nm_flags &= ~PRUC_RECONNECT; 1211 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1212 } 1213 1214 /* 1215 * Bind if we have to 1216 */ 1217 if (inp->inp_lport == 0) { 1218 error = in6_pcbbind(inp, NULL, td); 1219 if (error) 1220 goto out; 1221 } 1222 1223 /* 1224 * Cannot simply call in_pcbconnect, because there might be an 1225 * earlier incarnation of this same connection still in 1226 * TIME_WAIT state, creating an ADDRINUSE error. 1227 */ 1228 error = in6_pcbladdr(inp, nam, &addr6, td); 1229 if (error) 1230 goto out; 1231 1232 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */ 1233 1234 if (port != &curthread->td_msgport) { 1235 struct route *ro = &inp->inp_route; 1236 1237 /* 1238 * in_pcbladdr() may have allocated a route entry for us 1239 * on the current CPU, but we need a route entry on the 1240 * inpcb's owner CPU, so free it here. 1241 */ 1242 if (ro->ro_rt != NULL) 1243 RTFREE(ro->ro_rt); 1244 bzero(ro, sizeof(*ro)); 1245 1246 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]); 1247 sosetport(so, port); 1248 msg->connect.nm_flags |= PRUC_RECONNECT; 1249 msg->connect.base.nm_dispatch = tcp6_connect; 1250 1251 lwkt_forwardmsg(port, &msg->connect.base.lmsg); 1252 /* msg invalid now */ 1253 return; 1254 } 1255 error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags, 1256 &msg->connect.nm_m, sin6, addr6); 1257 /* nm_m may still be intact */ 1258 out: 1259 if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) { 1260 tcp_connect(msg); 1261 /* msg invalid now */ 1262 } else { 1263 if (msg->connect.nm_m) { 1264 m_freem(msg->connect.nm_m); 1265 msg->connect.nm_m = NULL; 1266 } 1267 if (msg->connect.nm_flags & PRUC_NAMALLOC) { 1268 kfree(msg->connect.nm_nam, M_LWKTMSG); 1269 msg->connect.nm_nam = NULL; 1270 } 1271 lwkt_replymsg(&msg->connect.base.lmsg, error); 1272 /* msg invalid now */ 1273 } 1274 } 1275 1276 static int 1277 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp, 1278 struct sockaddr_in6 *sin6, struct in6_addr *addr6) 1279 { 1280 struct mbuf *m = *mp; 1281 struct inpcb *inp = tp->t_inpcb; 1282 struct socket *so = inp->inp_socket; 1283 struct inpcb *oinp; 1284 1285 /* 1286 * Cannot simply call in_pcbconnect, because there might be an 1287 * earlier incarnation of this same connection still in 1288 * TIME_WAIT state, creating an ADDRINUSE error. 1289 */ 1290 oinp = in6_pcblookup_hash(inp->inp_pcbinfo, 1291 &sin6->sin6_addr, sin6->sin6_port, 1292 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ? 1293 addr6 : &inp->in6p_laddr), 1294 inp->inp_lport, 0, NULL); 1295 if (oinp) 1296 return (EADDRINUSE); 1297 1298 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1299 inp->in6p_laddr = *addr6; 1300 inp->in6p_faddr = sin6->sin6_addr; 1301 inp->inp_fport = sin6->sin6_port; 1302 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0) 1303 inp->in6p_flowinfo = sin6->sin6_flowinfo; 1304 in_pcbinsconnhash(inp); 1305 1306 /* 1307 * Now that no more errors can occur, change the protocol processing 1308 * port to the current thread (which is the correct thread). 1309 * 1310 * Create TCP timer message now; we are on the tcpcb's owner 1311 * CPU/thread. 1312 */ 1313 tcp_create_timermsg(tp, &curthread->td_msgport); 1314 1315 /* Compute window scaling to request. */ 1316 if (tp->request_r_scale < TCP_MIN_WINSHIFT) 1317 tp->request_r_scale = TCP_MIN_WINSHIFT; 1318 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 1319 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) { 1320 tp->request_r_scale++; 1321 } 1322 1323 soisconnecting(so); 1324 tcpstat.tcps_connattempt++; 1325 tp->t_state = TCPS_SYN_SENT; 1326 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 1327 tp->iss = tcp_new_isn(tp); 1328 tcp_sendseqinit(tp); 1329 if (m) { 1330 ssb_appendstream(&so->so_snd, m); 1331 *mp = NULL; 1332 if (flags & PRUS_OOB) 1333 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc; 1334 } 1335 1336 /* 1337 * Close the send side of the connection after 1338 * the data is sent if flagged. 1339 */ 1340 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) { 1341 socantsendmore(so); 1342 tp = tcp_usrclosed(tp); 1343 } 1344 return (tcp_output(tp)); 1345 } 1346 1347 #endif /* INET6 */ 1348 1349 /* 1350 * The new sockopt interface makes it possible for us to block in the 1351 * copyin/out step (if we take a page fault). Taking a page fault while 1352 * in a critical section is probably a Bad Thing. (Since sockets and pcbs 1353 * both now use TSM, there probably isn't any need for this function to 1354 * run in a critical section any more. This needs more examination.) 1355 */ 1356 void 1357 tcp_ctloutput(netmsg_t msg) 1358 { 1359 struct socket *so = msg->base.nm_so; 1360 struct sockopt *sopt = msg->ctloutput.nm_sopt; 1361 int error, opt, optval, opthz; 1362 struct inpcb *inp; 1363 struct tcpcb *tp; 1364 1365 error = 0; 1366 inp = so->so_pcb; 1367 if (inp == NULL) { 1368 error = ECONNRESET; 1369 goto done; 1370 } 1371 1372 if (sopt->sopt_level != IPPROTO_TCP) { 1373 switch (sopt->sopt_name) { 1374 case IP_MULTICAST_IF: 1375 case IP_MULTICAST_VIF: 1376 case IP_MULTICAST_TTL: 1377 case IP_MULTICAST_LOOP: 1378 case IP_ADD_MEMBERSHIP: 1379 case IP_DROP_MEMBERSHIP: 1380 /* 1381 * Multicast does not make sense on TCP sockets. 1382 */ 1383 error = EOPNOTSUPP; 1384 goto done; 1385 } 1386 #ifdef INET6 1387 if (INP_CHECK_SOCKAF(so, AF_INET6)) 1388 ip6_ctloutput_dispatch(msg); 1389 else 1390 #endif /* INET6 */ 1391 ip_ctloutput(msg); 1392 /* msg invalid now */ 1393 return; 1394 } 1395 tp = intotcpcb(inp); 1396 1397 switch (sopt->sopt_dir) { 1398 case SOPT_SET: 1399 error = soopt_to_kbuf(sopt, &optval, sizeof optval, 1400 sizeof optval); 1401 if (error) 1402 break; 1403 switch (sopt->sopt_name) { 1404 case TCP_FASTKEEP: 1405 if (optval > 0) 1406 tp->t_keepidle = tp->t_keepintvl; 1407 else 1408 tp->t_keepidle = tcp_keepidle; 1409 tcp_timer_keep_activity(tp, 0); 1410 break; 1411 #ifdef TCP_SIGNATURE 1412 case TCP_SIGNATURE_ENABLE: 1413 if (tp->t_state == TCPS_CLOSED) { 1414 /* 1415 * This is the only safe state that this 1416 * option could be changed. Some segments 1417 * could already have been sent in other 1418 * states. 1419 */ 1420 if (optval > 0) 1421 tp->t_flags |= TF_SIGNATURE; 1422 else 1423 tp->t_flags &= ~TF_SIGNATURE; 1424 } else { 1425 error = EOPNOTSUPP; 1426 } 1427 break; 1428 #endif /* TCP_SIGNATURE */ 1429 case TCP_NODELAY: 1430 case TCP_NOOPT: 1431 switch (sopt->sopt_name) { 1432 case TCP_NODELAY: 1433 opt = TF_NODELAY; 1434 break; 1435 case TCP_NOOPT: 1436 opt = TF_NOOPT; 1437 break; 1438 default: 1439 opt = 0; /* dead code to fool gcc */ 1440 break; 1441 } 1442 1443 if (optval) 1444 tp->t_flags |= opt; 1445 else 1446 tp->t_flags &= ~opt; 1447 break; 1448 1449 case TCP_NOPUSH: 1450 if (tcp_disable_nopush) 1451 break; 1452 if (optval) 1453 tp->t_flags |= TF_NOPUSH; 1454 else { 1455 tp->t_flags &= ~TF_NOPUSH; 1456 error = tcp_output(tp); 1457 } 1458 break; 1459 1460 case TCP_MAXSEG: 1461 /* 1462 * Must be between 0 and maxseg. If the requested 1463 * maxseg is too small to satisfy the desired minmss, 1464 * pump it up (silently so sysctl modifications of 1465 * minmss do not create unexpected program failures). 1466 * Handle degenerate cases. 1467 */ 1468 if (optval > 0 && optval <= tp->t_maxseg) { 1469 if (optval + 40 < tcp_minmss) { 1470 optval = tcp_minmss - 40; 1471 if (optval < 0) 1472 optval = 1; 1473 } 1474 tp->t_maxseg = optval; 1475 } else { 1476 error = EINVAL; 1477 } 1478 break; 1479 1480 case TCP_KEEPINIT: 1481 opthz = ((int64_t)optval * hz) / 1000; 1482 if (opthz >= 1) 1483 tp->t_keepinit = opthz; 1484 else 1485 error = EINVAL; 1486 break; 1487 1488 case TCP_KEEPIDLE: 1489 opthz = ((int64_t)optval * hz) / 1000; 1490 if (opthz >= 1) { 1491 tp->t_keepidle = opthz; 1492 tcp_timer_keep_activity(tp, 0); 1493 } else { 1494 error = EINVAL; 1495 } 1496 break; 1497 1498 case TCP_KEEPINTVL: 1499 opthz = ((int64_t)optval * hz) / 1000; 1500 if (opthz >= 1) { 1501 tp->t_keepintvl = opthz; 1502 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 1503 } else { 1504 error = EINVAL; 1505 } 1506 break; 1507 1508 case TCP_KEEPCNT: 1509 if (optval > 0) { 1510 tp->t_keepcnt = optval; 1511 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 1512 } else { 1513 error = EINVAL; 1514 } 1515 break; 1516 1517 default: 1518 error = ENOPROTOOPT; 1519 break; 1520 } 1521 break; 1522 1523 case SOPT_GET: 1524 switch (sopt->sopt_name) { 1525 #ifdef TCP_SIGNATURE 1526 case TCP_SIGNATURE_ENABLE: 1527 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0; 1528 break; 1529 #endif /* TCP_SIGNATURE */ 1530 case TCP_NODELAY: 1531 optval = tp->t_flags & TF_NODELAY; 1532 break; 1533 case TCP_MAXSEG: 1534 optval = tp->t_maxseg; 1535 break; 1536 case TCP_NOOPT: 1537 optval = tp->t_flags & TF_NOOPT; 1538 break; 1539 case TCP_NOPUSH: 1540 optval = tp->t_flags & TF_NOPUSH; 1541 break; 1542 case TCP_KEEPINIT: 1543 optval = ((int64_t)tp->t_keepinit * 1000) / hz; 1544 break; 1545 case TCP_KEEPIDLE: 1546 optval = ((int64_t)tp->t_keepidle * 1000) / hz; 1547 break; 1548 case TCP_KEEPINTVL: 1549 optval = ((int64_t)tp->t_keepintvl * 1000) / hz; 1550 break; 1551 case TCP_KEEPCNT: 1552 optval = tp->t_keepcnt; 1553 break; 1554 default: 1555 error = ENOPROTOOPT; 1556 break; 1557 } 1558 if (error == 0) 1559 soopt_from_kbuf(sopt, &optval, sizeof optval); 1560 break; 1561 } 1562 done: 1563 lwkt_replymsg(&msg->lmsg, error); 1564 } 1565 1566 /* 1567 * tcp_sendspace and tcp_recvspace are the default send and receive window 1568 * sizes, respectively. These are obsolescent (this information should 1569 * be set by the route). 1570 * 1571 * Use a default that does not require tcp window scaling to be turned 1572 * on. Individual programs or the administrator can increase the default. 1573 */ 1574 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1575 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, 1576 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); 1577 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */ 1578 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1579 &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); 1580 1581 /* 1582 * Attach TCP protocol to socket, allocating internet protocol control 1583 * block, tcp control block, buffer space, and entering CLOSED state. 1584 */ 1585 static int 1586 tcp_attach(struct socket *so, struct pru_attach_info *ai) 1587 { 1588 struct tcpcb *tp; 1589 struct inpcb *inp; 1590 int error; 1591 int cpu; 1592 #ifdef INET6 1593 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; 1594 #endif 1595 1596 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 1597 lwkt_gettoken(&so->so_rcv.ssb_token); 1598 error = soreserve(so, tcp_sendspace, tcp_recvspace, 1599 ai->sb_rlimit); 1600 lwkt_reltoken(&so->so_rcv.ssb_token); 1601 if (error) 1602 return (error); 1603 } 1604 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC); 1605 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC); 1606 cpu = mycpu->gd_cpuid; 1607 1608 /* 1609 * Set the default port for protocol processing. This will likely 1610 * change when we connect. 1611 */ 1612 error = in_pcballoc(so, &tcbinfo[cpu]); 1613 if (error) 1614 return (error); 1615 inp = so->so_pcb; 1616 #ifdef INET6 1617 if (isipv6) { 1618 inp->inp_vflag |= INP_IPV6; 1619 inp->in6p_hops = -1; /* use kernel default */ 1620 } 1621 else 1622 #endif 1623 inp->inp_vflag |= INP_IPV4; 1624 tp = tcp_newtcpcb(inp); 1625 if (tp == NULL) { 1626 /* 1627 * Make sure the socket is destroyed by the pcbdetach. 1628 */ 1629 soreference(so); 1630 #ifdef INET6 1631 if (isipv6) 1632 in6_pcbdetach(inp); 1633 else 1634 #endif 1635 in_pcbdetach(inp); 1636 sofree(so); /* from ref above */ 1637 return (ENOBUFS); 1638 } 1639 tp->t_state = TCPS_CLOSED; 1640 /* Keep a reference for asynchronized pru_rcvd */ 1641 soreference(so); 1642 return (0); 1643 } 1644 1645 /* 1646 * Initiate (or continue) disconnect. 1647 * If embryonic state, just send reset (once). 1648 * If in ``let data drain'' option and linger null, just drop. 1649 * Otherwise (hard), mark socket disconnecting and drop 1650 * current input data; switch states based on user close, and 1651 * send segment to peer (with FIN). 1652 */ 1653 static struct tcpcb * 1654 tcp_disconnect(struct tcpcb *tp) 1655 { 1656 struct socket *so = tp->t_inpcb->inp_socket; 1657 1658 if (tp->t_state < TCPS_ESTABLISHED) { 1659 tp = tcp_close(tp); 1660 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) { 1661 tp = tcp_drop(tp, 0); 1662 } else { 1663 lwkt_gettoken(&so->so_rcv.ssb_token); 1664 soisdisconnecting(so); 1665 sbflush(&so->so_rcv.sb); 1666 tp = tcp_usrclosed(tp); 1667 if (tp) 1668 tcp_output(tp); 1669 lwkt_reltoken(&so->so_rcv.ssb_token); 1670 } 1671 return (tp); 1672 } 1673 1674 /* 1675 * User issued close, and wish to trail through shutdown states: 1676 * if never received SYN, just forget it. If got a SYN from peer, 1677 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN. 1678 * If already got a FIN from peer, then almost done; go to LAST_ACK 1679 * state. In all other cases, have already sent FIN to peer (e.g. 1680 * after PRU_SHUTDOWN), and just have to play tedious game waiting 1681 * for peer to send FIN or not respond to keep-alives, etc. 1682 * We can let the user exit from the close as soon as the FIN is acked. 1683 */ 1684 static struct tcpcb * 1685 tcp_usrclosed(struct tcpcb *tp) 1686 { 1687 1688 switch (tp->t_state) { 1689 1690 case TCPS_CLOSED: 1691 case TCPS_LISTEN: 1692 tp->t_state = TCPS_CLOSED; 1693 tp = tcp_close(tp); 1694 break; 1695 1696 case TCPS_SYN_SENT: 1697 case TCPS_SYN_RECEIVED: 1698 tp->t_flags |= TF_NEEDFIN; 1699 break; 1700 1701 case TCPS_ESTABLISHED: 1702 tp->t_state = TCPS_FIN_WAIT_1; 1703 break; 1704 1705 case TCPS_CLOSE_WAIT: 1706 tp->t_state = TCPS_LAST_ACK; 1707 break; 1708 } 1709 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) { 1710 soisdisconnected(tp->t_inpcb->inp_socket); 1711 /* To prevent the connection hanging in FIN_WAIT_2 forever. */ 1712 if (tp->t_state == TCPS_FIN_WAIT_2) { 1713 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle, 1714 tcp_timer_2msl); 1715 } 1716 } 1717 return (tp); 1718 } 1719