1 /* dnsmasq is Copyright (c) 2000-2009 Simon Kelley 2 3 This program is free software; you can redistribute it and/or modify 4 it under the terms of the GNU General Public License as published by 5 the Free Software Foundation; version 2 dated June, 1991, or 6 (at your option) version 3 dated 29 June, 2007. 7 8 This program is distributed in the hope that it will be useful, 9 but WITHOUT ANY WARRANTY; without even the implied warranty of 10 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11 GNU General Public License for more details. 12 13 You should have received a copy of the GNU General Public License 14 along with this program. If not, see <http://www.gnu.org/licenses/>. 15 */ 16 17 #include "dnsmasq.h" 18 19 static struct frec *lookup_frec(unsigned short id, unsigned int crc); 20 static struct frec *lookup_frec_by_sender(unsigned short id, 21 union mysockaddr *addr, 22 unsigned int crc); 23 static unsigned short get_id(int force, unsigned short force_id, unsigned int crc); 24 static void free_frec(struct frec *f); 25 static struct randfd *allocate_rfd(int family); 26 27 /* Send a UDP packet with its source address set as "source" 28 unless nowild is true, when we just send it with the kernel default */ 29 static void send_from(int fd, int nowild, char *packet, size_t len, 30 union mysockaddr *to, struct all_addr *source, 31 unsigned int iface) 32 { 33 struct msghdr msg; 34 struct iovec iov[1]; 35 union { 36 struct cmsghdr align; /* this ensures alignment */ 37 #if defined(HAVE_LINUX_NETWORK) 38 char control[CMSG_SPACE(sizeof(struct in_pktinfo))]; 39 #elif defined(IP_SENDSRCADDR) 40 char control[CMSG_SPACE(sizeof(struct in_addr))]; 41 #endif 42 #ifdef HAVE_IPV6 43 char control6[CMSG_SPACE(sizeof(struct in6_pktinfo))]; 44 #endif 45 } control_u; 46 47 iov[0].iov_base = packet; 48 iov[0].iov_len = len; 49 50 msg.msg_control = NULL; 51 msg.msg_controllen = 0; 52 msg.msg_flags = 0; 53 msg.msg_name = to; 54 msg.msg_namelen = sa_len(to); 55 msg.msg_iov = iov; 56 msg.msg_iovlen = 1; 57 58 if (!nowild) 59 { 60 struct cmsghdr *cmptr; 61 msg.msg_control = &control_u; 62 msg.msg_controllen = sizeof(control_u); 63 cmptr = CMSG_FIRSTHDR(&msg); 64 65 if (to->sa.sa_family == AF_INET) 66 { 67 #if defined(HAVE_LINUX_NETWORK) 68 struct in_pktinfo *pkt = (struct in_pktinfo *)CMSG_DATA(cmptr); 69 pkt->ipi_ifindex = 0; 70 pkt->ipi_spec_dst = source->addr.addr4; 71 msg.msg_controllen = cmptr->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo)); 72 cmptr->cmsg_level = SOL_IP; 73 cmptr->cmsg_type = IP_PKTINFO; 74 #elif defined(IP_SENDSRCADDR) 75 struct in_addr *a = (struct in_addr *)CMSG_DATA(cmptr); 76 *a = source->addr.addr4; 77 msg.msg_controllen = cmptr->cmsg_len = CMSG_LEN(sizeof(struct in_addr)); 78 cmptr->cmsg_level = IPPROTO_IP; 79 cmptr->cmsg_type = IP_SENDSRCADDR; 80 #endif 81 } 82 else 83 #ifdef HAVE_IPV6 84 { 85 struct in6_pktinfo *pkt = (struct in6_pktinfo *)CMSG_DATA(cmptr); 86 pkt->ipi6_ifindex = iface; /* Need iface for IPv6 to handle link-local addrs */ 87 pkt->ipi6_addr = source->addr.addr6; 88 msg.msg_controllen = cmptr->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo)); 89 cmptr->cmsg_type = IPV6_PKTINFO; 90 cmptr->cmsg_level = IPV6_LEVEL; 91 } 92 #else 93 iface = 0; /* eliminate warning */ 94 #endif 95 } 96 97 retry: 98 if (sendmsg(fd, &msg, 0) == -1) 99 { 100 /* certain Linux kernels seem to object to setting the source address in the IPv6 stack 101 by returning EINVAL from sendmsg. In that case, try again without setting the 102 source address, since it will nearly alway be correct anyway. IPv6 stinks. */ 103 if (errno == EINVAL && msg.msg_controllen) 104 { 105 msg.msg_controllen = 0; 106 goto retry; 107 } 108 if (retry_send()) 109 goto retry; 110 } 111 } 112 113 static unsigned short search_servers(time_t now, struct all_addr **addrpp, 114 unsigned short qtype, char *qdomain, int *type, char **domain) 115 116 { 117 /* If the query ends in the domain in one of our servers, set 118 domain to point to that name. We find the largest match to allow both 119 domain.org and sub.domain.org to exist. */ 120 121 unsigned int namelen = strlen(qdomain); 122 unsigned int matchlen = 0; 123 struct server *serv; 124 unsigned short flags = 0; 125 126 for (serv = daemon->servers; serv; serv=serv->next) 127 /* domain matches take priority over NODOTS matches */ 128 if ((serv->flags & SERV_FOR_NODOTS) && *type != SERV_HAS_DOMAIN && !strchr(qdomain, '.') && namelen != 0) 129 { 130 unsigned short sflag = serv->addr.sa.sa_family == AF_INET ? F_IPV4 : F_IPV6; 131 *type = SERV_FOR_NODOTS; 132 if (serv->flags & SERV_NO_ADDR) 133 flags = F_NXDOMAIN; 134 else if (serv->flags & SERV_LITERAL_ADDRESS) 135 { 136 if (sflag & qtype) 137 { 138 flags = sflag; 139 if (serv->addr.sa.sa_family == AF_INET) 140 *addrpp = (struct all_addr *)&serv->addr.in.sin_addr; 141 #ifdef HAVE_IPV6 142 else 143 *addrpp = (struct all_addr *)&serv->addr.in6.sin6_addr; 144 #endif 145 } 146 else if (!flags || (flags & F_NXDOMAIN)) 147 flags = F_NOERR; 148 } 149 } 150 else if (serv->flags & SERV_HAS_DOMAIN) 151 { 152 unsigned int domainlen = strlen(serv->domain); 153 char *matchstart = qdomain + namelen - domainlen; 154 if (namelen >= domainlen && 155 hostname_isequal(matchstart, serv->domain) && 156 domainlen >= matchlen && 157 (domainlen == 0 || namelen == domainlen || *(serv->domain) == '.' || *(matchstart-1) == '.' )) 158 { 159 unsigned short sflag = serv->addr.sa.sa_family == AF_INET ? F_IPV4 : F_IPV6; 160 *type = SERV_HAS_DOMAIN; 161 *domain = serv->domain; 162 matchlen = domainlen; 163 if (serv->flags & SERV_NO_ADDR) 164 flags = F_NXDOMAIN; 165 else if (serv->flags & SERV_LITERAL_ADDRESS) 166 { 167 if (sflag & qtype) 168 { 169 flags = sflag; 170 if (serv->addr.sa.sa_family == AF_INET) 171 *addrpp = (struct all_addr *)&serv->addr.in.sin_addr; 172 #ifdef HAVE_IPV6 173 else 174 *addrpp = (struct all_addr *)&serv->addr.in6.sin6_addr; 175 #endif 176 } 177 else if (!flags || (flags & F_NXDOMAIN)) 178 flags = F_NOERR; 179 } 180 } 181 } 182 183 if (flags == 0 && !(qtype & F_BIGNAME) && 184 (daemon->options & OPT_NODOTS_LOCAL) && !strchr(qdomain, '.') && namelen != 0) 185 /* don't forward simple names, make exception for NS queries and empty name. */ 186 flags = F_NXDOMAIN; 187 188 if (flags == F_NXDOMAIN && check_for_local_domain(qdomain, now)) 189 flags = F_NOERR; 190 191 if (flags) 192 { 193 int logflags = 0; 194 195 if (flags == F_NXDOMAIN || flags == F_NOERR) 196 logflags = F_NEG | qtype; 197 198 log_query(logflags | flags | F_CONFIG | F_FORWARD, qdomain, *addrpp, NULL); 199 } 200 201 return flags; 202 } 203 204 static int forward_query(int udpfd, union mysockaddr *udpaddr, 205 struct all_addr *dst_addr, unsigned int dst_iface, 206 HEADER *header, size_t plen, time_t now, struct frec *forward) 207 { 208 char *domain = NULL; 209 int type = 0; 210 struct all_addr *addrp = NULL; 211 unsigned int crc = questions_crc(header, plen, daemon->namebuff); 212 unsigned short flags = 0; 213 unsigned short gotname = extract_request(header, plen, daemon->namebuff, NULL); 214 struct server *start = NULL; 215 216 /* may be no servers available. */ 217 if (!daemon->servers) 218 forward = NULL; 219 else if (forward || (forward = lookup_frec_by_sender(ntohs(header->id), udpaddr, crc))) 220 { 221 /* retry on existing query, send to all available servers */ 222 domain = forward->sentto->domain; 223 forward->sentto->failed_queries++; 224 if (!(daemon->options & OPT_ORDER)) 225 { 226 forward->forwardall = 1; 227 daemon->last_server = NULL; 228 } 229 type = forward->sentto->flags & SERV_TYPE; 230 if (!(start = forward->sentto->next)) 231 start = daemon->servers; /* at end of list, recycle */ 232 header->id = htons(forward->new_id); 233 } 234 else 235 { 236 if (gotname) 237 flags = search_servers(now, &addrp, gotname, daemon->namebuff, &type, &domain); 238 239 if (!flags && !(forward = get_new_frec(now, NULL))) 240 /* table full - server failure. */ 241 flags = F_NEG; 242 243 if (forward) 244 { 245 /* force unchanging id for signed packets */ 246 int is_sign; 247 find_pseudoheader(header, plen, NULL, NULL, &is_sign); 248 249 forward->source = *udpaddr; 250 forward->dest = *dst_addr; 251 forward->iface = dst_iface; 252 forward->orig_id = ntohs(header->id); 253 forward->new_id = get_id(is_sign, forward->orig_id, crc); 254 forward->fd = udpfd; 255 forward->crc = crc; 256 forward->forwardall = 0; 257 header->id = htons(forward->new_id); 258 259 /* In strict_order mode, or when using domain specific servers 260 always try servers in the order specified in resolv.conf, 261 otherwise, use the one last known to work. */ 262 263 if (type != 0 || (daemon->options & OPT_ORDER)) 264 start = daemon->servers; 265 else if (!(start = daemon->last_server) || 266 daemon->forwardcount++ > FORWARD_TEST || 267 difftime(now, daemon->forwardtime) > FORWARD_TIME) 268 { 269 start = daemon->servers; 270 forward->forwardall = 1; 271 daemon->forwardcount = 0; 272 daemon->forwardtime = now; 273 } 274 } 275 } 276 277 /* check for send errors here (no route to host) 278 if we fail to send to all nameservers, send back an error 279 packet straight away (helps modem users when offline) */ 280 281 if (!flags && forward) 282 { 283 struct server *firstsentto = start; 284 int forwarded = 0; 285 286 while (1) 287 { 288 /* only send to servers dealing with our domain. 289 domain may be NULL, in which case server->domain 290 must be NULL also. */ 291 292 if (type == (start->flags & SERV_TYPE) && 293 (type != SERV_HAS_DOMAIN || hostname_isequal(domain, start->domain)) && 294 !(start->flags & SERV_LITERAL_ADDRESS)) 295 { 296 int fd; 297 298 /* find server socket to use, may need to get random one. */ 299 if (start->sfd) 300 fd = start->sfd->fd; 301 else 302 { 303 #ifdef HAVE_IPV6 304 if (start->addr.sa.sa_family == AF_INET6) 305 { 306 if (!forward->rfd6 && 307 !(forward->rfd6 = allocate_rfd(AF_INET6))) 308 break; 309 daemon->rfd_save = forward->rfd6; 310 fd = forward->rfd6->fd; 311 } 312 else 313 #endif 314 { 315 if (!forward->rfd4 && 316 !(forward->rfd4 = allocate_rfd(AF_INET))) 317 break; 318 daemon->rfd_save = forward->rfd4; 319 fd = forward->rfd4->fd; 320 } 321 322 #ifdef ANDROID 323 // Mark the socket so it goes out on the correct network. Note 324 // that we never clear the mark, only re-set it the next time we 325 // allocate a new random fd. This is because we buffer DNS 326 // queries (in daemon->srv_save, daemon->packet_len) and socket 327 // file descriptors (in daemon->rfd_save) with the expectation of 328 // being able to use them again. 329 // 330 // Server fds are marked separately in allocate_sfd. 331 setsockopt(fd, SOL_SOCKET, SO_MARK, &start->mark, sizeof(start->mark)); 332 #endif 333 } 334 335 if (sendto(fd, (char *)header, plen, 0, 336 &start->addr.sa, 337 sa_len(&start->addr)) == -1) 338 { 339 if (retry_send()) 340 continue; 341 } 342 else 343 { 344 /* Keep info in case we want to re-send this packet */ 345 daemon->srv_save = start; 346 daemon->packet_len = plen; 347 348 if (!gotname) 349 strcpy(daemon->namebuff, "query"); 350 if (start->addr.sa.sa_family == AF_INET) 351 log_query(F_SERVER | F_IPV4 | F_FORWARD, daemon->namebuff, 352 (struct all_addr *)&start->addr.in.sin_addr, NULL); 353 #ifdef HAVE_IPV6 354 else 355 log_query(F_SERVER | F_IPV6 | F_FORWARD, daemon->namebuff, 356 (struct all_addr *)&start->addr.in6.sin6_addr, NULL); 357 #endif 358 start->queries++; 359 forwarded = 1; 360 forward->sentto = start; 361 if (!forward->forwardall) 362 break; 363 forward->forwardall++; 364 } 365 } 366 367 if (!(start = start->next)) 368 start = daemon->servers; 369 370 if (start == firstsentto) 371 break; 372 } 373 374 if (forwarded) 375 return 1; 376 377 /* could not send on, prepare to return */ 378 header->id = htons(forward->orig_id); 379 free_frec(forward); /* cancel */ 380 } 381 382 /* could not send on, return empty answer or address if known for whole domain */ 383 if (udpfd != -1) 384 { 385 plen = setup_reply(header, plen, addrp, flags, daemon->local_ttl); 386 send_from(udpfd, daemon->options & OPT_NOWILD, (char *)header, plen, udpaddr, dst_addr, dst_iface); 387 } 388 389 return 0; 390 } 391 392 static size_t process_reply(HEADER *header, time_t now, 393 struct server *server, size_t n) 394 { 395 unsigned char *pheader, *sizep; 396 int munged = 0, is_sign; 397 size_t plen; 398 399 /* If upstream is advertising a larger UDP packet size 400 than we allow, trim it so that we don't get overlarge 401 requests for the client. We can't do this for signed packets. */ 402 403 if ((pheader = find_pseudoheader(header, n, &plen, &sizep, &is_sign)) && !is_sign) 404 { 405 unsigned short udpsz; 406 unsigned char *psave = sizep; 407 408 GETSHORT(udpsz, sizep); 409 if (udpsz > daemon->edns_pktsz) 410 PUTSHORT(daemon->edns_pktsz, psave); 411 } 412 413 if (header->opcode != QUERY || (header->rcode != NOERROR && header->rcode != NXDOMAIN)) 414 return n; 415 416 /* Complain loudly if the upstream server is non-recursive. */ 417 if (!header->ra && header->rcode == NOERROR && ntohs(header->ancount) == 0 && 418 server && !(server->flags & SERV_WARNED_RECURSIVE)) 419 { 420 prettyprint_addr(&server->addr, daemon->namebuff); 421 my_syslog(LOG_WARNING, _("nameserver %s refused to do a recursive query"), daemon->namebuff); 422 if (!(daemon->options & OPT_LOG)) 423 server->flags |= SERV_WARNED_RECURSIVE; 424 } 425 426 if (daemon->bogus_addr && header->rcode != NXDOMAIN && 427 check_for_bogus_wildcard(header, n, daemon->namebuff, daemon->bogus_addr, now)) 428 { 429 munged = 1; 430 header->rcode = NXDOMAIN; 431 header->aa = 0; 432 } 433 else 434 { 435 if (header->rcode == NXDOMAIN && 436 extract_request(header, n, daemon->namebuff, NULL) && 437 check_for_local_domain(daemon->namebuff, now)) 438 { 439 /* if we forwarded a query for a locally known name (because it was for 440 an unknown type) and the answer is NXDOMAIN, convert that to NODATA, 441 since we know that the domain exists, even if upstream doesn't */ 442 munged = 1; 443 header->aa = 1; 444 header->rcode = NOERROR; 445 } 446 447 if (extract_addresses(header, n, daemon->namebuff, now)) 448 { 449 my_syslog(LOG_WARNING, _("possible DNS-rebind attack detected")); 450 munged = 1; 451 } 452 } 453 454 /* do this after extract_addresses. Ensure NODATA reply and remove 455 nameserver info. */ 456 457 if (munged) 458 { 459 header->ancount = htons(0); 460 header->nscount = htons(0); 461 header->arcount = htons(0); 462 } 463 464 /* the bogus-nxdomain stuff, doctor and NXDOMAIN->NODATA munging can all elide 465 sections of the packet. Find the new length here and put back pseudoheader 466 if it was removed. */ 467 return resize_packet(header, n, pheader, plen); 468 } 469 470 /* sets new last_server */ 471 void reply_query(int fd, int family, time_t now) 472 { 473 /* packet from peer server, extract data for cache, and send to 474 original requester */ 475 HEADER *header; 476 union mysockaddr serveraddr; 477 struct frec *forward; 478 socklen_t addrlen = sizeof(serveraddr); 479 ssize_t n = recvfrom(fd, daemon->packet, daemon->edns_pktsz, 0, &serveraddr.sa, &addrlen); 480 size_t nn; 481 struct server *server; 482 483 /* packet buffer overwritten */ 484 daemon->srv_save = NULL; 485 486 /* Determine the address of the server replying so that we can mark that as good */ 487 serveraddr.sa.sa_family = family; 488 #ifdef HAVE_IPV6 489 if (serveraddr.sa.sa_family == AF_INET6) 490 serveraddr.in6.sin6_flowinfo = 0; 491 #endif 492 493 /* spoof check: answer must come from known server, */ 494 for (server = daemon->servers; server; server = server->next) 495 if (!(server->flags & (SERV_LITERAL_ADDRESS | SERV_NO_ADDR)) && 496 sockaddr_isequal(&server->addr, &serveraddr)) 497 break; 498 499 header = (HEADER *)daemon->packet; 500 501 if (!server || 502 n < (int)sizeof(HEADER) || !header->qr || 503 !(forward = lookup_frec(ntohs(header->id), questions_crc(header, n, daemon->namebuff)))) 504 return; 505 506 server = forward->sentto; 507 508 if ((header->rcode == SERVFAIL || header->rcode == REFUSED) && 509 !(daemon->options & OPT_ORDER) && 510 forward->forwardall == 0) 511 /* for broken servers, attempt to send to another one. */ 512 { 513 unsigned char *pheader; 514 size_t plen; 515 int is_sign; 516 517 /* recreate query from reply */ 518 pheader = find_pseudoheader(header, (size_t)n, &plen, NULL, &is_sign); 519 if (!is_sign) 520 { 521 header->ancount = htons(0); 522 header->nscount = htons(0); 523 header->arcount = htons(0); 524 if ((nn = resize_packet(header, (size_t)n, pheader, plen))) 525 { 526 header->qr = 0; 527 header->tc = 0; 528 forward_query(-1, NULL, NULL, 0, header, nn, now, forward); 529 return; 530 } 531 } 532 } 533 534 if ((forward->sentto->flags & SERV_TYPE) == 0) 535 { 536 if (header->rcode == SERVFAIL || header->rcode == REFUSED) 537 server = NULL; 538 else 539 { 540 struct server *last_server; 541 542 /* find good server by address if possible, otherwise assume the last one we sent to */ 543 for (last_server = daemon->servers; last_server; last_server = last_server->next) 544 if (!(last_server->flags & (SERV_LITERAL_ADDRESS | SERV_HAS_DOMAIN | SERV_FOR_NODOTS | SERV_NO_ADDR)) && 545 sockaddr_isequal(&last_server->addr, &serveraddr)) 546 { 547 server = last_server; 548 break; 549 } 550 } 551 if (!(daemon->options & OPT_ALL_SERVERS)) 552 daemon->last_server = server; 553 } 554 555 /* If the answer is an error, keep the forward record in place in case 556 we get a good reply from another server. Kill it when we've 557 had replies from all to avoid filling the forwarding table when 558 everything is broken */ 559 if (forward->forwardall == 0 || --forward->forwardall == 1 || 560 (header->rcode != REFUSED && header->rcode != SERVFAIL)) 561 { 562 if ((nn = process_reply(header, now, server, (size_t)n))) 563 { 564 header->id = htons(forward->orig_id); 565 header->ra = 1; /* recursion if available */ 566 send_from(forward->fd, daemon->options & OPT_NOWILD, daemon->packet, nn, 567 &forward->source, &forward->dest, forward->iface); 568 } 569 free_frec(forward); /* cancel */ 570 } 571 } 572 573 574 void receive_query(struct listener *listen, time_t now) 575 { 576 HEADER *header = (HEADER *)daemon->packet; 577 union mysockaddr source_addr; 578 unsigned short type; 579 struct all_addr dst_addr; 580 struct in_addr netmask, dst_addr_4; 581 size_t m; 582 ssize_t n; 583 int if_index = 0; 584 struct iovec iov[1]; 585 struct msghdr msg; 586 struct cmsghdr *cmptr; 587 union { 588 struct cmsghdr align; /* this ensures alignment */ 589 #ifdef HAVE_IPV6 590 char control6[CMSG_SPACE(sizeof(struct in6_pktinfo))]; 591 #endif 592 #if defined(HAVE_LINUX_NETWORK) 593 char control[CMSG_SPACE(sizeof(struct in_pktinfo))]; 594 #elif defined(IP_RECVDSTADDR) && defined(HAVE_SOLARIS_NETWORK) 595 char control[CMSG_SPACE(sizeof(struct in_addr)) + 596 CMSG_SPACE(sizeof(unsigned int))]; 597 #elif defined(IP_RECVDSTADDR) 598 char control[CMSG_SPACE(sizeof(struct in_addr)) + 599 CMSG_SPACE(sizeof(struct sockaddr_dl))]; 600 #endif 601 } control_u; 602 603 /* packet buffer overwritten */ 604 daemon->srv_save = NULL; 605 606 if (listen->family == AF_INET && (daemon->options & OPT_NOWILD)) 607 { 608 dst_addr_4 = listen->iface->addr.in.sin_addr; 609 netmask = listen->iface->netmask; 610 } 611 else 612 { 613 dst_addr_4.s_addr = 0; 614 netmask.s_addr = 0; 615 } 616 617 iov[0].iov_base = daemon->packet; 618 iov[0].iov_len = daemon->edns_pktsz; 619 620 msg.msg_control = control_u.control; 621 msg.msg_controllen = sizeof(control_u); 622 msg.msg_flags = 0; 623 msg.msg_name = &source_addr; 624 msg.msg_namelen = sizeof(source_addr); 625 msg.msg_iov = iov; 626 msg.msg_iovlen = 1; 627 628 if ((n = recvmsg(listen->fd, &msg, 0)) == -1) 629 return; 630 631 if (n < (int)sizeof(HEADER) || 632 (msg.msg_flags & MSG_TRUNC) || 633 header->qr) 634 return; 635 636 source_addr.sa.sa_family = listen->family; 637 #ifdef HAVE_IPV6 638 if (listen->family == AF_INET6) 639 source_addr.in6.sin6_flowinfo = 0; 640 #endif 641 642 if (!(daemon->options & OPT_NOWILD)) 643 { 644 struct ifreq ifr; 645 646 if (msg.msg_controllen < sizeof(struct cmsghdr)) 647 return; 648 649 #if defined(HAVE_LINUX_NETWORK) 650 if (listen->family == AF_INET) 651 for (cmptr = CMSG_FIRSTHDR(&msg); cmptr; cmptr = CMSG_NXTHDR(&msg, cmptr)) 652 if (cmptr->cmsg_level == SOL_IP && cmptr->cmsg_type == IP_PKTINFO) 653 { 654 dst_addr_4 = dst_addr.addr.addr4 = ((struct in_pktinfo *)CMSG_DATA(cmptr))->ipi_spec_dst; 655 if_index = ((struct in_pktinfo *)CMSG_DATA(cmptr))->ipi_ifindex; 656 } 657 #elif defined(IP_RECVDSTADDR) && defined(IP_RECVIF) 658 if (listen->family == AF_INET) 659 { 660 for (cmptr = CMSG_FIRSTHDR(&msg); cmptr; cmptr = CMSG_NXTHDR(&msg, cmptr)) 661 if (cmptr->cmsg_level == IPPROTO_IP && cmptr->cmsg_type == IP_RECVDSTADDR) 662 dst_addr_4 = dst_addr.addr.addr4 = *((struct in_addr *)CMSG_DATA(cmptr)); 663 else if (cmptr->cmsg_level == IPPROTO_IP && cmptr->cmsg_type == IP_RECVIF) 664 #ifdef HAVE_SOLARIS_NETWORK 665 if_index = *((unsigned int *)CMSG_DATA(cmptr)); 666 #else 667 if_index = ((struct sockaddr_dl *)CMSG_DATA(cmptr))->sdl_index; 668 #endif 669 } 670 #endif 671 672 #ifdef HAVE_IPV6 673 if (listen->family == AF_INET6) 674 { 675 for (cmptr = CMSG_FIRSTHDR(&msg); cmptr; cmptr = CMSG_NXTHDR(&msg, cmptr)) 676 if (cmptr->cmsg_level == IPV6_LEVEL && cmptr->cmsg_type == IPV6_PKTINFO) 677 { 678 dst_addr.addr.addr6 = ((struct in6_pktinfo *)CMSG_DATA(cmptr))->ipi6_addr; 679 if_index =((struct in6_pktinfo *)CMSG_DATA(cmptr))->ipi6_ifindex; 680 } 681 } 682 #endif 683 684 /* enforce available interface configuration */ 685 686 if (!indextoname(listen->fd, if_index, ifr.ifr_name) || 687 !iface_check(listen->family, &dst_addr, ifr.ifr_name, &if_index)) 688 return; 689 690 if (listen->family == AF_INET && 691 (daemon->options & OPT_LOCALISE) && 692 ioctl(listen->fd, SIOCGIFNETMASK, &ifr) == -1) 693 return; 694 695 netmask = ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr; 696 } 697 698 if (extract_request(header, (size_t)n, daemon->namebuff, &type)) 699 { 700 char types[20]; 701 702 querystr(types, type); 703 704 if (listen->family == AF_INET) 705 log_query(F_QUERY | F_IPV4 | F_FORWARD, daemon->namebuff, 706 (struct all_addr *)&source_addr.in.sin_addr, types); 707 #ifdef HAVE_IPV6 708 else 709 log_query(F_QUERY | F_IPV6 | F_FORWARD, daemon->namebuff, 710 (struct all_addr *)&source_addr.in6.sin6_addr, types); 711 #endif 712 } 713 714 m = answer_request (header, ((char *) header) + PACKETSZ, (size_t)n, 715 dst_addr_4, netmask, now); 716 if (m >= 1) 717 { 718 send_from(listen->fd, daemon->options & OPT_NOWILD, (char *)header, 719 m, &source_addr, &dst_addr, if_index); 720 daemon->local_answer++; 721 } 722 else if (forward_query(listen->fd, &source_addr, &dst_addr, if_index, 723 header, (size_t)n, now, NULL)) 724 daemon->queries_forwarded++; 725 else 726 daemon->local_answer++; 727 } 728 729 /* The daemon forks before calling this: it should deal with one connection, 730 blocking as neccessary, and then return. Note, need to be a bit careful 731 about resources for debug mode, when the fork is suppressed: that's 732 done by the caller. */ 733 unsigned char *tcp_request(int confd, time_t now, 734 struct in_addr local_addr, struct in_addr netmask) 735 { 736 int size = 0; 737 size_t m; 738 unsigned short qtype, gotname; 739 unsigned char c1, c2; 740 /* Max TCP packet + slop */ 741 unsigned char *packet = whine_malloc(65536 + MAXDNAME + RRFIXEDSZ); 742 HEADER *header; 743 struct server *last_server; 744 745 while (1) 746 { 747 if (!packet || 748 !read_write(confd, &c1, 1, 1) || !read_write(confd, &c2, 1, 1) || 749 !(size = c1 << 8 | c2) || 750 !read_write(confd, packet, size, 1)) 751 return packet; 752 753 if (size < (int)sizeof(HEADER)) 754 continue; 755 756 header = (HEADER *)packet; 757 758 if ((gotname = extract_request(header, (unsigned int)size, daemon->namebuff, &qtype))) 759 { 760 union mysockaddr peer_addr; 761 socklen_t peer_len = sizeof(union mysockaddr); 762 763 if (getpeername(confd, (struct sockaddr *)&peer_addr, &peer_len) != -1) 764 { 765 char types[20]; 766 767 querystr(types, qtype); 768 769 if (peer_addr.sa.sa_family == AF_INET) 770 log_query(F_QUERY | F_IPV4 | F_FORWARD, daemon->namebuff, 771 (struct all_addr *)&peer_addr.in.sin_addr, types); 772 #ifdef HAVE_IPV6 773 else 774 log_query(F_QUERY | F_IPV6 | F_FORWARD, daemon->namebuff, 775 (struct all_addr *)&peer_addr.in6.sin6_addr, types); 776 #endif 777 } 778 } 779 780 /* m > 0 if answered from cache */ 781 m = answer_request(header, ((char *) header) + 65536, (unsigned int)size, 782 local_addr, netmask, now); 783 784 /* Do this by steam now we're not in the select() loop */ 785 check_log_writer(NULL); 786 787 if (m == 0) 788 { 789 unsigned short flags = 0; 790 struct all_addr *addrp = NULL; 791 int type = 0; 792 char *domain = NULL; 793 794 if (gotname) 795 flags = search_servers(now, &addrp, gotname, daemon->namebuff, &type, &domain); 796 797 if (type != 0 || (daemon->options & OPT_ORDER) || !daemon->last_server) 798 last_server = daemon->servers; 799 else 800 last_server = daemon->last_server; 801 802 if (!flags && last_server) 803 { 804 struct server *firstsendto = NULL; 805 unsigned int crc = questions_crc(header, (unsigned int)size, daemon->namebuff); 806 807 /* Loop round available servers until we succeed in connecting to one. 808 Note that this code subtley ensures that consecutive queries on this connection 809 which can go to the same server, do so. */ 810 while (1) 811 { 812 if (!firstsendto) 813 firstsendto = last_server; 814 else 815 { 816 if (!(last_server = last_server->next)) 817 last_server = daemon->servers; 818 819 if (last_server == firstsendto) 820 break; 821 } 822 823 /* server for wrong domain */ 824 if (type != (last_server->flags & SERV_TYPE) || 825 (type == SERV_HAS_DOMAIN && !hostname_isequal(domain, last_server->domain))) 826 continue; 827 828 if ((last_server->tcpfd == -1) && 829 (last_server->tcpfd = socket(last_server->addr.sa.sa_family, SOCK_STREAM, 0)) != -1 && 830 (!local_bind(last_server->tcpfd, &last_server->source_addr, 831 last_server->interface, last_server->mark, 1) || 832 connect(last_server->tcpfd, &last_server->addr.sa, sa_len(&last_server->addr)) == -1)) 833 { 834 close(last_server->tcpfd); 835 last_server->tcpfd = -1; 836 } 837 838 if (last_server->tcpfd == -1) 839 continue; 840 841 c1 = size >> 8; 842 c2 = size; 843 844 if (!read_write(last_server->tcpfd, &c1, 1, 0) || 845 !read_write(last_server->tcpfd, &c2, 1, 0) || 846 !read_write(last_server->tcpfd, packet, size, 0) || 847 !read_write(last_server->tcpfd, &c1, 1, 1) || 848 !read_write(last_server->tcpfd, &c2, 1, 1)) 849 { 850 close(last_server->tcpfd); 851 last_server->tcpfd = -1; 852 continue; 853 } 854 855 m = (c1 << 8) | c2; 856 if (!read_write(last_server->tcpfd, packet, m, 1)) 857 return packet; 858 859 if (!gotname) 860 strcpy(daemon->namebuff, "query"); 861 if (last_server->addr.sa.sa_family == AF_INET) 862 log_query(F_SERVER | F_IPV4 | F_FORWARD, daemon->namebuff, 863 (struct all_addr *)&last_server->addr.in.sin_addr, NULL); 864 #ifdef HAVE_IPV6 865 else 866 log_query(F_SERVER | F_IPV6 | F_FORWARD, daemon->namebuff, 867 (struct all_addr *)&last_server->addr.in6.sin6_addr, NULL); 868 #endif 869 870 /* There's no point in updating the cache, since this process will exit and 871 lose the information after a few queries. We make this call for the alias and 872 bogus-nxdomain side-effects. */ 873 /* If the crc of the question section doesn't match the crc we sent, then 874 someone might be attempting to insert bogus values into the cache by 875 sending replies containing questions and bogus answers. */ 876 if (crc == questions_crc(header, (unsigned int)m, daemon->namebuff)) 877 m = process_reply(header, now, last_server, (unsigned int)m); 878 879 break; 880 } 881 } 882 883 /* In case of local answer or no connections made. */ 884 if (m == 0) 885 m = setup_reply(header, (unsigned int)size, addrp, flags, daemon->local_ttl); 886 } 887 888 check_log_writer(NULL); 889 890 c1 = m>>8; 891 c2 = m; 892 if (!read_write(confd, &c1, 1, 0) || 893 !read_write(confd, &c2, 1, 0) || 894 !read_write(confd, packet, m, 0)) 895 return packet; 896 } 897 } 898 899 static struct frec *allocate_frec(time_t now) 900 { 901 struct frec *f; 902 903 if ((f = (struct frec *)whine_malloc(sizeof(struct frec)))) 904 { 905 f->next = daemon->frec_list; 906 f->time = now; 907 f->sentto = NULL; 908 f->rfd4 = NULL; 909 #ifdef HAVE_IPV6 910 f->rfd6 = NULL; 911 #endif 912 daemon->frec_list = f; 913 } 914 915 return f; 916 } 917 918 static struct randfd *allocate_rfd(int family) 919 { 920 static int finger = 0; 921 int i; 922 923 /* limit the number of sockets we have open to avoid starvation of 924 (eg) TFTP. Once we have a reasonable number, randomness should be OK */ 925 926 for (i = 0; i < RANDOM_SOCKS; i++) 927 if (daemon->randomsocks[i].refcount == 0) 928 { 929 if ((daemon->randomsocks[i].fd = random_sock(family)) == -1) 930 break; 931 932 daemon->randomsocks[i].refcount = 1; 933 daemon->randomsocks[i].family = family; 934 return &daemon->randomsocks[i]; 935 } 936 937 /* No free ones or cannot get new socket, grab an existing one */ 938 for (i = 0; i < RANDOM_SOCKS; i++) 939 { 940 int j = (i+finger) % RANDOM_SOCKS; 941 if (daemon->randomsocks[j].refcount != 0 && 942 daemon->randomsocks[j].family == family && 943 daemon->randomsocks[j].refcount != 0xffff) 944 { 945 finger = j; 946 daemon->randomsocks[j].refcount++; 947 return &daemon->randomsocks[j]; 948 } 949 } 950 951 return NULL; /* doom */ 952 } 953 954 static void free_frec(struct frec *f) 955 { 956 if (f->rfd4 && --(f->rfd4->refcount) == 0) 957 close(f->rfd4->fd); 958 959 f->rfd4 = NULL; 960 f->sentto = NULL; 961 962 #ifdef HAVE_IPV6 963 if (f->rfd6 && --(f->rfd6->refcount) == 0) 964 close(f->rfd6->fd); 965 966 f->rfd6 = NULL; 967 #endif 968 } 969 970 /* if wait==NULL return a free or older than TIMEOUT record. 971 else return *wait zero if one available, or *wait is delay to 972 when the oldest in-use record will expire. Impose an absolute 973 limit of 4*TIMEOUT before we wipe things (for random sockets) */ 974 struct frec *get_new_frec(time_t now, int *wait) 975 { 976 struct frec *f, *oldest, *target; 977 int count; 978 979 if (wait) 980 *wait = 0; 981 982 for (f = daemon->frec_list, oldest = NULL, target = NULL, count = 0; f; f = f->next, count++) 983 if (!f->sentto) 984 target = f; 985 else 986 { 987 if (difftime(now, f->time) >= 4*TIMEOUT) 988 { 989 free_frec(f); 990 target = f; 991 } 992 993 if (!oldest || difftime(f->time, oldest->time) <= 0) 994 oldest = f; 995 } 996 997 if (target) 998 { 999 target->time = now; 1000 return target; 1001 } 1002 1003 /* can't find empty one, use oldest if there is one 1004 and it's older than timeout */ 1005 if (oldest && ((int)difftime(now, oldest->time)) >= TIMEOUT) 1006 { 1007 /* keep stuff for twice timeout if we can by allocating a new 1008 record instead */ 1009 if (difftime(now, oldest->time) < 2*TIMEOUT && 1010 count <= daemon->ftabsize && 1011 (f = allocate_frec(now))) 1012 return f; 1013 1014 if (!wait) 1015 { 1016 free_frec(oldest); 1017 oldest->time = now; 1018 } 1019 return oldest; 1020 } 1021 1022 /* none available, calculate time 'till oldest record expires */ 1023 if (count > daemon->ftabsize) 1024 { 1025 if (oldest && wait) 1026 *wait = oldest->time + (time_t)TIMEOUT - now; 1027 return NULL; 1028 } 1029 1030 if (!(f = allocate_frec(now)) && wait) 1031 /* wait one second on malloc failure */ 1032 *wait = 1; 1033 1034 return f; /* OK if malloc fails and this is NULL */ 1035 } 1036 1037 /* crc is all-ones if not known. */ 1038 static struct frec *lookup_frec(unsigned short id, unsigned int crc) 1039 { 1040 struct frec *f; 1041 1042 for(f = daemon->frec_list; f; f = f->next) 1043 if (f->sentto && f->new_id == id && 1044 (f->crc == crc || crc == 0xffffffff)) 1045 return f; 1046 1047 return NULL; 1048 } 1049 1050 static struct frec *lookup_frec_by_sender(unsigned short id, 1051 union mysockaddr *addr, 1052 unsigned int crc) 1053 { 1054 struct frec *f; 1055 1056 for(f = daemon->frec_list; f; f = f->next) 1057 if (f->sentto && 1058 f->orig_id == id && 1059 f->crc == crc && 1060 sockaddr_isequal(&f->source, addr)) 1061 return f; 1062 1063 return NULL; 1064 } 1065 1066 /* A server record is going away, remove references to it */ 1067 void server_gone(struct server *server) 1068 { 1069 struct frec *f; 1070 1071 for (f = daemon->frec_list; f; f = f->next) 1072 if (f->sentto && f->sentto == server) 1073 free_frec(f); 1074 1075 if (daemon->last_server == server) 1076 daemon->last_server = NULL; 1077 1078 if (daemon->srv_save == server) 1079 daemon->srv_save = NULL; 1080 } 1081 1082 /* return unique random ids. 1083 For signed packets we can't change the ID without breaking the 1084 signing, so we keep the same one. In this case force is set, and this 1085 routine degenerates into killing any conflicting forward record. */ 1086 static unsigned short get_id(int force, unsigned short force_id, unsigned int crc) 1087 { 1088 unsigned short ret = 0; 1089 1090 if (force) 1091 { 1092 struct frec *f = lookup_frec(force_id, crc); 1093 if (f) 1094 free_frec(f); /* free */ 1095 ret = force_id; 1096 } 1097 else do 1098 ret = rand16(); 1099 while (lookup_frec(ret, crc)); 1100 1101 return ret; 1102 } 1103 1104 1105 1106 1107 1108
