src/dnsproxy.c

   1 /*
   2  *
   3  *  Connection Manager
   4  *
   5  *  Copyright (C) 2007-2012  Intel Corporation. All rights reserved.
   6  *
   7  *  This program is free software; you can redistribute it and/or modify
   8  *  it under the terms of the GNU General Public License version 2 as
   9  *  published by the Free Software Foundation.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, write to the Free Software
  18  *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  */
  21
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <errno.h>
  27 #include <stdlib.h>
  28 #include <unistd.h>
  29 #include <string.h>
  30 #include <stdint.h>
  31 #include <arpa/inet.h>
  32 #include <netinet/in.h>
  33 #include <sys/types.h>
  34 #include <sys/socket.h>
  35 #include <netdb.h>
  36 #include <resolv.h>
  37 #include <gweb/gresolv.h>
  38
  39 #include <glib.h>
  40
  41 #include "connman.h"
  42
  43 #if __BYTE_ORDER == __LITTLE_ENDIAN
  44 struct domain_hdr {
  45         uint16_t id;
  46         uint8_t rd:1;
  47         uint8_t tc:1;
  48         uint8_t aa:1;
  49         uint8_t opcode:4;
  50         uint8_t qr:1;
  51         uint8_t rcode:4;
  52         uint8_t z:3;
  53         uint8_t ra:1;
  54         uint16_t qdcount;
  55         uint16_t ancount;
  56         uint16_t nscount;
  57         uint16_t arcount;
  58 } __attribute__ ((packed));
  59 #elif __BYTE_ORDER == __BIG_ENDIAN
  60 struct domain_hdr {
  61         uint16_t id;
  62         uint8_t qr:1;
  63         uint8_t opcode:4;
  64         uint8_t aa:1;
  65         uint8_t tc:1;
  66         uint8_t rd:1;
  67         uint8_t ra:1;
  68         uint8_t z:3;
  69         uint8_t rcode:4;
  70         uint16_t qdcount;
  71         uint16_t ancount;
  72         uint16_t nscount;
  73         uint16_t arcount;
  74 } __attribute__ ((packed));
  75 #else
  76 #error "Unknown byte order"
  77 #endif
  78
  79 struct partial_reply {
  80         uint16_t len;
  81         uint16_t received;
  82         unsigned char buf[];
  83 };
  84
  85 struct server_data {
  86         char *interface;
  87         GList *domains;
  88         char *server;
  89         struct sockaddr server_addr;
  90         socklen_t server_addr_len;
  91         int protocol;
  92         GIOChannel *channel;
  93         guint watch;
  94         guint timeout;
  95         gboolean enabled;
  96         gboolean connected;
  97         struct partial_reply *incoming_reply;
  98 };
  99
 100 struct request_data {
 101         union {
 102                 struct sockaddr_in6 __sin6; /* Only for the length */
 103                 struct sockaddr sa;
 104         };
 105         socklen_t sa_len;
 106         int client_sk;
 107         int protocol;
 108         guint16 srcid;
 109         guint16 dstid;
 110         guint16 altid;
 111         guint timeout;
 112         guint watch;
 113         guint numserv;
 114         guint numresp;
 115         gpointer request;
 116         gsize request_len;
 117         gpointer name;
 118         gpointer resp;
 119         gsize resplen;
 120         struct listener_data *ifdata;
 121         gboolean append_domain;
 122 };
 123
 124 struct listener_data {
 125         char *ifname;
 126         GIOChannel *udp_listener_channel;
 127         guint udp_listener_watch;
 128         GIOChannel *tcp_listener_channel;
 129         guint tcp_listener_watch;
 130 };
 131
 132 struct cache_data {
 133         time_t inserted;
 134         time_t valid_until;
 135         time_t cache_until;
 136         int timeout;
 137         uint16_t type;
 138         uint16_t answers;
 139         unsigned int data_len;
 140         unsigned char *data; /* contains DNS header + body */
 141 };
 142
 143 struct cache_entry {
 144         char *key;
 145         int want_refresh;
 146         int hits;
 147         struct cache_data *ipv4;
 148         struct cache_data *ipv6;
 149 };
 150
 151 struct domain_question {
 152         uint16_t type;
 153         uint16_t class;
 154 } __attribute__ ((packed));
 155
 156 struct domain_rr {
 157         uint16_t type;
 158         uint16_t class;
 159         uint32_t ttl;
 160         uint16_t rdlen;
 161 } __attribute__ ((packed));
 162
 163 /*
 164  * We limit how long the cached DNS entry stays in the cache.
 165  * By default the TTL (time-to-live) of the DNS response is used
 166  * when setting the cache entry life time. The value is in seconds.
 167  */
 168 #define MAX_CACHE_TTL (60 * 30)
 169 /*
 170  * Also limit the other end, cache at least for 30 seconds.
 171  */
 172 #define MIN_CACHE_TTL (30)
 173
 174 /*
 175  * We limit the cache size to some sane value so that cached data does
 176  * not occupy too much memory. Each cached entry occupies on average
 177  * about 100 bytes memory (depending on DNS name length).
 178  * Example: caching www.connman.net uses 97 bytes memory.
 179  * The value is the max amount of cached DNS responses (count).
 180  */
 181 #define MAX_CACHE_SIZE 256
 182
 183 static int cache_size;
 184 static GHashTable *cache;
 185 static int cache_refcount;
 186 static GSList *server_list = NULL;
 187 static GSList *request_list = NULL;
 188 static GHashTable *listener_table = NULL;
 189 static time_t next_refresh;
 190
 191 static guint16 get_id()
 192 {
 193         return random();
 194 }
 195
 196 static int protocol_offset(int protocol)
 197 {
 198         switch (protocol) {
 199         case IPPROTO_UDP:
 200                 return 0;
 201
 202         case IPPROTO_TCP:
 203                 return 2;
 204
 205         default:
 206                 return -EINVAL;
 207         }
 208
 209 }
 210
 211 /*
 212  * There is a power and efficiency benefit to have entries
 213  * in our cache expire at the same time. To this extend,
 214  * we round down the cache valid time to common boundaries.
 215  */
 216 static time_t round_down_ttl(time_t end_time, int ttl)
 217 {
 218         if (ttl < 15)
 219                 return end_time;
 220
 221         /* Less than 5 minutes, round to 10 second boundary */
 222         if (ttl < 300) {
 223                 end_time = end_time / 10;
 224                 end_time = end_time * 10;
 225         } else { /* 5 or more minutes, round to 30 seconds */
 226                 end_time = end_time / 30;
 227                 end_time = end_time * 30;
 228         }
 229         return end_time;
 230 }
 231
 232 static struct request_data *find_request(guint16 id)
 233 {
 234         GSList *list;
 235
 236         for (list = request_list; list; list = list->next) {
 237                 struct request_data *req = list->data;
 238
 239                 if (req->dstid == id || req->altid == id)
 240                         return req;
 241         }
 242
 243         return NULL;
 244 }
 245
 246 static struct server_data *find_server(const char *interface,
 247                                         const char *server,
 248                                                 int protocol)
 249 {
 250         GSList *list;
 251
 252         DBG("interface %s server %s", interface, server);
 253
 254         for (list = server_list; list; list = list->next) {
 255                 struct server_data *data = list->data;
 256
 257                 if (interface == NULL && data->interface == NULL &&
 258                                 g_str_equal(data->server, server) == TRUE &&
 259                                 data->protocol == protocol)
 260                         return data;
 261
 262                 if (interface == NULL ||
 263                                 data->interface == NULL || data->server == NULL)
 264                         continue;
 265
 266                 if (g_str_equal(data->interface, interface) == TRUE &&
 267                                 g_str_equal(data->server, server) == TRUE &&
 268                                 data->protocol == protocol)
 269                         return data;
 270         }
 271
 272         return NULL;
 273 }
 274
 275 /* we can keep using the same resolve's */
 276 static GResolv *ipv4_resolve;
 277 static GResolv *ipv6_resolve;
 278
 279 static void dummy_resolve_func(GResolvResultStatus status,
 280                                         char **results, gpointer user_data)
 281 {
 282 }
 283
 284 /*
 285  * Refresh a DNS entry, but also age the hit count a bit */
 286 static void refresh_dns_entry(struct cache_entry *entry, char *name)
 287 {
 288         int age = 1;
 289
 290         if (ipv4_resolve == NULL) {
 291                 ipv4_resolve = g_resolv_new(0);
 292                 g_resolv_set_address_family(ipv4_resolve, AF_INET);
 293                 g_resolv_add_nameserver(ipv4_resolve, "127.0.0.1", 53, 0);
 294         }
 295
 296         if (ipv6_resolve == NULL) {
 297                 ipv6_resolve = g_resolv_new(0);
 298                 g_resolv_set_address_family(ipv6_resolve, AF_INET6);
 299                 g_resolv_add_nameserver(ipv6_resolve, "127.0.0.1", 53, 0);
 300         }
 301
 302         if (entry->ipv4 == NULL) {
 303                 DBG("Refresing A record for %s", name);
 304                 g_resolv_lookup_hostname(ipv4_resolve, name,
 305                                         dummy_resolve_func, NULL);
 306                 age = 4;
 307         }
 308
 309         if (entry->ipv6 == NULL) {
 310                 DBG("Refresing AAAA record for %s", name);
 311                 g_resolv_lookup_hostname(ipv6_resolve, name,
 312                                         dummy_resolve_func, NULL);
 313                 age = 4;
 314         }
 315
 316         entry->hits -= age;
 317         if (entry->hits < 0)
 318                 entry->hits = 0;
 319 }
 320
 321 static int dns_name_length(unsigned char *buf)
 322 {
 323         if ((buf[0] & NS_CMPRSFLGS) == NS_CMPRSFLGS) /* compressed name */
 324                 return 2;
 325         return strlen((char *)buf);
 326 }
 327
 328 static void update_cached_ttl(unsigned char *buf, int len, int new_ttl)
 329 {
 330         unsigned char *c;
 331         uint32_t *i;
 332         uint16_t *w;
 333         int l;
 334
 335         /* skip the header */
 336         c = buf + 12;
 337         len -= 12;
 338
 339         /* skip the query, which is a name and 2 16 bit words */
 340         l = dns_name_length(c);
 341         c += l;
 342         len -= l;
 343         c += 4;
 344         len -= 4;
 345
 346         /* now we get the answer records */
 347
 348         while (len > 0) {
 349                 /* first a name */
 350                 l = dns_name_length(c);
 351                 c += l;
 352                 len -= l;
 353                 if (len < 0)
 354                         break;
 355                 /* then type + class, 2 bytes each */
 356                 c += 4;
 357                 len -= 4;
 358                 if (len < 0)
 359                         break;
 360
 361                 /* now the 4 byte TTL field */
 362                 i = (uint32_t *)c;
 363                 *i = htonl(new_ttl);
 364                 c += 4;
 365                 len -= 4;
 366                 if (len < 0)
 367                         break;
 368
 369                 /* now the 2 byte rdlen field */
 370                 w = (uint16_t *)c;
 371                 c += ntohs(*w) + 2;
 372                 len -= ntohs(*w) + 2;
 373         }
 374 }
 375
 376 static void send_cached_response(int sk, unsigned char *buf, int len,
 377                                 const struct sockaddr *to, socklen_t tolen,
 378                                 int protocol, int id, uint16_t answers, int ttl)
 379 {
 380         struct domain_hdr *hdr;
 381         unsigned char *ptr = buf;
 382         int err, offset, dns_len, adj_len = len - 2;
 383
 384         /*
 385          * The cached packet contains always the TCP offset (two bytes)
 386          * so skip them for UDP.
 387          */
 388         switch (protocol) {
 389         case IPPROTO_UDP:
 390                 ptr += 2;
 391                 len -= 2;
 392                 dns_len = len;
 393                 offset = 0;
 394                 break;
 395         case IPPROTO_TCP:
 396                 offset = 2;
 397                 dns_len = ptr[0] * 256 + ptr[1];
 398                 break;
 399         default:
 400                 return;
 401         }
 402
 403         if (len < 12)
 404                 return;
 405
 406         hdr = (void *) (ptr + offset);
 407
 408         hdr->id = id;
 409         hdr->qr = 1;
 410         hdr->rcode = 0;
 411         hdr->ancount = htons(answers);
 412         hdr->nscount = 0;
 413         hdr->arcount = 0;
 414
 415         /* if this is a negative reply, we are authorative */
 416         if (answers == 0)
 417                 hdr->aa = 1;
 418         else
 419                 update_cached_ttl((unsigned char *)hdr, adj_len, ttl);
 420
 421         DBG("sk %d id 0x%04x answers %d ptr %p length %d dns %d",
 422                 sk, hdr->id, answers, ptr, len, dns_len);
 423
 424         err = sendto(sk, ptr, len, MSG_NOSIGNAL, to, tolen);
 425         if (err < 0) {
 426                 connman_error("Cannot send cached DNS response: %s",
 427                                 strerror(errno));
 428                 return;
 429         }
 430
 431         if (err != len || (dns_len != (len - 2) && protocol == IPPROTO_TCP) ||
 432                                 (dns_len != len && protocol == IPPROTO_UDP))
 433                 DBG("Packet length mismatch, sent %d wanted %d dns %d",
 434                         err, len, dns_len);
 435 }
 436
 437 static void send_response(int sk, unsigned char *buf, int len,
 438                                 const struct sockaddr *to, socklen_t tolen,
 439                                 int protocol)
 440 {
 441         struct domain_hdr *hdr;
 442         int err, offset = protocol_offset(protocol);
 443
 444         DBG("sk %d", sk);
 445
 446         if (offset < 0)
 447                 return;
 448
 449         if (len < 12)
 450                 return;
 451
 452         hdr = (void *) (buf + offset);
 453
 454         DBG("id 0x%04x qr %d opcode %d", hdr->id, hdr->qr, hdr->opcode);
 455
 456         hdr->qr = 1;
 457         hdr->rcode = 2;
 458
 459         hdr->ancount = 0;
 460         hdr->nscount = 0;
 461         hdr->arcount = 0;
 462
 463         err = sendto(sk, buf, len, MSG_NOSIGNAL, to, tolen);
 464         if (err < 0) {
 465                 connman_error("Failed to send DNS response to %d: %s",
 466                                 sk, strerror(errno));
 467                 return;
 468         }
 469 }
 470
 471 static gboolean request_timeout(gpointer user_data)
 472 {
 473         struct request_data *req = user_data;
 474         struct listener_data *ifdata;
 475
 476         DBG("id 0x%04x", req->srcid);
 477
 478         if (req == NULL)
 479                 return FALSE;
 480
 481         ifdata = req->ifdata;
 482
 483         request_list = g_slist_remove(request_list, req);
 484         req->numserv--;
 485
 486         if (req->resplen > 0 && req->resp != NULL) {
 487                 int sk, err;
 488
 489                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
 490
 491                 err = sendto(sk, req->resp, req->resplen, MSG_NOSIGNAL,
 492                                                 &req->sa, req->sa_len);
 493                 if (err < 0)
 494                         return FALSE;
 495         } else if (req->request && req->numserv == 0) {
 496                 struct domain_hdr *hdr;
 497
 498                 if (req->protocol == IPPROTO_TCP) {
 499                         hdr = (void *) (req->request + 2);
 500                         hdr->id = req->srcid;
 501                         send_response(req->client_sk, req->request,
 502                                 req->request_len, NULL, 0, IPPROTO_TCP);
 503
 504                 } else if (req->protocol == IPPROTO_UDP) {
 505                         int sk;
 506
 507                         hdr = (void *) (req->request);
 508                         hdr->id = req->srcid;
 509                         sk = g_io_channel_unix_get_fd(
 510                                                 ifdata->udp_listener_channel);
 511                         send_response(sk, req->request, req->request_len,
 512                                         &req->sa, req->sa_len, IPPROTO_UDP);
 513                 }
 514         }
 515
 516         g_free(req->resp);
 517         g_free(req);
 518
 519         return FALSE;
 520 }
 521
 522 static int append_query(unsigned char *buf, unsigned int size,
 523                                 const char *query, const char *domain)
 524 {
 525         unsigned char *ptr = buf;
 526         int len;
 527
 528         DBG("query %s domain %s", query, domain);
 529
 530         while (query != NULL) {
 531                 const char *tmp;
 532
 533                 tmp = strchr(query, '.');
 534                 if (tmp == NULL) {
 535                         len = strlen(query);
 536                         if (len == 0)
 537                                 break;
 538                         *ptr = len;
 539                         memcpy(ptr + 1, query, len);
 540                         ptr += len + 1;
 541                         break;
 542                 }
 543
 544                 *ptr = tmp - query;
 545                 memcpy(ptr + 1, query, tmp - query);
 546                 ptr += tmp - query + 1;
 547
 548                 query = tmp + 1;
 549         }
 550
 551         while (domain != NULL) {
 552                 const char *tmp;
 553
 554                 tmp = strchr(domain, '.');
 555                 if (tmp == NULL) {
 556                         len = strlen(domain);
 557                         if (len == 0)
 558                                 break;
 559                         *ptr = len;
 560                         memcpy(ptr + 1, domain, len);
 561                         ptr += len + 1;
 562                         break;
 563                 }
 564
 565                 *ptr = tmp - domain;
 566                 memcpy(ptr + 1, domain, tmp - domain);
 567                 ptr += tmp - domain + 1;
 568
 569                 domain = tmp + 1;
 570         }
 571
 572         *ptr++ = 0x00;
 573
 574         return ptr - buf;
 575 }
 576
 577 static gboolean cache_check_is_valid(struct cache_data *data,
 578                                 time_t current_time)
 579 {
 580         if (data == NULL)
 581                 return FALSE;
 582
 583         if (data->cache_until < current_time)
 584                 return FALSE;
 585
 586         return TRUE;
 587 }
 588
 589 /*
 590  * remove stale cached entries so that they can be refreshed
 591  */
 592 static void cache_enforce_validity(struct cache_entry *entry)
 593 {
 594         time_t current_time = time(NULL);
 595
 596         if (cache_check_is_valid(entry->ipv4, current_time) == FALSE
 597                                                         && entry->ipv4) {
 598                 DBG("cache timeout \"%s\" type A", entry->key);
 599                 g_free(entry->ipv4->data);
 600                 g_free(entry->ipv4);
 601                 entry->ipv4 = NULL;
 602
 603         }
 604
 605         if (cache_check_is_valid(entry->ipv6, current_time) == FALSE
 606                                                         && entry->ipv6) {
 607                 DBG("cache timeout \"%s\" type AAAA", entry->key);
 608                 g_free(entry->ipv6->data);
 609                 g_free(entry->ipv6);
 610                 entry->ipv6 = NULL;
 611         }
 612 }
 613
 614 static uint16_t cache_check_validity(char *question, uint16_t type,
 615                                 struct cache_entry *entry)
 616 {
 617         time_t current_time = time(NULL);
 618         int want_refresh = 0;
 619
 620         /*
 621          * if we have a popular entry, we want a refresh instead of
 622          * total destruction of the entry.
 623          */
 624         if (entry->hits > 2)
 625                 want_refresh = 1;
 626
 627         cache_enforce_validity(entry);
 628
 629         switch (type) {
 630         case 1:         /* IPv4 */
 631                 if (cache_check_is_valid(entry->ipv4, current_time) == FALSE) {
 632                         DBG("cache %s \"%s\" type A", entry->ipv4 ?
 633                                         "timeout" : "entry missing", question);
 634
 635                         if (want_refresh)
 636                                 entry->want_refresh = 1;
 637
 638                         /*
 639                          * We do not remove cache entry if there is still
 640                          * valid IPv6 entry found in the cache.
 641                          */
 642                         if (cache_check_is_valid(entry->ipv6, current_time)
 643                                         == FALSE && want_refresh == FALSE) {
 644                                 g_hash_table_remove(cache, question);
 645                                 type = 0;
 646                         }
 647                 }
 648                 break;
 649
 650         case 28:        /* IPv6 */
 651                 if (cache_check_is_valid(entry->ipv6, current_time) == FALSE) {
 652                         DBG("cache %s \"%s\" type AAAA", entry->ipv6 ?
 653                                         "timeout" : "entry missing", question);
 654
 655                         if (want_refresh)
 656                                 entry->want_refresh = 1;
 657
 658                         if (cache_check_is_valid(entry->ipv4, current_time)
 659                                         == FALSE && want_refresh == FALSE) {
 660                                 g_hash_table_remove(cache, question);
 661                                 type = 0;
 662                         }
 663                 }
 664                 break;
 665         }
 666
 667         return type;
 668 }
 669
 670 static struct cache_entry *cache_check(gpointer request, int *qtype, int proto)
 671 {
 672         char *question;
 673         struct cache_entry *entry;
 674         struct domain_question *q;
 675         uint16_t type;
 676         int offset, proto_offset;
 677
 678         if (request == NULL)
 679                 return NULL;
 680
 681         proto_offset = protocol_offset(proto);
 682         if (proto_offset < 0)
 683                 return NULL;
 684
 685         question = request + proto_offset + 12;
 686
 687         offset = strlen(question) + 1;
 688         q = (void *) (question + offset);
 689         type = ntohs(q->type);
 690
 691         /* We only cache either A (1) or AAAA (28) requests */
 692         if (type != 1 && type != 28)
 693                 return NULL;
 694
 695         entry = g_hash_table_lookup(cache, question);
 696         if (entry == NULL)
 697                 return NULL;
 698
 699         type = cache_check_validity(question, type, entry);
 700         if (type == 0)
 701                 return NULL;
 702
 703         *qtype = type;
 704         return entry;
 705 }
 706
 707 /*
 708  * Get a label/name from DNS resource record. The function decompresses the
 709  * label if necessary. The function does not convert the name to presentation
 710  * form. This means that the result string will contain label lengths instead
 711  * of dots between labels. We intentionally do not want to convert to dotted
 712  * format so that we can cache the wire format string directly.
 713  */
 714 static int get_name(int counter,
 715                 unsigned char *pkt, unsigned char *start, unsigned char *max,
 716                 unsigned char *output, int output_max, int *output_len,
 717                 unsigned char **end, char *name, int *name_len)
 718 {
 719         unsigned char *p;
 720
 721         /* Limit recursion to 10 (this means up to 10 labels in domain name) */
 722         if (counter > 10)
 723                 return -EINVAL;
 724
 725         p = start;
 726         while (*p) {
 727                 if ((*p & NS_CMPRSFLGS) == NS_CMPRSFLGS) {
 728                         uint16_t offset = (*p & 0x3F) * 256 + *(p + 1);
 729
 730                         if (offset >= max - pkt)
 731                                 return -ENOBUFS;
 732
 733                         if (*end == NULL)
 734                                 *end = p + 2;
 735
 736                         return get_name(counter + 1, pkt, pkt + offset, max,
 737                                         output, output_max, output_len, end,
 738                                         name, name_len);
 739                 } else {
 740                         unsigned label_len = *p;
 741
 742                         if (pkt + label_len > max)
 743                                 return -ENOBUFS;
 744
 745                         if (*output_len > output_max)
 746                                 return -ENOBUFS;
 747
 748                         /*
 749                          * We need the original name in order to check
 750                          * if this answer is the correct one.
 751                          */
 752                         name[(*name_len)++] = label_len;
 753                         memcpy(name + *name_len, p + 1, label_len + 1);
 754                         *name_len += label_len;
 755
 756                         /* We compress the result */
 757                         output[0] = NS_CMPRSFLGS;
 758                         output[1] = 0x0C;
 759                         *output_len = 2;
 760
 761                         p += label_len + 1;
 762
 763                         if (*end == NULL)
 764                                 *end = p;
 765
 766                         if (p >= max)
 767                                 return -ENOBUFS;
 768                 }
 769         }
 770
 771         return 0;
 772 }
 773
 774 static int parse_rr(unsigned char *buf, unsigned char *start,
 775                         unsigned char *max,
 776                         unsigned char *response, unsigned int *response_size,
 777                         uint16_t *type, uint16_t *class, int *ttl, int *rdlen,
 778                         unsigned char **end,
 779                         char *name)
 780 {
 781         struct domain_rr *rr;
 782         int err, offset;
 783         int name_len = 0, output_len = 0, max_rsp = *response_size;
 784
 785         err = get_name(0, buf, start, max, response, max_rsp,
 786                 &output_len, end, name, &name_len);
 787         if (err < 0)
 788                 return err;
 789
 790         offset = output_len;
 791
 792         if ((unsigned int) offset > *response_size)
 793                 return -ENOBUFS;
 794
 795         rr = (void *) (*end);
 796
 797         if (rr == NULL)
 798                 return -EINVAL;
 799
 800         *type = ntohs(rr->type);
 801         *class = ntohs(rr->class);
 802         *ttl = ntohl(rr->ttl);
 803         *rdlen = ntohs(rr->rdlen);
 804
 805         if (*ttl < 0)
 806                 return -EINVAL;
 807
 808         memcpy(response + offset, *end, sizeof(struct domain_rr));
 809
 810         offset += sizeof(struct domain_rr);
 811         *end += sizeof(struct domain_rr);
 812
 813         if ((unsigned int) (offset + *rdlen) > *response_size)
 814                 return -ENOBUFS;
 815
 816         memcpy(response + offset, *end, *rdlen);
 817
 818         *end += *rdlen;
 819
 820         *response_size = offset + *rdlen;
 821
 822         return 0;
 823 }
 824
 825 static gboolean check_alias(GSList *aliases, char *name)
 826 {
 827         GSList *list;
 828
 829         if (aliases != NULL) {
 830                 for (list = aliases; list; list = list->next) {
 831                         int len = strlen((char *)list->data);
 832                         if (strncmp((char *)list->data, name, len) == 0)
 833                                 return TRUE;
 834                 }
 835         }
 836
 837         return FALSE;
 838 }
 839
 840 static int parse_response(unsigned char *buf, int buflen,
 841                         char *question, int qlen,
 842                         uint16_t *type, uint16_t *class, int *ttl,
 843                         unsigned char *response, unsigned int *response_len,
 844                         uint16_t *answers)
 845 {
 846         struct domain_hdr *hdr = (void *) buf;
 847         struct domain_question *q;
 848         unsigned char *ptr;
 849         uint16_t qdcount = ntohs(hdr->qdcount);
 850         uint16_t ancount = ntohs(hdr->ancount);
 851         int err, i;
 852         uint16_t qtype, qclass;
 853         unsigned char *next = NULL;
 854         unsigned int maxlen = *response_len;
 855         GSList *aliases = NULL, *list;
 856         char name[NS_MAXDNAME + 1];
 857
 858         if (buflen < 12)
 859                 return -EINVAL;
 860
 861         DBG("qr %d qdcount %d", hdr->qr, qdcount);
 862
 863         /* We currently only cache responses where question count is 1 */
 864         if (hdr->qr != 1 || qdcount != 1)
 865                 return -EINVAL;
 866
 867         ptr = buf + sizeof(struct domain_hdr);
 868
 869         strncpy(question, (char *) ptr, qlen);
 870         qlen = strlen(question);
 871         ptr += qlen + 1; /* skip \0 */
 872
 873         q = (void *) ptr;
 874         qtype = ntohs(q->type);
 875
 876         /* We cache only A and AAAA records */
 877         if (qtype != 1 && qtype != 28)
 878                 return -ENOMSG;
 879
 880         qclass = ntohs(q->class);
 881
 882         ptr += 2 + 2; /* ptr points now to answers */
 883
 884         err = -ENOMSG;
 885         *response_len = 0;
 886         *answers = 0;
 887
 888         /*
 889          * We have a bunch of answers (like A, AAAA, CNAME etc) to
 890          * A or AAAA question. We traverse the answers and parse the
 891          * resource records. Only A and AAAA records are cached, all
 892          * the other records in answers are skipped.
 893          */
 894         for (i = 0; i < ancount; i++) {
 895                 /*
 896                  * Get one address at a time to this buffer.
 897                  * The max size of the answer is
 898                  *   2 (pointer) + 2 (type) + 2 (class) +
 899                  *   4 (ttl) + 2 (rdlen) + addr (16 or 4) = 28
 900                  * for A or AAAA record.
 901                  * For CNAME the size can be bigger.
 902                  */
 903                 unsigned char rsp[NS_MAXCDNAME];
 904                 unsigned int rsp_len = sizeof(rsp) - 1;
 905                 int ret, rdlen;
 906
 907                 memset(rsp, 0, sizeof(rsp));
 908
 909                 ret = parse_rr(buf, ptr, buf + buflen, rsp, &rsp_len,
 910                         type, class, ttl, &rdlen, &next, name);
 911                 if (ret != 0) {
 912                         err = ret;
 913                         goto out;
 914                 }
 915
 916                 /*
 917                  * Now rsp contains compressed or uncompressed resource
 918                  * record. Next we check if this record answers the question.
 919                  * The name var contains the uncompressed label.
 920                  * One tricky bit is the CNAME records as they alias
 921                  * the name we might be interested in.
 922                  */
 923
 924                 /*
 925                  * Go to next answer if the class is not the one we are
 926                  * looking for.
 927                  */
 928                 if (*class != qclass) {
 929                         ptr = next;
 930                         next = NULL;
 931                         continue;
 932                 }
 933
 934                 /*
 935                  * Try to resolve aliases also, type is CNAME(5).
 936                  * This is important as otherwise the aliased names would not
 937                  * be cached at all as the cache would not contain the aliased
 938                  * question.
 939                  *
 940                  * If any CNAME is found in DNS packet, then we cache the alias
 941                  * IP address instead of the question (as the server
 942                  * said that question has only an alias).
 943                  * This means in practice that if e.g., ipv6.google.com is
 944                  * queried, DNS server returns CNAME of that name which is
 945                  * ipv6.l.google.com. We then cache the address of the CNAME
 946                  * but return the question name to client. So the alias
 947                  * status of the name is not saved in cache and thus not
 948                  * returned to the client. We do not return DNS packets from
 949                  * cache to client saying that ipv6.google.com is an alias to
 950                  * ipv6.l.google.com but we return instead a DNS packet that
 951                  * says ipv6.google.com has address xxx which is in fact the
 952                  * address of ipv6.l.google.com. For caching purposes this
 953                  * should not cause any issues.
 954                  */
 955                 if (*type == 5 && strncmp(question, name, qlen) == 0) {
 956                         /*
 957                          * So now the alias answered the question. This is
 958                          * not very useful from caching point of view as
 959                          * the following A or AAAA records will not match the
 960                          * question. We need to find the real A/AAAA record
 961                          * of the alias and cache that.
 962                          */
 963                         unsigned char *end = NULL;
 964                         int name_len = 0, output_len;
 965
 966                         memset(rsp, 0, sizeof(rsp));
 967                         rsp_len = sizeof(rsp) - 1;
 968
 969                         /*
 970                          * Alias is in rdata part of the message,
 971                          * and next-rdlen points to it. So we need to get
 972                          * the real name of the alias.
 973                          */
 974                         ret = get_name(0, buf, next - rdlen, buf + buflen,
 975                                         rsp, rsp_len, &output_len, &end,
 976                                         name, &name_len);
 977                         if (ret != 0) {
 978                                 /* just ignore the error at this point */
 979                                 ptr = next;
 980                                 next = NULL;
 981                                 continue;
 982                         }
 983
 984                         /*
 985                          * We should now have the alias of the entry we might
 986                          * want to cache. Just remember it for a while.
 987                          * We check the alias list when we have parsed the
 988                          * A or AAAA record.
 989                          */
 990                         aliases = g_slist_prepend(aliases, g_strdup(name));
 991
 992                         ptr = next;
 993                         next = NULL;
 994                         continue;
 995                 }
 996
 997                 if (*type == qtype) {
 998                         /*
 999                          * We found correct type (A or AAAA)
1000                          */
1001                         if (check_alias(aliases, name) == TRUE ||
1002                                 (aliases == NULL && strncmp(question, name,
1003                                                         qlen) == 0)) {
1004                                 /*
1005                                  * We found an alias or the name of the rr
1006                                  * matches the question. If so, we append
1007                                  * the compressed label to the cache.
1008                                  * The end result is a response buffer that
1009                                  * will contain one or more cached and
1010                                  * compressed resource records.
1011                                  */
1012                                 if (*response_len + rsp_len > maxlen) {
1013                                         err = -ENOBUFS;
1014                                         goto out;
1015                                 }
1016                                 memcpy(response + *response_len, rsp, rsp_len);
1017                                 *response_len += rsp_len;
1018                                 (*answers)++;
1019                                 err = 0;
1020                         }
1021                 }
1022
1023                 ptr = next;
1024                 next = NULL;
1025         }
1026
1027 out:
1028         for (list = aliases; list; list = list->next)
1029                 g_free(list->data);
1030         g_slist_free(aliases);
1031
1032         return err;
1033 }
1034
1035 struct cache_timeout {
1036         time_t current_time;
1037         int max_timeout;
1038         int try_harder;
1039 };
1040
1041 static gboolean cache_check_entry(gpointer key, gpointer value,
1042                                         gpointer user_data)
1043 {
1044         struct cache_timeout *data = user_data;
1045         struct cache_entry *entry = value;
1046         int max_timeout;
1047
1048         /* Scale the number of hits by half as part of cache aging */
1049
1050         entry->hits /= 2;
1051
1052         /*
1053          * If either IPv4 or IPv6 cached entry has expired, we
1054          * remove both from the cache.
1055          */
1056
1057         if (entry->ipv4 != NULL && entry->ipv4->timeout > 0) {
1058                 max_timeout = entry->ipv4->cache_until;
1059                 if (max_timeout > data->max_timeout)
1060                         data->max_timeout = max_timeout;
1061
1062                 if (entry->ipv4->cache_until < data->current_time)
1063                         return TRUE;
1064         }
1065
1066         if (entry->ipv6 != NULL && entry->ipv6->timeout > 0) {
1067                 max_timeout = entry->ipv6->cache_until;
1068                 if (max_timeout > data->max_timeout)
1069                         data->max_timeout = max_timeout;
1070
1071                 if (entry->ipv6->cache_until < data->current_time)
1072                         return TRUE;
1073         }
1074
1075         /*
1076          * if we're asked to try harder, also remove entries that have
1077          * few hits
1078          */
1079         if (data->try_harder && entry->hits < 4)
1080                 return TRUE;
1081
1082         return FALSE;
1083 }
1084
1085 static void cache_cleanup(void)
1086 {
1087         static int max_timeout;
1088         struct cache_timeout data;
1089         int count = 0;
1090
1091         data.current_time = time(NULL);
1092         data.max_timeout = 0;
1093         data.try_harder = 0;
1094
1095         /*
1096          * In the first pass, we only remove entries that have timed out.
1097          * We use a cache of the first time to expire to do this only
1098          * when it makes sense.
1099          */
1100         if (max_timeout <= data.current_time) {
1101                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1102                                                 &data);
1103         }
1104         DBG("removed %d in the first pass", count);
1105
1106         /*
1107          * In the second pass, if the first pass turned up blank,
1108          * we also expire entries with a low hit count,
1109          * while aging the hit count at the same time.
1110          */
1111         data.try_harder = 1;
1112         if (count == 0)
1113                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1114                                                 &data);
1115
1116         if (count == 0)
1117                 /*
1118                  * If we could not remove anything, then remember
1119                  * what is the max timeout and do nothing if we
1120                  * have not yet reached it. This will prevent
1121                  * constant traversal of the cache if it is full.
1122                  */
1123                 max_timeout = data.max_timeout;
1124         else
1125                 max_timeout = 0;
1126 }
1127
1128 static gboolean cache_invalidate_entry(gpointer key, gpointer value,
1129                                         gpointer user_data)
1130 {
1131         struct cache_entry *entry = value;
1132
1133         /* first, delete any expired elements */
1134         cache_enforce_validity(entry);
1135
1136         /* if anything is not expired, mark the entry for refresh */
1137         if (entry->hits > 0 && (entry->ipv4 || entry->ipv6))
1138                 entry->want_refresh = 1;
1139
1140         /* delete the cached data */
1141         if (entry->ipv4) {
1142                 g_free(entry->ipv4->data);
1143                 g_free(entry->ipv4);
1144                 entry->ipv4 = NULL;
1145         }
1146
1147         if (entry->ipv6) {
1148                 g_free(entry->ipv6->data);
1149                 g_free(entry->ipv6);
1150                 entry->ipv6 = NULL;
1151         }
1152
1153         /* keep the entry if we want it refreshed, delete it otherwise */
1154         if (entry->want_refresh)
1155                 return FALSE;
1156         else
1157                 return TRUE;
1158 }
1159
1160 /*
1161  * cache_invalidate is called from places where the DNS landscape
1162  * has changed, say because connections are added or we entered a VPN.
1163  * The logic is to wipe all cache data, but mark all non-expired
1164  * parts of the cache for refresh rather than deleting the whole cache.
1165  */
1166 static void cache_invalidate(void)
1167 {
1168         DBG("Invalidating the DNS cache %p", cache);
1169
1170         if (cache == NULL)
1171                 return;
1172
1173         g_hash_table_foreach_remove(cache, cache_invalidate_entry, NULL);
1174 }
1175
1176 static void cache_refresh_entry(struct cache_entry *entry)
1177 {
1178
1179         cache_enforce_validity(entry);
1180
1181         if (entry->hits > 2 && entry->ipv4 == NULL)
1182                 entry->want_refresh = 1;
1183         if (entry->hits > 2 && entry->ipv6 == NULL)
1184                 entry->want_refresh = 1;
1185
1186         if (entry->want_refresh) {
1187                 char *c;
1188                 char dns_name[NS_MAXDNAME + 1];
1189                 entry->want_refresh = 0;
1190
1191                 /* turn a DNS name into a hostname with dots */
1192                 strncpy(dns_name, entry->key, NS_MAXDNAME);
1193                 c = dns_name;
1194                 while (c && *c) {
1195                         int jump;
1196                         jump = *c;
1197                         *c = '.';
1198                         c += jump + 1;
1199                 }
1200                 DBG("Refreshing %s\n", dns_name);
1201                 /* then refresh the hostname */
1202                 refresh_dns_entry(entry, &dns_name[1]);
1203         }
1204 }
1205
1206 static void cache_refresh_iterator(gpointer key, gpointer value,
1207                                         gpointer user_data)
1208 {
1209         struct cache_entry *entry = value;
1210
1211         cache_refresh_entry(entry);
1212 }
1213
1214 static void cache_refresh(void)
1215 {
1216         if (cache == NULL)
1217                 return;
1218
1219         g_hash_table_foreach(cache, cache_refresh_iterator, NULL);
1220 }
1221
1222 static int reply_query_type(unsigned char *msg, int len)
1223 {
1224         unsigned char *c;
1225         uint16_t *w;
1226         int l;
1227         int type;
1228
1229         /* skip the header */
1230         c = msg + sizeof(struct domain_hdr);
1231         len -= sizeof(struct domain_hdr);
1232
1233         if (len < 0)
1234                 return 0;
1235
1236         /* now the query, which is a name and 2 16 bit words */
1237         l = dns_name_length(c) + 1;
1238         c += l;
1239         w = (uint16_t *) c;
1240         type = ntohs(*w);
1241
1242         return type;
1243 }
1244
1245 static int cache_update(struct server_data *srv, unsigned char *msg,
1246                         unsigned int msg_len)
1247 {
1248         int offset = protocol_offset(srv->protocol);
1249         int err, qlen, ttl = 0;
1250         uint16_t answers = 0, type = 0, class = 0;
1251         struct domain_hdr *hdr = (void *)(msg + offset);
1252         struct domain_question *q;
1253         struct cache_entry *entry;
1254         struct cache_data *data;
1255         char question[NS_MAXDNAME + 1];
1256         unsigned char response[NS_MAXDNAME + 1];
1257         unsigned char *ptr;
1258         unsigned int rsplen;
1259         gboolean new_entry = TRUE;
1260         time_t current_time;
1261
1262         if (cache_size >= MAX_CACHE_SIZE) {
1263                 cache_cleanup();
1264                 if (cache_size >= MAX_CACHE_SIZE)
1265                         return 0;
1266         }
1267
1268         current_time = time(NULL);
1269
1270         /* don't do a cache refresh more than twice a minute */
1271         if (next_refresh < current_time) {
1272                 cache_refresh();
1273                 next_refresh = current_time + 30;
1274         }
1275
1276         if (offset < 0)
1277                 return 0;
1278
1279         DBG("offset %d hdr %p msg %p rcode %d", offset, hdr, msg, hdr->rcode);
1280
1281         /* Continue only if response code is 0 (=ok) */
1282         if (hdr->rcode != 0)
1283                 return 0;
1284
1285         rsplen = sizeof(response) - 1;
1286         question[sizeof(question) - 1] = '\0';
1287
1288         err = parse_response(msg + offset, msg_len - offset,
1289                                 question, sizeof(question) - 1,
1290                                 &type, &class, &ttl,
1291                                 response, &rsplen, &answers);
1292
1293         /*
1294          * special case: if we do a ipv6 lookup and get no result
1295          * for a record that's already in our ipv4 cache.. we want
1296          * to cache the negative response.
1297          */
1298         if ((err == -ENOMSG || err == -ENOBUFS) &&
1299                         reply_query_type(msg + offset,
1300                                         msg_len - offset) == 28) {
1301                 entry = g_hash_table_lookup(cache, question);
1302                 if (entry && entry->ipv4 && entry->ipv6 == NULL) {
1303                         int cache_offset = 0;
1304
1305                         data = g_try_new(struct cache_data, 1);
1306                         if (data == NULL)
1307                                 return -ENOMEM;
1308                         data->inserted = entry->ipv4->inserted;
1309                         data->type = type;
1310                         data->answers = hdr->ancount;
1311                         data->timeout = entry->ipv4->timeout;
1312                         if (srv->protocol == IPPROTO_UDP)
1313                                 cache_offset = 2;
1314                         data->data_len = msg_len + cache_offset;
1315                         data->data = ptr = g_malloc(data->data_len);
1316                         ptr[0] = (data->data_len - 2) / 256;
1317                         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1318                         if (srv->protocol == IPPROTO_UDP)
1319                                 ptr += 2;
1320                         data->valid_until = entry->ipv4->valid_until;
1321                         data->cache_until = entry->ipv4->cache_until;
1322                         memcpy(ptr, msg, msg_len);
1323                         entry->ipv6 = data;
1324                         /*
1325                          * we will get a "hit" when we serve the response
1326                          * out of the cache
1327                          */
1328                         entry->hits--;
1329                         if (entry->hits < 0)
1330                                 entry->hits = 0;
1331                         return 0;
1332                 }
1333         }
1334
1335         if (err < 0 || ttl == 0)
1336                 return 0;
1337
1338         qlen = strlen(question);
1339
1340         /*
1341          * If the cache contains already data, check if the
1342          * type of the cached data is the same and do not add
1343          * to cache if data is already there.
1344          * This is needed so that we can cache both A and AAAA
1345          * records for the same name.
1346          */
1347         entry = g_hash_table_lookup(cache, question);
1348         if (entry == NULL) {
1349                 entry = g_try_new(struct cache_entry, 1);
1350                 if (entry == NULL)
1351                         return -ENOMEM;
1352
1353                 data = g_try_new(struct cache_data, 1);
1354                 if (data == NULL) {
1355                         g_free(entry);
1356                         return -ENOMEM;
1357                 }
1358
1359                 entry->key = g_strdup(question);
1360                 entry->ipv4 = entry->ipv6 = NULL;
1361                 entry->want_refresh = 0;
1362                 entry->hits = 0;
1363
1364                 if (type == 1)
1365                         entry->ipv4 = data;
1366                 else
1367                         entry->ipv6 = data;
1368         } else {
1369                 if (type == 1 && entry->ipv4 != NULL)
1370                         return 0;
1371
1372                 if (type == 28 && entry->ipv6 != NULL)
1373                         return 0;
1374
1375                 data = g_try_new(struct cache_data, 1);
1376                 if (data == NULL)
1377                         return -ENOMEM;
1378
1379                 if (type == 1)
1380                         entry->ipv4 = data;
1381                 else
1382                         entry->ipv6 = data;
1383
1384                 /*
1385                  * compensate for the hit we'll get for serving
1386                  * the response out of the cache
1387                  */
1388                 entry->hits--;
1389                 if (entry->hits < 0)
1390                         entry->hits = 0;
1391
1392                 new_entry = FALSE;
1393         }
1394
1395         if (ttl < MIN_CACHE_TTL)
1396                 ttl = MIN_CACHE_TTL;
1397
1398         data->inserted = current_time;
1399         data->type = type;
1400         data->answers = answers;
1401         data->timeout = ttl;
1402         /*
1403          * The "2" in start of the length is the TCP offset. We allocate it
1404          * here even for UDP packet because it simplifies the sending
1405          * of cached packet.
1406          */
1407         data->data_len = 2 + 12 + qlen + 1 + 2 + 2 + rsplen;
1408         data->data = ptr = g_malloc(data->data_len);
1409         data->valid_until = current_time + ttl;
1410
1411         /*
1412          * Restrict the cached DNS record TTL to some sane value
1413          * in order to prevent data staying in the cache too long.
1414          */
1415         if (ttl > MAX_CACHE_TTL)
1416                 ttl = MAX_CACHE_TTL;
1417
1418         data->cache_until = round_down_ttl(current_time + ttl, ttl);
1419
1420         if (data->data == NULL) {
1421                 g_free(entry->key);
1422                 g_free(data);
1423                 g_free(entry);
1424                 return -ENOMEM;
1425         }
1426
1427         /*
1428          * We cache the two extra bytes at the start of the message
1429          * in a TCP packet. When sending UDP packet, we skip the first
1430          * two bytes. This way we do not need to know the format
1431          * (UDP/TCP) of the cached message.
1432          */
1433         ptr[0] = (data->data_len - 2) / 256;
1434         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1435         if (srv->protocol == IPPROTO_UDP)
1436                 ptr += 2;
1437
1438         memcpy(ptr, msg, offset + 12);
1439         memcpy(ptr + offset + 12, question, qlen + 1); /* copy also the \0 */
1440
1441         q = (void *) (ptr + offset + 12 + qlen + 1);
1442         q->type = htons(type);
1443         q->class = htons(class);
1444         memcpy(ptr + offset + 12 + qlen + 1 + sizeof(struct domain_question),
1445                 response, rsplen);
1446
1447         if (new_entry == TRUE) {
1448                 g_hash_table_replace(cache, entry->key, entry);
1449                 cache_size++;
1450         }
1451
1452         DBG("cache %d %squestion \"%s\" type %d ttl %d size %zd packet %u "
1453                                                                 "dns len %u",
1454                 cache_size, new_entry ? "new " : "old ",
1455                 question, type, ttl,
1456                 sizeof(*entry) + sizeof(*data) + data->data_len + qlen,
1457                 data->data_len,
1458                 srv->protocol == IPPROTO_TCP ?
1459                         (unsigned int)(data->data[0] * 256 + data->data[1]) :
1460                         data->data_len);
1461
1462         return 0;
1463 }
1464
1465 static int ns_resolv(struct server_data *server, struct request_data *req,
1466                                 gpointer request, gpointer name)
1467 {
1468         GList *list;
1469         int sk, err, type = 0;
1470         char *dot, *lookup = (char *) name;
1471         struct cache_entry *entry;
1472
1473         entry = cache_check(request, &type, req->protocol);
1474         if (entry != NULL) {
1475                 int ttl_left = 0;
1476                 struct cache_data *data;
1477
1478                 DBG("cache hit %s type %s", lookup, type == 1 ? "A" : "AAAA");
1479                 if (type == 1)
1480                         data = entry->ipv4;
1481                 else
1482                         data = entry->ipv6;
1483
1484                 if (data) {
1485                         ttl_left = data->valid_until - time(NULL);
1486                         entry->hits++;
1487                 }
1488
1489                 if (data != NULL && req->protocol == IPPROTO_TCP) {
1490                         send_cached_response(req->client_sk, data->data,
1491                                         data->data_len, NULL, 0, IPPROTO_TCP,
1492                                         req->srcid, data->answers, ttl_left);
1493                         return 1;
1494                 }
1495
1496                 if (data != NULL && req->protocol == IPPROTO_UDP) {
1497                         int udp_sk = g_io_channel_unix_get_fd(
1498                                         req->ifdata->udp_listener_channel);
1499
1500                         send_cached_response(udp_sk, data->data,
1501                                 data->data_len, &req->sa, req->sa_len,
1502                                 IPPROTO_UDP, req->srcid, data->answers,
1503                                 ttl_left);
1504                         return 1;
1505                 }
1506         }
1507
1508         sk = g_io_channel_unix_get_fd(server->channel);
1509
1510         err = sendto(sk, request, req->request_len, MSG_NOSIGNAL,
1511                         &server->server_addr, server->server_addr_len);
1512         if (err < 0) {
1513                 DBG("Cannot send message to server %s sock %d "
1514                         "protocol %d (%s/%d)",
1515                         server->server, sk, server->protocol,
1516                         strerror(errno), errno);
1517                 return -EIO;
1518         }
1519
1520         req->numserv++;
1521
1522         /* If we have more than one dot, we don't add domains */
1523         dot = strchr(lookup, '.');
1524         if (dot != NULL && dot != lookup + strlen(lookup) - 1)
1525                 return 0;
1526
1527         if (server->domains != NULL && server->domains->data != NULL)
1528                 req->append_domain = TRUE;
1529
1530         for (list = server->domains; list; list = list->next) {
1531                 char *domain;
1532                 unsigned char alt[1024];
1533                 struct domain_hdr *hdr = (void *) &alt;
1534                 int altlen, domlen, offset;
1535
1536                 domain = list->data;
1537
1538                 if (domain == NULL)
1539                         continue;
1540
1541                 offset = protocol_offset(server->protocol);
1542                 if (offset < 0)
1543                         return offset;
1544
1545                 domlen = strlen(domain) + 1;
1546                 if (domlen < 5)
1547                         return -EINVAL;
1548
1549                 alt[offset] = req->altid & 0xff;
1550                 alt[offset + 1] = req->altid >> 8;
1551
1552                 memcpy(alt + offset + 2, request + offset + 2, 10);
1553                 hdr->qdcount = htons(1);
1554
1555                 altlen = append_query(alt + offset + 12, sizeof(alt) - 12,
1556                                         name, domain);
1557                 if (altlen < 0)
1558                         return -EINVAL;
1559
1560                 altlen += 12;
1561
1562                 memcpy(alt + offset + altlen,
1563                         request + offset + altlen - domlen,
1564                                 req->request_len - altlen - offset + domlen);
1565
1566                 if (server->protocol == IPPROTO_TCP) {
1567                         int req_len = req->request_len + domlen - 2;
1568
1569                         alt[0] = (req_len >> 8) & 0xff;
1570                         alt[1] = req_len & 0xff;
1571                 }
1572
1573                 err = send(sk, alt, req->request_len + domlen, MSG_NOSIGNAL);
1574                 if (err < 0)
1575                         return -EIO;
1576
1577                 req->numserv++;
1578         }
1579
1580         return 0;
1581 }
1582
1583 static void destroy_request_data(struct request_data *req)
1584 {
1585         if (req->timeout > 0)
1586                 g_source_remove(req->timeout);
1587
1588         g_free(req->resp);
1589         g_free(req->request);
1590         g_free(req->name);
1591         g_free(req);
1592 }
1593
1594 static int forward_dns_reply(unsigned char *reply, int reply_len, int protocol,
1595                                 struct server_data *data)
1596 {
1597         struct domain_hdr *hdr;
1598         struct request_data *req;
1599         int dns_id, sk, err, offset = protocol_offset(protocol);
1600         struct listener_data *ifdata;
1601
1602         if (offset < 0)
1603                 return offset;
1604
1605         hdr = (void *)(reply + offset);
1606         dns_id = reply[offset] | reply[offset + 1] << 8;
1607
1608         DBG("Received %d bytes (id 0x%04x)", reply_len, dns_id);
1609
1610         req = find_request(dns_id);
1611         if (req == NULL)
1612                 return -EINVAL;
1613
1614         DBG("id 0x%04x rcode %d", hdr->id, hdr->rcode);
1615
1616         ifdata = req->ifdata;
1617
1618         reply[offset] = req->srcid & 0xff;
1619         reply[offset + 1] = req->srcid >> 8;
1620
1621         req->numresp++;
1622
1623         if (hdr->rcode == 0 || req->resp == NULL) {
1624
1625                 /*
1626                  * If the domain name was append
1627                  * remove it before forwarding the reply.
1628                  */
1629                 if (req->append_domain == TRUE) {
1630                         unsigned char *ptr;
1631                         uint8_t host_len;
1632                         unsigned int domain_len;
1633
1634                         /*
1635                          * ptr points to the first char of the hostname.
1636                          * ->hostname.domain.net
1637                          */
1638                         ptr = reply + offset + sizeof(struct domain_hdr);
1639                         host_len = *ptr;
1640                         domain_len = strlen((const char *)ptr + host_len + 1);
1641
1642                         /*
1643                          * Remove the domain name and replace it by the end
1644                          * of reply. Check if the domain is really there
1645                          * before trying to copy the data. The domain_len can
1646                          * be 0 because if the original query did not contain
1647                          * a domain name, then we are sending two packets,
1648                          * first without the domain name and the second packet
1649                          * with domain name. The append_domain is set to true
1650                          * even if we sent the first packet without domain
1651                          * name. In this case we end up in this branch.
1652                          */
1653                         if (domain_len > 0) {
1654                                 /*
1655                                  * Note that we must use memmove() here,
1656                                  * because the memory areas can overlap.
1657                                  */
1658                                 memmove(ptr + host_len + 1,
1659                                         ptr + host_len + domain_len + 1,
1660                                         reply_len - (ptr - reply + domain_len));
1661
1662                                 reply_len = reply_len - domain_len;
1663                         }
1664                 }
1665
1666                 g_free(req->resp);
1667                 req->resplen = 0;
1668
1669                 req->resp = g_try_malloc(reply_len);
1670                 if (req->resp == NULL)
1671                         return -ENOMEM;
1672
1673                 memcpy(req->resp, reply, reply_len);
1674                 req->resplen = reply_len;
1675
1676                 cache_update(data, reply, reply_len);
1677         }
1678
1679         if (hdr->rcode > 0 && req->numresp < req->numserv)
1680                 return -EINVAL;
1681
1682         request_list = g_slist_remove(request_list, req);
1683
1684         if (protocol == IPPROTO_UDP) {
1685                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
1686                 err = sendto(sk, req->resp, req->resplen, 0,
1687                              &req->sa, req->sa_len);
1688         } else {
1689                 sk = req->client_sk;
1690                 err = send(sk, req->resp, req->resplen, MSG_NOSIGNAL);
1691                 close(sk);
1692         }
1693
1694         if (err < 0)
1695                 DBG("Cannot send msg, sk %d proto %d errno %d/%s", sk,
1696                         protocol, errno, strerror(errno));
1697         else
1698                 DBG("proto %d sent %d bytes to %d", protocol, err, sk);
1699
1700         destroy_request_data(req);
1701
1702         return err;
1703 }
1704
1705 static void cache_element_destroy(gpointer value)
1706 {
1707         struct cache_entry *entry = value;
1708
1709         if (entry == NULL)
1710                 return;
1711
1712         if (entry->ipv4 != NULL) {
1713                 g_free(entry->ipv4->data);
1714                 g_free(entry->ipv4);
1715         }
1716
1717         if (entry->ipv6 != NULL) {
1718                 g_free(entry->ipv6->data);
1719                 g_free(entry->ipv6);
1720         }
1721
1722         g_free(entry->key);
1723         g_free(entry);
1724
1725         if (--cache_size < 0)
1726                 cache_size = 0;
1727 }
1728
1729 static gboolean try_remove_cache(gpointer user_data)
1730 {
1731         if (__sync_fetch_and_sub(&cache_refcount, 1) == 1) {
1732                 DBG("No cache users, removing it.");
1733
1734                 g_hash_table_destroy(cache);
1735                 cache = NULL;
1736         }
1737
1738         return FALSE;
1739 }
1740
1741 static void destroy_server(struct server_data *server)
1742 {
1743         GList *list;
1744
1745         DBG("interface %s server %s sock %d", server->interface, server->server,
1746                 g_io_channel_unix_get_fd(server->channel));
1747
1748         server_list = g_slist_remove(server_list, server);
1749
1750         if (server->watch > 0)
1751                 g_source_remove(server->watch);
1752
1753         if (server->timeout > 0)
1754                 g_source_remove(server->timeout);
1755
1756         g_io_channel_unref(server->channel);
1757
1758         if (server->protocol == IPPROTO_UDP)
1759                 DBG("Removing DNS server %s", server->server);
1760
1761         g_free(server->incoming_reply);
1762         g_free(server->server);
1763         for (list = server->domains; list; list = list->next) {
1764                 char *domain = list->data;
1765
1766                 server->domains = g_list_remove(server->domains, domain);
1767                 g_free(domain);
1768         }
1769         g_free(server->interface);
1770
1771         /*
1772          * We do not remove cache right away but delay it few seconds.
1773          * The idea is that when IPv6 DNS server is added via RDNSS, it has a
1774          * lifetime. When the lifetime expires we decrease the refcount so it
1775          * is possible that the cache is then removed. Because a new DNS server
1776          * is usually created almost immediately we would then loose the cache
1777          * without any good reason. The small delay allows the new RDNSS to
1778          * create a new DNS server instance and the refcount does not go to 0.
1779          */
1780         g_timeout_add_seconds(3, try_remove_cache, NULL);
1781
1782         g_free(server);
1783 }
1784
1785 static gboolean udp_server_event(GIOChannel *channel, GIOCondition condition,
1786                                                         gpointer user_data)
1787 {
1788         unsigned char buf[4096];
1789         int sk, err, len;
1790         struct server_data *data = user_data;
1791
1792         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1793                 connman_error("Error with UDP server %s", data->server);
1794                 data->watch = 0;
1795                 return FALSE;
1796         }
1797
1798         sk = g_io_channel_unix_get_fd(channel);
1799
1800         len = recv(sk, buf, sizeof(buf), 0);
1801         if (len < 12)
1802                 return TRUE;
1803
1804         err = forward_dns_reply(buf, len, IPPROTO_UDP, data);
1805         if (err < 0)
1806                 return TRUE;
1807
1808         return TRUE;
1809 }
1810
1811 static gboolean tcp_server_event(GIOChannel *channel, GIOCondition condition,
1812                                                         gpointer user_data)
1813 {
1814         int sk;
1815         struct server_data *server = user_data;
1816
1817         sk = g_io_channel_unix_get_fd(channel);
1818         if (sk == 0)
1819                 return FALSE;
1820
1821         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1822                 GSList *list;
1823 hangup:
1824                 DBG("TCP server channel closed, sk %d", sk);
1825
1826                 /*
1827                  * Discard any partial response which is buffered; better
1828                  * to get a proper response from a working server.
1829                  */
1830                 g_free(server->incoming_reply);
1831                 server->incoming_reply = NULL;
1832
1833                 for (list = request_list; list; list = list->next) {
1834                         struct request_data *req = list->data;
1835                         struct domain_hdr *hdr;
1836
1837                         if (req->protocol == IPPROTO_UDP)
1838                                 continue;
1839
1840                         if (req->request == NULL)
1841                                 continue;
1842
1843                         /*
1844                          * If we're not waiting for any further response
1845                          * from another name server, then we send an error
1846                          * response to the client.
1847                          */
1848                         if (req->numserv && --(req->numserv))
1849                                 continue;
1850
1851                         hdr = (void *) (req->request + 2);
1852                         hdr->id = req->srcid;
1853                         send_response(req->client_sk, req->request,
1854                                 req->request_len, NULL, 0, IPPROTO_TCP);
1855
1856                         request_list = g_slist_remove(request_list, req);
1857                 }
1858
1859                 destroy_server(server);
1860
1861                 return FALSE;
1862         }
1863
1864         if ((condition & G_IO_OUT) && !server->connected) {
1865                 GSList *list;
1866                 GList *domains;
1867                 int no_request_sent = TRUE;
1868                 struct server_data *udp_server;
1869
1870                 udp_server = find_server(server->interface, server->server,
1871                                                                 IPPROTO_UDP);
1872                 if (udp_server != NULL) {
1873                         for (domains = udp_server->domains; domains;
1874                                                 domains = domains->next) {
1875                                 char *dom = domains->data;
1876
1877                                 DBG("Adding domain %s to %s",
1878                                                 dom, server->server);
1879
1880                                 server->domains = g_list_append(server->domains,
1881                                                                 g_strdup(dom));
1882                         }
1883                 }
1884
1885                 server->connected = TRUE;
1886                 server_list = g_slist_append(server_list, server);
1887
1888                 if (server->timeout > 0) {
1889                         g_source_remove(server->timeout);
1890                         server->timeout = 0;
1891                 }
1892
1893                 for (list = request_list; list; ) {
1894                         struct request_data *req = list->data;
1895                         int status;
1896
1897                         if (req->protocol == IPPROTO_UDP) {
1898                                 list = list->next;
1899                                 continue;
1900                         }
1901
1902                         DBG("Sending req %s over TCP", (char *)req->name);
1903
1904                         status = ns_resolv(server, req,
1905                                                 req->request, req->name);
1906                         if (status > 0) {
1907                                 /*
1908                                  * A cached result was sent,
1909                                  * so the request can be released
1910                                  */
1911                                 list = list->next;
1912                                 request_list = g_slist_remove(request_list, req);
1913                                 destroy_request_data(req);
1914                                 continue;
1915                         }
1916
1917                         if (status < 0) {
1918                                 list = list->next;
1919                                 continue;
1920                         }
1921
1922                         no_request_sent = FALSE;
1923
1924                         if (req->timeout > 0)
1925                                 g_source_remove(req->timeout);
1926
1927                         req->timeout = g_timeout_add_seconds(30,
1928                                                 request_timeout, req);
1929                         list = list->next;
1930                 }
1931
1932                 if (no_request_sent == TRUE) {
1933                         destroy_server(server);
1934                         return FALSE;
1935                 }
1936
1937         } else if (condition & G_IO_IN) {
1938                 struct partial_reply *reply = server->incoming_reply;
1939                 int bytes_recv;
1940
1941                 if (!reply) {
1942                         unsigned char reply_len_buf[2];
1943                         uint16_t reply_len;
1944
1945                         bytes_recv = recv(sk, reply_len_buf, 2, MSG_PEEK);
1946                         if (!bytes_recv) {
1947                                 goto hangup;
1948                         } else if (bytes_recv < 0) {
1949                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1950                                         return TRUE;
1951
1952                                 connman_error("DNS proxy error %s",
1953                                                 strerror(errno));
1954                                 goto hangup;
1955                         } else if (bytes_recv < 2)
1956                                 return TRUE;
1957
1958                         reply_len = reply_len_buf[1] | reply_len_buf[0] << 8;
1959                         reply_len += 2;
1960
1961                         DBG("TCP reply %d bytes from %d", reply_len, sk);
1962
1963                         reply = g_try_malloc(sizeof(*reply) + reply_len + 2);
1964                         if (!reply)
1965                                 return TRUE;
1966
1967                         reply->len = reply_len;
1968                         reply->received = 0;
1969
1970                         server->incoming_reply = reply;
1971                 }
1972
1973                 while (reply->received < reply->len) {
1974                         bytes_recv = recv(sk, reply->buf + reply->received,
1975                                         reply->len - reply->received, 0);
1976                         if (!bytes_recv) {
1977                                 connman_error("DNS proxy TCP disconnect");
1978                                 break;
1979                         } else if (bytes_recv < 0) {
1980                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1981                                         return TRUE;
1982
1983                                 connman_error("DNS proxy error %s",
1984                                                 strerror(errno));
1985                                 break;
1986                         }
1987                         reply->received += bytes_recv;
1988                 }
1989
1990                 forward_dns_reply(reply->buf, reply->received, IPPROTO_TCP,
1991                                         server);
1992
1993                 g_free(reply);
1994                 server->incoming_reply = NULL;
1995
1996                 destroy_server(server);
1997
1998                 return FALSE;
1999         }
2000
2001         return TRUE;
2002 }
2003
2004 static gboolean tcp_idle_timeout(gpointer user_data)
2005 {
2006         struct server_data *server = user_data;
2007
2008         DBG("");
2009
2010         if (server == NULL)
2011                 return FALSE;
2012
2013         destroy_server(server);
2014
2015         return FALSE;
2016 }
2017
2018 static struct server_data *create_server(const char *interface,
2019                                         const char *domain, const char *server,
2020                                         int protocol)
2021 {
2022         struct addrinfo hints, *rp;
2023         struct server_data *data;
2024         int sk, ret;
2025
2026         DBG("interface %s server %s", interface, server);
2027
2028         memset(&hints, 0, sizeof(hints));
2029
2030         switch (protocol) {
2031         case IPPROTO_UDP:
2032                 hints.ai_socktype = SOCK_DGRAM;
2033                 break;
2034
2035         case IPPROTO_TCP:
2036                 hints.ai_socktype = SOCK_STREAM;
2037                 break;
2038
2039         default:
2040                 return NULL;
2041         }
2042         hints.ai_family = AF_UNSPEC;
2043         hints.ai_flags = AI_NUMERICSERV | AI_NUMERICHOST;
2044
2045         ret = getaddrinfo(server, "53", &hints, &rp);
2046         if (ret) {
2047                 connman_error("Failed to parse server %s address: %s\n",
2048                               server, gai_strerror(ret));
2049                 return NULL;
2050         }
2051         /* Do not blindly copy this code elsewhere; it doesn't loop over the
2052            results using ->ai_next as it should. That's OK in *this* case
2053            because it was a numeric lookup; we *know* there's only one. */
2054
2055         sk = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
2056         if (sk < 0) {
2057                 connman_error("Failed to create server %s socket", server);
2058                 freeaddrinfo(rp);
2059                 return NULL;
2060         }
2061
2062         DBG("sk %d", sk);
2063
2064         if (interface != NULL) {
2065                 if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2066                                 interface, strlen(interface) + 1) < 0) {
2067                         connman_error("Failed to bind server %s "
2068                                                 "to interface %s",
2069                                                         server, interface);
2070                         freeaddrinfo(rp);
2071                         close(sk);
2072                         return NULL;
2073                 }
2074         }
2075
2076         data = g_try_new0(struct server_data, 1);
2077         if (data == NULL) {
2078                 connman_error("Failed to allocate server %s data", server);
2079                 freeaddrinfo(rp);
2080                 close(sk);
2081                 return NULL;
2082         }
2083
2084         data->channel = g_io_channel_unix_new(sk);
2085         if (data->channel == NULL) {
2086                 connman_error("Failed to create server %s channel", server);
2087                 freeaddrinfo(rp);
2088                 close(sk);
2089                 g_free(data);
2090                 return NULL;
2091         }
2092
2093         g_io_channel_set_close_on_unref(data->channel, TRUE);
2094
2095         if (protocol == IPPROTO_TCP) {
2096                 g_io_channel_set_flags(data->channel, G_IO_FLAG_NONBLOCK, NULL);
2097                 data->watch = g_io_add_watch(data->channel,
2098                         G_IO_OUT | G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR,
2099                                                 tcp_server_event, data);
2100                 data->timeout = g_timeout_add_seconds(30, tcp_idle_timeout,
2101                                                                 data);
2102         } else
2103                 data->watch = g_io_add_watch(data->channel,
2104                         G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2105                                                 udp_server_event, data);
2106
2107         data->interface = g_strdup(interface);
2108         if (domain)
2109                 data->domains = g_list_append(data->domains, g_strdup(domain));
2110         data->server = g_strdup(server);
2111         data->protocol = protocol;
2112         data->server_addr_len = rp->ai_addrlen;
2113         memcpy(&data->server_addr, rp->ai_addr, rp->ai_addrlen);
2114
2115         ret = connect(sk, rp->ai_addr, rp->ai_addrlen);
2116         freeaddrinfo(rp);
2117         if (ret < 0) {
2118                 if ((protocol == IPPROTO_TCP && errno != EINPROGRESS) ||
2119                                 protocol == IPPROTO_UDP) {
2120                         GList *list;
2121
2122                         connman_error("Failed to connect to server %s", server);
2123                         if (data->watch > 0)
2124                                 g_source_remove(data->watch);
2125                         if (data->timeout > 0)
2126                                 g_source_remove(data->timeout);
2127
2128                         g_io_channel_unref(data->channel);
2129                         close(sk);
2130
2131                         g_free(data->server);
2132                         g_free(data->interface);
2133                         for (list = data->domains; list; list = list->next) {
2134                                 char *tmp_domain = list->data;
2135
2136                                 data->domains = g_list_remove(data->domains,
2137                                                                 tmp_domain);
2138                                 g_free(tmp_domain);
2139                         }
2140                         g_free(data);
2141                         return NULL;
2142                 }
2143         }
2144
2145         if (__sync_fetch_and_add(&cache_refcount, 1) == 0)
2146                 cache = g_hash_table_new_full(g_str_hash,
2147                                         g_str_equal,
2148                                         NULL,
2149                                         cache_element_destroy);
2150
2151         if (protocol == IPPROTO_UDP) {
2152                 /* Enable new servers by default */
2153                 data->enabled = TRUE;
2154                 DBG("Adding DNS server %s", data->server);
2155
2156                 server_list = g_slist_append(server_list, data);
2157         }
2158
2159         return data;
2160 }
2161
2162 static gboolean resolv(struct request_data *req,
2163                                 gpointer request, gpointer name)
2164 {
2165         GSList *list;
2166
2167         for (list = server_list; list; list = list->next) {
2168                 struct server_data *data = list->data;
2169
2170                 DBG("server %s enabled %d", data->server, data->enabled);
2171
2172                 if (data->enabled == FALSE)
2173                         continue;
2174
2175                 if (data->watch == 0 && data->protocol == IPPROTO_UDP)
2176                         data->watch = g_io_add_watch(data->channel,
2177                                 G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2178                                                 udp_server_event, data);
2179
2180                 if (ns_resolv(data, req, request, name) > 0)
2181                         return TRUE;
2182         }
2183
2184         return FALSE;
2185 }
2186
2187 static void append_domain(const char *interface, const char *domain)
2188 {
2189         GSList *list;
2190
2191         DBG("interface %s domain %s", interface, domain);
2192
2193         if (domain == NULL)
2194                 return;
2195
2196         for (list = server_list; list; list = list->next) {
2197                 struct server_data *data = list->data;
2198                 GList *dom_list;
2199                 char *dom;
2200                 gboolean dom_found = FALSE;
2201
2202                 if (data->interface == NULL)
2203                         continue;
2204
2205                 if (g_str_equal(data->interface, interface) == FALSE)
2206                         continue;
2207
2208                 for (dom_list = data->domains; dom_list;
2209                                 dom_list = dom_list->next) {
2210                         dom = dom_list->data;
2211
2212                         if (g_str_equal(dom, domain)) {
2213                                 dom_found = TRUE;
2214                                 break;
2215                         }
2216                 }
2217
2218                 if (dom_found == FALSE) {
2219                         data->domains =
2220                                 g_list_append(data->domains, g_strdup(domain));
2221                 }
2222         }
2223 }
2224
2225 int __connman_dnsproxy_append(const char *interface, const char *domain,
2226                                                         const char *server)
2227 {
2228         struct server_data *data;
2229
2230         DBG("interface %s server %s", interface, server);
2231
2232         if (server == NULL && domain == NULL)
2233                 return -EINVAL;
2234
2235         if (server == NULL) {
2236                 append_domain(interface, domain);
2237
2238                 return 0;
2239         }
2240
2241         if (g_str_equal(server, "127.0.0.1") == TRUE)
2242                 return -ENODEV;
2243
2244         data = find_server(interface, server, IPPROTO_UDP);
2245         if (data != NULL) {
2246                 append_domain(interface, domain);
2247                 return 0;
2248         }
2249
2250         data = create_server(interface, domain, server, IPPROTO_UDP);
2251         if (data == NULL)
2252                 return -EIO;
2253
2254         return 0;
2255 }
2256
2257 static void remove_server(const char *interface, const char *domain,
2258                         const char *server, int protocol)
2259 {
2260         struct server_data *data;
2261
2262         data = find_server(interface, server, protocol);
2263         if (data == NULL)
2264                 return;
2265
2266         destroy_server(data);
2267 }
2268
2269 int __connman_dnsproxy_remove(const char *interface, const char *domain,
2270                                                         const char *server)
2271 {
2272         DBG("interface %s server %s", interface, server);
2273
2274         if (server == NULL)
2275                 return -EINVAL;
2276
2277         if (g_str_equal(server, "127.0.0.1") == TRUE)
2278                 return -ENODEV;
2279
2280         remove_server(interface, domain, server, IPPROTO_UDP);
2281         remove_server(interface, domain, server, IPPROTO_TCP);
2282
2283         return 0;
2284 }
2285
2286 void __connman_dnsproxy_flush(void)
2287 {
2288         GSList *list;
2289
2290         list = request_list;
2291         while (list) {
2292                 struct request_data *req = list->data;
2293
2294                 list = list->next;
2295
2296                 if (resolv(req, req->request, req->name) == TRUE) {
2297                         /*
2298                          * A cached result was sent,
2299                          * so the request can be released
2300                          */
2301                         request_list =
2302                                 g_slist_remove(request_list, req);
2303                         destroy_request_data(req);
2304                         continue;
2305                 }
2306
2307                 if (req->timeout > 0)
2308                         g_source_remove(req->timeout);
2309                 req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2310         }
2311 }
2312
2313 static void dnsproxy_offline_mode(connman_bool_t enabled)
2314 {
2315         GSList *list;
2316
2317         DBG("enabled %d", enabled);
2318
2319         for (list = server_list; list; list = list->next) {
2320                 struct server_data *data = list->data;
2321
2322                 if (enabled == FALSE) {
2323                         DBG("Enabling DNS server %s", data->server);
2324                         data->enabled = TRUE;
2325                         cache_invalidate();
2326                         cache_refresh();
2327                 } else {
2328                         DBG("Disabling DNS server %s", data->server);
2329                         data->enabled = FALSE;
2330                         cache_invalidate();
2331                 }
2332         }
2333 }
2334
2335 static void dnsproxy_default_changed(struct connman_service *service)
2336 {
2337         GSList *list;
2338         char *interface;
2339
2340         DBG("service %p", service);
2341
2342         /* DNS has changed, invalidate the cache */
2343         cache_invalidate();
2344
2345         if (service == NULL) {
2346                 /* When no services are active, then disable DNS proxying */
2347                 dnsproxy_offline_mode(TRUE);
2348                 return;
2349         }
2350
2351         interface = connman_service_get_interface(service);
2352         if (interface == NULL)
2353                 return;
2354
2355         for (list = server_list; list; list = list->next) {
2356                 struct server_data *data = list->data;
2357
2358                 if (g_strcmp0(data->interface, interface) == 0) {
2359                         DBG("Enabling DNS server %s", data->server);
2360                         data->enabled = TRUE;
2361                 } else {
2362                         DBG("Disabling DNS server %s", data->server);
2363                         data->enabled = FALSE;
2364                 }
2365         }
2366
2367         g_free(interface);
2368         cache_refresh();
2369 }
2370
2371 static struct connman_notifier dnsproxy_notifier = {
2372         .name                   = "dnsproxy",
2373         .default_changed        = dnsproxy_default_changed,
2374         .offline_mode           = dnsproxy_offline_mode,
2375 };
2376
2377 static unsigned char opt_edns0_type[2] = { 0x00, 0x29 };
2378
2379 static int parse_request(unsigned char *buf, int len,
2380                                         char *name, unsigned int size)
2381 {
2382         struct domain_hdr *hdr = (void *) buf;
2383         uint16_t qdcount = ntohs(hdr->qdcount);
2384         uint16_t arcount = ntohs(hdr->arcount);
2385         unsigned char *ptr;
2386         char *last_label = NULL;
2387         unsigned int remain, used = 0;
2388
2389         if (len < 12)
2390                 return -EINVAL;
2391
2392         DBG("id 0x%04x qr %d opcode %d qdcount %d arcount %d",
2393                                         hdr->id, hdr->qr, hdr->opcode,
2394                                                         qdcount, arcount);
2395
2396         if (hdr->qr != 0 || qdcount != 1)
2397                 return -EINVAL;
2398
2399         name[0] = '\0';
2400
2401         ptr = buf + sizeof(struct domain_hdr);
2402         remain = len - sizeof(struct domain_hdr);
2403
2404         while (remain > 0) {
2405                 uint8_t label_len = *ptr;
2406
2407                 if (label_len == 0x00) {
2408                         last_label = (char *) (ptr + 1);
2409                         break;
2410                 }
2411
2412                 if (used + label_len + 1 > size)
2413                         return -ENOBUFS;
2414
2415                 strncat(name, (char *) (ptr + 1), label_len);
2416                 strcat(name, ".");
2417
2418                 used += label_len + 1;
2419
2420                 ptr += label_len + 1;
2421                 remain -= label_len + 1;
2422         }
2423
2424         if (last_label && arcount && remain >= 9 && last_label[4] == 0 &&
2425                                 !memcmp(last_label + 5, opt_edns0_type, 2)) {
2426                 uint16_t edns0_bufsize;
2427
2428                 edns0_bufsize = last_label[7] << 8 | last_label[8];
2429
2430                 DBG("EDNS0 buffer size %u", edns0_bufsize);
2431
2432                 /* This is an evil hack until full TCP support has been
2433                  * implemented.
2434                  *
2435                  * Somtimes the EDNS0 request gets send with a too-small
2436                  * buffer size. Since glibc doesn't seem to crash when it
2437                  * gets a response biffer then it requested, just bump
2438                  * the buffer size up to 4KiB.
2439                  */
2440                 if (edns0_bufsize < 0x1000) {
2441                         last_label[7] = 0x10;
2442                         last_label[8] = 0x00;
2443                 }
2444         }
2445
2446         DBG("query %s", name);
2447
2448         return 0;
2449 }
2450
2451 static gboolean tcp_listener_event(GIOChannel *channel, GIOCondition condition,
2452                                                         gpointer user_data)
2453 {
2454         unsigned char buf[768];
2455         char query[512];
2456         struct request_data *req;
2457         int sk, client_sk, len, err;
2458         struct sockaddr_in6 client_addr;
2459         socklen_t client_addr_len = sizeof(client_addr);
2460         GSList *list;
2461         struct listener_data *ifdata = user_data;
2462         int waiting_for_connect = FALSE, qtype = 0;
2463         struct cache_entry *entry;
2464
2465         DBG("condition 0x%x", condition);
2466
2467         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2468                 if (ifdata->tcp_listener_watch > 0)
2469                         g_source_remove(ifdata->tcp_listener_watch);
2470                 ifdata->tcp_listener_watch = 0;
2471
2472                 connman_error("Error with TCP listener channel");
2473
2474                 return FALSE;
2475         }
2476
2477         sk = g_io_channel_unix_get_fd(channel);
2478
2479         client_sk = accept(sk, (void *)&client_addr, &client_addr_len);
2480         if (client_sk < 0) {
2481                 connman_error("Accept failure on TCP listener");
2482                 ifdata->tcp_listener_watch = 0;
2483                 return FALSE;
2484         }
2485
2486         len = recv(client_sk, buf, sizeof(buf), 0);
2487         if (len < 2)
2488                 return TRUE;
2489
2490         DBG("Received %d bytes (id 0x%04x) from %d", len,
2491                 buf[2] | buf[3] << 8, client_sk);
2492
2493         err = parse_request(buf + 2, len - 2, query, sizeof(query));
2494         if (err < 0 || (g_slist_length(server_list) == 0)) {
2495                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2496                 return TRUE;
2497         }
2498
2499         req = g_try_new0(struct request_data, 1);
2500         if (req == NULL)
2501                 return TRUE;
2502
2503         memcpy(&req->sa, &client_addr, client_addr_len);
2504         req->sa_len = client_addr_len;
2505         req->client_sk = client_sk;
2506         req->protocol = IPPROTO_TCP;
2507
2508         req->srcid = buf[2] | (buf[3] << 8);
2509         req->dstid = get_id();
2510         req->altid = get_id();
2511         req->request_len = len;
2512
2513         buf[2] = req->dstid & 0xff;
2514         buf[3] = req->dstid >> 8;
2515
2516         req->numserv = 0;
2517         req->ifdata = (struct listener_data *) ifdata;
2518         req->append_domain = FALSE;
2519
2520         /*
2521          * Check if the answer is found in the cache before
2522          * creating sockets to the server.
2523          */
2524         entry = cache_check(buf, &qtype, IPPROTO_TCP);
2525         if (entry != NULL) {
2526                 int ttl_left = 0;
2527                 struct cache_data *data;
2528
2529                 DBG("cache hit %s type %s", query, qtype == 1 ? "A" : "AAAA");
2530                 if (qtype == 1)
2531                         data = entry->ipv4;
2532                 else
2533                         data = entry->ipv6;
2534
2535                 if (data != NULL) {
2536                         ttl_left = data->valid_until - time(NULL);
2537                         entry->hits++;
2538
2539                         send_cached_response(client_sk, data->data,
2540                                         data->data_len, NULL, 0, IPPROTO_TCP,
2541                                         req->srcid, data->answers, ttl_left);
2542
2543                         g_free(req);
2544                         return TRUE;
2545                 } else
2546                         DBG("data missing, ignoring cache for this query");
2547         }
2548
2549         for (list = server_list; list; list = list->next) {
2550                 struct server_data *data = list->data;
2551
2552                 if (data->protocol != IPPROTO_UDP || data->enabled == FALSE)
2553                         continue;
2554
2555                 if(create_server(data->interface, NULL,
2556                                         data->server, IPPROTO_TCP) == NULL)
2557                         continue;
2558
2559                 waiting_for_connect = TRUE;
2560         }
2561
2562         if (waiting_for_connect == FALSE) {
2563                 /* No server is waiting for connect */
2564                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2565                 g_free(req);
2566                 return TRUE;
2567         }
2568
2569         /*
2570          * The server is not connected yet.
2571          * Copy the relevant buffers.
2572          * The request will actually be sent once we're
2573          * properly connected over TCP to the nameserver.
2574          */
2575         req->request = g_try_malloc0(req->request_len);
2576         if (req->request == NULL) {
2577                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2578                 g_free(req);
2579                 return TRUE;
2580         }
2581         memcpy(req->request, buf, req->request_len);
2582
2583         req->name = g_try_malloc0(sizeof(query));
2584         if (req->name == NULL) {
2585                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2586                 g_free(req->request);
2587                 g_free(req);
2588                 return TRUE;
2589         }
2590         memcpy(req->name, query, sizeof(query));
2591
2592         req->timeout = g_timeout_add_seconds(30, request_timeout, req);
2593
2594         request_list = g_slist_append(request_list, req);
2595
2596         return TRUE;
2597 }
2598
2599 static gboolean udp_listener_event(GIOChannel *channel, GIOCondition condition,
2600                                                         gpointer user_data)
2601 {
2602         unsigned char buf[768];
2603         char query[512];
2604         struct request_data *req;
2605         struct sockaddr_in6 client_addr;
2606         socklen_t client_addr_len = sizeof(client_addr);
2607         int sk, err, len;
2608         struct listener_data *ifdata = user_data;
2609
2610         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2611                 connman_error("Error with UDP listener channel");
2612                 ifdata->udp_listener_watch = 0;
2613                 return FALSE;
2614         }
2615
2616         sk = g_io_channel_unix_get_fd(channel);
2617
2618         memset(&client_addr, 0, client_addr_len);
2619         len = recvfrom(sk, buf, sizeof(buf), 0, (void *)&client_addr,
2620                        &client_addr_len);
2621         if (len < 2)
2622                 return TRUE;
2623
2624         DBG("Received %d bytes (id 0x%04x)", len, buf[0] | buf[1] << 8);
2625
2626         err = parse_request(buf, len, query, sizeof(query));
2627         if (err < 0 || (g_slist_length(server_list) == 0)) {
2628                 send_response(sk, buf, len, (void *)&client_addr,
2629                                 client_addr_len, IPPROTO_UDP);
2630                 return TRUE;
2631         }
2632
2633         req = g_try_new0(struct request_data, 1);
2634         if (req == NULL)
2635                 return TRUE;
2636
2637         memcpy(&req->sa, &client_addr, client_addr_len);
2638         req->sa_len = client_addr_len;
2639         req->client_sk = 0;
2640         req->protocol = IPPROTO_UDP;
2641
2642         req->srcid = buf[0] | (buf[1] << 8);
2643         req->dstid = get_id();
2644         req->altid = get_id();
2645         req->request_len = len;
2646
2647         buf[0] = req->dstid & 0xff;
2648         buf[1] = req->dstid >> 8;
2649
2650         req->numserv = 0;
2651         req->ifdata = (struct listener_data *) ifdata;
2652         req->append_domain = FALSE;
2653
2654         if (resolv(req, buf, query) == TRUE) {
2655                 /* a cached result was sent, so the request can be released */
2656                 g_free(req);
2657                 return TRUE;
2658         }
2659
2660         req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2661         request_list = g_slist_append(request_list, req);
2662
2663         return TRUE;
2664 }
2665
2666 static int create_dns_listener(int protocol, struct listener_data *ifdata)
2667 {
2668         GIOChannel *channel;
2669         const char *proto;
2670         union {
2671                 struct sockaddr sa;
2672                 struct sockaddr_in6 sin6;
2673                 struct sockaddr_in sin;
2674         } s;
2675         socklen_t slen;
2676         int sk, type, v6only = 0;
2677         int family = AF_INET6;
2678
2679
2680         DBG("interface %s", ifdata->ifname);
2681
2682         switch (protocol) {
2683         case IPPROTO_UDP:
2684                 proto = "UDP";
2685                 type = SOCK_DGRAM | SOCK_CLOEXEC;
2686                 break;
2687
2688         case IPPROTO_TCP:
2689                 proto = "TCP";
2690                 type = SOCK_STREAM | SOCK_CLOEXEC;
2691                 break;
2692
2693         default:
2694                 return -EINVAL;
2695         }
2696
2697         sk = socket(family, type, protocol);
2698         if (sk < 0 && family == AF_INET6 && errno == EAFNOSUPPORT) {
2699                 connman_error("No IPv6 support; DNS proxy listening only on Legacy IP");
2700                 family = AF_INET;
2701                 sk = socket(family, type, protocol);
2702         }
2703         if (sk < 0) {
2704                 connman_error("Failed to create %s listener socket", proto);
2705                 return -EIO;
2706         }
2707
2708         if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2709                                         ifdata->ifname,
2710                                         strlen(ifdata->ifname) + 1) < 0) {
2711                 connman_error("Failed to bind %s listener interface", proto);
2712                 close(sk);
2713                 return -EIO;
2714         }
2715         /* Ensure it accepts Legacy IP connections too */
2716         if (family == AF_INET6 &&
2717                         setsockopt(sk, SOL_IPV6, IPV6_V6ONLY,
2718                                         &v6only, sizeof(v6only)) < 0) {
2719                 connman_error("Failed to clear V6ONLY on %s listener socket",
2720                               proto);
2721                 close(sk);
2722                 return -EIO;
2723         }
2724
2725         if (family == AF_INET) {
2726                 memset(&s.sin, 0, sizeof(s.sin));
2727                 s.sin.sin_family = AF_INET;
2728                 s.sin.sin_port = htons(53);
2729                 s.sin.sin_addr.s_addr = htonl(INADDR_ANY);
2730                 slen = sizeof(s.sin);
2731         } else {
2732                 memset(&s.sin6, 0, sizeof(s.sin6));
2733                 s.sin6.sin6_family = AF_INET6;
2734                 s.sin6.sin6_port = htons(53);
2735                 s.sin6.sin6_addr = in6addr_any;
2736                 slen = sizeof(s.sin6);
2737         }
2738
2739         if (bind(sk, &s.sa, slen) < 0) {
2740                 connman_error("Failed to bind %s listener socket", proto);
2741                 close(sk);
2742                 return -EIO;
2743         }
2744
2745         if (protocol == IPPROTO_TCP && listen(sk, 10) < 0) {
2746                 connman_error("Failed to listen on TCP socket");
2747                 close(sk);
2748                 return -EIO;
2749         }
2750
2751         channel = g_io_channel_unix_new(sk);
2752         if (channel == NULL) {
2753                 connman_error("Failed to create %s listener channel", proto);
2754                 close(sk);
2755                 return -EIO;
2756         }
2757
2758         g_io_channel_set_close_on_unref(channel, TRUE);
2759
2760         if (protocol == IPPROTO_TCP) {
2761                 ifdata->tcp_listener_channel = channel;
2762                 ifdata->tcp_listener_watch = g_io_add_watch(channel,
2763                                 G_IO_IN, tcp_listener_event, (gpointer) ifdata);
2764         } else {
2765                 ifdata->udp_listener_channel = channel;
2766                 ifdata->udp_listener_watch = g_io_add_watch(channel,
2767                                 G_IO_IN, udp_listener_event, (gpointer) ifdata);
2768         }
2769
2770         return 0;
2771 }
2772
2773 static void destroy_udp_listener(struct listener_data *ifdata)
2774 {
2775         DBG("interface %s", ifdata->ifname);
2776
2777         if (ifdata->udp_listener_watch > 0)
2778                 g_source_remove(ifdata->udp_listener_watch);
2779
2780         g_io_channel_unref(ifdata->udp_listener_channel);
2781 }
2782
2783 static void destroy_tcp_listener(struct listener_data *ifdata)
2784 {
2785         DBG("interface %s", ifdata->ifname);
2786
2787         if (ifdata->tcp_listener_watch > 0)
2788                 g_source_remove(ifdata->tcp_listener_watch);
2789
2790         g_io_channel_unref(ifdata->tcp_listener_channel);
2791 }
2792
2793 static int create_listener(struct listener_data *ifdata)
2794 {
2795         int err;
2796
2797         err = create_dns_listener(IPPROTO_UDP, ifdata);
2798         if (err < 0)
2799                 return err;
2800
2801         err = create_dns_listener(IPPROTO_TCP, ifdata);
2802         if (err < 0) {
2803                 destroy_udp_listener(ifdata);
2804                 return err;
2805         }
2806
2807         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2808                 __connman_resolvfile_append("lo", NULL, "127.0.0.1");
2809
2810         return 0;
2811 }
2812
2813 static void destroy_listener(struct listener_data *ifdata)
2814 {
2815         GSList *list;
2816
2817         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2818                 __connman_resolvfile_remove("lo", NULL, "127.0.0.1");
2819
2820         for (list = request_list; list; list = list->next) {
2821                 struct request_data *req = list->data;
2822
2823                 DBG("Dropping request (id 0x%04x -> 0x%04x)",
2824                                                 req->srcid, req->dstid);
2825                 destroy_request_data(req);
2826                 list->data = NULL;
2827         }
2828
2829         g_slist_free(request_list);
2830         request_list = NULL;
2831
2832         destroy_tcp_listener(ifdata);
2833         destroy_udp_listener(ifdata);
2834 }
2835
2836 int __connman_dnsproxy_add_listener(const char *interface)
2837 {
2838         struct listener_data *ifdata;
2839         int err;
2840
2841         DBG("interface %s", interface);
2842
2843         if (g_hash_table_lookup(listener_table, interface) != NULL)
2844                 return 0;
2845
2846         ifdata = g_try_new0(struct listener_data, 1);
2847         if (ifdata == NULL)
2848                 return -ENOMEM;
2849
2850         ifdata->ifname = g_strdup(interface);
2851         ifdata->udp_listener_channel = NULL;
2852         ifdata->udp_listener_watch = 0;
2853         ifdata->tcp_listener_channel = NULL;
2854         ifdata->tcp_listener_watch = 0;
2855
2856         err = create_listener(ifdata);
2857         if (err < 0) {
2858                 connman_error("Couldn't create listener for %s err %d",
2859                                 interface, err);
2860                 g_free(ifdata->ifname);
2861                 g_free(ifdata);
2862                 return err;
2863         }
2864         g_hash_table_insert(listener_table, ifdata->ifname, ifdata);
2865         return 0;
2866 }
2867
2868 void __connman_dnsproxy_remove_listener(const char *interface)
2869 {
2870         struct listener_data *ifdata;
2871
2872         DBG("interface %s", interface);
2873
2874         ifdata = g_hash_table_lookup(listener_table, interface);
2875         if (ifdata == NULL)
2876                 return;
2877
2878         destroy_listener(ifdata);
2879
2880         g_hash_table_remove(listener_table, interface);
2881 }
2882
2883 static void remove_listener(gpointer key, gpointer value, gpointer user_data)
2884 {
2885         const char *interface = key;
2886         struct listener_data *ifdata = value;
2887
2888         DBG("interface %s", interface);
2889
2890         destroy_listener(ifdata);
2891 }
2892
2893 int __connman_dnsproxy_init(void)
2894 {
2895         int err;
2896
2897         DBG("");
2898
2899         srandom(time(NULL));
2900
2901         listener_table = g_hash_table_new_full(g_str_hash, g_str_equal,
2902                                                         g_free, g_free);
2903         err = __connman_dnsproxy_add_listener("lo");
2904         if (err < 0)
2905                 return err;
2906
2907         err = connman_notifier_register(&dnsproxy_notifier);
2908         if (err < 0)
2909                 goto destroy;
2910
2911         return 0;
2912
2913 destroy:
2914         __connman_dnsproxy_remove_listener("lo");
2915         g_hash_table_destroy(listener_table);
2916
2917         return err;
2918 }
2919
2920 void __connman_dnsproxy_cleanup(void)
2921 {
2922         DBG("");
2923
2924         connman_notifier_unregister(&dnsproxy_notifier);
2925
2926         g_hash_table_foreach(listener_table, remove_listener, NULL);
2927
2928         g_hash_table_destroy(listener_table);
2929 }