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         int protocol;
  90         GIOChannel *channel;
  91         guint watch;
  92         guint timeout;
  93         gboolean enabled;
  94         gboolean connected;
  95         struct partial_reply *incoming_reply;
  96 };
  97
  98 struct request_data {
  99         union {
 100                 struct sockaddr_in6 __sin6; /* Only for the length */
 101                 struct sockaddr sa;
 102         };
 103         socklen_t sa_len;
 104         int client_sk;
 105         int protocol;
 106         guint16 srcid;
 107         guint16 dstid;
 108         guint16 altid;
 109         guint timeout;
 110         guint watch;
 111         guint numserv;
 112         guint numresp;
 113         gpointer request;
 114         gsize request_len;
 115         gpointer name;
 116         gpointer resp;
 117         gsize resplen;
 118         struct listener_data *ifdata;
 119         gboolean append_domain;
 120 };
 121
 122 struct listener_data {
 123         char *ifname;
 124         GIOChannel *udp_listener_channel;
 125         guint udp_listener_watch;
 126         GIOChannel *tcp_listener_channel;
 127         guint tcp_listener_watch;
 128 };
 129
 130 struct cache_data {
 131         time_t inserted;
 132         time_t valid_until;
 133         time_t cache_until;
 134         int timeout;
 135         uint16_t type;
 136         uint16_t answers;
 137         unsigned int data_len;
 138         unsigned char *data; /* contains DNS header + body */
 139 };
 140
 141 struct cache_entry {
 142         char *key;
 143         int want_refresh;
 144         int hits;
 145         struct cache_data *ipv4;
 146         struct cache_data *ipv6;
 147 };
 148
 149 struct domain_question {
 150         uint16_t type;
 151         uint16_t class;
 152 } __attribute__ ((packed));
 153
 154 struct domain_rr {
 155         uint16_t type;
 156         uint16_t class;
 157         uint32_t ttl;
 158         uint16_t rdlen;
 159 } __attribute__ ((packed));
 160
 161 /*
 162  * We limit how long the cached DNS entry stays in the cache.
 163  * By default the TTL (time-to-live) of the DNS response is used
 164  * when setting the cache entry life time. The value is in seconds.
 165  */
 166 #define MAX_CACHE_TTL (60 * 30)
 167 /*
 168  * Also limit the other end, cache at least for 30 seconds.
 169  */
 170 #define MIN_CACHE_TTL (30)
 171
 172 /*
 173  * We limit the cache size to some sane value so that cached data does
 174  * not occupy too much memory. Each cached entry occupies on average
 175  * about 100 bytes memory (depending on DNS name length).
 176  * Example: caching www.connman.net uses 97 bytes memory.
 177  * The value is the max amount of cached DNS responses (count).
 178  */
 179 #define MAX_CACHE_SIZE 256
 180
 181 static int cache_size;
 182 static GHashTable *cache;
 183 static int cache_refcount;
 184 static GSList *server_list = NULL;
 185 static GSList *request_list = NULL;
 186 static GHashTable *listener_table = NULL;
 187 static time_t next_refresh;
 188
 189 static guint16 get_id()
 190 {
 191         return random();
 192 }
 193
 194 static int protocol_offset(int protocol)
 195 {
 196         switch (protocol) {
 197         case IPPROTO_UDP:
 198                 return 0;
 199
 200         case IPPROTO_TCP:
 201                 return 2;
 202
 203         default:
 204                 return -EINVAL;
 205         }
 206
 207 }
 208
 209 /*
 210  * There is a power and efficiency benefit to have entries
 211  * in our cache expire at the same time. To this extend,
 212  * we round down the cache valid time to common boundaries.
 213  */
 214 static time_t round_down_ttl(time_t end_time, int ttl)
 215 {
 216         if (ttl < 15)
 217                 return end_time;
 218
 219         /* Less than 5 minutes, round to 10 second boundary */
 220         if (ttl < 300) {
 221                 end_time = end_time / 10;
 222                 end_time = end_time * 10;
 223         } else { /* 5 or more minutes, round to 30 seconds */
 224                 end_time = end_time / 30;
 225                 end_time = end_time * 30;
 226         }
 227         return end_time;
 228 }
 229
 230 static struct request_data *find_request(guint16 id)
 231 {
 232         GSList *list;
 233
 234         for (list = request_list; list; list = list->next) {
 235                 struct request_data *req = list->data;
 236
 237                 if (req->dstid == id || req->altid == id)
 238                         return req;
 239         }
 240
 241         return NULL;
 242 }
 243
 244 static struct server_data *find_server(const char *interface,
 245                                         const char *server,
 246                                                 int protocol)
 247 {
 248         GSList *list;
 249
 250         DBG("interface %s server %s", interface, server);
 251
 252         for (list = server_list; list; list = list->next) {
 253                 struct server_data *data = list->data;
 254
 255                 if (interface == NULL && data->interface == NULL &&
 256                                 g_str_equal(data->server, server) == TRUE &&
 257                                 data->protocol == protocol)
 258                         return data;
 259
 260                 if (interface == NULL ||
 261                                 data->interface == NULL || data->server == NULL)
 262                         continue;
 263
 264                 if (g_str_equal(data->interface, interface) == TRUE &&
 265                                 g_str_equal(data->server, server) == TRUE &&
 266                                 data->protocol == protocol)
 267                         return data;
 268         }
 269
 270         return NULL;
 271 }
 272
 273 /* we can keep using the same resolve's */
 274 static GResolv *ipv4_resolve;
 275 static GResolv *ipv6_resolve;
 276
 277 static void dummy_resolve_func(GResolvResultStatus status,
 278                                         char **results, gpointer user_data)
 279 {
 280 }
 281
 282 /*
 283  * Refresh a DNS entry, but also age the hit count a bit */
 284 static void refresh_dns_entry(struct cache_entry *entry, char *name)
 285 {
 286         int age = 1;
 287
 288         if (ipv4_resolve == NULL) {
 289                 ipv4_resolve = g_resolv_new(0);
 290                 g_resolv_set_address_family(ipv4_resolve, AF_INET);
 291                 g_resolv_add_nameserver(ipv4_resolve, "127.0.0.1", 53, 0);
 292         }
 293
 294         if (ipv6_resolve == NULL) {
 295                 ipv6_resolve = g_resolv_new(0);
 296                 g_resolv_set_address_family(ipv6_resolve, AF_INET6);
 297                 g_resolv_add_nameserver(ipv6_resolve, "127.0.0.1", 53, 0);
 298         }
 299
 300         if (entry->ipv4 == NULL) {
 301                 DBG("Refresing A record for %s", name);
 302                 g_resolv_lookup_hostname(ipv4_resolve, name,
 303                                         dummy_resolve_func, NULL);
 304                 age = 4;
 305         }
 306
 307         if (entry->ipv6 == NULL) {
 308                 DBG("Refresing AAAA record for %s", name);
 309                 g_resolv_lookup_hostname(ipv6_resolve, name,
 310                                         dummy_resolve_func, NULL);
 311                 age = 4;
 312         }
 313
 314         entry->hits -= age;
 315         if (entry->hits < 0)
 316                 entry->hits = 0;
 317 }
 318
 319 static int dns_name_length(unsigned char *buf)
 320 {
 321         if ((buf[0] & NS_CMPRSFLGS) == NS_CMPRSFLGS) /* compressed name */
 322                 return 2;
 323         return strlen((char *)buf);
 324 }
 325
 326 static void update_cached_ttl(unsigned char *buf, int len, int new_ttl)
 327 {
 328         unsigned char *c;
 329         uint32_t *i;
 330         uint16_t *w;
 331         int l;
 332
 333         /* skip the header */
 334         c = buf + 12;
 335         len -= 12;
 336
 337         /* skip the query, which is a name and 2 16 bit words */
 338         l = dns_name_length(c);
 339         c += l;
 340         len -= l;
 341         c += 4;
 342         len -= 4;
 343
 344         /* now we get the answer records */
 345
 346         while (len > 0) {
 347                 /* first a name */
 348                 l = dns_name_length(c);
 349                 c += l;
 350                 len -= l;
 351                 if (len < 0)
 352                         break;
 353                 /* then type + class, 2 bytes each */
 354                 c += 4;
 355                 len -= 4;
 356                 if (len < 0)
 357                         break;
 358
 359                 /* now the 4 byte TTL field */
 360                 i = (uint32_t *)c;
 361                 *i = htonl(new_ttl);
 362                 c += 4;
 363                 len -= 4;
 364                 if (len < 0)
 365                         break;
 366
 367                 /* now the 2 byte rdlen field */
 368                 w = (uint16_t *)c;
 369                 c += ntohs(*w) + 2;
 370                 len -= ntohs(*w) + 2;
 371         }
 372 }
 373
 374 static void send_cached_response(int sk, unsigned char *buf, int len,
 375                                 const struct sockaddr *to, socklen_t tolen,
 376                                 int protocol, int id, uint16_t answers, int ttl)
 377 {
 378         struct domain_hdr *hdr;
 379         unsigned char *ptr = buf;
 380         int err, offset, dns_len, adj_len = len - 2;
 381
 382         /*
 383          * The cached packet contains always the TCP offset (two bytes)
 384          * so skip them for UDP.
 385          */
 386         switch (protocol) {
 387         case IPPROTO_UDP:
 388                 ptr += 2;
 389                 len -= 2;
 390                 dns_len = len;
 391                 offset = 0;
 392                 break;
 393         case IPPROTO_TCP:
 394                 offset = 2;
 395                 dns_len = ptr[0] * 256 + ptr[1];
 396                 break;
 397         default:
 398                 return;
 399         }
 400
 401         if (len < 12)
 402                 return;
 403
 404         hdr = (void *) (ptr + offset);
 405
 406         hdr->id = id;
 407         hdr->qr = 1;
 408         hdr->rcode = 0;
 409         hdr->ancount = htons(answers);
 410         hdr->nscount = 0;
 411         hdr->arcount = 0;
 412
 413         /* if this is a negative reply, we are authorative */
 414         if (answers == 0)
 415                 hdr->aa = 1;
 416         else
 417                 update_cached_ttl((unsigned char *)hdr, adj_len, ttl);
 418
 419         DBG("sk %d id 0x%04x answers %d ptr %p length %d dns %d",
 420                 sk, hdr->id, answers, ptr, len, dns_len);
 421
 422         err = sendto(sk, ptr, len, MSG_NOSIGNAL, to, tolen);
 423         if (err < 0) {
 424                 connman_error("Cannot send cached DNS response: %s",
 425                                 strerror(errno));
 426                 return;
 427         }
 428
 429         if (err != len || (dns_len != (len - 2) && protocol == IPPROTO_TCP) ||
 430                                 (dns_len != len && protocol == IPPROTO_UDP))
 431                 DBG("Packet length mismatch, sent %d wanted %d dns %d",
 432                         err, len, dns_len);
 433 }
 434
 435 static void send_response(int sk, unsigned char *buf, int len,
 436                                 const struct sockaddr *to, socklen_t tolen,
 437                                 int protocol)
 438 {
 439         struct domain_hdr *hdr;
 440         int err, offset = protocol_offset(protocol);
 441
 442         DBG("");
 443
 444         if (offset < 0)
 445                 return;
 446
 447         if (len < 12)
 448                 return;
 449
 450         hdr = (void *) (buf + offset);
 451
 452         DBG("id 0x%04x qr %d opcode %d", hdr->id, hdr->qr, hdr->opcode);
 453
 454         hdr->qr = 1;
 455         hdr->rcode = 2;
 456
 457         hdr->ancount = 0;
 458         hdr->nscount = 0;
 459         hdr->arcount = 0;
 460
 461         err = sendto(sk, buf, len, MSG_NOSIGNAL, to, tolen);
 462         if (err < 0) {
 463                 connman_error("Failed to send DNS response: %s",
 464                                 strerror(errno));
 465                 return;
 466         }
 467 }
 468
 469 static gboolean request_timeout(gpointer user_data)
 470 {
 471         struct request_data *req = user_data;
 472         struct listener_data *ifdata;
 473
 474         DBG("id 0x%04x", req->srcid);
 475
 476         if (req == NULL)
 477                 return FALSE;
 478
 479         ifdata = req->ifdata;
 480
 481         request_list = g_slist_remove(request_list, req);
 482         req->numserv--;
 483
 484         if (req->resplen > 0 && req->resp != NULL) {
 485                 int sk, err;
 486
 487                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
 488
 489                 err = sendto(sk, req->resp, req->resplen, MSG_NOSIGNAL,
 490                                                 &req->sa, req->sa_len);
 491                 if (err < 0)
 492                         return FALSE;
 493         } else if (req->request && req->numserv == 0) {
 494                 struct domain_hdr *hdr;
 495
 496                 if (req->protocol == IPPROTO_TCP) {
 497                         hdr = (void *) (req->request + 2);
 498                         hdr->id = req->srcid;
 499                         send_response(req->client_sk, req->request,
 500                                 req->request_len, NULL, 0, IPPROTO_TCP);
 501
 502                 } else if (req->protocol == IPPROTO_UDP) {
 503                         int sk;
 504
 505                         hdr = (void *) (req->request);
 506                         hdr->id = req->srcid;
 507                         sk = g_io_channel_unix_get_fd(
 508                                                 ifdata->udp_listener_channel);
 509                         send_response(sk, req->request, req->request_len,
 510                                         &req->sa, req->sa_len, IPPROTO_UDP);
 511                 }
 512         }
 513
 514         g_free(req->resp);
 515         g_free(req);
 516
 517         return FALSE;
 518 }
 519
 520 static int append_query(unsigned char *buf, unsigned int size,
 521                                 const char *query, const char *domain)
 522 {
 523         unsigned char *ptr = buf;
 524         int len;
 525
 526         DBG("query %s domain %s", query, domain);
 527
 528         while (query != NULL) {
 529                 const char *tmp;
 530
 531                 tmp = strchr(query, '.');
 532                 if (tmp == NULL) {
 533                         len = strlen(query);
 534                         if (len == 0)
 535                                 break;
 536                         *ptr = len;
 537                         memcpy(ptr + 1, query, len);
 538                         ptr += len + 1;
 539                         break;
 540                 }
 541
 542                 *ptr = tmp - query;
 543                 memcpy(ptr + 1, query, tmp - query);
 544                 ptr += tmp - query + 1;
 545
 546                 query = tmp + 1;
 547         }
 548
 549         while (domain != NULL) {
 550                 const char *tmp;
 551
 552                 tmp = strchr(domain, '.');
 553                 if (tmp == NULL) {
 554                         len = strlen(domain);
 555                         if (len == 0)
 556                                 break;
 557                         *ptr = len;
 558                         memcpy(ptr + 1, domain, len);
 559                         ptr += len + 1;
 560                         break;
 561                 }
 562
 563                 *ptr = tmp - domain;
 564                 memcpy(ptr + 1, domain, tmp - domain);
 565                 ptr += tmp - domain + 1;
 566
 567                 domain = tmp + 1;
 568         }
 569
 570         *ptr++ = 0x00;
 571
 572         return ptr - buf;
 573 }
 574
 575 static gboolean cache_check_is_valid(struct cache_data *data,
 576                                 time_t current_time)
 577 {
 578         if (data == NULL)
 579                 return FALSE;
 580
 581         if (data->cache_until < current_time)
 582                 return FALSE;
 583
 584         return TRUE;
 585 }
 586
 587 /*
 588  * remove stale cached entries so that they can be refreshed
 589  */
 590 static void cache_enforce_validity(struct cache_entry *entry)
 591 {
 592         time_t current_time = time(NULL);
 593
 594         if (cache_check_is_valid(entry->ipv4, current_time) == FALSE
 595                                                         && entry->ipv4) {
 596                 DBG("cache timeout \"%s\" type A", entry->key);
 597                 g_free(entry->ipv4->data);
 598                 g_free(entry->ipv4);
 599                 entry->ipv4 = NULL;
 600
 601         }
 602
 603         if (cache_check_is_valid(entry->ipv6, current_time) == FALSE
 604                                                         && entry->ipv6) {
 605                 DBG("cache timeout \"%s\" type AAAA", entry->key);
 606                 g_free(entry->ipv6->data);
 607                 g_free(entry->ipv6);
 608                 entry->ipv6 = NULL;
 609         }
 610 }
 611
 612 static uint16_t cache_check_validity(char *question, uint16_t type,
 613                                 struct cache_entry *entry)
 614 {
 615         time_t current_time = time(NULL);
 616         int want_refresh = 0;
 617
 618         /*
 619          * if we have a popular entry, we want a refresh instead of
 620          * total destruction of the entry.
 621          */
 622         if (entry->hits > 2)
 623                 want_refresh = 1;
 624
 625         cache_enforce_validity(entry);
 626
 627         switch (type) {
 628         case 1:         /* IPv4 */
 629                 if (cache_check_is_valid(entry->ipv4, current_time) == FALSE) {
 630                         DBG("cache %s \"%s\" type A", entry->ipv4 ?
 631                                         "timeout" : "entry missing", question);
 632
 633                         if (want_refresh)
 634                                 entry->want_refresh = 1;
 635
 636                         /*
 637                          * We do not remove cache entry if there is still
 638                          * valid IPv6 entry found in the cache.
 639                          */
 640                         if (cache_check_is_valid(entry->ipv6, current_time)
 641                                         == FALSE && want_refresh == FALSE) {
 642                                 g_hash_table_remove(cache, question);
 643                                 type = 0;
 644                         }
 645                 }
 646                 break;
 647
 648         case 28:        /* IPv6 */
 649                 if (cache_check_is_valid(entry->ipv6, current_time) == FALSE) {
 650                         DBG("cache %s \"%s\" type AAAA", entry->ipv6 ?
 651                                         "timeout" : "entry missing", question);
 652
 653                         if (want_refresh)
 654                                 entry->want_refresh = 1;
 655
 656                         if (cache_check_is_valid(entry->ipv4, current_time)
 657                                         == FALSE && want_refresh == FALSE) {
 658                                 g_hash_table_remove(cache, question);
 659                                 type = 0;
 660                         }
 661                 }
 662                 break;
 663         }
 664
 665         return type;
 666 }
 667
 668 static struct cache_entry *cache_check(gpointer request, int *qtype, int proto)
 669 {
 670         char *question;
 671         struct cache_entry *entry;
 672         struct domain_question *q;
 673         uint16_t type;
 674         int offset, proto_offset;
 675
 676         if (request == NULL)
 677                 return NULL;
 678
 679         proto_offset = protocol_offset(proto);
 680         if (proto_offset < 0)
 681                 return NULL;
 682
 683         question = request + proto_offset + 12;
 684
 685         offset = strlen(question) + 1;
 686         q = (void *) (question + offset);
 687         type = ntohs(q->type);
 688
 689         /* We only cache either A (1) or AAAA (28) requests */
 690         if (type != 1 && type != 28)
 691                 return NULL;
 692
 693         entry = g_hash_table_lookup(cache, question);
 694         if (entry == NULL)
 695                 return NULL;
 696
 697         type = cache_check_validity(question, type, entry);
 698         if (type == 0)
 699                 return NULL;
 700
 701         *qtype = type;
 702         return entry;
 703 }
 704
 705 /*
 706  * Get a label/name from DNS resource record. The function decompresses the
 707  * label if necessary. The function does not convert the name to presentation
 708  * form. This means that the result string will contain label lengths instead
 709  * of dots between labels. We intentionally do not want to convert to dotted
 710  * format so that we can cache the wire format string directly.
 711  */
 712 static int get_name(int counter,
 713                 unsigned char *pkt, unsigned char *start, unsigned char *max,
 714                 unsigned char *output, int output_max, int *output_len,
 715                 unsigned char **end, char *name, int *name_len)
 716 {
 717         unsigned char *p;
 718
 719         /* Limit recursion to 10 (this means up to 10 labels in domain name) */
 720         if (counter > 10)
 721                 return -EINVAL;
 722
 723         p = start;
 724         while (*p) {
 725                 if ((*p & NS_CMPRSFLGS) == NS_CMPRSFLGS) {
 726                         uint16_t offset = (*p & 0x3F) * 256 + *(p + 1);
 727
 728                         if (offset >= max - pkt)
 729                                 return -ENOBUFS;
 730
 731                         if (*end == NULL)
 732                                 *end = p + 2;
 733
 734                         return get_name(counter + 1, pkt, pkt + offset, max,
 735                                         output, output_max, output_len, end,
 736                                         name, name_len);
 737                 } else {
 738                         unsigned label_len = *p;
 739
 740                         if (pkt + label_len > max)
 741                                 return -ENOBUFS;
 742
 743                         if (*output_len > output_max)
 744                                 return -ENOBUFS;
 745
 746                         /*
 747                          * We need the original name in order to check
 748                          * if this answer is the correct one.
 749                          */
 750                         name[(*name_len)++] = label_len;
 751                         memcpy(name + *name_len, p + 1, label_len + 1);
 752                         *name_len += label_len;
 753
 754                         /* We compress the result */
 755                         output[0] = NS_CMPRSFLGS;
 756                         output[1] = 0x0C;
 757                         *output_len = 2;
 758
 759                         p += label_len + 1;
 760
 761                         if (*end == NULL)
 762                                 *end = p;
 763
 764                         if (p >= max)
 765                                 return -ENOBUFS;
 766                 }
 767         }
 768
 769         return 0;
 770 }
 771
 772 static int parse_rr(unsigned char *buf, unsigned char *start,
 773                         unsigned char *max,
 774                         unsigned char *response, unsigned int *response_size,
 775                         uint16_t *type, uint16_t *class, int *ttl, int *rdlen,
 776                         unsigned char **end,
 777                         char *name)
 778 {
 779         struct domain_rr *rr;
 780         int err, offset;
 781         int name_len = 0, output_len = 0, max_rsp = *response_size;
 782
 783         err = get_name(0, buf, start, max, response, max_rsp,
 784                 &output_len, end, name, &name_len);
 785         if (err < 0)
 786                 return err;
 787
 788         offset = output_len;
 789
 790         if ((unsigned int) offset > *response_size)
 791                 return -ENOBUFS;
 792
 793         rr = (void *) (*end);
 794
 795         if (rr == NULL)
 796                 return -EINVAL;
 797
 798         *type = ntohs(rr->type);
 799         *class = ntohs(rr->class);
 800         *ttl = ntohl(rr->ttl);
 801         *rdlen = ntohs(rr->rdlen);
 802
 803         if (*ttl < 0)
 804                 return -EINVAL;
 805
 806         memcpy(response + offset, *end, sizeof(struct domain_rr));
 807
 808         offset += sizeof(struct domain_rr);
 809         *end += sizeof(struct domain_rr);
 810
 811         if ((unsigned int) (offset + *rdlen) > *response_size)
 812                 return -ENOBUFS;
 813
 814         memcpy(response + offset, *end, *rdlen);
 815
 816         *end += *rdlen;
 817
 818         *response_size = offset + *rdlen;
 819
 820         return 0;
 821 }
 822
 823 static gboolean check_alias(GSList *aliases, char *name)
 824 {
 825         GSList *list;
 826
 827         if (aliases != NULL) {
 828                 for (list = aliases; list; list = list->next) {
 829                         int len = strlen((char *)list->data);
 830                         if (strncmp((char *)list->data, name, len) == 0)
 831                                 return TRUE;
 832                 }
 833         }
 834
 835         return FALSE;
 836 }
 837
 838 static int parse_response(unsigned char *buf, int buflen,
 839                         char *question, int qlen,
 840                         uint16_t *type, uint16_t *class, int *ttl,
 841                         unsigned char *response, unsigned int *response_len,
 842                         uint16_t *answers)
 843 {
 844         struct domain_hdr *hdr = (void *) buf;
 845         struct domain_question *q;
 846         unsigned char *ptr;
 847         uint16_t qdcount = ntohs(hdr->qdcount);
 848         uint16_t ancount = ntohs(hdr->ancount);
 849         int err, i;
 850         uint16_t qtype, qclass;
 851         unsigned char *next = NULL;
 852         unsigned int maxlen = *response_len;
 853         GSList *aliases = NULL, *list;
 854         char name[NS_MAXDNAME + 1];
 855
 856         if (buflen < 12)
 857                 return -EINVAL;
 858
 859         DBG("qr %d qdcount %d", hdr->qr, qdcount);
 860
 861         /* We currently only cache responses where question count is 1 */
 862         if (hdr->qr != 1 || qdcount != 1)
 863                 return -EINVAL;
 864
 865         ptr = buf + sizeof(struct domain_hdr);
 866
 867         strncpy(question, (char *) ptr, qlen);
 868         qlen = strlen(question);
 869         ptr += qlen + 1; /* skip \0 */
 870
 871         q = (void *) ptr;
 872         qtype = ntohs(q->type);
 873
 874         /* We cache only A and AAAA records */
 875         if (qtype != 1 && qtype != 28)
 876                 return -ENOMSG;
 877
 878         qclass = ntohs(q->class);
 879
 880         ptr += 2 + 2; /* ptr points now to answers */
 881
 882         err = -ENOMSG;
 883         *response_len = 0;
 884         *answers = 0;
 885
 886         /*
 887          * We have a bunch of answers (like A, AAAA, CNAME etc) to
 888          * A or AAAA question. We traverse the answers and parse the
 889          * resource records. Only A and AAAA records are cached, all
 890          * the other records in answers are skipped.
 891          */
 892         for (i = 0; i < ancount; i++) {
 893                 /*
 894                  * Get one address at a time to this buffer.
 895                  * The max size of the answer is
 896                  *   2 (pointer) + 2 (type) + 2 (class) +
 897                  *   4 (ttl) + 2 (rdlen) + addr (16 or 4) = 28
 898                  * for A or AAAA record.
 899                  * For CNAME the size can be bigger.
 900                  */
 901                 unsigned char rsp[NS_MAXCDNAME];
 902                 unsigned int rsp_len = sizeof(rsp) - 1;
 903                 int ret, rdlen;
 904
 905                 memset(rsp, 0, sizeof(rsp));
 906
 907                 ret = parse_rr(buf, ptr, buf + buflen, rsp, &rsp_len,
 908                         type, class, ttl, &rdlen, &next, name);
 909                 if (ret != 0) {
 910                         err = ret;
 911                         goto out;
 912                 }
 913
 914                 /*
 915                  * Now rsp contains compressed or uncompressed resource
 916                  * record. Next we check if this record answers the question.
 917                  * The name var contains the uncompressed label.
 918                  * One tricky bit is the CNAME records as they alias
 919                  * the name we might be interested in.
 920                  */
 921
 922                 /*
 923                  * Go to next answer if the class is not the one we are
 924                  * looking for.
 925                  */
 926                 if (*class != qclass) {
 927                         ptr = next;
 928                         next = NULL;
 929                         continue;
 930                 }
 931
 932                 /*
 933                  * Try to resolve aliases also, type is CNAME(5).
 934                  * This is important as otherwise the aliased names would not
 935                  * be cached at all as the cache would not contain the aliased
 936                  * question.
 937                  *
 938                  * If any CNAME is found in DNS packet, then we cache the alias
 939                  * IP address instead of the question (as the server
 940                  * said that question has only an alias).
 941                  * This means in practice that if e.g., ipv6.google.com is
 942                  * queried, DNS server returns CNAME of that name which is
 943                  * ipv6.l.google.com. We then cache the address of the CNAME
 944                  * but return the question name to client. So the alias
 945                  * status of the name is not saved in cache and thus not
 946                  * returned to the client. We do not return DNS packets from
 947                  * cache to client saying that ipv6.google.com is an alias to
 948                  * ipv6.l.google.com but we return instead a DNS packet that
 949                  * says ipv6.google.com has address xxx which is in fact the
 950                  * address of ipv6.l.google.com. For caching purposes this
 951                  * should not cause any issues.
 952                  */
 953                 if (*type == 5 && strncmp(question, name, qlen) == 0) {
 954                         /*
 955                          * So now the alias answered the question. This is
 956                          * not very useful from caching point of view as
 957                          * the following A or AAAA records will not match the
 958                          * question. We need to find the real A/AAAA record
 959                          * of the alias and cache that.
 960                          */
 961                         unsigned char *end = NULL;
 962                         int name_len = 0, output_len;
 963
 964                         memset(rsp, 0, sizeof(rsp));
 965                         rsp_len = sizeof(rsp) - 1;
 966
 967                         /*
 968                          * Alias is in rdata part of the message,
 969                          * and next-rdlen points to it. So we need to get
 970                          * the real name of the alias.
 971                          */
 972                         ret = get_name(0, buf, next - rdlen, buf + buflen,
 973                                         rsp, rsp_len, &output_len, &end,
 974                                         name, &name_len);
 975                         if (ret != 0) {
 976                                 /* just ignore the error at this point */
 977                                 ptr = next;
 978                                 next = NULL;
 979                                 continue;
 980                         }
 981
 982                         /*
 983                          * We should now have the alias of the entry we might
 984                          * want to cache. Just remember it for a while.
 985                          * We check the alias list when we have parsed the
 986                          * A or AAAA record.
 987                          */
 988                         aliases = g_slist_prepend(aliases, g_strdup(name));
 989
 990                         ptr = next;
 991                         next = NULL;
 992                         continue;
 993                 }
 994
 995                 if (*type == qtype) {
 996                         /*
 997                          * We found correct type (A or AAAA)
 998                          */
 999                         if (check_alias(aliases, name) == TRUE ||
1000                                 (aliases == NULL && strncmp(question, name,
1001                                                         qlen) == 0)) {
1002                                 /*
1003                                  * We found an alias or the name of the rr
1004                                  * matches the question. If so, we append
1005                                  * the compressed label to the cache.
1006                                  * The end result is a response buffer that
1007                                  * will contain one or more cached and
1008                                  * compressed resource records.
1009                                  */
1010                                 if (*response_len + rsp_len > maxlen) {
1011                                         err = -ENOBUFS;
1012                                         goto out;
1013                                 }
1014                                 memcpy(response + *response_len, rsp, rsp_len);
1015                                 *response_len += rsp_len;
1016                                 (*answers)++;
1017                                 err = 0;
1018                         }
1019                 }
1020
1021                 ptr = next;
1022                 next = NULL;
1023         }
1024
1025 out:
1026         for (list = aliases; list; list = list->next)
1027                 g_free(list->data);
1028         g_slist_free(aliases);
1029
1030         return err;
1031 }
1032
1033 struct cache_timeout {
1034         time_t current_time;
1035         int max_timeout;
1036         int try_harder;
1037 };
1038
1039 static gboolean cache_check_entry(gpointer key, gpointer value,
1040                                         gpointer user_data)
1041 {
1042         struct cache_timeout *data = user_data;
1043         struct cache_entry *entry = value;
1044         int max_timeout;
1045
1046         /* Scale the number of hits by half as part of cache aging */
1047
1048         entry->hits /= 2;
1049
1050         /*
1051          * If either IPv4 or IPv6 cached entry has expired, we
1052          * remove both from the cache.
1053          */
1054
1055         if (entry->ipv4 != NULL && entry->ipv4->timeout > 0) {
1056                 max_timeout = entry->ipv4->cache_until;
1057                 if (max_timeout > data->max_timeout)
1058                         data->max_timeout = max_timeout;
1059
1060                 if (entry->ipv4->cache_until < data->current_time)
1061                         return TRUE;
1062         }
1063
1064         if (entry->ipv6 != NULL && entry->ipv6->timeout > 0) {
1065                 max_timeout = entry->ipv6->cache_until;
1066                 if (max_timeout > data->max_timeout)
1067                         data->max_timeout = max_timeout;
1068
1069                 if (entry->ipv6->cache_until < data->current_time)
1070                         return TRUE;
1071         }
1072
1073         /*
1074          * if we're asked to try harder, also remove entries that have
1075          * few hits
1076          */
1077         if (data->try_harder && entry->hits < 4)
1078                 return TRUE;
1079
1080         return FALSE;
1081 }
1082
1083 static void cache_cleanup(void)
1084 {
1085         static int max_timeout;
1086         struct cache_timeout data;
1087         int count = 0;
1088
1089         data.current_time = time(NULL);
1090         data.max_timeout = 0;
1091         data.try_harder = 0;
1092
1093         /*
1094          * In the first pass, we only remove entries that have timed out.
1095          * We use a cache of the first time to expire to do this only
1096          * when it makes sense.
1097          */
1098         if (max_timeout <= data.current_time) {
1099                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1100                                                 &data);
1101         }
1102         DBG("removed %d in the first pass", count);
1103
1104         /*
1105          * In the second pass, if the first pass turned up blank,
1106          * we also expire entries with a low hit count,
1107          * while aging the hit count at the same time.
1108          */
1109         data.try_harder = 1;
1110         if (count == 0)
1111                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1112                                                 &data);
1113
1114         if (count == 0)
1115                 /*
1116                  * If we could not remove anything, then remember
1117                  * what is the max timeout and do nothing if we
1118                  * have not yet reached it. This will prevent
1119                  * constant traversal of the cache if it is full.
1120                  */
1121                 max_timeout = data.max_timeout;
1122         else
1123                 max_timeout = 0;
1124 }
1125
1126 static gboolean cache_invalidate_entry(gpointer key, gpointer value,
1127                                         gpointer user_data)
1128 {
1129         struct cache_entry *entry = value;
1130
1131         /* first, delete any expired elements */
1132         cache_enforce_validity(entry);
1133
1134         /* if anything is not expired, mark the entry for refresh */
1135         if (entry->hits > 0 && (entry->ipv4 || entry->ipv6))
1136                 entry->want_refresh = 1;
1137
1138         /* delete the cached data */
1139         if (entry->ipv4) {
1140                 g_free(entry->ipv4->data);
1141                 g_free(entry->ipv4);
1142                 entry->ipv4 = NULL;
1143         }
1144
1145         if (entry->ipv6) {
1146                 g_free(entry->ipv6->data);
1147                 g_free(entry->ipv6);
1148                 entry->ipv6 = NULL;
1149         }
1150
1151         /* keep the entry if we want it refreshed, delete it otherwise */
1152         if (entry->want_refresh)
1153                 return FALSE;
1154         else
1155                 return TRUE;
1156 }
1157
1158 /*
1159  * cache_invalidate is called from places where the DNS landscape
1160  * has changed, say because connections are added or we entered a VPN.
1161  * The logic is to wipe all cache data, but mark all non-expired
1162  * parts of the cache for refresh rather than deleting the whole cache.
1163  */
1164 static void cache_invalidate(void)
1165 {
1166         DBG("Invalidating the DNS cache %p", cache);
1167
1168         if (cache == NULL)
1169                 return;
1170
1171         g_hash_table_foreach_remove(cache, cache_invalidate_entry, NULL);
1172 }
1173
1174 static void cache_refresh_entry(struct cache_entry *entry)
1175 {
1176
1177         cache_enforce_validity(entry);
1178
1179         if (entry->hits > 2 && entry->ipv4 == NULL)
1180                 entry->want_refresh = 1;
1181         if (entry->hits > 2 && entry->ipv6 == NULL)
1182                 entry->want_refresh = 1;
1183
1184         if (entry->want_refresh) {
1185                 char *c;
1186                 char dns_name[NS_MAXDNAME + 1];
1187                 entry->want_refresh = 0;
1188
1189                 /* turn a DNS name into a hostname with dots */
1190                 strncpy(dns_name, entry->key, NS_MAXDNAME);
1191                 c = dns_name;
1192                 while (c && *c) {
1193                         int jump;
1194                         jump = *c;
1195                         *c = '.';
1196                         c += jump + 1;
1197                 }
1198                 DBG("Refreshing %s\n", dns_name);
1199                 /* then refresh the hostname */
1200                 refresh_dns_entry(entry, &dns_name[1]);
1201         }
1202 }
1203
1204 static void cache_refresh_iterator(gpointer key, gpointer value,
1205                                         gpointer user_data)
1206 {
1207         struct cache_entry *entry = value;
1208
1209         cache_refresh_entry(entry);
1210 }
1211
1212 static void cache_refresh(void)
1213 {
1214         if (cache == NULL)
1215                 return;
1216
1217         g_hash_table_foreach(cache, cache_refresh_iterator, NULL);
1218 }
1219
1220 static int reply_query_type(unsigned char *msg, int len)
1221 {
1222         unsigned char *c;
1223         uint16_t *w;
1224         int l;
1225         int type;
1226
1227         /* skip the header */
1228         c = msg + sizeof(struct domain_hdr);
1229         len -= sizeof(struct domain_hdr);
1230
1231         if (len < 0)
1232                 return 0;
1233
1234         /* now the query, which is a name and 2 16 bit words */
1235         l = dns_name_length(c) + 1;
1236         c += l;
1237         w = (uint16_t *) c;
1238         type = ntohs(*w);
1239
1240         return type;
1241 }
1242
1243 static int cache_update(struct server_data *srv, unsigned char *msg,
1244                         unsigned int msg_len)
1245 {
1246         int offset = protocol_offset(srv->protocol);
1247         int err, qlen, ttl = 0;
1248         uint16_t answers = 0, type = 0, class = 0;
1249         struct domain_question *q;
1250         struct cache_entry *entry;
1251         struct cache_data *data;
1252         char question[NS_MAXDNAME + 1];
1253         unsigned char response[NS_MAXDNAME + 1];
1254         unsigned char *ptr;
1255         unsigned int rsplen;
1256         gboolean new_entry = TRUE;
1257         time_t current_time;
1258
1259         if (cache_size >= MAX_CACHE_SIZE) {
1260                 cache_cleanup();
1261                 if (cache_size >= MAX_CACHE_SIZE)
1262                         return 0;
1263         }
1264
1265         current_time = time(NULL);
1266
1267         /* don't do a cache refresh more than twice a minute */
1268         if (next_refresh < current_time) {
1269                 cache_refresh();
1270                 next_refresh = current_time + 30;
1271         }
1272
1273
1274         /* Continue only if response code is 0 (=ok) */
1275         if (msg[3] & 0x0f)
1276                 return 0;
1277
1278         if (offset < 0)
1279                 return 0;
1280
1281         rsplen = sizeof(response) - 1;
1282         question[sizeof(question) - 1] = '\0';
1283
1284         err = parse_response(msg + offset, msg_len - offset,
1285                                 question, sizeof(question) - 1,
1286                                 &type, &class, &ttl,
1287                                 response, &rsplen, &answers);
1288
1289         /*
1290          * special case: if we do a ipv6 lookup and get no result
1291          * for a record that's already in our ipv4 cache.. we want
1292          * to cache the negative response.
1293          */
1294         if ((err == -ENOMSG || err == -ENOBUFS) &&
1295                         reply_query_type(msg + offset,
1296                                         msg_len - offset) == 28) {
1297                 entry = g_hash_table_lookup(cache, question);
1298                 if (entry && entry->ipv4 && entry->ipv6 == NULL) {
1299                         int cache_offset = 0;
1300
1301                         data = g_try_new(struct cache_data, 1);
1302                         if (data == NULL)
1303                                 return -ENOMEM;
1304                         data->inserted = entry->ipv4->inserted;
1305                         data->type = type;
1306                         data->answers = msg[5];
1307                         data->timeout = entry->ipv4->timeout;
1308                         if (srv->protocol == IPPROTO_UDP)
1309                                 cache_offset = 2;
1310                         data->data_len = msg_len + cache_offset;
1311                         data->data = ptr = g_malloc(data->data_len);
1312                         ptr[0] = (data->data_len - 2) / 256;
1313                         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1314                         if (srv->protocol == IPPROTO_UDP)
1315                                 ptr += 2;
1316                         data->valid_until = entry->ipv4->valid_until;
1317                         data->cache_until = entry->ipv4->cache_until;
1318                         memcpy(ptr, msg, msg_len);
1319                         entry->ipv6 = data;
1320                         /*
1321                          * we will get a "hit" when we serve the response
1322                          * out of the cache
1323                          */
1324                         entry->hits--;
1325                         if (entry->hits < 0)
1326                                 entry->hits = 0;
1327                         return 0;
1328                 }
1329         }
1330
1331         if (err < 0 || ttl == 0)
1332                 return 0;
1333
1334         qlen = strlen(question);
1335
1336         /*
1337          * If the cache contains already data, check if the
1338          * type of the cached data is the same and do not add
1339          * to cache if data is already there.
1340          * This is needed so that we can cache both A and AAAA
1341          * records for the same name.
1342          */
1343         entry = g_hash_table_lookup(cache, question);
1344         if (entry == NULL) {
1345                 entry = g_try_new(struct cache_entry, 1);
1346                 if (entry == NULL)
1347                         return -ENOMEM;
1348
1349                 data = g_try_new(struct cache_data, 1);
1350                 if (data == NULL) {
1351                         g_free(entry);
1352                         return -ENOMEM;
1353                 }
1354
1355                 entry->key = g_strdup(question);
1356                 entry->ipv4 = entry->ipv6 = NULL;
1357                 entry->want_refresh = 0;
1358                 entry->hits = 0;
1359
1360                 if (type == 1)
1361                         entry->ipv4 = data;
1362                 else
1363                         entry->ipv6 = data;
1364         } else {
1365                 if (type == 1 && entry->ipv4 != NULL)
1366                         return 0;
1367
1368                 if (type == 28 && entry->ipv6 != NULL)
1369                         return 0;
1370
1371                 data = g_try_new(struct cache_data, 1);
1372                 if (data == NULL)
1373                         return -ENOMEM;
1374
1375                 if (type == 1)
1376                         entry->ipv4 = data;
1377                 else
1378                         entry->ipv6 = data;
1379
1380                 /*
1381                  * compensate for the hit we'll get for serving
1382                  * the response out of the cache
1383                  */
1384                 entry->hits--;
1385                 if (entry->hits < 0)
1386                         entry->hits = 0;
1387
1388                 new_entry = FALSE;
1389         }
1390
1391         if (ttl < MIN_CACHE_TTL)
1392                 ttl = MIN_CACHE_TTL;
1393
1394         data->inserted = current_time;
1395         data->type = type;
1396         data->answers = answers;
1397         data->timeout = ttl;
1398         /*
1399          * The "2" in start of the length is the TCP offset. We allocate it
1400          * here even for UDP packet because it simplifies the sending
1401          * of cached packet.
1402          */
1403         data->data_len = 2 + 12 + qlen + 1 + 2 + 2 + rsplen;
1404         data->data = ptr = g_malloc(data->data_len);
1405         data->valid_until = current_time + ttl;
1406
1407         /*
1408          * Restrict the cached DNS record TTL to some sane value
1409          * in order to prevent data staying in the cache too long.
1410          */
1411         if (ttl > MAX_CACHE_TTL)
1412                 ttl = MAX_CACHE_TTL;
1413
1414         data->cache_until = round_down_ttl(current_time + ttl, ttl);
1415
1416         if (data->data == NULL) {
1417                 g_free(entry->key);
1418                 g_free(data);
1419                 g_free(entry);
1420                 return -ENOMEM;
1421         }
1422
1423         /*
1424          * We cache the two extra bytes at the start of the message
1425          * in a TCP packet. When sending UDP packet, we skip the first
1426          * two bytes. This way we do not need to know the format
1427          * (UDP/TCP) of the cached message.
1428          */
1429         ptr[0] = (data->data_len - 2) / 256;
1430         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1431         if (srv->protocol == IPPROTO_UDP)
1432                 ptr += 2;
1433
1434         memcpy(ptr, msg, offset + 12);
1435         memcpy(ptr + offset + 12, question, qlen + 1); /* copy also the \0 */
1436
1437         q = (void *) (ptr + offset + 12 + qlen + 1);
1438         q->type = htons(type);
1439         q->class = htons(class);
1440         memcpy(ptr + offset + 12 + qlen + 1 + sizeof(struct domain_question),
1441                 response, rsplen);
1442
1443         if (new_entry == TRUE) {
1444                 g_hash_table_replace(cache, entry->key, entry);
1445                 cache_size++;
1446         }
1447
1448         DBG("cache %d %squestion \"%s\" type %d ttl %d size %zd packet %u "
1449                                                                 "dns len %u",
1450                 cache_size, new_entry ? "new " : "old ",
1451                 question, type, ttl,
1452                 sizeof(*entry) + sizeof(*data) + data->data_len + qlen,
1453                 data->data_len,
1454                 srv->protocol == IPPROTO_TCP ?
1455                         (unsigned int)(data->data[0] * 256 + data->data[1]) :
1456                         data->data_len);
1457
1458         return 0;
1459 }
1460
1461 static int ns_resolv(struct server_data *server, struct request_data *req,
1462                                 gpointer request, gpointer name)
1463 {
1464         GList *list;
1465         int sk, err, type = 0;
1466         char *dot, *lookup = (char *) name;
1467         struct cache_entry *entry;
1468
1469         entry = cache_check(request, &type, req->protocol);
1470         if (entry != NULL) {
1471                 int ttl_left = 0;
1472                 struct cache_data *data;
1473
1474                 DBG("cache hit %s type %s", lookup, type == 1 ? "A" : "AAAA");
1475                 if (type == 1)
1476                         data = entry->ipv4;
1477                 else
1478                         data = entry->ipv6;
1479
1480                 if (data) {
1481                         ttl_left = data->valid_until - time(NULL);
1482                         entry->hits++;
1483                 }
1484
1485                 if (data != NULL && req->protocol == IPPROTO_TCP) {
1486                         send_cached_response(req->client_sk, data->data,
1487                                         data->data_len, NULL, 0, IPPROTO_TCP,
1488                                         req->srcid, data->answers, ttl_left);
1489                         return 1;
1490                 }
1491
1492                 if (data != NULL && req->protocol == IPPROTO_UDP) {
1493                         int sk;
1494                         sk = g_io_channel_unix_get_fd(
1495                                         req->ifdata->udp_listener_channel);
1496
1497                         send_cached_response(sk, data->data,
1498                                 data->data_len, &req->sa, req->sa_len,
1499                                 IPPROTO_UDP, req->srcid, data->answers,
1500                                 ttl_left);
1501                         return 1;
1502                 }
1503         }
1504
1505         sk = g_io_channel_unix_get_fd(server->channel);
1506
1507         err = send(sk, request, req->request_len, MSG_NOSIGNAL);
1508         if (err < 0)
1509                 return -EIO;
1510
1511         req->numserv++;
1512
1513         /* If we have more than one dot, we don't add domains */
1514         dot = strchr(lookup, '.');
1515         if (dot != NULL && dot != lookup + strlen(lookup) - 1)
1516                 return 0;
1517
1518         if (server->domains != NULL && server->domains->data != NULL)
1519                 req->append_domain = TRUE;
1520
1521         for (list = server->domains; list; list = list->next) {
1522                 char *domain;
1523                 unsigned char alt[1024];
1524                 struct domain_hdr *hdr = (void *) &alt;
1525                 int altlen, domlen, offset;
1526
1527                 domain = list->data;
1528
1529                 if (domain == NULL)
1530                         continue;
1531
1532                 offset = protocol_offset(server->protocol);
1533                 if (offset < 0)
1534                         return offset;
1535
1536                 domlen = strlen(domain) + 1;
1537                 if (domlen < 5)
1538                         return -EINVAL;
1539
1540                 alt[offset] = req->altid & 0xff;
1541                 alt[offset + 1] = req->altid >> 8;
1542
1543                 memcpy(alt + offset + 2, request + offset + 2, 10);
1544                 hdr->qdcount = htons(1);
1545
1546                 altlen = append_query(alt + offset + 12, sizeof(alt) - 12,
1547                                         name, domain);
1548                 if (altlen < 0)
1549                         return -EINVAL;
1550
1551                 altlen += 12;
1552
1553                 memcpy(alt + offset + altlen,
1554                         request + offset + altlen - domlen,
1555                                 req->request_len - altlen - offset + domlen);
1556
1557                 if (server->protocol == IPPROTO_TCP) {
1558                         int req_len = req->request_len + domlen - 2;
1559
1560                         alt[0] = (req_len >> 8) & 0xff;
1561                         alt[1] = req_len & 0xff;
1562                 }
1563
1564                 err = send(sk, alt, req->request_len + domlen, MSG_NOSIGNAL);
1565                 if (err < 0)
1566                         return -EIO;
1567
1568                 req->numserv++;
1569         }
1570
1571         return 0;
1572 }
1573
1574 static void destroy_request_data(struct request_data *req)
1575 {
1576         if (req->timeout > 0)
1577                 g_source_remove(req->timeout);
1578
1579         g_free(req->resp);
1580         g_free(req->request);
1581         g_free(req->name);
1582         g_free(req);
1583 }
1584
1585 static int forward_dns_reply(unsigned char *reply, int reply_len, int protocol,
1586                                 struct server_data *data)
1587 {
1588         struct domain_hdr *hdr;
1589         struct request_data *req;
1590         int dns_id, sk, err, offset = protocol_offset(protocol);
1591         struct listener_data *ifdata;
1592
1593         if (offset < 0)
1594                 return offset;
1595
1596         hdr = (void *)(reply + offset);
1597         dns_id = reply[offset] | reply[offset + 1] << 8;
1598
1599         DBG("Received %d bytes (id 0x%04x)", reply_len, dns_id);
1600
1601         req = find_request(dns_id);
1602         if (req == NULL)
1603                 return -EINVAL;
1604
1605         DBG("id 0x%04x rcode %d", hdr->id, hdr->rcode);
1606
1607         ifdata = req->ifdata;
1608
1609         reply[offset] = req->srcid & 0xff;
1610         reply[offset + 1] = req->srcid >> 8;
1611
1612         req->numresp++;
1613
1614         if (hdr->rcode == 0 || req->resp == NULL) {
1615
1616                 /*
1617                  * If the domain name was append
1618                  * remove it before forwarding the reply.
1619                  */
1620                 if (req->append_domain == TRUE) {
1621                         unsigned char *ptr;
1622                         uint8_t host_len;
1623                         unsigned int domain_len;
1624
1625                         /*
1626                          * ptr points to the first char of the hostname.
1627                          * ->hostname.domain.net
1628                          */
1629                         ptr = reply + offset + sizeof(struct domain_hdr);
1630                         host_len = *ptr;
1631                         domain_len = strlen((const char *)ptr + host_len + 1);
1632
1633                         /*
1634                          * Remove the domain name and replace it by the end
1635                          * of reply. Check if the domain is really there
1636                          * before trying to copy the data. The domain_len can
1637                          * be 0 because if the original query did not contain
1638                          * a domain name, then we are sending two packets,
1639                          * first without the domain name and the second packet
1640                          * with domain name. The append_domain is set to true
1641                          * even if we sent the first packet without domain
1642                          * name. In this case we end up in this branch.
1643                          */
1644                         if (domain_len > 0) {
1645                                 /*
1646                                  * Note that we must use memmove() here,
1647                                  * because the memory areas can overlap.
1648                                  */
1649                                 memmove(ptr + host_len + 1,
1650                                         ptr + host_len + domain_len + 1,
1651                                         reply_len - (ptr - reply + domain_len));
1652
1653                                 reply_len = reply_len - domain_len;
1654                         }
1655                 }
1656
1657                 g_free(req->resp);
1658                 req->resplen = 0;
1659
1660                 req->resp = g_try_malloc(reply_len);
1661                 if (req->resp == NULL)
1662                         return -ENOMEM;
1663
1664                 memcpy(req->resp, reply, reply_len);
1665                 req->resplen = reply_len;
1666
1667                 cache_update(data, reply, reply_len);
1668         }
1669
1670         if (hdr->rcode > 0 && req->numresp < req->numserv)
1671                 return -EINVAL;
1672
1673         request_list = g_slist_remove(request_list, req);
1674
1675         if (protocol == IPPROTO_UDP) {
1676                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
1677                 err = sendto(sk, req->resp, req->resplen, 0,
1678                              &req->sa, req->sa_len);
1679         } else {
1680                 sk = req->client_sk;
1681                 err = send(sk, req->resp, req->resplen, MSG_NOSIGNAL);
1682                 close(sk);
1683         }
1684
1685         destroy_request_data(req);
1686
1687         return err;
1688 }
1689
1690 static void cache_element_destroy(gpointer value)
1691 {
1692         struct cache_entry *entry = value;
1693
1694         if (entry == NULL)
1695                 return;
1696
1697         if (entry->ipv4 != NULL) {
1698                 g_free(entry->ipv4->data);
1699                 g_free(entry->ipv4);
1700         }
1701
1702         if (entry->ipv6 != NULL) {
1703                 g_free(entry->ipv6->data);
1704                 g_free(entry->ipv6);
1705         }
1706
1707         g_free(entry->key);
1708         g_free(entry);
1709
1710         if (--cache_size < 0)
1711                 cache_size = 0;
1712 }
1713
1714 static gboolean try_remove_cache(gpointer user_data)
1715 {
1716         if (__sync_fetch_and_sub(&cache_refcount, 1) == 1) {
1717                 DBG("No cache users, removing it.");
1718
1719                 g_hash_table_destroy(cache);
1720                 cache = NULL;
1721         }
1722
1723         return FALSE;
1724 }
1725
1726 static void destroy_server(struct server_data *server)
1727 {
1728         GList *list;
1729
1730         DBG("interface %s server %s", server->interface, server->server);
1731
1732         server_list = g_slist_remove(server_list, server);
1733
1734         if (server->watch > 0)
1735                 g_source_remove(server->watch);
1736
1737         if (server->timeout > 0)
1738                 g_source_remove(server->timeout);
1739
1740         g_io_channel_unref(server->channel);
1741
1742         if (server->protocol == IPPROTO_UDP)
1743                 DBG("Removing DNS server %s", server->server);
1744
1745         g_free(server->incoming_reply);
1746         g_free(server->server);
1747         for (list = server->domains; list; list = list->next) {
1748                 char *domain = list->data;
1749
1750                 server->domains = g_list_remove(server->domains, domain);
1751                 g_free(domain);
1752         }
1753         g_free(server->interface);
1754
1755         /*
1756          * We do not remove cache right away but delay it few seconds.
1757          * The idea is that when IPv6 DNS server is added via RDNSS, it has a
1758          * lifetime. When the lifetime expires we decrease the refcount so it
1759          * is possible that the cache is then removed. Because a new DNS server
1760          * is usually created almost immediately we would then loose the cache
1761          * without any good reason. The small delay allows the new RDNSS to
1762          * create a new DNS server instance and the refcount does not go to 0.
1763          */
1764         g_timeout_add_seconds(3, try_remove_cache, NULL);
1765
1766         g_free(server);
1767 }
1768
1769 static gboolean udp_server_event(GIOChannel *channel, GIOCondition condition,
1770                                                         gpointer user_data)
1771 {
1772         unsigned char buf[4096];
1773         int sk, err, len;
1774         struct server_data *data = user_data;
1775
1776         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1777                 connman_error("Error with UDP server %s", data->server);
1778                 data->watch = 0;
1779                 return FALSE;
1780         }
1781
1782         sk = g_io_channel_unix_get_fd(channel);
1783
1784         len = recv(sk, buf, sizeof(buf), 0);
1785         if (len < 12)
1786                 return TRUE;
1787
1788         err = forward_dns_reply(buf, len, IPPROTO_UDP, data);
1789         if (err < 0)
1790                 return TRUE;
1791
1792         return TRUE;
1793 }
1794
1795 static gboolean tcp_server_event(GIOChannel *channel, GIOCondition condition,
1796                                                         gpointer user_data)
1797 {
1798         int sk;
1799         struct server_data *server = user_data;
1800
1801         sk = g_io_channel_unix_get_fd(channel);
1802         if (sk == 0)
1803                 return FALSE;
1804
1805         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1806                 GSList *list;
1807 hangup:
1808                 DBG("TCP server channel closed");
1809
1810                 /*
1811                  * Discard any partial response which is buffered; better
1812                  * to get a proper response from a working server.
1813                  */
1814                 g_free(server->incoming_reply);
1815                 server->incoming_reply = NULL;
1816
1817                 for (list = request_list; list; list = list->next) {
1818                         struct request_data *req = list->data;
1819                         struct domain_hdr *hdr;
1820
1821                         if (req->protocol == IPPROTO_UDP)
1822                                 continue;
1823
1824                         if (req->request == NULL)
1825                                 continue;
1826
1827                         /*
1828                          * If we're not waiting for any further response
1829                          * from another name server, then we send an error
1830                          * response to the client.
1831                          */
1832                         if (req->numserv && --(req->numserv))
1833                                 continue;
1834
1835                         hdr = (void *) (req->request + 2);
1836                         hdr->id = req->srcid;
1837                         send_response(req->client_sk, req->request,
1838                                 req->request_len, NULL, 0, IPPROTO_TCP);
1839
1840                         request_list = g_slist_remove(request_list, req);
1841                 }
1842
1843                 destroy_server(server);
1844
1845                 return FALSE;
1846         }
1847
1848         if ((condition & G_IO_OUT) && !server->connected) {
1849                 GSList *list;
1850                 GList *domains;
1851                 int no_request_sent = TRUE;
1852                 struct server_data *udp_server;
1853
1854                 udp_server = find_server(server->interface, server->server,
1855                                                                 IPPROTO_UDP);
1856                 if (udp_server != NULL) {
1857                         for (domains = udp_server->domains; domains;
1858                                                 domains = domains->next) {
1859                                 char *dom = domains->data;
1860
1861                                 DBG("Adding domain %s to %s",
1862                                                 dom, server->server);
1863
1864                                 server->domains = g_list_append(server->domains,
1865                                                                 g_strdup(dom));
1866                         }
1867                 }
1868
1869                 server->connected = TRUE;
1870                 server_list = g_slist_append(server_list, server);
1871
1872                 if (server->timeout > 0) {
1873                         g_source_remove(server->timeout);
1874                         server->timeout = 0;
1875                 }
1876
1877                 for (list = request_list; list; ) {
1878                         struct request_data *req = list->data;
1879                         int status;
1880
1881                         if (req->protocol == IPPROTO_UDP) {
1882                                 list = list->next;
1883                                 continue;
1884                         }
1885
1886                         DBG("Sending req %s over TCP", (char *)req->name);
1887
1888                         status = ns_resolv(server, req,
1889                                                 req->request, req->name);
1890                         if (status > 0) {
1891                                 /*
1892                                  * A cached result was sent,
1893                                  * so the request can be released
1894                                  */
1895                                 list = list->next;
1896                                 request_list = g_slist_remove(request_list, req);
1897                                 destroy_request_data(req);
1898                                 continue;
1899                         }
1900
1901                         if (status < 0) {
1902                                 list = list->next;
1903                                 continue;
1904                         }
1905
1906                         no_request_sent = FALSE;
1907
1908                         if (req->timeout > 0)
1909                                 g_source_remove(req->timeout);
1910
1911                         req->timeout = g_timeout_add_seconds(30,
1912                                                 request_timeout, req);
1913                         list = list->next;
1914                 }
1915
1916                 if (no_request_sent == TRUE) {
1917                         destroy_server(server);
1918                         return FALSE;
1919                 }
1920
1921         } else if (condition & G_IO_IN) {
1922                 struct partial_reply *reply = server->incoming_reply;
1923                 int bytes_recv;
1924
1925                 if (!reply) {
1926                         unsigned char reply_len_buf[2];
1927                         uint16_t reply_len;
1928
1929                         bytes_recv = recv(sk, reply_len_buf, 2, MSG_PEEK);
1930                         if (!bytes_recv) {
1931                                 goto hangup;
1932                         } else if (bytes_recv < 0) {
1933                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1934                                         return TRUE;
1935
1936                                 connman_error("DNS proxy error %s",
1937                                                 strerror(errno));
1938                                 goto hangup;
1939                         } else if (bytes_recv < 2)
1940                                 return TRUE;
1941
1942                         reply_len = reply_len_buf[1] | reply_len_buf[0] << 8;
1943                         reply_len += 2;
1944
1945                         DBG("TCP reply %d bytes", reply_len);
1946
1947                         reply = g_try_malloc(sizeof(*reply) + reply_len + 2);
1948                         if (!reply)
1949                                 return TRUE;
1950
1951                         reply->len = reply_len;
1952                         reply->received = 0;
1953
1954                         server->incoming_reply = reply;
1955                 }
1956
1957                 while (reply->received < reply->len) {
1958                         bytes_recv = recv(sk, reply->buf + reply->received,
1959                                         reply->len - reply->received, 0);
1960                         if (!bytes_recv) {
1961                                 connman_error("DNS proxy TCP disconnect");
1962                                 break;
1963                         } else if (bytes_recv < 0) {
1964                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1965                                         return TRUE;
1966
1967                                 connman_error("DNS proxy error %s",
1968                                                 strerror(errno));
1969                                 break;
1970                         }
1971                         reply->received += bytes_recv;
1972                 }
1973
1974                 forward_dns_reply(reply->buf, reply->received, IPPROTO_TCP,
1975                                         server);
1976
1977                 g_free(reply);
1978                 server->incoming_reply = NULL;
1979
1980                 destroy_server(server);
1981
1982                 return FALSE;
1983         }
1984
1985         return TRUE;
1986 }
1987
1988 static gboolean tcp_idle_timeout(gpointer user_data)
1989 {
1990         struct server_data *server = user_data;
1991
1992         DBG("");
1993
1994         if (server == NULL)
1995                 return FALSE;
1996
1997         destroy_server(server);
1998
1999         return FALSE;
2000 }
2001
2002 static struct server_data *create_server(const char *interface,
2003                                         const char *domain, const char *server,
2004                                         int protocol)
2005 {
2006         struct addrinfo hints, *rp;
2007         struct server_data *data;
2008         int sk, ret;
2009
2010         DBG("interface %s server %s", interface, server);
2011
2012         memset(&hints, 0, sizeof(hints));
2013
2014         switch (protocol) {
2015         case IPPROTO_UDP:
2016                 hints.ai_socktype = SOCK_DGRAM;
2017                 break;
2018
2019         case IPPROTO_TCP:
2020                 hints.ai_socktype = SOCK_STREAM;
2021                 break;
2022
2023         default:
2024                 return NULL;
2025         }
2026         hints.ai_family = AF_UNSPEC;
2027         hints.ai_flags = AI_PASSIVE | AI_NUMERICSERV | AI_NUMERICHOST;
2028
2029         ret = getaddrinfo(server, "53", &hints, &rp);
2030         if (ret) {
2031                 connman_error("Failed to parse server %s address: %s\n",
2032                               server, gai_strerror(ret));
2033                 return NULL;
2034         }
2035         /* Do not blindly copy this code elsewhere; it doesn't loop over the
2036            results using ->ai_next as it should. That's OK in *this* case
2037            because it was a numeric lookup; we *know* there's only one. */
2038
2039         sk = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
2040         if (sk < 0) {
2041                 connman_error("Failed to create server %s socket", server);
2042                 freeaddrinfo(rp);
2043                 return NULL;
2044         }
2045
2046         if (interface != NULL) {
2047                 if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2048                                 interface, strlen(interface) + 1) < 0) {
2049                         connman_error("Failed to bind server %s "
2050                                                 "to interface %s",
2051                                                         server, interface);
2052                         freeaddrinfo(rp);
2053                         close(sk);
2054                         return NULL;
2055                 }
2056         }
2057
2058         data = g_try_new0(struct server_data, 1);
2059         if (data == NULL) {
2060                 connman_error("Failed to allocate server %s data", server);
2061                 freeaddrinfo(rp);
2062                 close(sk);
2063                 return NULL;
2064         }
2065
2066         data->channel = g_io_channel_unix_new(sk);
2067         if (data->channel == NULL) {
2068                 connman_error("Failed to create server %s channel", server);
2069                 freeaddrinfo(rp);
2070                 close(sk);
2071                 g_free(data);
2072                 return NULL;
2073         }
2074
2075         g_io_channel_set_close_on_unref(data->channel, TRUE);
2076
2077         if (protocol == IPPROTO_TCP) {
2078                 g_io_channel_set_flags(data->channel, G_IO_FLAG_NONBLOCK, NULL);
2079                 data->watch = g_io_add_watch(data->channel,
2080                         G_IO_OUT | G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR,
2081                                                 tcp_server_event, data);
2082                 data->timeout = g_timeout_add_seconds(30, tcp_idle_timeout,
2083                                                                 data);
2084         } else
2085                 data->watch = g_io_add_watch(data->channel,
2086                         G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2087                                                 udp_server_event, data);
2088
2089         data->interface = g_strdup(interface);
2090         if (domain)
2091                 data->domains = g_list_append(data->domains, g_strdup(domain));
2092         data->server = g_strdup(server);
2093         data->protocol = protocol;
2094
2095         ret = connect(sk, rp->ai_addr, rp->ai_addrlen);
2096         freeaddrinfo(rp);
2097         if (ret < 0) {
2098                 if ((protocol == IPPROTO_TCP && errno != EINPROGRESS) ||
2099                                 protocol == IPPROTO_UDP) {
2100                         GList *list;
2101
2102                         connman_error("Failed to connect to server %s", server);
2103                         if (data->watch > 0)
2104                                 g_source_remove(data->watch);
2105                         if (data->timeout > 0)
2106                                 g_source_remove(data->timeout);
2107
2108                         g_io_channel_unref(data->channel);
2109                         close(sk);
2110
2111                         g_free(data->server);
2112                         g_free(data->interface);
2113                         for (list = data->domains; list; list = list->next) {
2114                                 char *domain = list->data;
2115
2116                                 data->domains = g_list_remove(data->domains,
2117                                                                         domain);
2118                                 g_free(domain);
2119                         }
2120                         g_free(data);
2121                         return NULL;
2122                 }
2123         }
2124
2125         if (__sync_fetch_and_add(&cache_refcount, 1) == 0)
2126                 cache = g_hash_table_new_full(g_str_hash,
2127                                         g_str_equal,
2128                                         NULL,
2129                                         cache_element_destroy);
2130
2131         if (protocol == IPPROTO_UDP) {
2132                 /* Enable new servers by default */
2133                 data->enabled = TRUE;
2134                 DBG("Adding DNS server %s", data->server);
2135
2136                 server_list = g_slist_append(server_list, data);
2137         }
2138
2139         return data;
2140 }
2141
2142 static gboolean resolv(struct request_data *req,
2143                                 gpointer request, gpointer name)
2144 {
2145         GSList *list;
2146
2147         for (list = server_list; list; list = list->next) {
2148                 struct server_data *data = list->data;
2149
2150                 DBG("server %s enabled %d", data->server, data->enabled);
2151
2152                 if (data->enabled == FALSE)
2153                         continue;
2154
2155                 if (data->watch == 0 && data->protocol == IPPROTO_UDP)
2156                         data->watch = g_io_add_watch(data->channel,
2157                                 G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2158                                                 udp_server_event, data);
2159
2160                 if (ns_resolv(data, req, request, name) > 0)
2161                         return TRUE;
2162         }
2163
2164         return FALSE;
2165 }
2166
2167 static void append_domain(const char *interface, const char *domain)
2168 {
2169         GSList *list;
2170
2171         DBG("interface %s domain %s", interface, domain);
2172
2173         if (domain == NULL)
2174                 return;
2175
2176         for (list = server_list; list; list = list->next) {
2177                 struct server_data *data = list->data;
2178                 GList *dom_list;
2179                 char *dom;
2180                 gboolean dom_found = FALSE;
2181
2182                 if (data->interface == NULL)
2183                         continue;
2184
2185                 if (g_str_equal(data->interface, interface) == FALSE)
2186                         continue;
2187
2188                 for (dom_list = data->domains; dom_list;
2189                                 dom_list = dom_list->next) {
2190                         dom = dom_list->data;
2191
2192                         if (g_str_equal(dom, domain)) {
2193                                 dom_found = TRUE;
2194                                 break;
2195                         }
2196                 }
2197
2198                 if (dom_found == FALSE) {
2199                         data->domains =
2200                                 g_list_append(data->domains, g_strdup(domain));
2201                 }
2202         }
2203 }
2204
2205 int __connman_dnsproxy_append(const char *interface, const char *domain,
2206                                                         const char *server)
2207 {
2208         struct server_data *data;
2209
2210         DBG("interface %s server %s", interface, server);
2211
2212         if (server == NULL && domain == NULL)
2213                 return -EINVAL;
2214
2215         if (server == NULL) {
2216                 append_domain(interface, domain);
2217
2218                 return 0;
2219         }
2220
2221         if (g_str_equal(server, "127.0.0.1") == TRUE)
2222                 return -ENODEV;
2223
2224         data = find_server(interface, server, IPPROTO_UDP);
2225         if (data != NULL) {
2226                 append_domain(interface, domain);
2227                 return 0;
2228         }
2229
2230         data = create_server(interface, domain, server, IPPROTO_UDP);
2231         if (data == NULL)
2232                 return -EIO;
2233
2234         return 0;
2235 }
2236
2237 static void remove_server(const char *interface, const char *domain,
2238                         const char *server, int protocol)
2239 {
2240         struct server_data *data;
2241
2242         data = find_server(interface, server, protocol);
2243         if (data == NULL)
2244                 return;
2245
2246         destroy_server(data);
2247 }
2248
2249 int __connman_dnsproxy_remove(const char *interface, const char *domain,
2250                                                         const char *server)
2251 {
2252         DBG("interface %s server %s", interface, server);
2253
2254         if (server == NULL)
2255                 return -EINVAL;
2256
2257         if (g_str_equal(server, "127.0.0.1") == TRUE)
2258                 return -ENODEV;
2259
2260         remove_server(interface, domain, server, IPPROTO_UDP);
2261         remove_server(interface, domain, server, IPPROTO_TCP);
2262
2263         return 0;
2264 }
2265
2266 void __connman_dnsproxy_flush(void)
2267 {
2268         GSList *list;
2269
2270         list = request_list;
2271         while (list) {
2272                 struct request_data *req = list->data;
2273
2274                 list = list->next;
2275
2276                 if (resolv(req, req->request, req->name) == TRUE) {
2277                         /*
2278                          * A cached result was sent,
2279                          * so the request can be released
2280                          */
2281                         request_list =
2282                                 g_slist_remove(request_list, req);
2283                         destroy_request_data(req);
2284                         continue;
2285                 }
2286
2287                 if (req->timeout > 0)
2288                         g_source_remove(req->timeout);
2289                 req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2290         }
2291 }
2292
2293 static void dnsproxy_offline_mode(connman_bool_t enabled)
2294 {
2295         GSList *list;
2296
2297         DBG("enabled %d", enabled);
2298
2299         for (list = server_list; list; list = list->next) {
2300                 struct server_data *data = list->data;
2301
2302                 if (enabled == FALSE) {
2303                         DBG("Enabling DNS server %s", data->server);
2304                         data->enabled = TRUE;
2305                         cache_invalidate();
2306                         cache_refresh();
2307                 } else {
2308                         DBG("Disabling DNS server %s", data->server);
2309                         data->enabled = FALSE;
2310                         cache_invalidate();
2311                 }
2312         }
2313 }
2314
2315 static void dnsproxy_default_changed(struct connman_service *service)
2316 {
2317         GSList *list;
2318         char *interface;
2319
2320         DBG("service %p", service);
2321
2322         /* DNS has changed, invalidate the cache */
2323         cache_invalidate();
2324
2325         if (service == NULL) {
2326                 /* When no services are active, then disable DNS proxying */
2327                 dnsproxy_offline_mode(TRUE);
2328                 return;
2329         }
2330
2331         interface = connman_service_get_interface(service);
2332         if (interface == NULL)
2333                 return;
2334
2335         for (list = server_list; list; list = list->next) {
2336                 struct server_data *data = list->data;
2337
2338                 if (g_strcmp0(data->interface, interface) == 0) {
2339                         DBG("Enabling DNS server %s", data->server);
2340                         data->enabled = TRUE;
2341                 } else {
2342                         DBG("Disabling DNS server %s", data->server);
2343                         data->enabled = FALSE;
2344                 }
2345         }
2346
2347         g_free(interface);
2348         cache_refresh();
2349 }
2350
2351 static struct connman_notifier dnsproxy_notifier = {
2352         .name                   = "dnsproxy",
2353         .default_changed        = dnsproxy_default_changed,
2354         .offline_mode           = dnsproxy_offline_mode,
2355 };
2356
2357 static unsigned char opt_edns0_type[2] = { 0x00, 0x29 };
2358
2359 static int parse_request(unsigned char *buf, int len,
2360                                         char *name, unsigned int size)
2361 {
2362         struct domain_hdr *hdr = (void *) buf;
2363         uint16_t qdcount = ntohs(hdr->qdcount);
2364         uint16_t arcount = ntohs(hdr->arcount);
2365         unsigned char *ptr;
2366         char *last_label = NULL;
2367         unsigned int remain, used = 0;
2368
2369         if (len < 12)
2370                 return -EINVAL;
2371
2372         DBG("id 0x%04x qr %d opcode %d qdcount %d arcount %d",
2373                                         hdr->id, hdr->qr, hdr->opcode,
2374                                                         qdcount, arcount);
2375
2376         if (hdr->qr != 0 || qdcount != 1)
2377                 return -EINVAL;
2378
2379         name[0] = '\0';
2380
2381         ptr = buf + sizeof(struct domain_hdr);
2382         remain = len - sizeof(struct domain_hdr);
2383
2384         while (remain > 0) {
2385                 uint8_t len = *ptr;
2386
2387                 if (len == 0x00) {
2388                         last_label = (char *) (ptr + 1);
2389                         break;
2390                 }
2391
2392                 if (used + len + 1 > size)
2393                         return -ENOBUFS;
2394
2395                 strncat(name, (char *) (ptr + 1), len);
2396                 strcat(name, ".");
2397
2398                 used += len + 1;
2399
2400                 ptr += len + 1;
2401                 remain -= len + 1;
2402         }
2403
2404         if (last_label && arcount && remain >= 9 && last_label[4] == 0 &&
2405                                 !memcmp(last_label + 5, opt_edns0_type, 2)) {
2406                 uint16_t edns0_bufsize;
2407
2408                 edns0_bufsize = last_label[7] << 8 | last_label[8];
2409
2410                 DBG("EDNS0 buffer size %u", edns0_bufsize);
2411
2412                 /* This is an evil hack until full TCP support has been
2413                  * implemented.
2414                  *
2415                  * Somtimes the EDNS0 request gets send with a too-small
2416                  * buffer size. Since glibc doesn't seem to crash when it
2417                  * gets a response biffer then it requested, just bump
2418                  * the buffer size up to 4KiB.
2419                  */
2420                 if (edns0_bufsize < 0x1000) {
2421                         last_label[7] = 0x10;
2422                         last_label[8] = 0x00;
2423                 }
2424         }
2425
2426         DBG("query %s", name);
2427
2428         return 0;
2429 }
2430
2431 static gboolean tcp_listener_event(GIOChannel *channel, GIOCondition condition,
2432                                                         gpointer user_data)
2433 {
2434         unsigned char buf[768];
2435         char query[512];
2436         struct request_data *req;
2437         int sk, client_sk, len, err;
2438         struct sockaddr_in6 client_addr;
2439         socklen_t client_addr_len = sizeof(client_addr);
2440         GSList *list;
2441         struct listener_data *ifdata = user_data;
2442         int waiting_for_connect = FALSE;
2443
2444         DBG("condition 0x%x", condition);
2445
2446         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2447                 if (ifdata->tcp_listener_watch > 0)
2448                         g_source_remove(ifdata->tcp_listener_watch);
2449                 ifdata->tcp_listener_watch = 0;
2450
2451                 connman_error("Error with TCP listener channel");
2452
2453                 return FALSE;
2454         }
2455
2456         sk = g_io_channel_unix_get_fd(channel);
2457
2458         client_sk = accept(sk, (void *)&client_addr, &client_addr_len);
2459         if (client_sk < 0) {
2460                 connman_error("Accept failure on TCP listener");
2461                 ifdata->tcp_listener_watch = 0;
2462                 return FALSE;
2463         }
2464
2465         len = recv(client_sk, buf, sizeof(buf), 0);
2466         if (len < 2)
2467                 return TRUE;
2468
2469         DBG("Received %d bytes (id 0x%04x)", len, buf[2] | buf[3] << 8);
2470
2471         err = parse_request(buf + 2, len - 2, query, sizeof(query));
2472         if (err < 0 || (g_slist_length(server_list) == 0)) {
2473                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2474                 return TRUE;
2475         }
2476
2477         req = g_try_new0(struct request_data, 1);
2478         if (req == NULL)
2479                 return TRUE;
2480
2481         memcpy(&req->sa, &client_addr, client_addr_len);
2482         req->sa_len = client_addr_len;
2483         req->client_sk = client_sk;
2484         req->protocol = IPPROTO_TCP;
2485
2486         req->srcid = buf[2] | (buf[3] << 8);
2487         req->dstid = get_id();
2488         req->altid = get_id();
2489         req->request_len = len;
2490
2491         buf[2] = req->dstid & 0xff;
2492         buf[3] = req->dstid >> 8;
2493
2494         req->numserv = 0;
2495         req->ifdata = (struct listener_data *) ifdata;
2496         req->append_domain = FALSE;
2497
2498         for (list = server_list; list; list = list->next) {
2499                 struct server_data *data = list->data;
2500
2501                 if (data->protocol != IPPROTO_UDP || data->enabled == FALSE)
2502                         continue;
2503
2504                 if(create_server(data->interface, NULL,
2505                                         data->server, IPPROTO_TCP) == NULL)
2506                         continue;
2507
2508                 waiting_for_connect = TRUE;
2509         }
2510
2511         if (waiting_for_connect == FALSE) {
2512                 /* No server is waiting for connect */
2513                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2514                 g_free(req);
2515                 return TRUE;
2516         }
2517
2518         /*
2519          * The server is not connected yet.
2520          * Copy the relevant buffers.
2521          * The request will actually be sent once we're
2522          * properly connected over TCP to the nameserver.
2523          */
2524         req->request = g_try_malloc0(req->request_len);
2525         if (req->request == NULL) {
2526                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2527                 g_free(req);
2528                 return TRUE;
2529         }
2530         memcpy(req->request, buf, req->request_len);
2531
2532         req->name = g_try_malloc0(sizeof(query));
2533         if (req->name == NULL) {
2534                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2535                 g_free(req->request);
2536                 g_free(req);
2537                 return TRUE;
2538         }
2539         memcpy(req->name, query, sizeof(query));
2540
2541         req->timeout = g_timeout_add_seconds(30, request_timeout, req);
2542
2543         request_list = g_slist_append(request_list, req);
2544
2545         return TRUE;
2546 }
2547
2548 static gboolean udp_listener_event(GIOChannel *channel, GIOCondition condition,
2549                                                         gpointer user_data)
2550 {
2551         unsigned char buf[768];
2552         char query[512];
2553         struct request_data *req;
2554         struct sockaddr_in6 client_addr;
2555         socklen_t client_addr_len = sizeof(client_addr);
2556         int sk, err, len;
2557         struct listener_data *ifdata = user_data;
2558
2559         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2560                 connman_error("Error with UDP listener channel");
2561                 ifdata->udp_listener_watch = 0;
2562                 return FALSE;
2563         }
2564
2565         sk = g_io_channel_unix_get_fd(channel);
2566
2567         memset(&client_addr, 0, client_addr_len);
2568         len = recvfrom(sk, buf, sizeof(buf), 0, (void *)&client_addr,
2569                        &client_addr_len);
2570         if (len < 2)
2571                 return TRUE;
2572
2573         DBG("Received %d bytes (id 0x%04x)", len, buf[0] | buf[1] << 8);
2574
2575         err = parse_request(buf, len, query, sizeof(query));
2576         if (err < 0 || (g_slist_length(server_list) == 0)) {
2577                 send_response(sk, buf, len, (void *)&client_addr,
2578                                 client_addr_len, IPPROTO_UDP);
2579                 return TRUE;
2580         }
2581
2582         req = g_try_new0(struct request_data, 1);
2583         if (req == NULL)
2584                 return TRUE;
2585
2586         memcpy(&req->sa, &client_addr, client_addr_len);
2587         req->sa_len = client_addr_len;
2588         req->client_sk = 0;
2589         req->protocol = IPPROTO_UDP;
2590
2591         req->srcid = buf[0] | (buf[1] << 8);
2592         req->dstid = get_id();
2593         req->altid = get_id();
2594         req->request_len = len;
2595
2596         buf[0] = req->dstid & 0xff;
2597         buf[1] = req->dstid >> 8;
2598
2599         req->numserv = 0;
2600         req->ifdata = (struct listener_data *) ifdata;
2601         req->append_domain = FALSE;
2602
2603         if (resolv(req, buf, query) == TRUE) {
2604                 /* a cached result was sent, so the request can be released */
2605                 g_free(req);
2606                 return TRUE;
2607         }
2608
2609         req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2610         request_list = g_slist_append(request_list, req);
2611
2612         return TRUE;
2613 }
2614
2615 static int create_dns_listener(int protocol, struct listener_data *ifdata)
2616 {
2617         GIOChannel *channel;
2618         const char *proto;
2619         union {
2620                 struct sockaddr sa;
2621                 struct sockaddr_in6 sin6;
2622                 struct sockaddr_in sin;
2623         } s;
2624         socklen_t slen;
2625         int sk, type, v6only = 0;
2626         int family = AF_INET6;
2627
2628
2629         DBG("interface %s", ifdata->ifname);
2630
2631         switch (protocol) {
2632         case IPPROTO_UDP:
2633                 proto = "UDP";
2634                 type = SOCK_DGRAM | SOCK_CLOEXEC;
2635                 break;
2636
2637         case IPPROTO_TCP:
2638                 proto = "TCP";
2639                 type = SOCK_STREAM | SOCK_CLOEXEC;
2640                 break;
2641
2642         default:
2643                 return -EINVAL;
2644         }
2645
2646         sk = socket(family, type, protocol);
2647         if (sk < 0 && family == AF_INET6 && errno == EAFNOSUPPORT) {
2648                 connman_error("No IPv6 support; DNS proxy listening only on Legacy IP");
2649                 family = AF_INET;
2650                 sk = socket(family, type, protocol);
2651         }
2652         if (sk < 0) {
2653                 connman_error("Failed to create %s listener socket", proto);
2654                 return -EIO;
2655         }
2656
2657         if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2658                                         ifdata->ifname,
2659                                         strlen(ifdata->ifname) + 1) < 0) {
2660                 connman_error("Failed to bind %s listener interface", proto);
2661                 close(sk);
2662                 return -EIO;
2663         }
2664         /* Ensure it accepts Legacy IP connections too */
2665         if (family == AF_INET6 &&
2666                         setsockopt(sk, SOL_IPV6, IPV6_V6ONLY,
2667                                         &v6only, sizeof(v6only)) < 0) {
2668                 connman_error("Failed to clear V6ONLY on %s listener socket",
2669                               proto);
2670                 close(sk);
2671                 return -EIO;
2672         }
2673
2674         if (family == AF_INET) {
2675                 memset(&s.sin, 0, sizeof(s.sin));
2676                 s.sin.sin_family = AF_INET;
2677                 s.sin.sin_port = htons(53);
2678                 s.sin.sin_addr.s_addr = htonl(INADDR_ANY);
2679                 slen = sizeof(s.sin);
2680         } else {
2681                 memset(&s.sin6, 0, sizeof(s.sin6));
2682                 s.sin6.sin6_family = AF_INET6;
2683                 s.sin6.sin6_port = htons(53);
2684                 s.sin6.sin6_addr = in6addr_any;
2685                 slen = sizeof(s.sin6);
2686         }
2687
2688         if (bind(sk, &s.sa, slen) < 0) {
2689                 connman_error("Failed to bind %s listener socket", proto);
2690                 close(sk);
2691                 return -EIO;
2692         }
2693
2694         if (protocol == IPPROTO_TCP && listen(sk, 10) < 0) {
2695                 connman_error("Failed to listen on TCP socket");
2696                 close(sk);
2697                 return -EIO;
2698         }
2699
2700         channel = g_io_channel_unix_new(sk);
2701         if (channel == NULL) {
2702                 connman_error("Failed to create %s listener channel", proto);
2703                 close(sk);
2704                 return -EIO;
2705         }
2706
2707         g_io_channel_set_close_on_unref(channel, TRUE);
2708
2709         if (protocol == IPPROTO_TCP) {
2710                 ifdata->tcp_listener_channel = channel;
2711                 ifdata->tcp_listener_watch = g_io_add_watch(channel,
2712                                 G_IO_IN, tcp_listener_event, (gpointer) ifdata);
2713         } else {
2714                 ifdata->udp_listener_channel = channel;
2715                 ifdata->udp_listener_watch = g_io_add_watch(channel,
2716                                 G_IO_IN, udp_listener_event, (gpointer) ifdata);
2717         }
2718
2719         return 0;
2720 }
2721
2722 static void destroy_udp_listener(struct listener_data *ifdata)
2723 {
2724         DBG("interface %s", ifdata->ifname);
2725
2726         if (ifdata->udp_listener_watch > 0)
2727                 g_source_remove(ifdata->udp_listener_watch);
2728
2729         g_io_channel_unref(ifdata->udp_listener_channel);
2730 }
2731
2732 static void destroy_tcp_listener(struct listener_data *ifdata)
2733 {
2734         DBG("interface %s", ifdata->ifname);
2735
2736         if (ifdata->tcp_listener_watch > 0)
2737                 g_source_remove(ifdata->tcp_listener_watch);
2738
2739         g_io_channel_unref(ifdata->tcp_listener_channel);
2740 }
2741
2742 static int create_listener(struct listener_data *ifdata)
2743 {
2744         int err;
2745
2746         err = create_dns_listener(IPPROTO_UDP, ifdata);
2747         if (err < 0)
2748                 return err;
2749
2750         err = create_dns_listener(IPPROTO_TCP, ifdata);
2751         if (err < 0) {
2752                 destroy_udp_listener(ifdata);
2753                 return err;
2754         }
2755
2756         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2757                 __connman_resolvfile_append("lo", NULL, "127.0.0.1");
2758
2759         return 0;
2760 }
2761
2762 static void destroy_listener(struct listener_data *ifdata)
2763 {
2764         GSList *list;
2765
2766         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2767                 __connman_resolvfile_remove("lo", NULL, "127.0.0.1");
2768
2769         for (list = request_list; list; list = list->next) {
2770                 struct request_data *req = list->data;
2771
2772                 DBG("Dropping request (id 0x%04x -> 0x%04x)",
2773                                                 req->srcid, req->dstid);
2774                 destroy_request_data(req);
2775                 list->data = NULL;
2776         }
2777
2778         g_slist_free(request_list);
2779         request_list = NULL;
2780
2781         destroy_tcp_listener(ifdata);
2782         destroy_udp_listener(ifdata);
2783 }
2784
2785 int __connman_dnsproxy_add_listener(const char *interface)
2786 {
2787         struct listener_data *ifdata;
2788         int err;
2789
2790         DBG("interface %s", interface);
2791
2792         if (g_hash_table_lookup(listener_table, interface) != NULL)
2793                 return 0;
2794
2795         ifdata = g_try_new0(struct listener_data, 1);
2796         if (ifdata == NULL)
2797                 return -ENOMEM;
2798
2799         ifdata->ifname = g_strdup(interface);
2800         ifdata->udp_listener_channel = NULL;
2801         ifdata->udp_listener_watch = 0;
2802         ifdata->tcp_listener_channel = NULL;
2803         ifdata->tcp_listener_watch = 0;
2804
2805         err = create_listener(ifdata);
2806         if (err < 0) {
2807                 connman_error("Couldn't create listener for %s err %d",
2808                                 interface, err);
2809                 g_free(ifdata->ifname);
2810                 g_free(ifdata);
2811                 return err;
2812         }
2813         g_hash_table_insert(listener_table, ifdata->ifname, ifdata);
2814         return 0;
2815 }
2816
2817 void __connman_dnsproxy_remove_listener(const char *interface)
2818 {
2819         struct listener_data *ifdata;
2820
2821         DBG("interface %s", interface);
2822
2823         ifdata = g_hash_table_lookup(listener_table, interface);
2824         if (ifdata == NULL)
2825                 return;
2826
2827         destroy_listener(ifdata);
2828
2829         g_hash_table_remove(listener_table, interface);
2830 }
2831
2832 static void remove_listener(gpointer key, gpointer value, gpointer user_data)
2833 {
2834         const char *interface = key;
2835         struct listener_data *ifdata = value;
2836
2837         DBG("interface %s", interface);
2838
2839         destroy_listener(ifdata);
2840 }
2841
2842 int __connman_dnsproxy_init(void)
2843 {
2844         int err;
2845
2846         DBG("");
2847
2848         srandom(time(NULL));
2849
2850         listener_table = g_hash_table_new_full(g_str_hash, g_str_equal,
2851                                                         g_free, g_free);
2852         err = __connman_dnsproxy_add_listener("lo");
2853         if (err < 0)
2854                 return err;
2855
2856         err = connman_notifier_register(&dnsproxy_notifier);
2857         if (err < 0)
2858                 goto destroy;
2859
2860         return 0;
2861
2862 destroy:
2863         __connman_dnsproxy_remove_listener("lo");
2864         g_hash_table_destroy(listener_table);
2865
2866         return err;
2867 }
2868
2869 void __connman_dnsproxy_cleanup(void)
2870 {
2871         DBG("");
2872
2873         connman_notifier_unregister(&dnsproxy_notifier);
2874
2875         g_hash_table_foreach(listener_table, remove_listener, NULL);
2876
2877         g_hash_table_destroy(listener_table);
2878 }