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         int index;
  87         GList *domains;
  88         char *server;
  89         struct sockaddr *server_addr;
  90         socklen_t server_addr_len;
  91         int protocol;
  92         GIOChannel *channel;
  93         guint watch;
  94         guint timeout;
  95         gboolean enabled;
  96         gboolean connected;
  97         struct partial_reply *incoming_reply;
  98 };
  99
 100 struct request_data {
 101         union {
 102                 struct sockaddr_in6 __sin6; /* Only for the length */
 103                 struct sockaddr sa;
 104         };
 105         socklen_t sa_len;
 106         int client_sk;
 107         int protocol;
 108         guint16 srcid;
 109         guint16 dstid;
 110         guint16 altid;
 111         guint timeout;
 112         guint watch;
 113         guint numserv;
 114         guint numresp;
 115         gpointer request;
 116         gsize request_len;
 117         gpointer name;
 118         gpointer resp;
 119         gsize resplen;
 120         struct listener_data *ifdata;
 121         gboolean append_domain;
 122 };
 123
 124 struct listener_data {
 125         int index;
 126         GIOChannel *udp_listener_channel;
 127         guint udp_listener_watch;
 128         GIOChannel *tcp_listener_channel;
 129         guint tcp_listener_watch;
 130 };
 131
 132 struct cache_data {
 133         time_t inserted;
 134         time_t valid_until;
 135         time_t cache_until;
 136         int timeout;
 137         uint16_t type;
 138         uint16_t answers;
 139         unsigned int data_len;
 140         unsigned char *data; /* contains DNS header + body */
 141 };
 142
 143 struct cache_entry {
 144         char *key;
 145         int want_refresh;
 146         int hits;
 147         struct cache_data *ipv4;
 148         struct cache_data *ipv6;
 149 };
 150
 151 struct domain_question {
 152         uint16_t type;
 153         uint16_t class;
 154 } __attribute__ ((packed));
 155
 156 struct domain_rr {
 157         uint16_t type;
 158         uint16_t class;
 159         uint32_t ttl;
 160         uint16_t rdlen;
 161 } __attribute__ ((packed));
 162
 163 /*
 164  * We limit how long the cached DNS entry stays in the cache.
 165  * By default the TTL (time-to-live) of the DNS response is used
 166  * when setting the cache entry life time. The value is in seconds.
 167  */
 168 #define MAX_CACHE_TTL (60 * 30)
 169 /*
 170  * Also limit the other end, cache at least for 30 seconds.
 171  */
 172 #define MIN_CACHE_TTL (30)
 173
 174 /*
 175  * We limit the cache size to some sane value so that cached data does
 176  * not occupy too much memory. Each cached entry occupies on average
 177  * about 100 bytes memory (depending on DNS name length).
 178  * Example: caching www.connman.net uses 97 bytes memory.
 179  * The value is the max amount of cached DNS responses (count).
 180  */
 181 #define MAX_CACHE_SIZE 256
 182
 183 static int cache_size;
 184 static GHashTable *cache;
 185 static int cache_refcount;
 186 static GSList *server_list = NULL;
 187 static GSList *request_list = NULL;
 188 static GHashTable *listener_table = NULL;
 189 static time_t next_refresh;
 190
 191 static guint16 get_id()
 192 {
 193         return random();
 194 }
 195
 196 static int protocol_offset(int protocol)
 197 {
 198         switch (protocol) {
 199         case IPPROTO_UDP:
 200                 return 0;
 201
 202         case IPPROTO_TCP:
 203                 return 2;
 204
 205         default:
 206                 return -EINVAL;
 207         }
 208
 209 }
 210
 211 /*
 212  * There is a power and efficiency benefit to have entries
 213  * in our cache expire at the same time. To this extend,
 214  * we round down the cache valid time to common boundaries.
 215  */
 216 static time_t round_down_ttl(time_t end_time, int ttl)
 217 {
 218         if (ttl < 15)
 219                 return end_time;
 220
 221         /* Less than 5 minutes, round to 10 second boundary */
 222         if (ttl < 300) {
 223                 end_time = end_time / 10;
 224                 end_time = end_time * 10;
 225         } else { /* 5 or more minutes, round to 30 seconds */
 226                 end_time = end_time / 30;
 227                 end_time = end_time * 30;
 228         }
 229         return end_time;
 230 }
 231
 232 static struct request_data *find_request(guint16 id)
 233 {
 234         GSList *list;
 235
 236         for (list = request_list; list; list = list->next) {
 237                 struct request_data *req = list->data;
 238
 239                 if (req->dstid == id || req->altid == id)
 240                         return req;
 241         }
 242
 243         return NULL;
 244 }
 245
 246 static struct server_data *find_server(int index,
 247                                         const char *server,
 248                                                 int protocol)
 249 {
 250         GSList *list;
 251
 252         DBG("index %d server %s proto %d", index, server, protocol);
 253
 254         for (list = server_list; list; list = list->next) {
 255                 struct server_data *data = list->data;
 256
 257                 if (index < 0 && data->index < 0 &&
 258                                 g_str_equal(data->server, server) == TRUE &&
 259                                 data->protocol == protocol)
 260                         return data;
 261
 262                 if (index < 0 ||
 263                                 data->index < 0 || data->server == NULL)
 264                         continue;
 265
 266                 if (data->index == index &&
 267                                 g_str_equal(data->server, server) == TRUE &&
 268                                 data->protocol == protocol)
 269                         return data;
 270         }
 271
 272         return NULL;
 273 }
 274
 275 /* we can keep using the same resolve's */
 276 static GResolv *ipv4_resolve;
 277 static GResolv *ipv6_resolve;
 278
 279 static void dummy_resolve_func(GResolvResultStatus status,
 280                                         char **results, gpointer user_data)
 281 {
 282 }
 283
 284 /*
 285  * Refresh a DNS entry, but also age the hit count a bit */
 286 static void refresh_dns_entry(struct cache_entry *entry, char *name)
 287 {
 288         int age = 1;
 289
 290         if (ipv4_resolve == NULL) {
 291                 ipv4_resolve = g_resolv_new(0);
 292                 g_resolv_set_address_family(ipv4_resolve, AF_INET);
 293                 g_resolv_add_nameserver(ipv4_resolve, "127.0.0.1", 53, 0);
 294         }
 295
 296         if (ipv6_resolve == NULL) {
 297                 ipv6_resolve = g_resolv_new(0);
 298                 g_resolv_set_address_family(ipv6_resolve, AF_INET6);
 299                 g_resolv_add_nameserver(ipv6_resolve, "127.0.0.1", 53, 0);
 300         }
 301
 302         if (entry->ipv4 == NULL) {
 303                 DBG("Refresing A record for %s", name);
 304                 g_resolv_lookup_hostname(ipv4_resolve, name,
 305                                         dummy_resolve_func, NULL);
 306                 age = 4;
 307         }
 308
 309         if (entry->ipv6 == NULL) {
 310                 DBG("Refresing AAAA record for %s", name);
 311                 g_resolv_lookup_hostname(ipv6_resolve, name,
 312                                         dummy_resolve_func, NULL);
 313                 age = 4;
 314         }
 315
 316         entry->hits -= age;
 317         if (entry->hits < 0)
 318                 entry->hits = 0;
 319 }
 320
 321 static int dns_name_length(unsigned char *buf)
 322 {
 323         if ((buf[0] & NS_CMPRSFLGS) == NS_CMPRSFLGS) /* compressed name */
 324                 return 2;
 325         return strlen((char *)buf);
 326 }
 327
 328 static void update_cached_ttl(unsigned char *buf, int len, int new_ttl)
 329 {
 330         unsigned char *c;
 331         uint32_t *i;
 332         uint16_t *w;
 333         int l;
 334
 335         /* skip the header */
 336         c = buf + 12;
 337         len -= 12;
 338
 339         /* skip the query, which is a name and 2 16 bit words */
 340         l = dns_name_length(c);
 341         c += l;
 342         len -= l;
 343         c += 4;
 344         len -= 4;
 345
 346         /* now we get the answer records */
 347
 348         while (len > 0) {
 349                 /* first a name */
 350                 l = dns_name_length(c);
 351                 c += l;
 352                 len -= l;
 353                 if (len < 0)
 354                         break;
 355                 /* then type + class, 2 bytes each */
 356                 c += 4;
 357                 len -= 4;
 358                 if (len < 0)
 359                         break;
 360
 361                 /* now the 4 byte TTL field */
 362                 i = (uint32_t *)c;
 363                 *i = htonl(new_ttl);
 364                 c += 4;
 365                 len -= 4;
 366                 if (len < 0)
 367                         break;
 368
 369                 /* now the 2 byte rdlen field */
 370                 w = (uint16_t *)c;
 371                 c += ntohs(*w) + 2;
 372                 len -= ntohs(*w) + 2;
 373         }
 374 }
 375
 376 static void send_cached_response(int sk, unsigned char *buf, int len,
 377                                 const struct sockaddr *to, socklen_t tolen,
 378                                 int protocol, int id, uint16_t answers, int ttl)
 379 {
 380         struct domain_hdr *hdr;
 381         unsigned char *ptr = buf;
 382         int err, offset, dns_len, adj_len = len - 2;
 383
 384         /*
 385          * The cached packet contains always the TCP offset (two bytes)
 386          * so skip them for UDP.
 387          */
 388         switch (protocol) {
 389         case IPPROTO_UDP:
 390                 ptr += 2;
 391                 len -= 2;
 392                 dns_len = len;
 393                 offset = 0;
 394                 break;
 395         case IPPROTO_TCP:
 396                 offset = 2;
 397                 dns_len = ptr[0] * 256 + ptr[1];
 398                 break;
 399         default:
 400                 return;
 401         }
 402
 403         if (len < 12)
 404                 return;
 405
 406         hdr = (void *) (ptr + offset);
 407
 408         hdr->id = id;
 409         hdr->qr = 1;
 410         hdr->rcode = 0;
 411         hdr->ancount = htons(answers);
 412         hdr->nscount = 0;
 413         hdr->arcount = 0;
 414
 415         /* if this is a negative reply, we are authorative */
 416         if (answers == 0)
 417                 hdr->aa = 1;
 418         else
 419                 update_cached_ttl((unsigned char *)hdr, adj_len, ttl);
 420
 421         DBG("sk %d id 0x%04x answers %d ptr %p length %d dns %d",
 422                 sk, hdr->id, answers, ptr, len, dns_len);
 423
 424         err = sendto(sk, ptr, len, MSG_NOSIGNAL, to, tolen);
 425         if (err < 0) {
 426                 connman_error("Cannot send cached DNS response: %s",
 427                                 strerror(errno));
 428                 return;
 429         }
 430
 431         if (err != len || (dns_len != (len - 2) && protocol == IPPROTO_TCP) ||
 432                                 (dns_len != len && protocol == IPPROTO_UDP))
 433                 DBG("Packet length mismatch, sent %d wanted %d dns %d",
 434                         err, len, dns_len);
 435 }
 436
 437 static void send_response(int sk, unsigned char *buf, int len,
 438                                 const struct sockaddr *to, socklen_t tolen,
 439                                 int protocol)
 440 {
 441         struct domain_hdr *hdr;
 442         int err, offset = protocol_offset(protocol);
 443
 444         DBG("sk %d", sk);
 445
 446         if (offset < 0)
 447                 return;
 448
 449         if (len < 12)
 450                 return;
 451
 452         hdr = (void *) (buf + offset);
 453
 454         DBG("id 0x%04x qr %d opcode %d", hdr->id, hdr->qr, hdr->opcode);
 455
 456         hdr->qr = 1;
 457         hdr->rcode = 2;
 458
 459         hdr->ancount = 0;
 460         hdr->nscount = 0;
 461         hdr->arcount = 0;
 462
 463         err = sendto(sk, buf, len, MSG_NOSIGNAL, to, tolen);
 464         if (err < 0) {
 465                 connman_error("Failed to send DNS response to %d: %s",
 466                                 sk, strerror(errno));
 467                 return;
 468         }
 469 }
 470
 471 static gboolean request_timeout(gpointer user_data)
 472 {
 473         struct request_data *req = user_data;
 474         struct listener_data *ifdata;
 475
 476         DBG("id 0x%04x", req->srcid);
 477
 478         if (req == NULL)
 479                 return FALSE;
 480
 481         ifdata = req->ifdata;
 482
 483         request_list = g_slist_remove(request_list, req);
 484         req->numserv--;
 485
 486         if (req->resplen > 0 && req->resp != NULL) {
 487                 int sk, err;
 488
 489                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
 490
 491                 err = sendto(sk, req->resp, req->resplen, MSG_NOSIGNAL,
 492                                                 &req->sa, req->sa_len);
 493                 if (err < 0)
 494                         return FALSE;
 495         } else if (req->request && req->numserv == 0) {
 496                 struct domain_hdr *hdr;
 497
 498                 if (req->protocol == IPPROTO_TCP) {
 499                         hdr = (void *) (req->request + 2);
 500                         hdr->id = req->srcid;
 501                         send_response(req->client_sk, req->request,
 502                                 req->request_len, NULL, 0, IPPROTO_TCP);
 503
 504                 } else if (req->protocol == IPPROTO_UDP) {
 505                         int sk;
 506
 507                         hdr = (void *) (req->request);
 508                         hdr->id = req->srcid;
 509                         sk = g_io_channel_unix_get_fd(
 510                                                 ifdata->udp_listener_channel);
 511                         send_response(sk, req->request, req->request_len,
 512                                         &req->sa, req->sa_len, IPPROTO_UDP);
 513                 }
 514         }
 515
 516         g_free(req->resp);
 517         g_free(req);
 518
 519         return FALSE;
 520 }
 521
 522 static int append_query(unsigned char *buf, unsigned int size,
 523                                 const char *query, const char *domain)
 524 {
 525         unsigned char *ptr = buf;
 526         int len;
 527
 528         DBG("query %s domain %s", query, domain);
 529
 530         while (query != NULL) {
 531                 const char *tmp;
 532
 533                 tmp = strchr(query, '.');
 534                 if (tmp == NULL) {
 535                         len = strlen(query);
 536                         if (len == 0)
 537                                 break;
 538                         *ptr = len;
 539                         memcpy(ptr + 1, query, len);
 540                         ptr += len + 1;
 541                         break;
 542                 }
 543
 544                 *ptr = tmp - query;
 545                 memcpy(ptr + 1, query, tmp - query);
 546                 ptr += tmp - query + 1;
 547
 548                 query = tmp + 1;
 549         }
 550
 551         while (domain != NULL) {
 552                 const char *tmp;
 553
 554                 tmp = strchr(domain, '.');
 555                 if (tmp == NULL) {
 556                         len = strlen(domain);
 557                         if (len == 0)
 558                                 break;
 559                         *ptr = len;
 560                         memcpy(ptr + 1, domain, len);
 561                         ptr += len + 1;
 562                         break;
 563                 }
 564
 565                 *ptr = tmp - domain;
 566                 memcpy(ptr + 1, domain, tmp - domain);
 567                 ptr += tmp - domain + 1;
 568
 569                 domain = tmp + 1;
 570         }
 571
 572         *ptr++ = 0x00;
 573
 574         return ptr - buf;
 575 }
 576
 577 static gboolean cache_check_is_valid(struct cache_data *data,
 578                                 time_t current_time)
 579 {
 580         if (data == NULL)
 581                 return FALSE;
 582
 583         if (data->cache_until < current_time)
 584                 return FALSE;
 585
 586         return TRUE;
 587 }
 588
 589 /*
 590  * remove stale cached entries so that they can be refreshed
 591  */
 592 static void cache_enforce_validity(struct cache_entry *entry)
 593 {
 594         time_t current_time = time(NULL);
 595
 596         if (cache_check_is_valid(entry->ipv4, current_time) == FALSE
 597                                                         && entry->ipv4) {
 598                 DBG("cache timeout \"%s\" type A", entry->key);
 599                 g_free(entry->ipv4->data);
 600                 g_free(entry->ipv4);
 601                 entry->ipv4 = NULL;
 602
 603         }
 604
 605         if (cache_check_is_valid(entry->ipv6, current_time) == FALSE
 606                                                         && entry->ipv6) {
 607                 DBG("cache timeout \"%s\" type AAAA", entry->key);
 608                 g_free(entry->ipv6->data);
 609                 g_free(entry->ipv6);
 610                 entry->ipv6 = NULL;
 611         }
 612 }
 613
 614 static uint16_t cache_check_validity(char *question, uint16_t type,
 615                                 struct cache_entry *entry)
 616 {
 617         time_t current_time = time(NULL);
 618         int want_refresh = 0;
 619
 620         /*
 621          * if we have a popular entry, we want a refresh instead of
 622          * total destruction of the entry.
 623          */
 624         if (entry->hits > 2)
 625                 want_refresh = 1;
 626
 627         cache_enforce_validity(entry);
 628
 629         switch (type) {
 630         case 1:         /* IPv4 */
 631                 if (cache_check_is_valid(entry->ipv4, current_time) == FALSE) {
 632                         DBG("cache %s \"%s\" type A", entry->ipv4 ?
 633                                         "timeout" : "entry missing", question);
 634
 635                         if (want_refresh)
 636                                 entry->want_refresh = 1;
 637
 638                         /*
 639                          * We do not remove cache entry if there is still
 640                          * valid IPv6 entry found in the cache.
 641                          */
 642                         if (cache_check_is_valid(entry->ipv6, current_time)
 643                                         == FALSE && want_refresh == FALSE) {
 644                                 g_hash_table_remove(cache, question);
 645                                 type = 0;
 646                         }
 647                 }
 648                 break;
 649
 650         case 28:        /* IPv6 */
 651                 if (cache_check_is_valid(entry->ipv6, current_time) == FALSE) {
 652                         DBG("cache %s \"%s\" type AAAA", entry->ipv6 ?
 653                                         "timeout" : "entry missing", question);
 654
 655                         if (want_refresh)
 656                                 entry->want_refresh = 1;
 657
 658                         if (cache_check_is_valid(entry->ipv4, current_time)
 659                                         == FALSE && want_refresh == FALSE) {
 660                                 g_hash_table_remove(cache, question);
 661                                 type = 0;
 662                         }
 663                 }
 664                 break;
 665         }
 666
 667         return type;
 668 }
 669
 670 static struct cache_entry *cache_check(gpointer request, int *qtype, int proto)
 671 {
 672         char *question;
 673         struct cache_entry *entry;
 674         struct domain_question *q;
 675         uint16_t type;
 676         int offset, proto_offset;
 677
 678         if (request == NULL)
 679                 return NULL;
 680
 681         proto_offset = protocol_offset(proto);
 682         if (proto_offset < 0)
 683                 return NULL;
 684
 685         question = request + proto_offset + 12;
 686
 687         offset = strlen(question) + 1;
 688         q = (void *) (question + offset);
 689         type = ntohs(q->type);
 690
 691         /* We only cache either A (1) or AAAA (28) requests */
 692         if (type != 1 && type != 28)
 693                 return NULL;
 694
 695         entry = g_hash_table_lookup(cache, question);
 696         if (entry == NULL)
 697                 return NULL;
 698
 699         type = cache_check_validity(question, type, entry);
 700         if (type == 0)
 701                 return NULL;
 702
 703         *qtype = type;
 704         return entry;
 705 }
 706
 707 /*
 708  * Get a label/name from DNS resource record. The function decompresses the
 709  * label if necessary. The function does not convert the name to presentation
 710  * form. This means that the result string will contain label lengths instead
 711  * of dots between labels. We intentionally do not want to convert to dotted
 712  * format so that we can cache the wire format string directly.
 713  */
 714 static int get_name(int counter,
 715                 unsigned char *pkt, unsigned char *start, unsigned char *max,
 716                 unsigned char *output, int output_max, int *output_len,
 717                 unsigned char **end, char *name, int *name_len)
 718 {
 719         unsigned char *p;
 720
 721         /* Limit recursion to 10 (this means up to 10 labels in domain name) */
 722         if (counter > 10)
 723                 return -EINVAL;
 724
 725         p = start;
 726         while (*p) {
 727                 if ((*p & NS_CMPRSFLGS) == NS_CMPRSFLGS) {
 728                         uint16_t offset = (*p & 0x3F) * 256 + *(p + 1);
 729
 730                         if (offset >= max - pkt)
 731                                 return -ENOBUFS;
 732
 733                         if (*end == NULL)
 734                                 *end = p + 2;
 735
 736                         return get_name(counter + 1, pkt, pkt + offset, max,
 737                                         output, output_max, output_len, end,
 738                                         name, name_len);
 739                 } else {
 740                         unsigned label_len = *p;
 741
 742                         if (pkt + label_len > max)
 743                                 return -ENOBUFS;
 744
 745                         if (*output_len > output_max)
 746                                 return -ENOBUFS;
 747
 748                         /*
 749                          * We need the original name in order to check
 750                          * if this answer is the correct one.
 751                          */
 752                         name[(*name_len)++] = label_len;
 753                         memcpy(name + *name_len, p + 1, label_len + 1);
 754                         *name_len += label_len;
 755
 756                         /* We compress the result */
 757                         output[0] = NS_CMPRSFLGS;
 758                         output[1] = 0x0C;
 759                         *output_len = 2;
 760
 761                         p += label_len + 1;
 762
 763                         if (*end == NULL)
 764                                 *end = p;
 765
 766                         if (p >= max)
 767                                 return -ENOBUFS;
 768                 }
 769         }
 770
 771         return 0;
 772 }
 773
 774 static int parse_rr(unsigned char *buf, unsigned char *start,
 775                         unsigned char *max,
 776                         unsigned char *response, unsigned int *response_size,
 777                         uint16_t *type, uint16_t *class, int *ttl, int *rdlen,
 778                         unsigned char **end,
 779                         char *name)
 780 {
 781         struct domain_rr *rr;
 782         int err, offset;
 783         int name_len = 0, output_len = 0, max_rsp = *response_size;
 784
 785         err = get_name(0, buf, start, max, response, max_rsp,
 786                 &output_len, end, name, &name_len);
 787         if (err < 0)
 788                 return err;
 789
 790         offset = output_len;
 791
 792         if ((unsigned int) offset > *response_size)
 793                 return -ENOBUFS;
 794
 795         rr = (void *) (*end);
 796
 797         if (rr == NULL)
 798                 return -EINVAL;
 799
 800         *type = ntohs(rr->type);
 801         *class = ntohs(rr->class);
 802         *ttl = ntohl(rr->ttl);
 803         *rdlen = ntohs(rr->rdlen);
 804
 805         if (*ttl < 0)
 806                 return -EINVAL;
 807
 808         memcpy(response + offset, *end, sizeof(struct domain_rr));
 809
 810         offset += sizeof(struct domain_rr);
 811         *end += sizeof(struct domain_rr);
 812
 813         if ((unsigned int) (offset + *rdlen) > *response_size)
 814                 return -ENOBUFS;
 815
 816         memcpy(response + offset, *end, *rdlen);
 817
 818         *end += *rdlen;
 819
 820         *response_size = offset + *rdlen;
 821
 822         return 0;
 823 }
 824
 825 static gboolean check_alias(GSList *aliases, char *name)
 826 {
 827         GSList *list;
 828
 829         if (aliases != NULL) {
 830                 for (list = aliases; list; list = list->next) {
 831                         int len = strlen((char *)list->data);
 832                         if (strncmp((char *)list->data, name, len) == 0)
 833                                 return TRUE;
 834                 }
 835         }
 836
 837         return FALSE;
 838 }
 839
 840 static int parse_response(unsigned char *buf, int buflen,
 841                         char *question, int qlen,
 842                         uint16_t *type, uint16_t *class, int *ttl,
 843                         unsigned char *response, unsigned int *response_len,
 844                         uint16_t *answers)
 845 {
 846         struct domain_hdr *hdr = (void *) buf;
 847         struct domain_question *q;
 848         unsigned char *ptr;
 849         uint16_t qdcount = ntohs(hdr->qdcount);
 850         uint16_t ancount = ntohs(hdr->ancount);
 851         int err, i;
 852         uint16_t qtype, qclass;
 853         unsigned char *next = NULL;
 854         unsigned int maxlen = *response_len;
 855         GSList *aliases = NULL, *list;
 856         char name[NS_MAXDNAME + 1];
 857
 858         if (buflen < 12)
 859                 return -EINVAL;
 860
 861         DBG("qr %d qdcount %d", hdr->qr, qdcount);
 862
 863         /* We currently only cache responses where question count is 1 */
 864         if (hdr->qr != 1 || qdcount != 1)
 865                 return -EINVAL;
 866
 867         ptr = buf + sizeof(struct domain_hdr);
 868
 869         strncpy(question, (char *) ptr, qlen);
 870         qlen = strlen(question);
 871         ptr += qlen + 1; /* skip \0 */
 872
 873         q = (void *) ptr;
 874         qtype = ntohs(q->type);
 875
 876         /* We cache only A and AAAA records */
 877         if (qtype != 1 && qtype != 28)
 878                 return -ENOMSG;
 879
 880         qclass = ntohs(q->class);
 881
 882         ptr += 2 + 2; /* ptr points now to answers */
 883
 884         err = -ENOMSG;
 885         *response_len = 0;
 886         *answers = 0;
 887
 888         /*
 889          * We have a bunch of answers (like A, AAAA, CNAME etc) to
 890          * A or AAAA question. We traverse the answers and parse the
 891          * resource records. Only A and AAAA records are cached, all
 892          * the other records in answers are skipped.
 893          */
 894         for (i = 0; i < ancount; i++) {
 895                 /*
 896                  * Get one address at a time to this buffer.
 897                  * The max size of the answer is
 898                  *   2 (pointer) + 2 (type) + 2 (class) +
 899                  *   4 (ttl) + 2 (rdlen) + addr (16 or 4) = 28
 900                  * for A or AAAA record.
 901                  * For CNAME the size can be bigger.
 902                  */
 903                 unsigned char rsp[NS_MAXCDNAME];
 904                 unsigned int rsp_len = sizeof(rsp) - 1;
 905                 int ret, rdlen;
 906
 907                 memset(rsp, 0, sizeof(rsp));
 908
 909                 ret = parse_rr(buf, ptr, buf + buflen, rsp, &rsp_len,
 910                         type, class, ttl, &rdlen, &next, name);
 911                 if (ret != 0) {
 912                         err = ret;
 913                         goto out;
 914                 }
 915
 916                 /*
 917                  * Now rsp contains compressed or uncompressed resource
 918                  * record. Next we check if this record answers the question.
 919                  * The name var contains the uncompressed label.
 920                  * One tricky bit is the CNAME records as they alias
 921                  * the name we might be interested in.
 922                  */
 923
 924                 /*
 925                  * Go to next answer if the class is not the one we are
 926                  * looking for.
 927                  */
 928                 if (*class != qclass) {
 929                         ptr = next;
 930                         next = NULL;
 931                         continue;
 932                 }
 933
 934                 /*
 935                  * Try to resolve aliases also, type is CNAME(5).
 936                  * This is important as otherwise the aliased names would not
 937                  * be cached at all as the cache would not contain the aliased
 938                  * question.
 939                  *
 940                  * If any CNAME is found in DNS packet, then we cache the alias
 941                  * IP address instead of the question (as the server
 942                  * said that question has only an alias).
 943                  * This means in practice that if e.g., ipv6.google.com is
 944                  * queried, DNS server returns CNAME of that name which is
 945                  * ipv6.l.google.com. We then cache the address of the CNAME
 946                  * but return the question name to client. So the alias
 947                  * status of the name is not saved in cache and thus not
 948                  * returned to the client. We do not return DNS packets from
 949                  * cache to client saying that ipv6.google.com is an alias to
 950                  * ipv6.l.google.com but we return instead a DNS packet that
 951                  * says ipv6.google.com has address xxx which is in fact the
 952                  * address of ipv6.l.google.com. For caching purposes this
 953                  * should not cause any issues.
 954                  */
 955                 if (*type == 5 && strncmp(question, name, qlen) == 0) {
 956                         /*
 957                          * So now the alias answered the question. This is
 958                          * not very useful from caching point of view as
 959                          * the following A or AAAA records will not match the
 960                          * question. We need to find the real A/AAAA record
 961                          * of the alias and cache that.
 962                          */
 963                         unsigned char *end = NULL;
 964                         int name_len = 0, output_len;
 965
 966                         memset(rsp, 0, sizeof(rsp));
 967                         rsp_len = sizeof(rsp) - 1;
 968
 969                         /*
 970                          * Alias is in rdata part of the message,
 971                          * and next-rdlen points to it. So we need to get
 972                          * the real name of the alias.
 973                          */
 974                         ret = get_name(0, buf, next - rdlen, buf + buflen,
 975                                         rsp, rsp_len, &output_len, &end,
 976                                         name, &name_len);
 977                         if (ret != 0) {
 978                                 /* just ignore the error at this point */
 979                                 ptr = next;
 980                                 next = NULL;
 981                                 continue;
 982                         }
 983
 984                         /*
 985                          * We should now have the alias of the entry we might
 986                          * want to cache. Just remember it for a while.
 987                          * We check the alias list when we have parsed the
 988                          * A or AAAA record.
 989                          */
 990                         aliases = g_slist_prepend(aliases, g_strdup(name));
 991
 992                         ptr = next;
 993                         next = NULL;
 994                         continue;
 995                 }
 996
 997                 if (*type == qtype) {
 998                         /*
 999                          * We found correct type (A or AAAA)
1000                          */
1001                         if (check_alias(aliases, name) == TRUE ||
1002                                 (aliases == NULL && strncmp(question, name,
1003                                                         qlen) == 0)) {
1004                                 /*
1005                                  * We found an alias or the name of the rr
1006                                  * matches the question. If so, we append
1007                                  * the compressed label to the cache.
1008                                  * The end result is a response buffer that
1009                                  * will contain one or more cached and
1010                                  * compressed resource records.
1011                                  */
1012                                 if (*response_len + rsp_len > maxlen) {
1013                                         err = -ENOBUFS;
1014                                         goto out;
1015                                 }
1016                                 memcpy(response + *response_len, rsp, rsp_len);
1017                                 *response_len += rsp_len;
1018                                 (*answers)++;
1019                                 err = 0;
1020                         }
1021                 }
1022
1023                 ptr = next;
1024                 next = NULL;
1025         }
1026
1027 out:
1028         for (list = aliases; list; list = list->next)
1029                 g_free(list->data);
1030         g_slist_free(aliases);
1031
1032         return err;
1033 }
1034
1035 struct cache_timeout {
1036         time_t current_time;
1037         int max_timeout;
1038         int try_harder;
1039 };
1040
1041 static gboolean cache_check_entry(gpointer key, gpointer value,
1042                                         gpointer user_data)
1043 {
1044         struct cache_timeout *data = user_data;
1045         struct cache_entry *entry = value;
1046         int max_timeout;
1047
1048         /* Scale the number of hits by half as part of cache aging */
1049
1050         entry->hits /= 2;
1051
1052         /*
1053          * If either IPv4 or IPv6 cached entry has expired, we
1054          * remove both from the cache.
1055          */
1056
1057         if (entry->ipv4 != NULL && entry->ipv4->timeout > 0) {
1058                 max_timeout = entry->ipv4->cache_until;
1059                 if (max_timeout > data->max_timeout)
1060                         data->max_timeout = max_timeout;
1061
1062                 if (entry->ipv4->cache_until < data->current_time)
1063                         return TRUE;
1064         }
1065
1066         if (entry->ipv6 != NULL && entry->ipv6->timeout > 0) {
1067                 max_timeout = entry->ipv6->cache_until;
1068                 if (max_timeout > data->max_timeout)
1069                         data->max_timeout = max_timeout;
1070
1071                 if (entry->ipv6->cache_until < data->current_time)
1072                         return TRUE;
1073         }
1074
1075         /*
1076          * if we're asked to try harder, also remove entries that have
1077          * few hits
1078          */
1079         if (data->try_harder && entry->hits < 4)
1080                 return TRUE;
1081
1082         return FALSE;
1083 }
1084
1085 static void cache_cleanup(void)
1086 {
1087         static int max_timeout;
1088         struct cache_timeout data;
1089         int count = 0;
1090
1091         data.current_time = time(NULL);
1092         data.max_timeout = 0;
1093         data.try_harder = 0;
1094
1095         /*
1096          * In the first pass, we only remove entries that have timed out.
1097          * We use a cache of the first time to expire to do this only
1098          * when it makes sense.
1099          */
1100         if (max_timeout <= data.current_time) {
1101                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1102                                                 &data);
1103         }
1104         DBG("removed %d in the first pass", count);
1105
1106         /*
1107          * In the second pass, if the first pass turned up blank,
1108          * we also expire entries with a low hit count,
1109          * while aging the hit count at the same time.
1110          */
1111         data.try_harder = 1;
1112         if (count == 0)
1113                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1114                                                 &data);
1115
1116         if (count == 0)
1117                 /*
1118                  * If we could not remove anything, then remember
1119                  * what is the max timeout and do nothing if we
1120                  * have not yet reached it. This will prevent
1121                  * constant traversal of the cache if it is full.
1122                  */
1123                 max_timeout = data.max_timeout;
1124         else
1125                 max_timeout = 0;
1126 }
1127
1128 static gboolean cache_invalidate_entry(gpointer key, gpointer value,
1129                                         gpointer user_data)
1130 {
1131         struct cache_entry *entry = value;
1132
1133         /* first, delete any expired elements */
1134         cache_enforce_validity(entry);
1135
1136         /* if anything is not expired, mark the entry for refresh */
1137         if (entry->hits > 0 && (entry->ipv4 || entry->ipv6))
1138                 entry->want_refresh = 1;
1139
1140         /* delete the cached data */
1141         if (entry->ipv4) {
1142                 g_free(entry->ipv4->data);
1143                 g_free(entry->ipv4);
1144                 entry->ipv4 = NULL;
1145         }
1146
1147         if (entry->ipv6) {
1148                 g_free(entry->ipv6->data);
1149                 g_free(entry->ipv6);
1150                 entry->ipv6 = NULL;
1151         }
1152
1153         /* keep the entry if we want it refreshed, delete it otherwise */
1154         if (entry->want_refresh)
1155                 return FALSE;
1156         else
1157                 return TRUE;
1158 }
1159
1160 /*
1161  * cache_invalidate is called from places where the DNS landscape
1162  * has changed, say because connections are added or we entered a VPN.
1163  * The logic is to wipe all cache data, but mark all non-expired
1164  * parts of the cache for refresh rather than deleting the whole cache.
1165  */
1166 static void cache_invalidate(void)
1167 {
1168         DBG("Invalidating the DNS cache %p", cache);
1169
1170         if (cache == NULL)
1171                 return;
1172
1173         g_hash_table_foreach_remove(cache, cache_invalidate_entry, NULL);
1174 }
1175
1176 static void cache_refresh_entry(struct cache_entry *entry)
1177 {
1178
1179         cache_enforce_validity(entry);
1180
1181         if (entry->hits > 2 && entry->ipv4 == NULL)
1182                 entry->want_refresh = 1;
1183         if (entry->hits > 2 && entry->ipv6 == NULL)
1184                 entry->want_refresh = 1;
1185
1186         if (entry->want_refresh) {
1187                 char *c;
1188                 char dns_name[NS_MAXDNAME + 1];
1189                 entry->want_refresh = 0;
1190
1191                 /* turn a DNS name into a hostname with dots */
1192                 strncpy(dns_name, entry->key, NS_MAXDNAME);
1193                 c = dns_name;
1194                 while (c && *c) {
1195                         int jump;
1196                         jump = *c;
1197                         *c = '.';
1198                         c += jump + 1;
1199                 }
1200                 DBG("Refreshing %s\n", dns_name);
1201                 /* then refresh the hostname */
1202                 refresh_dns_entry(entry, &dns_name[1]);
1203         }
1204 }
1205
1206 static void cache_refresh_iterator(gpointer key, gpointer value,
1207                                         gpointer user_data)
1208 {
1209         struct cache_entry *entry = value;
1210
1211         cache_refresh_entry(entry);
1212 }
1213
1214 static void cache_refresh(void)
1215 {
1216         if (cache == NULL)
1217                 return;
1218
1219         g_hash_table_foreach(cache, cache_refresh_iterator, NULL);
1220 }
1221
1222 static int reply_query_type(unsigned char *msg, int len)
1223 {
1224         unsigned char *c;
1225         uint16_t *w;
1226         int l;
1227         int type;
1228
1229         /* skip the header */
1230         c = msg + sizeof(struct domain_hdr);
1231         len -= sizeof(struct domain_hdr);
1232
1233         if (len < 0)
1234                 return 0;
1235
1236         /* now the query, which is a name and 2 16 bit words */
1237         l = dns_name_length(c) + 1;
1238         c += l;
1239         w = (uint16_t *) c;
1240         type = ntohs(*w);
1241
1242         return type;
1243 }
1244
1245 static int cache_update(struct server_data *srv, unsigned char *msg,
1246                         unsigned int msg_len)
1247 {
1248         int offset = protocol_offset(srv->protocol);
1249         int err, qlen, ttl = 0;
1250         uint16_t answers = 0, type = 0, class = 0;
1251         struct domain_hdr *hdr = (void *)(msg + offset);
1252         struct domain_question *q;
1253         struct cache_entry *entry;
1254         struct cache_data *data;
1255         char question[NS_MAXDNAME + 1];
1256         unsigned char response[NS_MAXDNAME + 1];
1257         unsigned char *ptr;
1258         unsigned int rsplen;
1259         gboolean new_entry = TRUE;
1260         time_t current_time;
1261
1262         if (cache_size >= MAX_CACHE_SIZE) {
1263                 cache_cleanup();
1264                 if (cache_size >= MAX_CACHE_SIZE)
1265                         return 0;
1266         }
1267
1268         current_time = time(NULL);
1269
1270         /* don't do a cache refresh more than twice a minute */
1271         if (next_refresh < current_time) {
1272                 cache_refresh();
1273                 next_refresh = current_time + 30;
1274         }
1275
1276         if (offset < 0)
1277                 return 0;
1278
1279         DBG("offset %d hdr %p msg %p rcode %d", offset, hdr, msg, hdr->rcode);
1280
1281         /* Continue only if response code is 0 (=ok) */
1282         if (hdr->rcode != 0)
1283                 return 0;
1284
1285         rsplen = sizeof(response) - 1;
1286         question[sizeof(question) - 1] = '\0';
1287
1288         err = parse_response(msg + offset, msg_len - offset,
1289                                 question, sizeof(question) - 1,
1290                                 &type, &class, &ttl,
1291                                 response, &rsplen, &answers);
1292
1293         /*
1294          * special case: if we do a ipv6 lookup and get no result
1295          * for a record that's already in our ipv4 cache.. we want
1296          * to cache the negative response.
1297          */
1298         if ((err == -ENOMSG || err == -ENOBUFS) &&
1299                         reply_query_type(msg + offset,
1300                                         msg_len - offset) == 28) {
1301                 entry = g_hash_table_lookup(cache, question);
1302                 if (entry && entry->ipv4 && entry->ipv6 == NULL) {
1303                         int cache_offset = 0;
1304
1305                         data = g_try_new(struct cache_data, 1);
1306                         if (data == NULL)
1307                                 return -ENOMEM;
1308                         data->inserted = entry->ipv4->inserted;
1309                         data->type = type;
1310                         data->answers = ntohs(hdr->ancount);
1311                         data->timeout = entry->ipv4->timeout;
1312                         if (srv->protocol == IPPROTO_UDP)
1313                                 cache_offset = 2;
1314                         data->data_len = msg_len + cache_offset;
1315                         data->data = ptr = g_malloc(data->data_len);
1316                         ptr[0] = (data->data_len - 2) / 256;
1317                         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1318                         if (srv->protocol == IPPROTO_UDP)
1319                                 ptr += 2;
1320                         data->valid_until = entry->ipv4->valid_until;
1321                         data->cache_until = entry->ipv4->cache_until;
1322                         memcpy(ptr, msg, msg_len);
1323                         entry->ipv6 = data;
1324                         /*
1325                          * we will get a "hit" when we serve the response
1326                          * out of the cache
1327                          */
1328                         entry->hits--;
1329                         if (entry->hits < 0)
1330                                 entry->hits = 0;
1331                         return 0;
1332                 }
1333         }
1334
1335         if (err < 0 || ttl == 0)
1336                 return 0;
1337
1338         qlen = strlen(question);
1339
1340         /*
1341          * If the cache contains already data, check if the
1342          * type of the cached data is the same and do not add
1343          * to cache if data is already there.
1344          * This is needed so that we can cache both A and AAAA
1345          * records for the same name.
1346          */
1347         entry = g_hash_table_lookup(cache, question);
1348         if (entry == NULL) {
1349                 entry = g_try_new(struct cache_entry, 1);
1350                 if (entry == NULL)
1351                         return -ENOMEM;
1352
1353                 data = g_try_new(struct cache_data, 1);
1354                 if (data == NULL) {
1355                         g_free(entry);
1356                         return -ENOMEM;
1357                 }
1358
1359                 entry->key = g_strdup(question);
1360                 entry->ipv4 = entry->ipv6 = NULL;
1361                 entry->want_refresh = 0;
1362                 entry->hits = 0;
1363
1364                 if (type == 1)
1365                         entry->ipv4 = data;
1366                 else
1367                         entry->ipv6 = data;
1368         } else {
1369                 if (type == 1 && entry->ipv4 != NULL)
1370                         return 0;
1371
1372                 if (type == 28 && entry->ipv6 != NULL)
1373                         return 0;
1374
1375                 data = g_try_new(struct cache_data, 1);
1376                 if (data == NULL)
1377                         return -ENOMEM;
1378
1379                 if (type == 1)
1380                         entry->ipv4 = data;
1381                 else
1382                         entry->ipv6 = data;
1383
1384                 /*
1385                  * compensate for the hit we'll get for serving
1386                  * the response out of the cache
1387                  */
1388                 entry->hits--;
1389                 if (entry->hits < 0)
1390                         entry->hits = 0;
1391
1392                 new_entry = FALSE;
1393         }
1394
1395         if (ttl < MIN_CACHE_TTL)
1396                 ttl = MIN_CACHE_TTL;
1397
1398         data->inserted = current_time;
1399         data->type = type;
1400         data->answers = answers;
1401         data->timeout = ttl;
1402         /*
1403          * The "2" in start of the length is the TCP offset. We allocate it
1404          * here even for UDP packet because it simplifies the sending
1405          * of cached packet.
1406          */
1407         data->data_len = 2 + 12 + qlen + 1 + 2 + 2 + rsplen;
1408         data->data = ptr = g_malloc(data->data_len);
1409         data->valid_until = current_time + ttl;
1410
1411         /*
1412          * Restrict the cached DNS record TTL to some sane value
1413          * in order to prevent data staying in the cache too long.
1414          */
1415         if (ttl > MAX_CACHE_TTL)
1416                 ttl = MAX_CACHE_TTL;
1417
1418         data->cache_until = round_down_ttl(current_time + ttl, ttl);
1419
1420         if (data->data == NULL) {
1421                 g_free(entry->key);
1422                 g_free(data);
1423                 g_free(entry);
1424                 return -ENOMEM;
1425         }
1426
1427         /*
1428          * We cache the two extra bytes at the start of the message
1429          * in a TCP packet. When sending UDP packet, we skip the first
1430          * two bytes. This way we do not need to know the format
1431          * (UDP/TCP) of the cached message.
1432          */
1433         ptr[0] = (data->data_len - 2) / 256;
1434         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1435         if (srv->protocol == IPPROTO_UDP)
1436                 ptr += 2;
1437
1438         memcpy(ptr, msg, offset + 12);
1439         memcpy(ptr + offset + 12, question, qlen + 1); /* copy also the \0 */
1440
1441         q = (void *) (ptr + offset + 12 + qlen + 1);
1442         q->type = htons(type);
1443         q->class = htons(class);
1444         memcpy(ptr + offset + 12 + qlen + 1 + sizeof(struct domain_question),
1445                 response, rsplen);
1446
1447         if (new_entry == TRUE) {
1448                 g_hash_table_replace(cache, entry->key, entry);
1449                 cache_size++;
1450         }
1451
1452         DBG("cache %d %squestion \"%s\" type %d ttl %d size %zd packet %u "
1453                                                                 "dns len %u",
1454                 cache_size, new_entry ? "new " : "old ",
1455                 question, type, ttl,
1456                 sizeof(*entry) + sizeof(*data) + data->data_len + qlen,
1457                 data->data_len,
1458                 srv->protocol == IPPROTO_TCP ?
1459                         (unsigned int)(data->data[0] * 256 + data->data[1]) :
1460                         data->data_len);
1461
1462         return 0;
1463 }
1464
1465 static int ns_resolv(struct server_data *server, struct request_data *req,
1466                                 gpointer request, gpointer name)
1467 {
1468         GList *list;
1469         int sk, err, type = 0;
1470         char *dot, *lookup = (char *) name;
1471         struct cache_entry *entry;
1472
1473         entry = cache_check(request, &type, req->protocol);
1474         if (entry != NULL) {
1475                 int ttl_left = 0;
1476                 struct cache_data *data;
1477
1478                 DBG("cache hit %s type %s", lookup, type == 1 ? "A" : "AAAA");
1479                 if (type == 1)
1480                         data = entry->ipv4;
1481                 else
1482                         data = entry->ipv6;
1483
1484                 if (data) {
1485                         ttl_left = data->valid_until - time(NULL);
1486                         entry->hits++;
1487                 }
1488
1489                 if (data != NULL && req->protocol == IPPROTO_TCP) {
1490                         send_cached_response(req->client_sk, data->data,
1491                                         data->data_len, NULL, 0, IPPROTO_TCP,
1492                                         req->srcid, data->answers, ttl_left);
1493                         return 1;
1494                 }
1495
1496                 if (data != NULL && req->protocol == IPPROTO_UDP) {
1497                         int udp_sk = g_io_channel_unix_get_fd(
1498                                         req->ifdata->udp_listener_channel);
1499
1500                         send_cached_response(udp_sk, data->data,
1501                                 data->data_len, &req->sa, req->sa_len,
1502                                 IPPROTO_UDP, req->srcid, data->answers,
1503                                 ttl_left);
1504                         return 1;
1505                 }
1506         }
1507
1508         sk = g_io_channel_unix_get_fd(server->channel);
1509
1510         err = sendto(sk, request, req->request_len, MSG_NOSIGNAL,
1511                         server->server_addr, server->server_addr_len);
1512         if (err < 0) {
1513                 DBG("Cannot send message to server %s sock %d "
1514                         "protocol %d (%s/%d)",
1515                         server->server, sk, server->protocol,
1516                         strerror(errno), errno);
1517                 return -EIO;
1518         }
1519
1520         req->numserv++;
1521
1522         /* If we have more than one dot, we don't add domains */
1523         dot = strchr(lookup, '.');
1524         if (dot != NULL && dot != lookup + strlen(lookup) - 1)
1525                 return 0;
1526
1527         if (server->domains != NULL && server->domains->data != NULL)
1528                 req->append_domain = TRUE;
1529
1530         for (list = server->domains; list; list = list->next) {
1531                 char *domain;
1532                 unsigned char alt[1024];
1533                 struct domain_hdr *hdr = (void *) &alt;
1534                 int altlen, domlen, offset;
1535
1536                 domain = list->data;
1537
1538                 if (domain == NULL)
1539                         continue;
1540
1541                 offset = protocol_offset(server->protocol);
1542                 if (offset < 0)
1543                         return offset;
1544
1545                 domlen = strlen(domain) + 1;
1546                 if (domlen < 5)
1547                         return -EINVAL;
1548
1549                 alt[offset] = req->altid & 0xff;
1550                 alt[offset + 1] = req->altid >> 8;
1551
1552                 memcpy(alt + offset + 2, request + offset + 2, 10);
1553                 hdr->qdcount = htons(1);
1554
1555                 altlen = append_query(alt + offset + 12, sizeof(alt) - 12,
1556                                         name, domain);
1557                 if (altlen < 0)
1558                         return -EINVAL;
1559
1560                 altlen += 12;
1561
1562                 memcpy(alt + offset + altlen,
1563                         request + offset + altlen - domlen,
1564                                 req->request_len - altlen - offset + domlen);
1565
1566                 if (server->protocol == IPPROTO_TCP) {
1567                         int req_len = req->request_len + domlen - 2;
1568
1569                         alt[0] = (req_len >> 8) & 0xff;
1570                         alt[1] = req_len & 0xff;
1571                 }
1572
1573                 DBG("req %p dstid 0x%04x altid 0x%04x", req, req->dstid,
1574                                 req->altid);
1575
1576                 err = send(sk, alt, req->request_len + domlen, MSG_NOSIGNAL);
1577                 if (err < 0)
1578                         return -EIO;
1579
1580                 req->numserv++;
1581         }
1582
1583         return 0;
1584 }
1585
1586 static void destroy_request_data(struct request_data *req)
1587 {
1588         if (req->timeout > 0)
1589                 g_source_remove(req->timeout);
1590
1591         g_free(req->resp);
1592         g_free(req->request);
1593         g_free(req->name);
1594         g_free(req);
1595 }
1596
1597 static int forward_dns_reply(unsigned char *reply, int reply_len, int protocol,
1598                                 struct server_data *data)
1599 {
1600         struct domain_hdr *hdr;
1601         struct request_data *req;
1602         int dns_id, sk, err, offset = protocol_offset(protocol);
1603         struct listener_data *ifdata;
1604
1605         if (offset < 0)
1606                 return offset;
1607
1608         hdr = (void *)(reply + offset);
1609         dns_id = reply[offset] | reply[offset + 1] << 8;
1610
1611         DBG("Received %d bytes (id 0x%04x)", reply_len, dns_id);
1612
1613         req = find_request(dns_id);
1614         if (req == NULL)
1615                 return -EINVAL;
1616
1617         DBG("req %p dstid 0x%04x altid 0x%04x rcode %d",
1618                         req, req->dstid, req->altid, hdr->rcode);
1619
1620         ifdata = req->ifdata;
1621
1622         reply[offset] = req->srcid & 0xff;
1623         reply[offset + 1] = req->srcid >> 8;
1624
1625         req->numresp++;
1626
1627         if (hdr->rcode == 0 || req->resp == NULL) {
1628
1629                 /*
1630                  * If the domain name was append
1631                  * remove it before forwarding the reply.
1632                  */
1633                 if (req->append_domain == TRUE) {
1634                         unsigned int domain_len = 0;
1635                         unsigned char *ptr;
1636                         uint8_t host_len;
1637                         unsigned int header_len;
1638
1639                         /*
1640                          * ptr points to the first char of the hostname.
1641                          * ->hostname.domain.net
1642                          */
1643                         header_len = offset + sizeof(struct domain_hdr);
1644                         ptr = reply + header_len;
1645                         host_len = *ptr;
1646                         if (host_len > 0)
1647                                 domain_len = strnlen((const char *)ptr + 1 +
1648                                                 host_len,
1649                                                 reply_len - header_len);
1650
1651
1652                         DBG("host len %d domain len %d", host_len, domain_len);
1653
1654                         /*
1655                          * Remove the domain name and replace it by the end
1656                          * of reply. Check if the domain is really there
1657                          * before trying to copy the data. The domain_len can
1658                          * be 0 because if the original query did not contain
1659                          * a domain name, then we are sending two packets,
1660                          * first without the domain name and the second packet
1661                          * with domain name. The append_domain is set to true
1662                          * even if we sent the first packet without domain
1663                          * name. In this case we end up in this branch.
1664                          */
1665                         if (domain_len > 0) {
1666                                 /*
1667                                  * Note that we must use memmove() here,
1668                                  * because the memory areas can overlap.
1669                                  */
1670                                 memmove(ptr + host_len + 1,
1671                                         ptr + host_len + domain_len + 1,
1672                                         reply_len - header_len - domain_len);
1673
1674                                 reply_len = reply_len - domain_len;
1675                         }
1676                 }
1677
1678                 g_free(req->resp);
1679                 req->resplen = 0;
1680
1681                 req->resp = g_try_malloc(reply_len);
1682                 if (req->resp == NULL)
1683                         return -ENOMEM;
1684
1685                 memcpy(req->resp, reply, reply_len);
1686                 req->resplen = reply_len;
1687
1688                 cache_update(data, reply, reply_len);
1689         }
1690
1691         if (hdr->rcode > 0 && req->numresp < req->numserv)
1692                 return -EINVAL;
1693
1694         request_list = g_slist_remove(request_list, req);
1695
1696         if (protocol == IPPROTO_UDP) {
1697                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
1698                 err = sendto(sk, req->resp, req->resplen, 0,
1699                              &req->sa, req->sa_len);
1700         } else {
1701                 sk = req->client_sk;
1702                 err = send(sk, req->resp, req->resplen, MSG_NOSIGNAL);
1703                 close(sk);
1704         }
1705
1706         if (err < 0)
1707                 DBG("Cannot send msg, sk %d proto %d errno %d/%s", sk,
1708                         protocol, errno, strerror(errno));
1709         else
1710                 DBG("proto %d sent %d bytes to %d", protocol, err, sk);
1711
1712         destroy_request_data(req);
1713
1714         return err;
1715 }
1716
1717 static void cache_element_destroy(gpointer value)
1718 {
1719         struct cache_entry *entry = value;
1720
1721         if (entry == NULL)
1722                 return;
1723
1724         if (entry->ipv4 != NULL) {
1725                 g_free(entry->ipv4->data);
1726                 g_free(entry->ipv4);
1727         }
1728
1729         if (entry->ipv6 != NULL) {
1730                 g_free(entry->ipv6->data);
1731                 g_free(entry->ipv6);
1732         }
1733
1734         g_free(entry->key);
1735         g_free(entry);
1736
1737         if (--cache_size < 0)
1738                 cache_size = 0;
1739 }
1740
1741 static gboolean try_remove_cache(gpointer user_data)
1742 {
1743         if (__sync_fetch_and_sub(&cache_refcount, 1) == 1) {
1744                 DBG("No cache users, removing it.");
1745
1746                 g_hash_table_destroy(cache);
1747                 cache = NULL;
1748         }
1749
1750         return FALSE;
1751 }
1752
1753 static void server_destroy_socket(struct server_data *data)
1754 {
1755         DBG("index %d server %s proto %d", data->index,
1756                                         data->server, data->protocol);
1757
1758         if (data->watch > 0) {
1759                 g_source_remove(data->watch);
1760                 data->watch = 0;
1761         }
1762
1763         if (data->timeout > 0) {
1764                 g_source_remove(data->timeout);
1765                 data->timeout = 0;
1766         }
1767
1768         if (data->channel != NULL) {
1769                 g_io_channel_shutdown(data->channel, TRUE, NULL);
1770                 g_io_channel_unref(data->channel);
1771                 data->channel = NULL;
1772         }
1773
1774         g_free(data->incoming_reply);
1775         data->incoming_reply = NULL;
1776 }
1777
1778 static void destroy_server(struct server_data *server)
1779 {
1780         GList *list;
1781
1782         DBG("index %d server %s sock %d", server->index, server->server,
1783                         server->channel != NULL ?
1784                         g_io_channel_unix_get_fd(server->channel): -1);
1785
1786         server_list = g_slist_remove(server_list, server);
1787         server_destroy_socket(server);
1788
1789         if (server->protocol == IPPROTO_UDP && server->enabled)
1790                 DBG("Removing DNS server %s", server->server);
1791
1792         g_free(server->server);
1793         for (list = server->domains; list; list = list->next) {
1794                 char *domain = list->data;
1795
1796                 server->domains = g_list_remove(server->domains, domain);
1797                 g_free(domain);
1798         }
1799         g_free(server->server_addr);
1800
1801         /*
1802          * We do not remove cache right away but delay it few seconds.
1803          * The idea is that when IPv6 DNS server is added via RDNSS, it has a
1804          * lifetime. When the lifetime expires we decrease the refcount so it
1805          * is possible that the cache is then removed. Because a new DNS server
1806          * is usually created almost immediately we would then loose the cache
1807          * without any good reason. The small delay allows the new RDNSS to
1808          * create a new DNS server instance and the refcount does not go to 0.
1809          */
1810         g_timeout_add_seconds(3, try_remove_cache, NULL);
1811
1812         g_free(server);
1813 }
1814
1815 static gboolean udp_server_event(GIOChannel *channel, GIOCondition condition,
1816                                                         gpointer user_data)
1817 {
1818         unsigned char buf[4096];
1819         int sk, err, len;
1820         struct server_data *data = user_data;
1821
1822         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1823                 connman_error("Error with UDP server %s", data->server);
1824                 server_destroy_socket(data);
1825                 return FALSE;
1826         }
1827
1828         sk = g_io_channel_unix_get_fd(channel);
1829
1830         len = recv(sk, buf, sizeof(buf), 0);
1831         if (len < 12)
1832                 return TRUE;
1833
1834         err = forward_dns_reply(buf, len, IPPROTO_UDP, data);
1835         if (err < 0)
1836                 return TRUE;
1837
1838         return TRUE;
1839 }
1840
1841 static gboolean tcp_server_event(GIOChannel *channel, GIOCondition condition,
1842                                                         gpointer user_data)
1843 {
1844         int sk;
1845         struct server_data *server = user_data;
1846
1847         sk = g_io_channel_unix_get_fd(channel);
1848         if (sk == 0)
1849                 return FALSE;
1850
1851         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1852                 GSList *list;
1853 hangup:
1854                 DBG("TCP server channel closed, sk %d", sk);
1855
1856                 /*
1857                  * Discard any partial response which is buffered; better
1858                  * to get a proper response from a working server.
1859                  */
1860                 g_free(server->incoming_reply);
1861                 server->incoming_reply = NULL;
1862
1863                 for (list = request_list; list; list = list->next) {
1864                         struct request_data *req = list->data;
1865                         struct domain_hdr *hdr;
1866
1867                         if (req->protocol == IPPROTO_UDP)
1868                                 continue;
1869
1870                         if (req->request == NULL)
1871                                 continue;
1872
1873                         /*
1874                          * If we're not waiting for any further response
1875                          * from another name server, then we send an error
1876                          * response to the client.
1877                          */
1878                         if (req->numserv && --(req->numserv))
1879                                 continue;
1880
1881                         hdr = (void *) (req->request + 2);
1882                         hdr->id = req->srcid;
1883                         send_response(req->client_sk, req->request,
1884                                 req->request_len, NULL, 0, IPPROTO_TCP);
1885
1886                         request_list = g_slist_remove(request_list, req);
1887                 }
1888
1889                 destroy_server(server);
1890
1891                 return FALSE;
1892         }
1893
1894         if ((condition & G_IO_OUT) && !server->connected) {
1895                 GSList *list;
1896                 GList *domains;
1897                 int no_request_sent = TRUE;
1898                 struct server_data *udp_server;
1899
1900                 udp_server = find_server(server->index, server->server,
1901                                                                 IPPROTO_UDP);
1902                 if (udp_server != NULL) {
1903                         for (domains = udp_server->domains; domains;
1904                                                 domains = domains->next) {
1905                                 char *dom = domains->data;
1906
1907                                 DBG("Adding domain %s to %s",
1908                                                 dom, server->server);
1909
1910                                 server->domains = g_list_append(server->domains,
1911                                                                 g_strdup(dom));
1912                         }
1913                 }
1914
1915                 server->connected = TRUE;
1916                 server_list = g_slist_append(server_list, server);
1917
1918                 if (server->timeout > 0) {
1919                         g_source_remove(server->timeout);
1920                         server->timeout = 0;
1921                 }
1922
1923                 for (list = request_list; list; ) {
1924                         struct request_data *req = list->data;
1925                         int status;
1926
1927                         if (req->protocol == IPPROTO_UDP) {
1928                                 list = list->next;
1929                                 continue;
1930                         }
1931
1932                         DBG("Sending req %s over TCP", (char *)req->name);
1933
1934                         status = ns_resolv(server, req,
1935                                                 req->request, req->name);
1936                         if (status > 0) {
1937                                 /*
1938                                  * A cached result was sent,
1939                                  * so the request can be released
1940                                  */
1941                                 list = list->next;
1942                                 request_list = g_slist_remove(request_list, req);
1943                                 destroy_request_data(req);
1944                                 continue;
1945                         }
1946
1947                         if (status < 0) {
1948                                 list = list->next;
1949                                 continue;
1950                         }
1951
1952                         no_request_sent = FALSE;
1953
1954                         if (req->timeout > 0)
1955                                 g_source_remove(req->timeout);
1956
1957                         req->timeout = g_timeout_add_seconds(30,
1958                                                 request_timeout, req);
1959                         list = list->next;
1960                 }
1961
1962                 if (no_request_sent == TRUE) {
1963                         destroy_server(server);
1964                         return FALSE;
1965                 }
1966
1967         } else if (condition & G_IO_IN) {
1968                 struct partial_reply *reply = server->incoming_reply;
1969                 int bytes_recv;
1970
1971                 if (!reply) {
1972                         unsigned char reply_len_buf[2];
1973                         uint16_t reply_len;
1974
1975                         bytes_recv = recv(sk, reply_len_buf, 2, MSG_PEEK);
1976                         if (!bytes_recv) {
1977                                 goto hangup;
1978                         } else if (bytes_recv < 0) {
1979                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1980                                         return TRUE;
1981
1982                                 connman_error("DNS proxy error %s",
1983                                                 strerror(errno));
1984                                 goto hangup;
1985                         } else if (bytes_recv < 2)
1986                                 return TRUE;
1987
1988                         reply_len = reply_len_buf[1] | reply_len_buf[0] << 8;
1989                         reply_len += 2;
1990
1991                         DBG("TCP reply %d bytes from %d", reply_len, sk);
1992
1993                         reply = g_try_malloc(sizeof(*reply) + reply_len + 2);
1994                         if (!reply)
1995                                 return TRUE;
1996
1997                         reply->len = reply_len;
1998                         reply->received = 0;
1999
2000                         server->incoming_reply = reply;
2001                 }
2002
2003                 while (reply->received < reply->len) {
2004                         bytes_recv = recv(sk, reply->buf + reply->received,
2005                                         reply->len - reply->received, 0);
2006                         if (!bytes_recv) {
2007                                 connman_error("DNS proxy TCP disconnect");
2008                                 break;
2009                         } else if (bytes_recv < 0) {
2010                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
2011                                         return TRUE;
2012
2013                                 connman_error("DNS proxy error %s",
2014                                                 strerror(errno));
2015                                 break;
2016                         }
2017                         reply->received += bytes_recv;
2018                 }
2019
2020                 forward_dns_reply(reply->buf, reply->received, IPPROTO_TCP,
2021                                         server);
2022
2023                 g_free(reply);
2024                 server->incoming_reply = NULL;
2025
2026                 destroy_server(server);
2027
2028                 return FALSE;
2029         }
2030
2031         return TRUE;
2032 }
2033
2034 static gboolean tcp_idle_timeout(gpointer user_data)
2035 {
2036         struct server_data *server = user_data;
2037
2038         DBG("");
2039
2040         if (server == NULL)
2041                 return FALSE;
2042
2043         destroy_server(server);
2044
2045         return FALSE;
2046 }
2047
2048 static int server_create_socket(struct server_data *data)
2049 {
2050         int sk, err;
2051         char *interface;
2052
2053         DBG("index %d server %s proto %d", data->index,
2054                                         data->server, data->protocol);
2055
2056         sk = socket(data->server_addr->sa_family,
2057                 data->protocol == IPPROTO_TCP ? SOCK_STREAM : SOCK_DGRAM,
2058                 data->protocol);
2059         if (sk < 0) {
2060                 err = errno;
2061                 connman_error("Failed to create server %s socket",
2062                                                         data->server);
2063                 server_destroy_socket(data);
2064                 return -err;
2065         }
2066
2067         DBG("sk %d", sk);
2068
2069         interface = connman_inet_ifname(data->index);
2070         if (interface != NULL) {
2071                 if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2072                                         interface,
2073                                         strlen(interface) + 1) < 0) {
2074                         err = errno;
2075                         connman_error("Failed to bind server %s "
2076                                                 "to interface %s",
2077                                                 data->server, interface);
2078                         close(sk);
2079                         server_destroy_socket(data);
2080                         g_free(interface);
2081                         return -err;
2082                 }
2083                 g_free(interface);
2084         }
2085
2086         data->channel = g_io_channel_unix_new(sk);
2087         if (data->channel == NULL) {
2088                 connman_error("Failed to create server %s channel",
2089                                                         data->server);
2090                 close(sk);
2091                 server_destroy_socket(data);
2092                 return -ENOMEM;
2093         }
2094
2095         g_io_channel_set_close_on_unref(data->channel, TRUE);
2096
2097         if (data->protocol == IPPROTO_TCP) {
2098                 g_io_channel_set_flags(data->channel, G_IO_FLAG_NONBLOCK, NULL);
2099                 data->watch = g_io_add_watch(data->channel,
2100                         G_IO_OUT | G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR,
2101                                                 tcp_server_event, data);
2102                 data->timeout = g_timeout_add_seconds(30, tcp_idle_timeout,
2103                                                                 data);
2104         } else
2105                 data->watch = g_io_add_watch(data->channel,
2106                         G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2107                                                 udp_server_event, data);
2108
2109         if (connect(sk, data->server_addr, data->server_addr_len) < 0) {
2110                 err = errno;
2111
2112                 if ((data->protocol == IPPROTO_TCP && errno != EINPROGRESS) ||
2113                                 data->protocol == IPPROTO_UDP) {
2114
2115                         connman_error("Failed to connect to server %s",
2116                                                                 data->server);
2117                         server_destroy_socket(data);
2118                         return -err;
2119                 }
2120         }
2121
2122         if (__sync_fetch_and_add(&cache_refcount, 1) == 0)
2123                 cache = g_hash_table_new_full(g_str_hash,
2124                                         g_str_equal,
2125                                         NULL,
2126                                         cache_element_destroy);
2127
2128         return 0;
2129 }
2130
2131 static struct server_data *create_server(int index,
2132                                         const char *domain, const char *server,
2133                                         int protocol)
2134 {
2135         struct server_data *data;
2136         struct addrinfo hints, *rp;
2137         int ret;
2138
2139         DBG("index %d server %s", index, server);
2140
2141         data = g_try_new0(struct server_data, 1);
2142         if (data == NULL) {
2143                 connman_error("Failed to allocate server %s data", server);
2144                 return NULL;
2145         }
2146
2147         data->index = index;
2148         if (domain)
2149                 data->domains = g_list_append(data->domains, g_strdup(domain));
2150         data->server = g_strdup(server);
2151         data->protocol = protocol;
2152
2153         memset(&hints, 0, sizeof(hints));
2154
2155         switch (protocol) {
2156         case IPPROTO_UDP:
2157                 hints.ai_socktype = SOCK_DGRAM;
2158                 break;
2159
2160         case IPPROTO_TCP:
2161                 hints.ai_socktype = SOCK_STREAM;
2162                 break;
2163
2164         default:
2165                 destroy_server(data);
2166                 return NULL;
2167         }
2168         hints.ai_family = AF_UNSPEC;
2169         hints.ai_flags = AI_NUMERICSERV | AI_NUMERICHOST;
2170
2171         ret = getaddrinfo(data->server, "53", &hints, &rp);
2172         if (ret) {
2173                 connman_error("Failed to parse server %s address: %s\n",
2174                               data->server, gai_strerror(ret));
2175                 destroy_server(data);
2176                 return NULL;
2177         }
2178
2179         /* Do not blindly copy this code elsewhere; it doesn't loop over the
2180            results using ->ai_next as it should. That's OK in *this* case
2181            because it was a numeric lookup; we *know* there's only one. */
2182
2183         data->server_addr_len = rp->ai_addrlen;
2184
2185         switch (rp->ai_family) {
2186         case AF_INET:
2187                 data->server_addr = (struct sockaddr *)
2188                                         g_try_new0(struct sockaddr_in, 1);
2189                 break;
2190         case AF_INET6:
2191                 data->server_addr = (struct sockaddr *)
2192                                         g_try_new0(struct sockaddr_in6, 1);
2193                 break;
2194         default:
2195                 connman_error("Wrong address family %d", rp->ai_family);
2196                 break;
2197         }
2198         if (data->server_addr == NULL) {
2199                 freeaddrinfo(rp);
2200                 destroy_server(data);
2201                 return NULL;
2202         }
2203         memcpy(data->server_addr, rp->ai_addr, rp->ai_addrlen);
2204         freeaddrinfo(rp);
2205
2206         if (server_create_socket(data) != 0) {
2207                 destroy_server(data);
2208                 return NULL;
2209         }
2210
2211         if (protocol == IPPROTO_UDP) {
2212                 /* Enable new servers by default */
2213                 data->enabled = TRUE;
2214                 DBG("Adding DNS server %s", data->server);
2215
2216                 server_list = g_slist_append(server_list, data);
2217         }
2218
2219         return data;
2220 }
2221
2222 static gboolean resolv(struct request_data *req,
2223                                 gpointer request, gpointer name)
2224 {
2225         GSList *list;
2226
2227         for (list = server_list; list; list = list->next) {
2228                 struct server_data *data = list->data;
2229
2230                 if (data->protocol == IPPROTO_TCP) {
2231                         DBG("server %s ignored proto TCP", data->server);
2232                         continue;
2233                 }
2234
2235                 DBG("server %s enabled %d", data->server, data->enabled);
2236
2237                 if (data->enabled == FALSE)
2238                         continue;
2239
2240                 if (data->channel == NULL && data->protocol == IPPROTO_UDP) {
2241                         if (server_create_socket(data) < 0) {
2242                                 DBG("socket creation failed while resolving");
2243                                 continue;
2244                         }
2245                 }
2246
2247                 if (ns_resolv(data, req, request, name) > 0)
2248                         return TRUE;
2249         }
2250
2251         return FALSE;
2252 }
2253
2254 static void append_domain(int index, const char *domain)
2255 {
2256         GSList *list;
2257
2258         DBG("index %d domain %s", index, domain);
2259
2260         if (domain == NULL)
2261                 return;
2262
2263         for (list = server_list; list; list = list->next) {
2264                 struct server_data *data = list->data;
2265                 GList *dom_list;
2266                 char *dom;
2267                 gboolean dom_found = FALSE;
2268
2269                 if (data->index < 0)
2270                         continue;
2271
2272                 if (data->index != index)
2273                         continue;
2274
2275                 for (dom_list = data->domains; dom_list;
2276                                 dom_list = dom_list->next) {
2277                         dom = dom_list->data;
2278
2279                         if (g_str_equal(dom, domain)) {
2280                                 dom_found = TRUE;
2281                                 break;
2282                         }
2283                 }
2284
2285                 if (dom_found == FALSE) {
2286                         data->domains =
2287                                 g_list_append(data->domains, g_strdup(domain));
2288                 }
2289         }
2290 }
2291
2292 int __connman_dnsproxy_append(int index, const char *domain,
2293                                                         const char *server)
2294 {
2295         struct server_data *data;
2296
2297         DBG("index %d server %s", index, server);
2298
2299         if (server == NULL && domain == NULL)
2300                 return -EINVAL;
2301
2302         if (server == NULL) {
2303                 append_domain(index, domain);
2304
2305                 return 0;
2306         }
2307
2308         if (g_str_equal(server, "127.0.0.1") == TRUE)
2309                 return -ENODEV;
2310
2311         data = find_server(index, server, IPPROTO_UDP);
2312         if (data != NULL) {
2313                 append_domain(index, domain);
2314                 return 0;
2315         }
2316
2317         data = create_server(index, domain, server, IPPROTO_UDP);
2318         if (data == NULL)
2319                 return -EIO;
2320
2321         return 0;
2322 }
2323
2324 static void remove_server(int index, const char *domain,
2325                         const char *server, int protocol)
2326 {
2327         struct server_data *data;
2328
2329         data = find_server(index, server, protocol);
2330         if (data == NULL)
2331                 return;
2332
2333         destroy_server(data);
2334 }
2335
2336 int __connman_dnsproxy_remove(int index, const char *domain,
2337                                                         const char *server)
2338 {
2339         DBG("index %d server %s", index, server);
2340
2341         if (server == NULL)
2342                 return -EINVAL;
2343
2344         if (g_str_equal(server, "127.0.0.1") == TRUE)
2345                 return -ENODEV;
2346
2347         remove_server(index, domain, server, IPPROTO_UDP);
2348         remove_server(index, domain, server, IPPROTO_TCP);
2349
2350         return 0;
2351 }
2352
2353 void __connman_dnsproxy_flush(void)
2354 {
2355         GSList *list;
2356
2357         list = request_list;
2358         while (list) {
2359                 struct request_data *req = list->data;
2360
2361                 list = list->next;
2362
2363                 if (resolv(req, req->request, req->name) == TRUE) {
2364                         /*
2365                          * A cached result was sent,
2366                          * so the request can be released
2367                          */
2368                         request_list =
2369                                 g_slist_remove(request_list, req);
2370                         destroy_request_data(req);
2371                         continue;
2372                 }
2373
2374                 if (req->timeout > 0)
2375                         g_source_remove(req->timeout);
2376                 req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2377         }
2378 }
2379
2380 static void dnsproxy_offline_mode(connman_bool_t enabled)
2381 {
2382         GSList *list;
2383
2384         DBG("enabled %d", enabled);
2385
2386         for (list = server_list; list; list = list->next) {
2387                 struct server_data *data = list->data;
2388
2389                 if (enabled == FALSE) {
2390                         DBG("Enabling DNS server %s", data->server);
2391                         data->enabled = TRUE;
2392                         cache_invalidate();
2393                         cache_refresh();
2394                 } else {
2395                         DBG("Disabling DNS server %s", data->server);
2396                         data->enabled = FALSE;
2397                         cache_invalidate();
2398                 }
2399         }
2400 }
2401
2402 static void dnsproxy_default_changed(struct connman_service *service)
2403 {
2404         GSList *list;
2405         int index;
2406
2407         DBG("service %p", service);
2408
2409         /* DNS has changed, invalidate the cache */
2410         cache_invalidate();
2411
2412         if (service == NULL) {
2413                 /* When no services are active, then disable DNS proxying */
2414                 dnsproxy_offline_mode(TRUE);
2415                 return;
2416         }
2417
2418         index = __connman_service_get_index(service);
2419         if (index < 0)
2420                 return;
2421
2422         for (list = server_list; list; list = list->next) {
2423                 struct server_data *data = list->data;
2424
2425                 if (data->index == index) {
2426                         DBG("Enabling DNS server %s", data->server);
2427                         data->enabled = TRUE;
2428                 } else {
2429                         DBG("Disabling DNS server %s", data->server);
2430                         data->enabled = FALSE;
2431                 }
2432         }
2433
2434         cache_refresh();
2435 }
2436
2437 static struct connman_notifier dnsproxy_notifier = {
2438         .name                   = "dnsproxy",
2439         .default_changed        = dnsproxy_default_changed,
2440         .offline_mode           = dnsproxy_offline_mode,
2441 };
2442
2443 static unsigned char opt_edns0_type[2] = { 0x00, 0x29 };
2444
2445 static int parse_request(unsigned char *buf, int len,
2446                                         char *name, unsigned int size)
2447 {
2448         struct domain_hdr *hdr = (void *) buf;
2449         uint16_t qdcount = ntohs(hdr->qdcount);
2450         uint16_t arcount = ntohs(hdr->arcount);
2451         unsigned char *ptr;
2452         char *last_label = NULL;
2453         unsigned int remain, used = 0;
2454
2455         if (len < 12)
2456                 return -EINVAL;
2457
2458         DBG("id 0x%04x qr %d opcode %d qdcount %d arcount %d",
2459                                         hdr->id, hdr->qr, hdr->opcode,
2460                                                         qdcount, arcount);
2461
2462         if (hdr->qr != 0 || qdcount != 1)
2463                 return -EINVAL;
2464
2465         name[0] = '\0';
2466
2467         ptr = buf + sizeof(struct domain_hdr);
2468         remain = len - sizeof(struct domain_hdr);
2469
2470         while (remain > 0) {
2471                 uint8_t label_len = *ptr;
2472
2473                 if (label_len == 0x00) {
2474                         last_label = (char *) (ptr + 1);
2475                         break;
2476                 }
2477
2478                 if (used + label_len + 1 > size)
2479                         return -ENOBUFS;
2480
2481                 strncat(name, (char *) (ptr + 1), label_len);
2482                 strcat(name, ".");
2483
2484                 used += label_len + 1;
2485
2486                 ptr += label_len + 1;
2487                 remain -= label_len + 1;
2488         }
2489
2490         if (last_label && arcount && remain >= 9 && last_label[4] == 0 &&
2491                                 !memcmp(last_label + 5, opt_edns0_type, 2)) {
2492                 uint16_t edns0_bufsize;
2493
2494                 edns0_bufsize = last_label[7] << 8 | last_label[8];
2495
2496                 DBG("EDNS0 buffer size %u", edns0_bufsize);
2497
2498                 /* This is an evil hack until full TCP support has been
2499                  * implemented.
2500                  *
2501                  * Somtimes the EDNS0 request gets send with a too-small
2502                  * buffer size. Since glibc doesn't seem to crash when it
2503                  * gets a response biffer then it requested, just bump
2504                  * the buffer size up to 4KiB.
2505                  */
2506                 if (edns0_bufsize < 0x1000) {
2507                         last_label[7] = 0x10;
2508                         last_label[8] = 0x00;
2509                 }
2510         }
2511
2512         DBG("query %s", name);
2513
2514         return 0;
2515 }
2516
2517 static gboolean tcp_listener_event(GIOChannel *channel, GIOCondition condition,
2518                                                         gpointer user_data)
2519 {
2520         unsigned char buf[768];
2521         char query[512];
2522         struct request_data *req;
2523         int sk, client_sk, len, err;
2524         struct sockaddr_in6 client_addr;
2525         socklen_t client_addr_len = sizeof(client_addr);
2526         GSList *list;
2527         struct listener_data *ifdata = user_data;
2528         int waiting_for_connect = FALSE, qtype = 0;
2529         struct cache_entry *entry;
2530
2531         DBG("condition 0x%x", condition);
2532
2533         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2534                 if (ifdata->tcp_listener_watch > 0)
2535                         g_source_remove(ifdata->tcp_listener_watch);
2536                 ifdata->tcp_listener_watch = 0;
2537
2538                 connman_error("Error with TCP listener channel");
2539
2540                 return FALSE;
2541         }
2542
2543         sk = g_io_channel_unix_get_fd(channel);
2544
2545         client_sk = accept(sk, (void *)&client_addr, &client_addr_len);
2546         if (client_sk < 0) {
2547                 connman_error("Accept failure on TCP listener");
2548                 ifdata->tcp_listener_watch = 0;
2549                 return FALSE;
2550         }
2551
2552         len = recv(client_sk, buf, sizeof(buf), 0);
2553         if (len < 2)
2554                 return TRUE;
2555
2556         DBG("Received %d bytes (id 0x%04x) from %d", len,
2557                 buf[2] | buf[3] << 8, client_sk);
2558
2559         err = parse_request(buf + 2, len - 2, query, sizeof(query));
2560         if (err < 0 || (g_slist_length(server_list) == 0)) {
2561                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2562                 return TRUE;
2563         }
2564
2565         req = g_try_new0(struct request_data, 1);
2566         if (req == NULL)
2567                 return TRUE;
2568
2569         memcpy(&req->sa, &client_addr, client_addr_len);
2570         req->sa_len = client_addr_len;
2571         req->client_sk = client_sk;
2572         req->protocol = IPPROTO_TCP;
2573
2574         req->srcid = buf[2] | (buf[3] << 8);
2575         req->dstid = get_id();
2576         req->altid = get_id();
2577         req->request_len = len;
2578
2579         buf[2] = req->dstid & 0xff;
2580         buf[3] = req->dstid >> 8;
2581
2582         req->numserv = 0;
2583         req->ifdata = (struct listener_data *) ifdata;
2584         req->append_domain = FALSE;
2585
2586         /*
2587          * Check if the answer is found in the cache before
2588          * creating sockets to the server.
2589          */
2590         entry = cache_check(buf, &qtype, IPPROTO_TCP);
2591         if (entry != NULL) {
2592                 int ttl_left = 0;
2593                 struct cache_data *data;
2594
2595                 DBG("cache hit %s type %s", query, qtype == 1 ? "A" : "AAAA");
2596                 if (qtype == 1)
2597                         data = entry->ipv4;
2598                 else
2599                         data = entry->ipv6;
2600
2601                 if (data != NULL) {
2602                         ttl_left = data->valid_until - time(NULL);
2603                         entry->hits++;
2604
2605                         send_cached_response(client_sk, data->data,
2606                                         data->data_len, NULL, 0, IPPROTO_TCP,
2607                                         req->srcid, data->answers, ttl_left);
2608
2609                         g_free(req);
2610                         return TRUE;
2611                 } else
2612                         DBG("data missing, ignoring cache for this query");
2613         }
2614
2615         for (list = server_list; list; list = list->next) {
2616                 struct server_data *data = list->data;
2617
2618                 if (data->protocol != IPPROTO_UDP || data->enabled == FALSE)
2619                         continue;
2620
2621                 if(create_server(data->index, NULL,
2622                                         data->server, IPPROTO_TCP) == NULL)
2623                         continue;
2624
2625                 waiting_for_connect = TRUE;
2626         }
2627
2628         if (waiting_for_connect == FALSE) {
2629                 /* No server is waiting for connect */
2630                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2631                 g_free(req);
2632                 return TRUE;
2633         }
2634
2635         /*
2636          * The server is not connected yet.
2637          * Copy the relevant buffers.
2638          * The request will actually be sent once we're
2639          * properly connected over TCP to the nameserver.
2640          */
2641         req->request = g_try_malloc0(req->request_len);
2642         if (req->request == NULL) {
2643                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2644                 g_free(req);
2645                 return TRUE;
2646         }
2647         memcpy(req->request, buf, req->request_len);
2648
2649         req->name = g_try_malloc0(sizeof(query));
2650         if (req->name == NULL) {
2651                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2652                 g_free(req->request);
2653                 g_free(req);
2654                 return TRUE;
2655         }
2656         memcpy(req->name, query, sizeof(query));
2657
2658         req->timeout = g_timeout_add_seconds(30, request_timeout, req);
2659
2660         request_list = g_slist_append(request_list, req);
2661
2662         return TRUE;
2663 }
2664
2665 static gboolean udp_listener_event(GIOChannel *channel, GIOCondition condition,
2666                                                         gpointer user_data)
2667 {
2668         unsigned char buf[768];
2669         char query[512];
2670         struct request_data *req;
2671         struct sockaddr_in6 client_addr;
2672         socklen_t client_addr_len = sizeof(client_addr);
2673         int sk, err, len;
2674         struct listener_data *ifdata = user_data;
2675
2676         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2677                 connman_error("Error with UDP listener channel");
2678                 ifdata->udp_listener_watch = 0;
2679                 return FALSE;
2680         }
2681
2682         sk = g_io_channel_unix_get_fd(channel);
2683
2684         memset(&client_addr, 0, client_addr_len);
2685         len = recvfrom(sk, buf, sizeof(buf), 0, (void *)&client_addr,
2686                        &client_addr_len);
2687         if (len < 2)
2688                 return TRUE;
2689
2690         DBG("Received %d bytes (id 0x%04x)", len, buf[0] | buf[1] << 8);
2691
2692         err = parse_request(buf, len, query, sizeof(query));
2693         if (err < 0 || (g_slist_length(server_list) == 0)) {
2694                 send_response(sk, buf, len, (void *)&client_addr,
2695                                 client_addr_len, IPPROTO_UDP);
2696                 return TRUE;
2697         }
2698
2699         req = g_try_new0(struct request_data, 1);
2700         if (req == NULL)
2701                 return TRUE;
2702
2703         memcpy(&req->sa, &client_addr, client_addr_len);
2704         req->sa_len = client_addr_len;
2705         req->client_sk = 0;
2706         req->protocol = IPPROTO_UDP;
2707
2708         req->srcid = buf[0] | (buf[1] << 8);
2709         req->dstid = get_id();
2710         req->altid = get_id();
2711         req->request_len = len;
2712
2713         buf[0] = req->dstid & 0xff;
2714         buf[1] = req->dstid >> 8;
2715
2716         req->numserv = 0;
2717         req->ifdata = (struct listener_data *) ifdata;
2718         req->append_domain = FALSE;
2719
2720         if (resolv(req, buf, query) == TRUE) {
2721                 /* a cached result was sent, so the request can be released */
2722                 g_free(req);
2723                 return TRUE;
2724         }
2725
2726         req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2727         request_list = g_slist_append(request_list, req);
2728
2729         return TRUE;
2730 }
2731
2732 static int create_dns_listener(int protocol, struct listener_data *ifdata)
2733 {
2734         GIOChannel *channel;
2735         const char *proto;
2736         union {
2737                 struct sockaddr sa;
2738                 struct sockaddr_in6 sin6;
2739                 struct sockaddr_in sin;
2740         } s;
2741         socklen_t slen;
2742         int sk, type, v6only = 0;
2743         int family = AF_INET6;
2744         char *interface;
2745
2746         DBG("index %d", ifdata->index);
2747
2748         switch (protocol) {
2749         case IPPROTO_UDP:
2750                 proto = "UDP";
2751                 type = SOCK_DGRAM | SOCK_CLOEXEC;
2752                 break;
2753
2754         case IPPROTO_TCP:
2755                 proto = "TCP";
2756                 type = SOCK_STREAM | SOCK_CLOEXEC;
2757                 break;
2758
2759         default:
2760                 return -EINVAL;
2761         }
2762
2763         sk = socket(family, type, protocol);
2764         if (sk < 0 && family == AF_INET6 && errno == EAFNOSUPPORT) {
2765                 connman_error("No IPv6 support; DNS proxy listening only on Legacy IP");
2766                 family = AF_INET;
2767                 sk = socket(family, type, protocol);
2768         }
2769         if (sk < 0) {
2770                 connman_error("Failed to create %s listener socket", proto);
2771                 return -EIO;
2772         }
2773
2774         interface = connman_inet_ifname(ifdata->index);
2775         if (interface == NULL || setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2776                                         interface,
2777                                         strlen(interface) + 1) < 0) {
2778                 connman_error("Failed to bind %s listener interface", proto);
2779                 close(sk);
2780                 g_free(interface);
2781                 return -EIO;
2782         }
2783         g_free(interface);
2784
2785         /* Ensure it accepts Legacy IP connections too */
2786         if (family == AF_INET6 &&
2787                         setsockopt(sk, SOL_IPV6, IPV6_V6ONLY,
2788                                         &v6only, sizeof(v6only)) < 0) {
2789                 connman_error("Failed to clear V6ONLY on %s listener socket",
2790                               proto);
2791                 close(sk);
2792                 return -EIO;
2793         }
2794
2795         if (family == AF_INET) {
2796                 memset(&s.sin, 0, sizeof(s.sin));
2797                 s.sin.sin_family = AF_INET;
2798                 s.sin.sin_port = htons(53);
2799                 s.sin.sin_addr.s_addr = htonl(INADDR_ANY);
2800                 slen = sizeof(s.sin);
2801         } else {
2802                 memset(&s.sin6, 0, sizeof(s.sin6));
2803                 s.sin6.sin6_family = AF_INET6;
2804                 s.sin6.sin6_port = htons(53);
2805                 s.sin6.sin6_addr = in6addr_any;
2806                 slen = sizeof(s.sin6);
2807         }
2808
2809         if (bind(sk, &s.sa, slen) < 0) {
2810                 connman_error("Failed to bind %s listener socket", proto);
2811                 close(sk);
2812                 return -EIO;
2813         }
2814
2815         if (protocol == IPPROTO_TCP && listen(sk, 10) < 0) {
2816                 connman_error("Failed to listen on TCP socket");
2817                 close(sk);
2818                 return -EIO;
2819         }
2820
2821         channel = g_io_channel_unix_new(sk);
2822         if (channel == NULL) {
2823                 connman_error("Failed to create %s listener channel", proto);
2824                 close(sk);
2825                 return -EIO;
2826         }
2827
2828         g_io_channel_set_close_on_unref(channel, TRUE);
2829
2830         if (protocol == IPPROTO_TCP) {
2831                 ifdata->tcp_listener_channel = channel;
2832                 ifdata->tcp_listener_watch = g_io_add_watch(channel,
2833                                 G_IO_IN, tcp_listener_event, (gpointer) ifdata);
2834         } else {
2835                 ifdata->udp_listener_channel = channel;
2836                 ifdata->udp_listener_watch = g_io_add_watch(channel,
2837                                 G_IO_IN, udp_listener_event, (gpointer) ifdata);
2838         }
2839
2840         return 0;
2841 }
2842
2843 static void destroy_udp_listener(struct listener_data *ifdata)
2844 {
2845         DBG("index %d", ifdata->index);
2846
2847         if (ifdata->udp_listener_watch > 0)
2848                 g_source_remove(ifdata->udp_listener_watch);
2849
2850         g_io_channel_unref(ifdata->udp_listener_channel);
2851 }
2852
2853 static void destroy_tcp_listener(struct listener_data *ifdata)
2854 {
2855         DBG("index %d", ifdata->index);
2856
2857         if (ifdata->tcp_listener_watch > 0)
2858                 g_source_remove(ifdata->tcp_listener_watch);
2859
2860         g_io_channel_unref(ifdata->tcp_listener_channel);
2861 }
2862
2863 static int create_listener(struct listener_data *ifdata)
2864 {
2865         int err, index;
2866
2867         err = create_dns_listener(IPPROTO_UDP, ifdata);
2868         if (err < 0)
2869                 return err;
2870
2871         err = create_dns_listener(IPPROTO_TCP, ifdata);
2872         if (err < 0) {
2873                 destroy_udp_listener(ifdata);
2874                 return err;
2875         }
2876
2877         index = connman_inet_ifindex("lo");
2878         if (ifdata->index == index)
2879                 __connman_resolvfile_append(index, NULL, "127.0.0.1");
2880
2881         return 0;
2882 }
2883
2884 static void destroy_listener(struct listener_data *ifdata)
2885 {
2886         int index;
2887         GSList *list;
2888
2889         index = connman_inet_ifindex("lo");
2890         if (ifdata->index == index)
2891                 __connman_resolvfile_remove(index, NULL, "127.0.0.1");
2892
2893         for (list = request_list; list; list = list->next) {
2894                 struct request_data *req = list->data;
2895
2896                 DBG("Dropping request (id 0x%04x -> 0x%04x)",
2897                                                 req->srcid, req->dstid);
2898                 destroy_request_data(req);
2899                 list->data = NULL;
2900         }
2901
2902         g_slist_free(request_list);
2903         request_list = NULL;
2904
2905         destroy_tcp_listener(ifdata);
2906         destroy_udp_listener(ifdata);
2907 }
2908
2909 int __connman_dnsproxy_add_listener(int index)
2910 {
2911         struct listener_data *ifdata;
2912         int err;
2913
2914         DBG("index %d", index);
2915
2916         if (index < 0)
2917                 return -EINVAL;
2918
2919         if (g_hash_table_lookup(listener_table, GINT_TO_POINTER(index)) != NULL)
2920                 return 0;
2921
2922         ifdata = g_try_new0(struct listener_data, 1);
2923         if (ifdata == NULL)
2924                 return -ENOMEM;
2925
2926         ifdata->index = index;
2927         ifdata->udp_listener_channel = NULL;
2928         ifdata->udp_listener_watch = 0;
2929         ifdata->tcp_listener_channel = NULL;
2930         ifdata->tcp_listener_watch = 0;
2931
2932         err = create_listener(ifdata);
2933         if (err < 0) {
2934                 connman_error("Couldn't create listener for index %d err %d",
2935                                 index, err);
2936                 g_free(ifdata);
2937                 return err;
2938         }
2939         g_hash_table_insert(listener_table, GINT_TO_POINTER(ifdata->index),
2940                         ifdata);
2941         return 0;
2942 }
2943
2944 void __connman_dnsproxy_remove_listener(int index)
2945 {
2946         struct listener_data *ifdata;
2947
2948         DBG("index %d", index);
2949
2950         ifdata = g_hash_table_lookup(listener_table, GINT_TO_POINTER(index));
2951         if (ifdata == NULL)
2952                 return;
2953
2954         destroy_listener(ifdata);
2955
2956         g_hash_table_remove(listener_table, GINT_TO_POINTER(index));
2957 }
2958
2959 static void remove_listener(gpointer key, gpointer value, gpointer user_data)
2960 {
2961         int index = GPOINTER_TO_INT(key);
2962         struct listener_data *ifdata = value;
2963
2964         DBG("index %d", index);
2965
2966         destroy_listener(ifdata);
2967 }
2968
2969 int __connman_dnsproxy_init(void)
2970 {
2971         int err, index;
2972
2973         DBG("");
2974
2975         srandom(time(NULL));
2976
2977         listener_table = g_hash_table_new_full(g_direct_hash, g_direct_equal,
2978                                                         NULL, g_free);
2979
2980         index = connman_inet_ifindex("lo");
2981         err = __connman_dnsproxy_add_listener(index);
2982         if (err < 0)
2983                 return err;
2984
2985         err = connman_notifier_register(&dnsproxy_notifier);
2986         if (err < 0)
2987                 goto destroy;
2988
2989         return 0;
2990
2991 destroy:
2992         __connman_dnsproxy_remove_listener(index);
2993         g_hash_table_destroy(listener_table);
2994
2995         return err;
2996 }
2997
2998 void __connman_dnsproxy_cleanup(void)
2999 {
3000         DBG("");
3001
3002         connman_notifier_unregister(&dnsproxy_notifier);
3003
3004         g_hash_table_foreach(listener_table, remove_listener, NULL);
3005
3006         g_hash_table_destroy(listener_table);
3007 }