src/dnsproxy.c

   1 /*
   2  *
   3  *  Connection Manager
   4  *
   5  *  Copyright (C) 2007-2012  Intel Corporation. All rights reserved.
   6  *
   7  *  This program is free software; you can redistribute it and/or modify
   8  *  it under the terms of the GNU General Public License version 2 as
   9  *  published by the Free Software Foundation.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, write to the Free Software
  18  *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  */
  21
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <errno.h>
  27 #include <stdlib.h>
  28 #include <unistd.h>
  29 #include <string.h>
  30 #include <stdint.h>
  31 #include <arpa/inet.h>
  32 #include <netinet/in.h>
  33 #include <sys/types.h>
  34 #include <sys/socket.h>
  35 #include <netdb.h>
  36 #include <resolv.h>
  37 #include <gweb/gresolv.h>
  38
  39 #include <glib.h>
  40
  41 #include "connman.h"
  42
  43 #if __BYTE_ORDER == __LITTLE_ENDIAN
  44 struct domain_hdr {
  45         uint16_t id;
  46         uint8_t rd:1;
  47         uint8_t tc:1;
  48         uint8_t aa:1;
  49         uint8_t opcode:4;
  50         uint8_t qr:1;
  51         uint8_t rcode:4;
  52         uint8_t z:3;
  53         uint8_t ra:1;
  54         uint16_t qdcount;
  55         uint16_t ancount;
  56         uint16_t nscount;
  57         uint16_t arcount;
  58 } __attribute__ ((packed));
  59 #elif __BYTE_ORDER == __BIG_ENDIAN
  60 struct domain_hdr {
  61         uint16_t id;
  62         uint8_t qr:1;
  63         uint8_t opcode:4;
  64         uint8_t aa:1;
  65         uint8_t tc:1;
  66         uint8_t rd:1;
  67         uint8_t ra:1;
  68         uint8_t z:3;
  69         uint8_t rcode:4;
  70         uint16_t qdcount;
  71         uint16_t ancount;
  72         uint16_t nscount;
  73         uint16_t arcount;
  74 } __attribute__ ((packed));
  75 #else
  76 #error "Unknown byte order"
  77 #endif
  78
  79 struct partial_reply {
  80         uint16_t len;
  81         uint16_t received;
  82         unsigned char buf[];
  83 };
  84
  85 struct server_data {
  86         char *interface;
  87         GList *domains;
  88         char *server;
  89         int protocol;
  90         GIOChannel *channel;
  91         guint watch;
  92         guint timeout;
  93         gboolean enabled;
  94         gboolean connected;
  95         struct partial_reply *incoming_reply;
  96 };
  97
  98 struct request_data {
  99         union {
 100                 struct sockaddr_in6 __sin6; /* Only for the length */
 101                 struct sockaddr sa;
 102         };
 103         socklen_t sa_len;
 104         int client_sk;
 105         int protocol;
 106         guint16 srcid;
 107         guint16 dstid;
 108         guint16 altid;
 109         guint timeout;
 110         guint watch;
 111         guint numserv;
 112         guint numresp;
 113         gpointer request;
 114         gsize request_len;
 115         gpointer name;
 116         gpointer resp;
 117         gsize resplen;
 118         struct listener_data *ifdata;
 119         gboolean append_domain;
 120 };
 121
 122 struct listener_data {
 123         char *ifname;
 124         GIOChannel *udp_listener_channel;
 125         guint udp_listener_watch;
 126         GIOChannel *tcp_listener_channel;
 127         guint tcp_listener_watch;
 128 };
 129
 130 struct cache_data {
 131         time_t inserted;
 132         time_t valid_until;
 133         time_t cache_until;
 134         int timeout;
 135         uint16_t type;
 136         uint16_t answers;
 137         unsigned int data_len;
 138         unsigned char *data; /* contains DNS header + body */
 139 };
 140
 141 struct cache_entry {
 142         char *key;
 143         int want_refresh;
 144         int hits;
 145         struct cache_data *ipv4;
 146         struct cache_data *ipv6;
 147 };
 148
 149 struct domain_question {
 150         uint16_t type;
 151         uint16_t class;
 152 } __attribute__ ((packed));
 153
 154 struct domain_rr {
 155         uint16_t type;
 156         uint16_t class;
 157         uint32_t ttl;
 158         uint16_t rdlen;
 159 } __attribute__ ((packed));
 160
 161 /*
 162  * We limit how long the cached DNS entry stays in the cache.
 163  * By default the TTL (time-to-live) of the DNS response is used
 164  * when setting the cache entry life time. The value is in seconds.
 165  */
 166 #define MAX_CACHE_TTL (60 * 30)
 167 /*
 168  * Also limit the other end, cache at least for 30 seconds.
 169  */
 170 #define MIN_CACHE_TTL (30)
 171
 172 /*
 173  * We limit the cache size to some sane value so that cached data does
 174  * not occupy too much memory. Each cached entry occupies on average
 175  * about 100 bytes memory (depending on DNS name length).
 176  * Example: caching www.connman.net uses 97 bytes memory.
 177  * The value is the max amount of cached DNS responses (count).
 178  */
 179 #define MAX_CACHE_SIZE 256
 180
 181 static int cache_size;
 182 static GHashTable *cache;
 183 static int cache_refcount;
 184 static GSList *server_list = NULL;
 185 static GSList *request_list = NULL;
 186 static GHashTable *listener_table = NULL;
 187 static time_t next_refresh;
 188
 189 static guint16 get_id()
 190 {
 191         return random();
 192 }
 193
 194 static int protocol_offset(int protocol)
 195 {
 196         switch (protocol) {
 197         case IPPROTO_UDP:
 198                 return 0;
 199
 200         case IPPROTO_TCP:
 201                 return 2;
 202
 203         default:
 204                 return -EINVAL;
 205         }
 206
 207 }
 208
 209 /*
 210  * There is a power and efficiency benefit to have entries
 211  * in our cache expire at the same time. To this extend,
 212  * we round down the cache valid time to common boundaries.
 213  */
 214 static time_t round_down_ttl(time_t end_time, int ttl)
 215 {
 216         if (ttl < 15)
 217                 return end_time;
 218
 219         /* Less than 5 minutes, round to 10 second boundary */
 220         if (ttl < 300) {
 221                 end_time = end_time / 10;
 222                 end_time = end_time * 10;
 223         } else { /* 5 or more minutes, round to 30 seconds */
 224                 end_time = end_time / 30;
 225                 end_time = end_time * 30;
 226         }
 227         return end_time;
 228 }
 229
 230 static struct request_data *find_request(guint16 id)
 231 {
 232         GSList *list;
 233
 234         for (list = request_list; list; list = list->next) {
 235                 struct request_data *req = list->data;
 236
 237                 if (req->dstid == id || req->altid == id)
 238                         return req;
 239         }
 240
 241         return NULL;
 242 }
 243
 244 static struct server_data *find_server(const char *interface,
 245                                         const char *server,
 246                                                 int protocol)
 247 {
 248         GSList *list;
 249
 250         DBG("interface %s server %s", interface, server);
 251
 252         for (list = server_list; list; list = list->next) {
 253                 struct server_data *data = list->data;
 254
 255                 if (interface == NULL && data->interface == NULL &&
 256                                 g_str_equal(data->server, server) == TRUE &&
 257                                 data->protocol == protocol)
 258                         return data;
 259
 260                 if (interface == NULL ||
 261                                 data->interface == NULL || data->server == NULL)
 262                         continue;
 263
 264                 if (g_str_equal(data->interface, interface) == TRUE &&
 265                                 g_str_equal(data->server, server) == TRUE &&
 266                                 data->protocol == protocol)
 267                         return data;
 268         }
 269
 270         return NULL;
 271 }
 272
 273 /* we can keep using the same resolve's */
 274 static GResolv *ipv4_resolve;
 275 static GResolv *ipv6_resolve;
 276
 277 static void dummy_resolve_func(GResolvResultStatus status,
 278                                         char **results, gpointer user_data)
 279 {
 280 }
 281
 282 /*
 283  * Refresh a DNS entry, but also age the hit count a bit */
 284 static void refresh_dns_entry(struct cache_entry *entry, char *name)
 285 {
 286         int age = 1;
 287
 288         if (ipv4_resolve == NULL) {
 289                 ipv4_resolve = g_resolv_new(0);
 290                 g_resolv_set_address_family(ipv4_resolve, AF_INET);
 291                 g_resolv_add_nameserver(ipv4_resolve, "127.0.0.1", 53, 0);
 292         }
 293
 294         if (ipv6_resolve == NULL) {
 295                 ipv6_resolve = g_resolv_new(0);
 296                 g_resolv_set_address_family(ipv6_resolve, AF_INET6);
 297                 g_resolv_add_nameserver(ipv6_resolve, "127.0.0.1", 53, 0);
 298         }
 299
 300         if (entry->ipv4 == NULL) {
 301                 DBG("Refresing A record for %s", name);
 302                 g_resolv_lookup_hostname(ipv4_resolve, name,
 303                                         dummy_resolve_func, NULL);
 304                 age = 4;
 305         }
 306
 307         if (entry->ipv6 == NULL) {
 308                 DBG("Refresing AAAA record for %s", name);
 309                 g_resolv_lookup_hostname(ipv6_resolve, name,
 310                                         dummy_resolve_func, NULL);
 311                 age = 4;
 312         }
 313
 314         entry->hits -= age;
 315         if (entry->hits < 0)
 316                 entry->hits = 0;
 317 }
 318
 319 static int dns_name_length(unsigned char *buf)
 320 {
 321         if ((buf[0] & NS_CMPRSFLGS) == NS_CMPRSFLGS) /* compressed name */
 322                 return 2;
 323         return strlen((char *)buf);
 324 }
 325
 326 static void update_cached_ttl(unsigned char *buf, int len, int new_ttl)
 327 {
 328         unsigned char *c;
 329         uint32_t *i;
 330         uint16_t *w;
 331         int l;
 332
 333         /* skip the header */
 334         c = buf + 12;
 335         len -= 12;
 336
 337         /* skip the query, which is a name and 2 16 bit words */
 338         l = dns_name_length(c);
 339         c += l;
 340         len -= l;
 341         c += 4;
 342         len -= 4;
 343
 344         /* now we get the answer records */
 345
 346         while (len > 0) {
 347                 /* first a name */
 348                 l = dns_name_length(c);
 349                 c += l;
 350                 len -= l;
 351                 if (len < 0)
 352                         break;
 353                 /* then type + class, 2 bytes each */
 354                 c += 4;
 355                 len -= 4;
 356                 if (len < 0)
 357                         break;
 358
 359                 /* now the 4 byte TTL field */
 360                 i = (uint32_t *)c;
 361                 *i = htonl(new_ttl);
 362                 c += 4;
 363                 len -= 4;
 364                 if (len < 0)
 365                         break;
 366
 367                 /* now the 2 byte rdlen field */
 368                 w = (uint16_t *)c;
 369                 c += ntohs(*w) + 2;
 370                 len -= ntohs(*w) + 2;
 371         }
 372 }
 373
 374 static void send_cached_response(int sk, unsigned char *buf, int len,
 375                                 const struct sockaddr *to, socklen_t tolen,
 376                                 int protocol, int id, uint16_t answers, int ttl)
 377 {
 378         struct domain_hdr *hdr;
 379         unsigned char *ptr = buf;
 380         int err, offset, dns_len, adj_len = len - 2;
 381
 382         /*
 383          * The cached packet contains always the TCP offset (two bytes)
 384          * so skip them for UDP.
 385          */
 386         switch (protocol) {
 387         case IPPROTO_UDP:
 388                 ptr += 2;
 389                 len -= 2;
 390                 dns_len = len;
 391                 offset = 0;
 392                 break;
 393         case IPPROTO_TCP:
 394                 offset = 2;
 395                 dns_len = ptr[0] * 256 + ptr[1];
 396                 break;
 397         default:
 398                 return;
 399         }
 400
 401         if (len < 12)
 402                 return;
 403
 404         hdr = (void *) (ptr + offset);
 405
 406         hdr->id = id;
 407         hdr->qr = 1;
 408         hdr->rcode = 0;
 409         hdr->ancount = htons(answers);
 410         hdr->nscount = 0;
 411         hdr->arcount = 0;
 412
 413         /* if this is a negative reply, we are authorative */
 414         if (answers == 0)
 415                 hdr->aa = 1;
 416         else
 417                 update_cached_ttl((unsigned char *)hdr, adj_len, ttl);
 418
 419         DBG("sk %d id 0x%04x answers %d ptr %p length %d dns %d",
 420                 sk, hdr->id, answers, ptr, len, dns_len);
 421
 422         err = sendto(sk, ptr, len, MSG_NOSIGNAL, to, tolen);
 423         if (err < 0) {
 424                 connman_error("Cannot send cached DNS response: %s",
 425                                 strerror(errno));
 426                 return;
 427         }
 428
 429         if (err != len || (dns_len != (len - 2) && protocol == IPPROTO_TCP) ||
 430                                 (dns_len != len && protocol == IPPROTO_UDP))
 431                 DBG("Packet length mismatch, sent %d wanted %d dns %d",
 432                         err, len, dns_len);
 433 }
 434
 435 static void send_response(int sk, unsigned char *buf, int len,
 436                                 const struct sockaddr *to, socklen_t tolen,
 437                                 int protocol)
 438 {
 439         struct domain_hdr *hdr;
 440         int err, offset = protocol_offset(protocol);
 441
 442         DBG("");
 443
 444         if (offset < 0)
 445                 return;
 446
 447         if (len < 12)
 448                 return;
 449
 450         hdr = (void *) (buf + offset);
 451
 452         DBG("id 0x%04x qr %d opcode %d", hdr->id, hdr->qr, hdr->opcode);
 453
 454         hdr->qr = 1;
 455         hdr->rcode = 2;
 456
 457         hdr->ancount = 0;
 458         hdr->nscount = 0;
 459         hdr->arcount = 0;
 460
 461         err = sendto(sk, buf, len, MSG_NOSIGNAL, to, tolen);
 462         if (err < 0) {
 463                 connman_error("Failed to send DNS response: %s",
 464                                 strerror(errno));
 465                 return;
 466         }
 467 }
 468
 469 static gboolean request_timeout(gpointer user_data)
 470 {
 471         struct request_data *req = user_data;
 472         struct listener_data *ifdata;
 473
 474         DBG("id 0x%04x", req->srcid);
 475
 476         if (req == NULL)
 477                 return FALSE;
 478
 479         ifdata = req->ifdata;
 480
 481         request_list = g_slist_remove(request_list, req);
 482         req->numserv--;
 483
 484         if (req->resplen > 0 && req->resp != NULL) {
 485                 int sk, err;
 486
 487                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
 488
 489                 err = sendto(sk, req->resp, req->resplen, MSG_NOSIGNAL,
 490                                                 &req->sa, req->sa_len);
 491                 if (err < 0)
 492                         return FALSE;
 493         } else if (req->request && req->numserv == 0) {
 494                 struct domain_hdr *hdr;
 495
 496                 if (req->protocol == IPPROTO_TCP) {
 497                         hdr = (void *) (req->request + 2);
 498                         hdr->id = req->srcid;
 499                         send_response(req->client_sk, req->request,
 500                                 req->request_len, NULL, 0, IPPROTO_TCP);
 501
 502                 } else if (req->protocol == IPPROTO_UDP) {
 503                         int sk;
 504
 505                         hdr = (void *) (req->request);
 506                         hdr->id = req->srcid;
 507                         sk = g_io_channel_unix_get_fd(
 508                                                 ifdata->udp_listener_channel);
 509                         send_response(sk, req->request, req->request_len,
 510                                         &req->sa, req->sa_len, IPPROTO_UDP);
 511                 }
 512         }
 513
 514         g_free(req->resp);
 515         g_free(req);
 516
 517         return FALSE;
 518 }
 519
 520 static int append_query(unsigned char *buf, unsigned int size,
 521                                 const char *query, const char *domain)
 522 {
 523         unsigned char *ptr = buf;
 524         int len;
 525
 526         DBG("query %s domain %s", query, domain);
 527
 528         while (query != NULL) {
 529                 const char *tmp;
 530
 531                 tmp = strchr(query, '.');
 532                 if (tmp == NULL) {
 533                         len = strlen(query);
 534                         if (len == 0)
 535                                 break;
 536                         *ptr = len;
 537                         memcpy(ptr + 1, query, len);
 538                         ptr += len + 1;
 539                         break;
 540                 }
 541
 542                 *ptr = tmp - query;
 543                 memcpy(ptr + 1, query, tmp - query);
 544                 ptr += tmp - query + 1;
 545
 546                 query = tmp + 1;
 547         }
 548
 549         while (domain != NULL) {
 550                 const char *tmp;
 551
 552                 tmp = strchr(domain, '.');
 553                 if (tmp == NULL) {
 554                         len = strlen(domain);
 555                         if (len == 0)
 556                                 break;
 557                         *ptr = len;
 558                         memcpy(ptr + 1, domain, len);
 559                         ptr += len + 1;
 560                         break;
 561                 }
 562
 563                 *ptr = tmp - domain;
 564                 memcpy(ptr + 1, domain, tmp - domain);
 565                 ptr += tmp - domain + 1;
 566
 567                 domain = tmp + 1;
 568         }
 569
 570         *ptr++ = 0x00;
 571
 572         return ptr - buf;
 573 }
 574
 575 static gboolean cache_check_is_valid(struct cache_data *data,
 576                                 time_t current_time)
 577 {
 578         if (data == NULL)
 579                 return FALSE;
 580
 581         if (data->cache_until < current_time)
 582                 return FALSE;
 583
 584         return TRUE;
 585 }
 586
 587 /*
 588  * remove stale cached entries so that they can be refreshed
 589  */
 590 static void cache_enforce_validity(struct cache_entry *entry)
 591 {
 592         time_t current_time = time(NULL);
 593
 594         if (cache_check_is_valid(entry->ipv4, current_time) == FALSE
 595                                                         && entry->ipv4) {
 596                 DBG("cache timeout \"%s\" type A", entry->key);
 597                 g_free(entry->ipv4->data);
 598                 g_free(entry->ipv4);
 599                 entry->ipv4 = NULL;
 600
 601         }
 602
 603         if (cache_check_is_valid(entry->ipv6, current_time) == FALSE
 604                                                         && entry->ipv6) {
 605                 DBG("cache timeout \"%s\" type AAAA", entry->key);
 606                 g_free(entry->ipv6->data);
 607                 g_free(entry->ipv6);
 608                 entry->ipv6 = NULL;
 609         }
 610 }
 611
 612 static uint16_t cache_check_validity(char *question, uint16_t type,
 613                                 struct cache_entry *entry)
 614 {
 615         time_t current_time = time(NULL);
 616         int want_refresh = 0;
 617
 618         /*
 619          * if we have a popular entry, we want a refresh instead of
 620          * total destruction of the entry.
 621          */
 622         if (entry->hits > 2)
 623                 want_refresh = 1;
 624
 625         cache_enforce_validity(entry);
 626
 627         switch (type) {
 628         case 1:         /* IPv4 */
 629                 if (cache_check_is_valid(entry->ipv4, current_time) == FALSE) {
 630                         DBG("cache %s \"%s\" type A", entry->ipv4 ?
 631                                         "timeout" : "entry missing", question);
 632
 633                         if (want_refresh)
 634                                 entry->want_refresh = 1;
 635
 636                         /*
 637                          * We do not remove cache entry if there is still
 638                          * valid IPv6 entry found in the cache.
 639                          */
 640                         if (cache_check_is_valid(entry->ipv6, current_time)
 641                                         == FALSE && want_refresh == FALSE) {
 642                                 g_hash_table_remove(cache, question);
 643                                 type = 0;
 644                         }
 645                 }
 646                 break;
 647
 648         case 28:        /* IPv6 */
 649                 if (cache_check_is_valid(entry->ipv6, current_time) == FALSE) {
 650                         DBG("cache %s \"%s\" type AAAA", entry->ipv6 ?
 651                                         "timeout" : "entry missing", question);
 652
 653                         if (want_refresh)
 654                                 entry->want_refresh = 1;
 655
 656                         if (cache_check_is_valid(entry->ipv4, current_time)
 657                                         == FALSE && want_refresh == FALSE) {
 658                                 g_hash_table_remove(cache, question);
 659                                 type = 0;
 660                         }
 661                 }
 662                 break;
 663         }
 664
 665         return type;
 666 }
 667
 668 static struct cache_entry *cache_check(gpointer request, int *qtype)
 669 {
 670         char *question = request + 12;
 671         struct cache_entry *entry;
 672         struct domain_question *q;
 673         uint16_t type;
 674         int offset;
 675
 676         offset = strlen(question) + 1;
 677         q = (void *) (question + offset);
 678         type = ntohs(q->type);
 679
 680         /* We only cache either A (1) or AAAA (28) requests */
 681         if (type != 1 && type != 28)
 682                 return NULL;
 683
 684         entry = g_hash_table_lookup(cache, question);
 685         if (entry == NULL)
 686                 return NULL;
 687
 688         type = cache_check_validity(question, type, entry);
 689         if (type == 0)
 690                 return NULL;
 691
 692         *qtype = type;
 693         return entry;
 694 }
 695
 696 /*
 697  * Get a label/name from DNS resource record. The function decompresses the
 698  * label if necessary. The function does not convert the name to presentation
 699  * form. This means that the result string will contain label lengths instead
 700  * of dots between labels. We intentionally do not want to convert to dotted
 701  * format so that we can cache the wire format string directly.
 702  */
 703 static int get_name(int counter,
 704                 unsigned char *pkt, unsigned char *start, unsigned char *max,
 705                 unsigned char *output, int output_max, int *output_len,
 706                 unsigned char **end, char *name, int *name_len)
 707 {
 708         unsigned char *p;
 709
 710         /* Limit recursion to 10 (this means up to 10 labels in domain name) */
 711         if (counter > 10)
 712                 return -EINVAL;
 713
 714         p = start;
 715         while (*p) {
 716                 if ((*p & NS_CMPRSFLGS) == NS_CMPRSFLGS) {
 717                         uint16_t offset = (*p & 0x3F) * 256 + *(p + 1);
 718
 719                         if (offset >= max - pkt)
 720                                 return -ENOBUFS;
 721
 722                         if (*end == NULL)
 723                                 *end = p + 2;
 724
 725                         return get_name(counter + 1, pkt, pkt + offset, max,
 726                                         output, output_max, output_len, end,
 727                                         name, name_len);
 728                 } else {
 729                         unsigned label_len = *p;
 730
 731                         if (pkt + label_len > max)
 732                                 return -ENOBUFS;
 733
 734                         if (*output_len > output_max)
 735                                 return -ENOBUFS;
 736
 737                         /*
 738                          * We need the original name in order to check
 739                          * if this answer is the correct one.
 740                          */
 741                         name[(*name_len)++] = label_len;
 742                         memcpy(name + *name_len, p + 1, label_len + 1);
 743                         *name_len += label_len;
 744
 745                         /* We compress the result */
 746                         output[0] = NS_CMPRSFLGS;
 747                         output[1] = 0x0C;
 748                         *output_len = 2;
 749
 750                         p += label_len + 1;
 751
 752                         if (*end == NULL)
 753                                 *end = p;
 754
 755                         if (p >= max)
 756                                 return -ENOBUFS;
 757                 }
 758         }
 759
 760         return 0;
 761 }
 762
 763 static int parse_rr(unsigned char *buf, unsigned char *start,
 764                         unsigned char *max,
 765                         unsigned char *response, unsigned int *response_size,
 766                         uint16_t *type, uint16_t *class, int *ttl, int *rdlen,
 767                         unsigned char **end,
 768                         char *name)
 769 {
 770         struct domain_rr *rr;
 771         int err, offset;
 772         int name_len = 0, output_len = 0, max_rsp = *response_size;
 773
 774         err = get_name(0, buf, start, max, response, max_rsp,
 775                 &output_len, end, name, &name_len);
 776         if (err < 0)
 777                 return err;
 778
 779         offset = output_len;
 780
 781         if ((unsigned int) offset > *response_size)
 782                 return -ENOBUFS;
 783
 784         rr = (void *) (*end);
 785
 786         if (rr == NULL)
 787                 return -EINVAL;
 788
 789         *type = ntohs(rr->type);
 790         *class = ntohs(rr->class);
 791         *ttl = ntohl(rr->ttl);
 792         *rdlen = ntohs(rr->rdlen);
 793
 794         if (*ttl < 0)
 795                 return -EINVAL;
 796
 797         memcpy(response + offset, *end, sizeof(struct domain_rr));
 798
 799         offset += sizeof(struct domain_rr);
 800         *end += sizeof(struct domain_rr);
 801
 802         if ((unsigned int) (offset + *rdlen) > *response_size)
 803                 return -ENOBUFS;
 804
 805         memcpy(response + offset, *end, *rdlen);
 806
 807         *end += *rdlen;
 808
 809         *response_size = offset + *rdlen;
 810
 811         return 0;
 812 }
 813
 814 static gboolean check_alias(GSList *aliases, char *name)
 815 {
 816         GSList *list;
 817
 818         if (aliases != NULL) {
 819                 for (list = aliases; list; list = list->next) {
 820                         int len = strlen((char *)list->data);
 821                         if (strncmp((char *)list->data, name, len) == 0)
 822                                 return TRUE;
 823                 }
 824         }
 825
 826         return FALSE;
 827 }
 828
 829 static int parse_response(unsigned char *buf, int buflen,
 830                         char *question, int qlen,
 831                         uint16_t *type, uint16_t *class, int *ttl,
 832                         unsigned char *response, unsigned int *response_len,
 833                         uint16_t *answers)
 834 {
 835         struct domain_hdr *hdr = (void *) buf;
 836         struct domain_question *q;
 837         unsigned char *ptr;
 838         uint16_t qdcount = ntohs(hdr->qdcount);
 839         uint16_t ancount = ntohs(hdr->ancount);
 840         int err, i;
 841         uint16_t qtype, qclass;
 842         unsigned char *next = NULL;
 843         unsigned int maxlen = *response_len;
 844         GSList *aliases = NULL, *list;
 845         char name[NS_MAXDNAME + 1];
 846
 847         if (buflen < 12)
 848                 return -EINVAL;
 849
 850         DBG("qr %d qdcount %d", hdr->qr, qdcount);
 851
 852         /* We currently only cache responses where question count is 1 */
 853         if (hdr->qr != 1 || qdcount != 1)
 854                 return -EINVAL;
 855
 856         ptr = buf + sizeof(struct domain_hdr);
 857
 858         strncpy(question, (char *) ptr, qlen);
 859         qlen = strlen(question);
 860         ptr += qlen + 1; /* skip \0 */
 861
 862         q = (void *) ptr;
 863         qtype = ntohs(q->type);
 864
 865         /* We cache only A and AAAA records */
 866         if (qtype != 1 && qtype != 28)
 867                 return -ENOMSG;
 868
 869         qclass = ntohs(q->class);
 870
 871         ptr += 2 + 2; /* ptr points now to answers */
 872
 873         err = -ENOMSG;
 874         *response_len = 0;
 875         *answers = 0;
 876
 877         /*
 878          * We have a bunch of answers (like A, AAAA, CNAME etc) to
 879          * A or AAAA question. We traverse the answers and parse the
 880          * resource records. Only A and AAAA records are cached, all
 881          * the other records in answers are skipped.
 882          */
 883         for (i = 0; i < ancount; i++) {
 884                 /*
 885                  * Get one address at a time to this buffer.
 886                  * The max size of the answer is
 887                  *   2 (pointer) + 2 (type) + 2 (class) +
 888                  *   4 (ttl) + 2 (rdlen) + addr (16 or 4) = 28
 889                  * for A or AAAA record.
 890                  * For CNAME the size can be bigger.
 891                  */
 892                 unsigned char rsp[NS_MAXCDNAME];
 893                 unsigned int rsp_len = sizeof(rsp) - 1;
 894                 int ret, rdlen;
 895
 896                 memset(rsp, 0, sizeof(rsp));
 897
 898                 ret = parse_rr(buf, ptr, buf + buflen, rsp, &rsp_len,
 899                         type, class, ttl, &rdlen, &next, name);
 900                 if (ret != 0) {
 901                         err = ret;
 902                         goto out;
 903                 }
 904
 905                 /*
 906                  * Now rsp contains compressed or uncompressed resource
 907                  * record. Next we check if this record answers the question.
 908                  * The name var contains the uncompressed label.
 909                  * One tricky bit is the CNAME records as they alias
 910                  * the name we might be interested in.
 911                  */
 912
 913                 /*
 914                  * Go to next answer if the class is not the one we are
 915                  * looking for.
 916                  */
 917                 if (*class != qclass) {
 918                         ptr = next;
 919                         next = NULL;
 920                         continue;
 921                 }
 922
 923                 /*
 924                  * Try to resolve aliases also, type is CNAME(5).
 925                  * This is important as otherwise the aliased names would not
 926                  * be cached at all as the cache would not contain the aliased
 927                  * question.
 928                  *
 929                  * If any CNAME is found in DNS packet, then we cache the alias
 930                  * IP address instead of the question (as the server
 931                  * said that question has only an alias).
 932                  * This means in practice that if e.g., ipv6.google.com is
 933                  * queried, DNS server returns CNAME of that name which is
 934                  * ipv6.l.google.com. We then cache the address of the CNAME
 935                  * but return the question name to client. So the alias
 936                  * status of the name is not saved in cache and thus not
 937                  * returned to the client. We do not return DNS packets from
 938                  * cache to client saying that ipv6.google.com is an alias to
 939                  * ipv6.l.google.com but we return instead a DNS packet that
 940                  * says ipv6.google.com has address xxx which is in fact the
 941                  * address of ipv6.l.google.com. For caching purposes this
 942                  * should not cause any issues.
 943                  */
 944                 if (*type == 5 && strncmp(question, name, qlen) == 0) {
 945                         /*
 946                          * So now the alias answered the question. This is
 947                          * not very useful from caching point of view as
 948                          * the following A or AAAA records will not match the
 949                          * question. We need to find the real A/AAAA record
 950                          * of the alias and cache that.
 951                          */
 952                         unsigned char *end = NULL;
 953                         int name_len = 0, output_len;
 954
 955                         memset(rsp, 0, sizeof(rsp));
 956                         rsp_len = sizeof(rsp) - 1;
 957
 958                         /*
 959                          * Alias is in rdata part of the message,
 960                          * and next-rdlen points to it. So we need to get
 961                          * the real name of the alias.
 962                          */
 963                         ret = get_name(0, buf, next - rdlen, buf + buflen,
 964                                         rsp, rsp_len, &output_len, &end,
 965                                         name, &name_len);
 966                         if (ret != 0) {
 967                                 /* just ignore the error at this point */
 968                                 ptr = next;
 969                                 next = NULL;
 970                                 continue;
 971                         }
 972
 973                         /*
 974                          * We should now have the alias of the entry we might
 975                          * want to cache. Just remember it for a while.
 976                          * We check the alias list when we have parsed the
 977                          * A or AAAA record.
 978                          */
 979                         aliases = g_slist_prepend(aliases, g_strdup(name));
 980
 981                         ptr = next;
 982                         next = NULL;
 983                         continue;
 984                 }
 985
 986                 if (*type == qtype) {
 987                         /*
 988                          * We found correct type (A or AAAA)
 989                          */
 990                         if (check_alias(aliases, name) == TRUE ||
 991                                 (aliases == NULL && strncmp(question, name,
 992                                                         qlen) == 0)) {
 993                                 /*
 994                                  * We found an alias or the name of the rr
 995                                  * matches the question. If so, we append
 996                                  * the compressed label to the cache.
 997                                  * The end result is a response buffer that
 998                                  * will contain one or more cached and
 999                                  * compressed resource records.
1000                                  */
1001                                 if (*response_len + rsp_len > maxlen) {
1002                                         err = -ENOBUFS;
1003                                         goto out;
1004                                 }
1005                                 memcpy(response + *response_len, rsp, rsp_len);
1006                                 *response_len += rsp_len;
1007                                 (*answers)++;
1008                                 err = 0;
1009                         }
1010                 }
1011
1012                 ptr = next;
1013                 next = NULL;
1014         }
1015
1016 out:
1017         for (list = aliases; list; list = list->next)
1018                 g_free(list->data);
1019         g_slist_free(aliases);
1020
1021         return err;
1022 }
1023
1024 struct cache_timeout {
1025         time_t current_time;
1026         int max_timeout;
1027         int try_harder;
1028 };
1029
1030 static gboolean cache_check_entry(gpointer key, gpointer value,
1031                                         gpointer user_data)
1032 {
1033         struct cache_timeout *data = user_data;
1034         struct cache_entry *entry = value;
1035         int max_timeout;
1036
1037         /* Scale the number of hits by half as part of cache aging */
1038
1039         entry->hits /= 2;
1040
1041         /*
1042          * If either IPv4 or IPv6 cached entry has expired, we
1043          * remove both from the cache.
1044          */
1045
1046         if (entry->ipv4 != NULL && entry->ipv4->timeout > 0) {
1047                 max_timeout = entry->ipv4->cache_until;
1048                 if (max_timeout > data->max_timeout)
1049                         data->max_timeout = max_timeout;
1050
1051                 if (entry->ipv4->cache_until < data->current_time)
1052                         return TRUE;
1053         }
1054
1055         if (entry->ipv6 != NULL && entry->ipv6->timeout > 0) {
1056                 max_timeout = entry->ipv6->cache_until;
1057                 if (max_timeout > data->max_timeout)
1058                         data->max_timeout = max_timeout;
1059
1060                 if (entry->ipv6->cache_until < data->current_time)
1061                         return TRUE;
1062         }
1063
1064         /*
1065          * if we're asked to try harder, also remove entries that have
1066          * few hits
1067          */
1068         if (data->try_harder && entry->hits < 4)
1069                 return TRUE;
1070
1071         return FALSE;
1072 }
1073
1074 static void cache_cleanup(void)
1075 {
1076         static int max_timeout;
1077         struct cache_timeout data;
1078         int count = 0;
1079
1080         data.current_time = time(NULL);
1081         data.max_timeout = 0;
1082         data.try_harder = 0;
1083
1084         /*
1085          * In the first pass, we only remove entries that have timed out.
1086          * We use a cache of the first time to expire to do this only
1087          * when it makes sense.
1088          */
1089         if (max_timeout <= data.current_time) {
1090                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1091                                                 &data);
1092         }
1093         DBG("removed %d in the first pass", count);
1094
1095         /*
1096          * In the second pass, if the first pass turned up blank,
1097          * we also expire entries with a low hit count,
1098          * while aging the hit count at the same time.
1099          */
1100         data.try_harder = 1;
1101         if (count == 0)
1102                 count = g_hash_table_foreach_remove(cache, cache_check_entry,
1103                                                 &data);
1104
1105         if (count == 0)
1106                 /*
1107                  * If we could not remove anything, then remember
1108                  * what is the max timeout and do nothing if we
1109                  * have not yet reached it. This will prevent
1110                  * constant traversal of the cache if it is full.
1111                  */
1112                 max_timeout = data.max_timeout;
1113         else
1114                 max_timeout = 0;
1115 }
1116
1117 static gboolean cache_invalidate_entry(gpointer key, gpointer value,
1118                                         gpointer user_data)
1119 {
1120         struct cache_entry *entry = value;
1121
1122         /* first, delete any expired elements */
1123         cache_enforce_validity(entry);
1124
1125         /* if anything is not expired, mark the entry for refresh */
1126         if (entry->hits > 0 && (entry->ipv4 || entry->ipv6))
1127                 entry->want_refresh = 1;
1128
1129         /* delete the cached data */
1130         if (entry->ipv4) {
1131                 g_free(entry->ipv4->data);
1132                 g_free(entry->ipv4);
1133                 entry->ipv4 = NULL;
1134         }
1135
1136         if (entry->ipv6) {
1137                 g_free(entry->ipv6->data);
1138                 g_free(entry->ipv6);
1139                 entry->ipv6 = NULL;
1140         }
1141
1142         /* keep the entry if we want it refreshed, delete it otherwise */
1143         if (entry->want_refresh)
1144                 return FALSE;
1145         else
1146                 return TRUE;
1147 }
1148
1149 /*
1150  * cache_invalidate is called from places where the DNS landscape
1151  * has changed, say because connections are added or we entered a VPN.
1152  * The logic is to wipe all cache data, but mark all non-expired
1153  * parts of the cache for refresh rather than deleting the whole cache.
1154  */
1155 static void cache_invalidate(void)
1156 {
1157         DBG("Invalidating the DNS cache %p", cache);
1158
1159         if (cache == NULL)
1160                 return;
1161
1162         g_hash_table_foreach_remove(cache, cache_invalidate_entry, NULL);
1163 }
1164
1165 static void cache_refresh_entry(struct cache_entry *entry)
1166 {
1167
1168         cache_enforce_validity(entry);
1169
1170         if (entry->hits > 2 && entry->ipv4 == NULL)
1171                 entry->want_refresh = 1;
1172         if (entry->hits > 2 && entry->ipv6 == NULL)
1173                 entry->want_refresh = 1;
1174
1175         if (entry->want_refresh) {
1176                 char *c;
1177                 char dns_name[NS_MAXDNAME + 1];
1178                 entry->want_refresh = 0;
1179
1180                 /* turn a DNS name into a hostname with dots */
1181                 strncpy(dns_name, entry->key, NS_MAXDNAME);
1182                 c = dns_name;
1183                 while (c && *c) {
1184                         int jump;
1185                         jump = *c;
1186                         *c = '.';
1187                         c += jump + 1;
1188                 }
1189                 DBG("Refreshing %s\n", dns_name);
1190                 /* then refresh the hostname */
1191                 refresh_dns_entry(entry, &dns_name[1]);
1192         }
1193 }
1194
1195 static void cache_refresh_iterator(gpointer key, gpointer value,
1196                                         gpointer user_data)
1197 {
1198         struct cache_entry *entry = value;
1199
1200         cache_refresh_entry(entry);
1201 }
1202
1203 static void cache_refresh(void)
1204 {
1205         if (cache == NULL)
1206                 return;
1207
1208         g_hash_table_foreach(cache, cache_refresh_iterator, NULL);
1209 }
1210
1211 static int reply_query_type(unsigned char *msg, int len)
1212 {
1213         unsigned char *c;
1214         uint16_t *w;
1215         int l;
1216         int type;
1217
1218         /* skip the header */
1219         c = msg + sizeof(struct domain_hdr);
1220         len -= sizeof(struct domain_hdr);
1221
1222         if (len < 0)
1223                 return 0;
1224
1225         /* now the query, which is a name and 2 16 bit words */
1226         l = dns_name_length(c) + 1;
1227         c += l;
1228         w = (uint16_t *) c;
1229         type = ntohs(*w);
1230
1231         return type;
1232 }
1233
1234 static int cache_update(struct server_data *srv, unsigned char *msg,
1235                         unsigned int msg_len)
1236 {
1237         int offset = protocol_offset(srv->protocol);
1238         int err, qlen, ttl = 0;
1239         uint16_t answers = 0, type = 0, class = 0;
1240         struct domain_question *q;
1241         struct cache_entry *entry;
1242         struct cache_data *data;
1243         char question[NS_MAXDNAME + 1];
1244         unsigned char response[NS_MAXDNAME + 1];
1245         unsigned char *ptr;
1246         unsigned int rsplen;
1247         gboolean new_entry = TRUE;
1248         time_t current_time;
1249
1250         if (cache_size >= MAX_CACHE_SIZE) {
1251                 cache_cleanup();
1252                 if (cache_size >= MAX_CACHE_SIZE)
1253                         return 0;
1254         }
1255
1256         current_time = time(NULL);
1257
1258         /* don't do a cache refresh more than twice a minute */
1259         if (next_refresh < current_time) {
1260                 cache_refresh();
1261                 next_refresh = current_time + 30;
1262         }
1263
1264
1265         /* Continue only if response code is 0 (=ok) */
1266         if (msg[3] & 0x0f)
1267                 return 0;
1268
1269         if (offset < 0)
1270                 return 0;
1271
1272         rsplen = sizeof(response) - 1;
1273         question[sizeof(question) - 1] = '\0';
1274
1275         err = parse_response(msg + offset, msg_len - offset,
1276                                 question, sizeof(question) - 1,
1277                                 &type, &class, &ttl,
1278                                 response, &rsplen, &answers);
1279
1280         /*
1281          * special case: if we do a ipv6 lookup and get no result
1282          * for a record that's already in our ipv4 cache.. we want
1283          * to cache the negative response.
1284          */
1285         if ((err == -ENOMSG || err == -ENOBUFS) &&
1286                         reply_query_type(msg + offset,
1287                                         msg_len - offset) == 28) {
1288                 entry = g_hash_table_lookup(cache, question);
1289                 if (entry && entry->ipv4 && entry->ipv6 == NULL) {
1290                         int cache_offset = 0;
1291
1292                         data = g_try_new(struct cache_data, 1);
1293                         if (data == NULL)
1294                                 return -ENOMEM;
1295                         data->inserted = entry->ipv4->inserted;
1296                         data->type = type;
1297                         data->answers = msg[5];
1298                         data->timeout = entry->ipv4->timeout;
1299                         if (srv->protocol == IPPROTO_UDP)
1300                                 cache_offset = 2;
1301                         data->data_len = msg_len + cache_offset;
1302                         data->data = ptr = g_malloc(data->data_len);
1303                         ptr[0] = (data->data_len - 2) / 256;
1304                         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1305                         if (srv->protocol == IPPROTO_UDP)
1306                                 ptr += 2;
1307                         data->valid_until = entry->ipv4->valid_until;
1308                         data->cache_until = entry->ipv4->cache_until;
1309                         memcpy(ptr, msg, msg_len);
1310                         entry->ipv6 = data;
1311                         /*
1312                          * we will get a "hit" when we serve the response
1313                          * out of the cache
1314                          */
1315                         entry->hits--;
1316                         if (entry->hits < 0)
1317                                 entry->hits = 0;
1318                         return 0;
1319                 }
1320         }
1321
1322         if (err < 0 || ttl == 0)
1323                 return 0;
1324
1325         qlen = strlen(question);
1326
1327         /*
1328          * If the cache contains already data, check if the
1329          * type of the cached data is the same and do not add
1330          * to cache if data is already there.
1331          * This is needed so that we can cache both A and AAAA
1332          * records for the same name.
1333          */
1334         entry = g_hash_table_lookup(cache, question);
1335         if (entry == NULL) {
1336                 entry = g_try_new(struct cache_entry, 1);
1337                 if (entry == NULL)
1338                         return -ENOMEM;
1339
1340                 data = g_try_new(struct cache_data, 1);
1341                 if (data == NULL) {
1342                         g_free(entry);
1343                         return -ENOMEM;
1344                 }
1345
1346                 entry->key = g_strdup(question);
1347                 entry->ipv4 = entry->ipv6 = NULL;
1348                 entry->want_refresh = 0;
1349                 entry->hits = 0;
1350
1351                 if (type == 1)
1352                         entry->ipv4 = data;
1353                 else
1354                         entry->ipv6 = data;
1355         } else {
1356                 if (type == 1 && entry->ipv4 != NULL)
1357                         return 0;
1358
1359                 if (type == 28 && entry->ipv6 != NULL)
1360                         return 0;
1361
1362                 data = g_try_new(struct cache_data, 1);
1363                 if (data == NULL)
1364                         return -ENOMEM;
1365
1366                 if (type == 1)
1367                         entry->ipv4 = data;
1368                 else
1369                         entry->ipv6 = data;
1370
1371                 /*
1372                  * compensate for the hit we'll get for serving
1373                  * the response out of the cache
1374                  */
1375                 entry->hits--;
1376                 if (entry->hits < 0)
1377                         entry->hits = 0;
1378
1379                 new_entry = FALSE;
1380         }
1381
1382         if (ttl < MIN_CACHE_TTL)
1383                 ttl = MIN_CACHE_TTL;
1384
1385         data->inserted = current_time;
1386         data->type = type;
1387         data->answers = answers;
1388         data->timeout = ttl;
1389         /*
1390          * The "2" in start of the length is the TCP offset. We allocate it
1391          * here even for UDP packet because it simplifies the sending
1392          * of cached packet.
1393          */
1394         data->data_len = 2 + 12 + qlen + 1 + 2 + 2 + rsplen;
1395         data->data = ptr = g_malloc(data->data_len);
1396         data->valid_until = current_time + ttl;
1397
1398         /*
1399          * Restrict the cached DNS record TTL to some sane value
1400          * in order to prevent data staying in the cache too long.
1401          */
1402         if (ttl > MAX_CACHE_TTL)
1403                 ttl = MAX_CACHE_TTL;
1404
1405         data->cache_until = round_down_ttl(current_time + ttl, ttl);
1406
1407         if (data->data == NULL) {
1408                 g_free(entry->key);
1409                 g_free(data);
1410                 g_free(entry);
1411                 return -ENOMEM;
1412         }
1413
1414         /*
1415          * We cache the two extra bytes at the start of the message
1416          * in a TCP packet. When sending UDP packet, we skip the first
1417          * two bytes. This way we do not need to know the format
1418          * (UDP/TCP) of the cached message.
1419          */
1420         ptr[0] = (data->data_len - 2) / 256;
1421         ptr[1] = (data->data_len - 2) - ptr[0] * 256;
1422         if (srv->protocol == IPPROTO_UDP)
1423                 ptr += 2;
1424
1425         memcpy(ptr, msg, offset + 12);
1426         memcpy(ptr + offset + 12, question, qlen + 1); /* copy also the \0 */
1427
1428         q = (void *) (ptr + offset + 12 + qlen + 1);
1429         q->type = htons(type);
1430         q->class = htons(class);
1431         memcpy(ptr + offset + 12 + qlen + 1 + sizeof(struct domain_question),
1432                 response, rsplen);
1433
1434         if (new_entry == TRUE) {
1435                 g_hash_table_replace(cache, entry->key, entry);
1436                 cache_size++;
1437         }
1438
1439         DBG("cache %d %squestion \"%s\" type %d ttl %d size %zd packet %u "
1440                                                                 "dns len %u",
1441                 cache_size, new_entry ? "new " : "old ",
1442                 question, type, ttl,
1443                 sizeof(*entry) + sizeof(*data) + data->data_len + qlen,
1444                 data->data_len,
1445                 srv->protocol == IPPROTO_TCP ?
1446                         (unsigned int)(data->data[0] * 256 + data->data[1]) :
1447                         data->data_len);
1448
1449         return 0;
1450 }
1451
1452 static int ns_resolv(struct server_data *server, struct request_data *req,
1453                                 gpointer request, gpointer name)
1454 {
1455         GList *list;
1456         int sk, err, type = 0;
1457         char *dot, *lookup = (char *) name;
1458         struct cache_entry *entry;
1459
1460         entry = cache_check(request, &type);
1461         if (entry != NULL) {
1462                 int ttl_left = 0;
1463                 struct cache_data *data;
1464
1465                 DBG("cache hit %s type %s", lookup, type == 1 ? "A" : "AAAA");
1466                 if (type == 1)
1467                         data = entry->ipv4;
1468                 else
1469                         data = entry->ipv6;
1470
1471                 if (data) {
1472                         ttl_left = data->valid_until - time(NULL);
1473                         entry->hits++;
1474                 }
1475
1476                 if (data != NULL && req->protocol == IPPROTO_TCP) {
1477                         send_cached_response(req->client_sk, data->data,
1478                                         data->data_len, NULL, 0, IPPROTO_TCP,
1479                                         req->srcid, data->answers, ttl_left);
1480                         return 1;
1481                 }
1482
1483                 if (data != NULL && req->protocol == IPPROTO_UDP) {
1484                         int sk;
1485                         sk = g_io_channel_unix_get_fd(
1486                                         req->ifdata->udp_listener_channel);
1487
1488                         send_cached_response(sk, data->data,
1489                                 data->data_len, &req->sa, req->sa_len,
1490                                 IPPROTO_UDP, req->srcid, data->answers,
1491                                 ttl_left);
1492                         return 1;
1493                 }
1494         }
1495
1496         sk = g_io_channel_unix_get_fd(server->channel);
1497
1498         err = send(sk, request, req->request_len, MSG_NOSIGNAL);
1499         if (err < 0)
1500                 return -EIO;
1501
1502         req->numserv++;
1503
1504         /* If we have more than one dot, we don't add domains */
1505         dot = strchr(lookup, '.');
1506         if (dot != NULL && dot != lookup + strlen(lookup) - 1)
1507                 return 0;
1508
1509         if (server->domains != NULL && server->domains->data != NULL)
1510                 req->append_domain = TRUE;
1511
1512         for (list = server->domains; list; list = list->next) {
1513                 char *domain;
1514                 unsigned char alt[1024];
1515                 struct domain_hdr *hdr = (void *) &alt;
1516                 int altlen, domlen, offset;
1517
1518                 domain = list->data;
1519
1520                 if (domain == NULL)
1521                         continue;
1522
1523                 offset = protocol_offset(server->protocol);
1524                 if (offset < 0)
1525                         return offset;
1526
1527                 domlen = strlen(domain) + 1;
1528                 if (domlen < 5)
1529                         return -EINVAL;
1530
1531                 alt[offset] = req->altid & 0xff;
1532                 alt[offset + 1] = req->altid >> 8;
1533
1534                 memcpy(alt + offset + 2, request + offset + 2, 10);
1535                 hdr->qdcount = htons(1);
1536
1537                 altlen = append_query(alt + offset + 12, sizeof(alt) - 12,
1538                                         name, domain);
1539                 if (altlen < 0)
1540                         return -EINVAL;
1541
1542                 altlen += 12;
1543
1544                 memcpy(alt + offset + altlen,
1545                         request + offset + altlen - domlen,
1546                                 req->request_len - altlen - offset + domlen);
1547
1548                 if (server->protocol == IPPROTO_TCP) {
1549                         int req_len = req->request_len + domlen - 2;
1550
1551                         alt[0] = (req_len >> 8) & 0xff;
1552                         alt[1] = req_len & 0xff;
1553                 }
1554
1555                 err = send(sk, alt, req->request_len + domlen, MSG_NOSIGNAL);
1556                 if (err < 0)
1557                         return -EIO;
1558
1559                 req->numserv++;
1560         }
1561
1562         return 0;
1563 }
1564
1565 static void destroy_request_data(struct request_data *req)
1566 {
1567         if (req->timeout > 0)
1568                 g_source_remove(req->timeout);
1569
1570         g_free(req->resp);
1571         g_free(req->request);
1572         g_free(req->name);
1573         g_free(req);
1574 }
1575
1576 static int forward_dns_reply(unsigned char *reply, int reply_len, int protocol,
1577                                 struct server_data *data)
1578 {
1579         struct domain_hdr *hdr;
1580         struct request_data *req;
1581         int dns_id, sk, err, offset = protocol_offset(protocol);
1582         struct listener_data *ifdata;
1583
1584         if (offset < 0)
1585                 return offset;
1586
1587         hdr = (void *)(reply + offset);
1588         dns_id = reply[offset] | reply[offset + 1] << 8;
1589
1590         DBG("Received %d bytes (id 0x%04x)", reply_len, dns_id);
1591
1592         req = find_request(dns_id);
1593         if (req == NULL)
1594                 return -EINVAL;
1595
1596         DBG("id 0x%04x rcode %d", hdr->id, hdr->rcode);
1597
1598         ifdata = req->ifdata;
1599
1600         reply[offset] = req->srcid & 0xff;
1601         reply[offset + 1] = req->srcid >> 8;
1602
1603         req->numresp++;
1604
1605         if (hdr->rcode == 0 || req->resp == NULL) {
1606
1607                 /*
1608                  * If the domain name was append
1609                  * remove it before forwarding the reply.
1610                  */
1611                 if (req->append_domain == TRUE) {
1612                         unsigned char *ptr;
1613                         uint8_t host_len;
1614                         unsigned int domain_len;
1615
1616                         /*
1617                          * ptr points to the first char of the hostname.
1618                          * ->hostname.domain.net
1619                          */
1620                         ptr = reply + offset + sizeof(struct domain_hdr);
1621                         host_len = *ptr;
1622                         domain_len = strlen((const char *)ptr + host_len + 1);
1623
1624                         /*
1625                          * Remove the domain name and replace it by the end
1626                          * of reply. Check if the domain is really there
1627                          * before trying to copy the data. The domain_len can
1628                          * be 0 because if the original query did not contain
1629                          * a domain name, then we are sending two packets,
1630                          * first without the domain name and the second packet
1631                          * with domain name. The append_domain is set to true
1632                          * even if we sent the first packet without domain
1633                          * name. In this case we end up in this branch.
1634                          */
1635                         if (domain_len > 0) {
1636                                 /*
1637                                  * Note that we must use memmove() here,
1638                                  * because the memory areas can overlap.
1639                                  */
1640                                 memmove(ptr + host_len + 1,
1641                                         ptr + host_len + domain_len + 1,
1642                                         reply_len - (ptr - reply + domain_len));
1643
1644                                 reply_len = reply_len - domain_len;
1645                         }
1646                 }
1647
1648                 g_free(req->resp);
1649                 req->resplen = 0;
1650
1651                 req->resp = g_try_malloc(reply_len);
1652                 if (req->resp == NULL)
1653                         return -ENOMEM;
1654
1655                 memcpy(req->resp, reply, reply_len);
1656                 req->resplen = reply_len;
1657
1658                 cache_update(data, reply, reply_len);
1659         }
1660
1661         if (hdr->rcode > 0 && req->numresp < req->numserv)
1662                 return -EINVAL;
1663
1664         request_list = g_slist_remove(request_list, req);
1665
1666         if (protocol == IPPROTO_UDP) {
1667                 sk = g_io_channel_unix_get_fd(ifdata->udp_listener_channel);
1668                 err = sendto(sk, req->resp, req->resplen, 0,
1669                              &req->sa, req->sa_len);
1670         } else {
1671                 sk = req->client_sk;
1672                 err = send(sk, req->resp, req->resplen, MSG_NOSIGNAL);
1673                 close(sk);
1674         }
1675
1676         destroy_request_data(req);
1677
1678         return err;
1679 }
1680
1681 static void cache_element_destroy(gpointer value)
1682 {
1683         struct cache_entry *entry = value;
1684
1685         if (entry == NULL)
1686                 return;
1687
1688         if (entry->ipv4 != NULL) {
1689                 g_free(entry->ipv4->data);
1690                 g_free(entry->ipv4);
1691         }
1692
1693         if (entry->ipv6 != NULL) {
1694                 g_free(entry->ipv6->data);
1695                 g_free(entry->ipv6);
1696         }
1697
1698         g_free(entry->key);
1699         g_free(entry);
1700
1701         if (--cache_size < 0)
1702                 cache_size = 0;
1703 }
1704
1705 static gboolean try_remove_cache(gpointer user_data)
1706 {
1707         if (__sync_fetch_and_sub(&cache_refcount, 1) == 1) {
1708                 DBG("No cache users, removing it.");
1709
1710                 g_hash_table_destroy(cache);
1711                 cache = NULL;
1712         }
1713
1714         return FALSE;
1715 }
1716
1717 static void destroy_server(struct server_data *server)
1718 {
1719         GList *list;
1720
1721         DBG("interface %s server %s", server->interface, server->server);
1722
1723         server_list = g_slist_remove(server_list, server);
1724
1725         if (server->watch > 0)
1726                 g_source_remove(server->watch);
1727
1728         if (server->timeout > 0)
1729                 g_source_remove(server->timeout);
1730
1731         g_io_channel_unref(server->channel);
1732
1733         if (server->protocol == IPPROTO_UDP)
1734                 DBG("Removing DNS server %s", server->server);
1735
1736         g_free(server->incoming_reply);
1737         g_free(server->server);
1738         for (list = server->domains; list; list = list->next) {
1739                 char *domain = list->data;
1740
1741                 server->domains = g_list_remove(server->domains, domain);
1742                 g_free(domain);
1743         }
1744         g_free(server->interface);
1745
1746         /*
1747          * We do not remove cache right away but delay it few seconds.
1748          * The idea is that when IPv6 DNS server is added via RDNSS, it has a
1749          * lifetime. When the lifetime expires we decrease the refcount so it
1750          * is possible that the cache is then removed. Because a new DNS server
1751          * is usually created almost immediately we would then loose the cache
1752          * without any good reason. The small delay allows the new RDNSS to
1753          * create a new DNS server instance and the refcount does not go to 0.
1754          */
1755         g_timeout_add_seconds(3, try_remove_cache, NULL);
1756
1757         g_free(server);
1758 }
1759
1760 static gboolean udp_server_event(GIOChannel *channel, GIOCondition condition,
1761                                                         gpointer user_data)
1762 {
1763         unsigned char buf[4096];
1764         int sk, err, len;
1765         struct server_data *data = user_data;
1766
1767         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1768                 connman_error("Error with UDP server %s", data->server);
1769                 data->watch = 0;
1770                 return FALSE;
1771         }
1772
1773         sk = g_io_channel_unix_get_fd(channel);
1774
1775         len = recv(sk, buf, sizeof(buf), 0);
1776         if (len < 12)
1777                 return TRUE;
1778
1779         err = forward_dns_reply(buf, len, IPPROTO_UDP, data);
1780         if (err < 0)
1781                 return TRUE;
1782
1783         return TRUE;
1784 }
1785
1786 static gboolean tcp_server_event(GIOChannel *channel, GIOCondition condition,
1787                                                         gpointer user_data)
1788 {
1789         int sk;
1790         struct server_data *server = user_data;
1791
1792         sk = g_io_channel_unix_get_fd(channel);
1793         if (sk == 0)
1794                 return FALSE;
1795
1796         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
1797                 GSList *list;
1798 hangup:
1799                 DBG("TCP server channel closed");
1800
1801                 /*
1802                  * Discard any partial response which is buffered; better
1803                  * to get a proper response from a working server.
1804                  */
1805                 g_free(server->incoming_reply);
1806                 server->incoming_reply = NULL;
1807
1808                 for (list = request_list; list; list = list->next) {
1809                         struct request_data *req = list->data;
1810                         struct domain_hdr *hdr;
1811
1812                         if (req->protocol == IPPROTO_UDP)
1813                                 continue;
1814
1815                         if (req->request == NULL)
1816                                 continue;
1817
1818                         /*
1819                          * If we're not waiting for any further response
1820                          * from another name server, then we send an error
1821                          * response to the client.
1822                          */
1823                         if (req->numserv && --(req->numserv))
1824                                 continue;
1825
1826                         hdr = (void *) (req->request + 2);
1827                         hdr->id = req->srcid;
1828                         send_response(req->client_sk, req->request,
1829                                 req->request_len, NULL, 0, IPPROTO_TCP);
1830
1831                         request_list = g_slist_remove(request_list, req);
1832                 }
1833
1834                 destroy_server(server);
1835
1836                 return FALSE;
1837         }
1838
1839         if ((condition & G_IO_OUT) && !server->connected) {
1840                 GSList *list;
1841                 GList *domains;
1842                 int no_request_sent = TRUE;
1843                 struct server_data *udp_server;
1844
1845                 udp_server = find_server(server->interface, server->server,
1846                                                                 IPPROTO_UDP);
1847                 if (udp_server != NULL) {
1848                         for (domains = udp_server->domains; domains;
1849                                                 domains = domains->next) {
1850                                 char *dom = domains->data;
1851
1852                                 DBG("Adding domain %s to %s",
1853                                                 dom, server->server);
1854
1855                                 server->domains = g_list_append(server->domains,
1856                                                                 g_strdup(dom));
1857                         }
1858                 }
1859
1860                 server->connected = TRUE;
1861                 server_list = g_slist_append(server_list, server);
1862
1863                 if (server->timeout > 0) {
1864                         g_source_remove(server->timeout);
1865                         server->timeout = 0;
1866                 }
1867
1868                 for (list = request_list; list; ) {
1869                         struct request_data *req = list->data;
1870                         int status;
1871
1872                         if (req->protocol == IPPROTO_UDP) {
1873                                 list = list->next;
1874                                 continue;
1875                         }
1876
1877                         DBG("Sending req %s over TCP", (char *)req->name);
1878
1879                         status = ns_resolv(server, req,
1880                                                 req->request, req->name);
1881                         if (status > 0) {
1882                                 /*
1883                                  * A cached result was sent,
1884                                  * so the request can be released
1885                                  */
1886                                 list = list->next;
1887                                 request_list = g_slist_remove(request_list, req);
1888                                 destroy_request_data(req);
1889                                 continue;
1890                         }
1891
1892                         if (status < 0) {
1893                                 list = list->next;
1894                                 continue;
1895                         }
1896
1897                         no_request_sent = FALSE;
1898
1899                         if (req->timeout > 0)
1900                                 g_source_remove(req->timeout);
1901
1902                         req->timeout = g_timeout_add_seconds(30,
1903                                                 request_timeout, req);
1904                         list = list->next;
1905                 }
1906
1907                 if (no_request_sent == TRUE) {
1908                         destroy_server(server);
1909                         return FALSE;
1910                 }
1911
1912         } else if (condition & G_IO_IN) {
1913                 struct partial_reply *reply = server->incoming_reply;
1914                 int bytes_recv;
1915
1916                 if (!reply) {
1917                         unsigned char reply_len_buf[2];
1918                         uint16_t reply_len;
1919
1920                         bytes_recv = recv(sk, reply_len_buf, 2, MSG_PEEK);
1921                         if (!bytes_recv) {
1922                                 goto hangup;
1923                         } else if (bytes_recv < 0) {
1924                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1925                                         return TRUE;
1926
1927                                 connman_error("DNS proxy error %s",
1928                                                 strerror(errno));
1929                                 goto hangup;
1930                         } else if (bytes_recv < 2)
1931                                 return TRUE;
1932
1933                         reply_len = reply_len_buf[1] | reply_len_buf[0] << 8;
1934                         reply_len += 2;
1935
1936                         DBG("TCP reply %d bytes", reply_len);
1937
1938                         reply = g_try_malloc(sizeof(*reply) + reply_len + 2);
1939                         if (!reply)
1940                                 return TRUE;
1941
1942                         reply->len = reply_len;
1943                         reply->received = 0;
1944
1945                         server->incoming_reply = reply;
1946                 }
1947
1948                 while (reply->received < reply->len) {
1949                         bytes_recv = recv(sk, reply->buf + reply->received,
1950                                         reply->len - reply->received, 0);
1951                         if (!bytes_recv) {
1952                                 connman_error("DNS proxy TCP disconnect");
1953                                 break;
1954                         } else if (bytes_recv < 0) {
1955                                 if (errno == EAGAIN || errno == EWOULDBLOCK)
1956                                         return TRUE;
1957
1958                                 connman_error("DNS proxy error %s",
1959                                                 strerror(errno));
1960                                 break;
1961                         }
1962                         reply->received += bytes_recv;
1963                 }
1964
1965                 forward_dns_reply(reply->buf, reply->received, IPPROTO_TCP,
1966                                         server);
1967
1968                 g_free(reply);
1969                 server->incoming_reply = NULL;
1970
1971                 destroy_server(server);
1972
1973                 return FALSE;
1974         }
1975
1976         return TRUE;
1977 }
1978
1979 static gboolean tcp_idle_timeout(gpointer user_data)
1980 {
1981         struct server_data *server = user_data;
1982
1983         DBG("");
1984
1985         if (server == NULL)
1986                 return FALSE;
1987
1988         destroy_server(server);
1989
1990         return FALSE;
1991 }
1992
1993 static struct server_data *create_server(const char *interface,
1994                                         const char *domain, const char *server,
1995                                         int protocol)
1996 {
1997         struct addrinfo hints, *rp;
1998         struct server_data *data;
1999         int sk, ret;
2000
2001         DBG("interface %s server %s", interface, server);
2002
2003         memset(&hints, 0, sizeof(hints));
2004
2005         switch (protocol) {
2006         case IPPROTO_UDP:
2007                 hints.ai_socktype = SOCK_DGRAM;
2008                 break;
2009
2010         case IPPROTO_TCP:
2011                 hints.ai_socktype = SOCK_STREAM;
2012                 break;
2013
2014         default:
2015                 return NULL;
2016         }
2017         hints.ai_family = AF_UNSPEC;
2018         hints.ai_flags = AI_PASSIVE | AI_NUMERICSERV | AI_NUMERICHOST;
2019
2020         ret = getaddrinfo(server, "53", &hints, &rp);
2021         if (ret) {
2022                 connman_error("Failed to parse server %s address: %s\n",
2023                               server, gai_strerror(ret));
2024                 return NULL;
2025         }
2026         /* Do not blindly copy this code elsewhere; it doesn't loop over the
2027            results using ->ai_next as it should. That's OK in *this* case
2028            because it was a numeric lookup; we *know* there's only one. */
2029
2030         sk = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
2031         if (sk < 0) {
2032                 connman_error("Failed to create server %s socket", server);
2033                 freeaddrinfo(rp);
2034                 return NULL;
2035         }
2036
2037         if (interface != NULL) {
2038                 if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2039                                 interface, strlen(interface) + 1) < 0) {
2040                         connman_error("Failed to bind server %s "
2041                                                 "to interface %s",
2042                                                         server, interface);
2043                         freeaddrinfo(rp);
2044                         close(sk);
2045                         return NULL;
2046                 }
2047         }
2048
2049         data = g_try_new0(struct server_data, 1);
2050         if (data == NULL) {
2051                 connman_error("Failed to allocate server %s data", server);
2052                 freeaddrinfo(rp);
2053                 close(sk);
2054                 return NULL;
2055         }
2056
2057         data->channel = g_io_channel_unix_new(sk);
2058         if (data->channel == NULL) {
2059                 connman_error("Failed to create server %s channel", server);
2060                 freeaddrinfo(rp);
2061                 close(sk);
2062                 g_free(data);
2063                 return NULL;
2064         }
2065
2066         g_io_channel_set_close_on_unref(data->channel, TRUE);
2067
2068         if (protocol == IPPROTO_TCP) {
2069                 g_io_channel_set_flags(data->channel, G_IO_FLAG_NONBLOCK, NULL);
2070                 data->watch = g_io_add_watch(data->channel,
2071                         G_IO_OUT | G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR,
2072                                                 tcp_server_event, data);
2073                 data->timeout = g_timeout_add_seconds(30, tcp_idle_timeout,
2074                                                                 data);
2075         } else
2076                 data->watch = g_io_add_watch(data->channel,
2077                         G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2078                                                 udp_server_event, data);
2079
2080         data->interface = g_strdup(interface);
2081         if (domain)
2082                 data->domains = g_list_append(data->domains, g_strdup(domain));
2083         data->server = g_strdup(server);
2084         data->protocol = protocol;
2085
2086         ret = connect(sk, rp->ai_addr, rp->ai_addrlen);
2087         freeaddrinfo(rp);
2088         if (ret < 0) {
2089                 if ((protocol == IPPROTO_TCP && errno != EINPROGRESS) ||
2090                                 protocol == IPPROTO_UDP) {
2091                         GList *list;
2092
2093                         connman_error("Failed to connect to server %s", server);
2094                         if (data->watch > 0)
2095                                 g_source_remove(data->watch);
2096                         if (data->timeout > 0)
2097                                 g_source_remove(data->timeout);
2098
2099                         g_io_channel_unref(data->channel);
2100                         close(sk);
2101
2102                         g_free(data->server);
2103                         g_free(data->interface);
2104                         for (list = data->domains; list; list = list->next) {
2105                                 char *domain = list->data;
2106
2107                                 data->domains = g_list_remove(data->domains,
2108                                                                         domain);
2109                                 g_free(domain);
2110                         }
2111                         g_free(data);
2112                         return NULL;
2113                 }
2114         }
2115
2116         if (__sync_fetch_and_add(&cache_refcount, 1) == 0)
2117                 cache = g_hash_table_new_full(g_str_hash,
2118                                         g_str_equal,
2119                                         NULL,
2120                                         cache_element_destroy);
2121
2122         if (protocol == IPPROTO_UDP) {
2123                 /* Enable new servers by default */
2124                 data->enabled = TRUE;
2125                 DBG("Adding DNS server %s", data->server);
2126
2127                 server_list = g_slist_append(server_list, data);
2128         }
2129
2130         return data;
2131 }
2132
2133 static gboolean resolv(struct request_data *req,
2134                                 gpointer request, gpointer name)
2135 {
2136         GSList *list;
2137
2138         for (list = server_list; list; list = list->next) {
2139                 struct server_data *data = list->data;
2140
2141                 DBG("server %s enabled %d", data->server, data->enabled);
2142
2143                 if (data->enabled == FALSE)
2144                         continue;
2145
2146                 if (data->watch == 0 && data->protocol == IPPROTO_UDP)
2147                         data->watch = g_io_add_watch(data->channel,
2148                                 G_IO_IN | G_IO_NVAL | G_IO_ERR | G_IO_HUP,
2149                                                 udp_server_event, data);
2150
2151                 if (ns_resolv(data, req, request, name) > 0)
2152                         return TRUE;
2153         }
2154
2155         return FALSE;
2156 }
2157
2158 static void append_domain(const char *interface, const char *domain)
2159 {
2160         GSList *list;
2161
2162         DBG("interface %s domain %s", interface, domain);
2163
2164         if (domain == NULL)
2165                 return;
2166
2167         for (list = server_list; list; list = list->next) {
2168                 struct server_data *data = list->data;
2169                 GList *dom_list;
2170                 char *dom;
2171                 gboolean dom_found = FALSE;
2172
2173                 if (data->interface == NULL)
2174                         continue;
2175
2176                 if (g_str_equal(data->interface, interface) == FALSE)
2177                         continue;
2178
2179                 for (dom_list = data->domains; dom_list;
2180                                 dom_list = dom_list->next) {
2181                         dom = dom_list->data;
2182
2183                         if (g_str_equal(dom, domain)) {
2184                                 dom_found = TRUE;
2185                                 break;
2186                         }
2187                 }
2188
2189                 if (dom_found == FALSE) {
2190                         data->domains =
2191                                 g_list_append(data->domains, g_strdup(domain));
2192                 }
2193         }
2194 }
2195
2196 int __connman_dnsproxy_append(const char *interface, const char *domain,
2197                                                         const char *server)
2198 {
2199         struct server_data *data;
2200
2201         DBG("interface %s server %s", interface, server);
2202
2203         if (server == NULL && domain == NULL)
2204                 return -EINVAL;
2205
2206         if (server == NULL) {
2207                 append_domain(interface, domain);
2208
2209                 return 0;
2210         }
2211
2212         if (g_str_equal(server, "127.0.0.1") == TRUE)
2213                 return -ENODEV;
2214
2215         data = find_server(interface, server, IPPROTO_UDP);
2216         if (data != NULL) {
2217                 append_domain(interface, domain);
2218                 return 0;
2219         }
2220
2221         data = create_server(interface, domain, server, IPPROTO_UDP);
2222         if (data == NULL)
2223                 return -EIO;
2224
2225         return 0;
2226 }
2227
2228 static void remove_server(const char *interface, const char *domain,
2229                         const char *server, int protocol)
2230 {
2231         struct server_data *data;
2232
2233         data = find_server(interface, server, protocol);
2234         if (data == NULL)
2235                 return;
2236
2237         destroy_server(data);
2238 }
2239
2240 int __connman_dnsproxy_remove(const char *interface, const char *domain,
2241                                                         const char *server)
2242 {
2243         DBG("interface %s server %s", interface, server);
2244
2245         if (server == NULL)
2246                 return -EINVAL;
2247
2248         if (g_str_equal(server, "127.0.0.1") == TRUE)
2249                 return -ENODEV;
2250
2251         remove_server(interface, domain, server, IPPROTO_UDP);
2252         remove_server(interface, domain, server, IPPROTO_TCP);
2253
2254         return 0;
2255 }
2256
2257 void __connman_dnsproxy_flush(void)
2258 {
2259         GSList *list;
2260
2261         list = request_list;
2262         while (list) {
2263                 struct request_data *req = list->data;
2264
2265                 list = list->next;
2266
2267                 if (resolv(req, req->request, req->name) == TRUE) {
2268                         /*
2269                          * A cached result was sent,
2270                          * so the request can be released
2271                          */
2272                         request_list =
2273                                 g_slist_remove(request_list, req);
2274                         destroy_request_data(req);
2275                         continue;
2276                 }
2277
2278                 if (req->timeout > 0)
2279                         g_source_remove(req->timeout);
2280                 req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2281         }
2282 }
2283
2284 static void dnsproxy_offline_mode(connman_bool_t enabled)
2285 {
2286         GSList *list;
2287
2288         DBG("enabled %d", enabled);
2289
2290         for (list = server_list; list; list = list->next) {
2291                 struct server_data *data = list->data;
2292
2293                 if (enabled == FALSE) {
2294                         DBG("Enabling DNS server %s", data->server);
2295                         data->enabled = TRUE;
2296                         cache_invalidate();
2297                         cache_refresh();
2298                 } else {
2299                         DBG("Disabling DNS server %s", data->server);
2300                         data->enabled = FALSE;
2301                         cache_invalidate();
2302                 }
2303         }
2304 }
2305
2306 static void dnsproxy_default_changed(struct connman_service *service)
2307 {
2308         GSList *list;
2309         char *interface;
2310
2311         DBG("service %p", service);
2312
2313         /* DNS has changed, invalidate the cache */
2314         cache_invalidate();
2315
2316         if (service == NULL) {
2317                 /* When no services are active, then disable DNS proxying */
2318                 dnsproxy_offline_mode(TRUE);
2319                 return;
2320         }
2321
2322         interface = connman_service_get_interface(service);
2323         if (interface == NULL)
2324                 return;
2325
2326         for (list = server_list; list; list = list->next) {
2327                 struct server_data *data = list->data;
2328
2329                 if (g_strcmp0(data->interface, interface) == 0) {
2330                         DBG("Enabling DNS server %s", data->server);
2331                         data->enabled = TRUE;
2332                 } else {
2333                         DBG("Disabling DNS server %s", data->server);
2334                         data->enabled = FALSE;
2335                 }
2336         }
2337
2338         g_free(interface);
2339         cache_refresh();
2340 }
2341
2342 static struct connman_notifier dnsproxy_notifier = {
2343         .name                   = "dnsproxy",
2344         .default_changed        = dnsproxy_default_changed,
2345         .offline_mode           = dnsproxy_offline_mode,
2346 };
2347
2348 static unsigned char opt_edns0_type[2] = { 0x00, 0x29 };
2349
2350 static int parse_request(unsigned char *buf, int len,
2351                                         char *name, unsigned int size)
2352 {
2353         struct domain_hdr *hdr = (void *) buf;
2354         uint16_t qdcount = ntohs(hdr->qdcount);
2355         uint16_t arcount = ntohs(hdr->arcount);
2356         unsigned char *ptr;
2357         char *last_label = NULL;
2358         unsigned int remain, used = 0;
2359
2360         if (len < 12)
2361                 return -EINVAL;
2362
2363         DBG("id 0x%04x qr %d opcode %d qdcount %d arcount %d",
2364                                         hdr->id, hdr->qr, hdr->opcode,
2365                                                         qdcount, arcount);
2366
2367         if (hdr->qr != 0 || qdcount != 1)
2368                 return -EINVAL;
2369
2370         name[0] = '\0';
2371
2372         ptr = buf + sizeof(struct domain_hdr);
2373         remain = len - sizeof(struct domain_hdr);
2374
2375         while (remain > 0) {
2376                 uint8_t len = *ptr;
2377
2378                 if (len == 0x00) {
2379                         last_label = (char *) (ptr + 1);
2380                         break;
2381                 }
2382
2383                 if (used + len + 1 > size)
2384                         return -ENOBUFS;
2385
2386                 strncat(name, (char *) (ptr + 1), len);
2387                 strcat(name, ".");
2388
2389                 used += len + 1;
2390
2391                 ptr += len + 1;
2392                 remain -= len + 1;
2393         }
2394
2395         if (last_label && arcount && remain >= 9 && last_label[4] == 0 &&
2396                                 !memcmp(last_label + 5, opt_edns0_type, 2)) {
2397                 uint16_t edns0_bufsize;
2398
2399                 edns0_bufsize = last_label[7] << 8 | last_label[8];
2400
2401                 DBG("EDNS0 buffer size %u", edns0_bufsize);
2402
2403                 /* This is an evil hack until full TCP support has been
2404                  * implemented.
2405                  *
2406                  * Somtimes the EDNS0 request gets send with a too-small
2407                  * buffer size. Since glibc doesn't seem to crash when it
2408                  * gets a response biffer then it requested, just bump
2409                  * the buffer size up to 4KiB.
2410                  */
2411                 if (edns0_bufsize < 0x1000) {
2412                         last_label[7] = 0x10;
2413                         last_label[8] = 0x00;
2414                 }
2415         }
2416
2417         DBG("query %s", name);
2418
2419         return 0;
2420 }
2421
2422 static gboolean tcp_listener_event(GIOChannel *channel, GIOCondition condition,
2423                                                         gpointer user_data)
2424 {
2425         unsigned char buf[768];
2426         char query[512];
2427         struct request_data *req;
2428         int sk, client_sk, len, err;
2429         struct sockaddr_in6 client_addr;
2430         socklen_t client_addr_len = sizeof(client_addr);
2431         GSList *list;
2432         struct listener_data *ifdata = user_data;
2433         int waiting_for_connect = FALSE;
2434
2435         DBG("condition 0x%x", condition);
2436
2437         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2438                 if (ifdata->tcp_listener_watch > 0)
2439                         g_source_remove(ifdata->tcp_listener_watch);
2440                 ifdata->tcp_listener_watch = 0;
2441
2442                 connman_error("Error with TCP listener channel");
2443
2444                 return FALSE;
2445         }
2446
2447         sk = g_io_channel_unix_get_fd(channel);
2448
2449         client_sk = accept(sk, (void *)&client_addr, &client_addr_len);
2450         if (client_sk < 0) {
2451                 connman_error("Accept failure on TCP listener");
2452                 ifdata->tcp_listener_watch = 0;
2453                 return FALSE;
2454         }
2455
2456         len = recv(client_sk, buf, sizeof(buf), 0);
2457         if (len < 2)
2458                 return TRUE;
2459
2460         DBG("Received %d bytes (id 0x%04x)", len, buf[2] | buf[3] << 8);
2461
2462         err = parse_request(buf + 2, len - 2, query, sizeof(query));
2463         if (err < 0 || (g_slist_length(server_list) == 0)) {
2464                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2465                 return TRUE;
2466         }
2467
2468         req = g_try_new0(struct request_data, 1);
2469         if (req == NULL)
2470                 return TRUE;
2471
2472         memcpy(&req->sa, &client_addr, client_addr_len);
2473         req->sa_len = client_addr_len;
2474         req->client_sk = client_sk;
2475         req->protocol = IPPROTO_TCP;
2476
2477         req->srcid = buf[2] | (buf[3] << 8);
2478         req->dstid = get_id();
2479         req->altid = get_id();
2480         req->request_len = len;
2481
2482         buf[2] = req->dstid & 0xff;
2483         buf[3] = req->dstid >> 8;
2484
2485         req->numserv = 0;
2486         req->ifdata = (struct listener_data *) ifdata;
2487         req->append_domain = FALSE;
2488
2489         for (list = server_list; list; list = list->next) {
2490                 struct server_data *data = list->data;
2491
2492                 if (data->protocol != IPPROTO_UDP || data->enabled == FALSE)
2493                         continue;
2494
2495                 if(create_server(data->interface, NULL,
2496                                         data->server, IPPROTO_TCP) == NULL)
2497                         continue;
2498
2499                 waiting_for_connect = TRUE;
2500         }
2501
2502         if (waiting_for_connect == FALSE) {
2503                 /* No server is waiting for connect */
2504                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2505                 g_free(req);
2506                 return TRUE;
2507         }
2508
2509         /*
2510          * The server is not connected yet.
2511          * Copy the relevant buffers.
2512          * The request will actually be sent once we're
2513          * properly connected over TCP to the nameserver.
2514          */
2515         req->request = g_try_malloc0(req->request_len);
2516         if (req->request == NULL) {
2517                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2518                 g_free(req);
2519                 return TRUE;
2520         }
2521         memcpy(req->request, buf, req->request_len);
2522
2523         req->name = g_try_malloc0(sizeof(query));
2524         if (req->name == NULL) {
2525                 send_response(client_sk, buf, len, NULL, 0, IPPROTO_TCP);
2526                 g_free(req->request);
2527                 g_free(req);
2528                 return TRUE;
2529         }
2530         memcpy(req->name, query, sizeof(query));
2531
2532         req->timeout = g_timeout_add_seconds(30, request_timeout, req);
2533
2534         request_list = g_slist_append(request_list, req);
2535
2536         return TRUE;
2537 }
2538
2539 static gboolean udp_listener_event(GIOChannel *channel, GIOCondition condition,
2540                                                         gpointer user_data)
2541 {
2542         unsigned char buf[768];
2543         char query[512];
2544         struct request_data *req;
2545         struct sockaddr_in6 client_addr;
2546         socklen_t client_addr_len = sizeof(client_addr);
2547         int sk, err, len;
2548         struct listener_data *ifdata = user_data;
2549
2550         if (condition & (G_IO_NVAL | G_IO_ERR | G_IO_HUP)) {
2551                 connman_error("Error with UDP listener channel");
2552                 ifdata->udp_listener_watch = 0;
2553                 return FALSE;
2554         }
2555
2556         sk = g_io_channel_unix_get_fd(channel);
2557
2558         memset(&client_addr, 0, client_addr_len);
2559         len = recvfrom(sk, buf, sizeof(buf), 0, (void *)&client_addr,
2560                        &client_addr_len);
2561         if (len < 2)
2562                 return TRUE;
2563
2564         DBG("Received %d bytes (id 0x%04x)", len, buf[0] | buf[1] << 8);
2565
2566         err = parse_request(buf, len, query, sizeof(query));
2567         if (err < 0 || (g_slist_length(server_list) == 0)) {
2568                 send_response(sk, buf, len, (void *)&client_addr,
2569                                 client_addr_len, IPPROTO_UDP);
2570                 return TRUE;
2571         }
2572
2573         req = g_try_new0(struct request_data, 1);
2574         if (req == NULL)
2575                 return TRUE;
2576
2577         memcpy(&req->sa, &client_addr, client_addr_len);
2578         req->sa_len = client_addr_len;
2579         req->client_sk = 0;
2580         req->protocol = IPPROTO_UDP;
2581
2582         req->srcid = buf[0] | (buf[1] << 8);
2583         req->dstid = get_id();
2584         req->altid = get_id();
2585         req->request_len = len;
2586
2587         buf[0] = req->dstid & 0xff;
2588         buf[1] = req->dstid >> 8;
2589
2590         req->numserv = 0;
2591         req->ifdata = (struct listener_data *) ifdata;
2592         req->append_domain = FALSE;
2593
2594         if (resolv(req, buf, query) == TRUE) {
2595                 /* a cached result was sent, so the request can be released */
2596                 g_free(req);
2597                 return TRUE;
2598         }
2599
2600         req->timeout = g_timeout_add_seconds(5, request_timeout, req);
2601         request_list = g_slist_append(request_list, req);
2602
2603         return TRUE;
2604 }
2605
2606 static int create_dns_listener(int protocol, struct listener_data *ifdata)
2607 {
2608         GIOChannel *channel;
2609         const char *proto;
2610         union {
2611                 struct sockaddr sa;
2612                 struct sockaddr_in6 sin6;
2613                 struct sockaddr_in sin;
2614         } s;
2615         socklen_t slen;
2616         int sk, type, v6only = 0;
2617         int family = AF_INET6;
2618
2619
2620         DBG("interface %s", ifdata->ifname);
2621
2622         switch (protocol) {
2623         case IPPROTO_UDP:
2624                 proto = "UDP";
2625                 type = SOCK_DGRAM | SOCK_CLOEXEC;
2626                 break;
2627
2628         case IPPROTO_TCP:
2629                 proto = "TCP";
2630                 type = SOCK_STREAM | SOCK_CLOEXEC;
2631                 break;
2632
2633         default:
2634                 return -EINVAL;
2635         }
2636
2637         sk = socket(family, type, protocol);
2638         if (sk < 0 && family == AF_INET6 && errno == EAFNOSUPPORT) {
2639                 connman_error("No IPv6 support; DNS proxy listening only on Legacy IP");
2640                 family = AF_INET;
2641                 sk = socket(family, type, protocol);
2642         }
2643         if (sk < 0) {
2644                 connman_error("Failed to create %s listener socket", proto);
2645                 return -EIO;
2646         }
2647
2648         if (setsockopt(sk, SOL_SOCKET, SO_BINDTODEVICE,
2649                                         ifdata->ifname,
2650                                         strlen(ifdata->ifname) + 1) < 0) {
2651                 connman_error("Failed to bind %s listener interface", proto);
2652                 close(sk);
2653                 return -EIO;
2654         }
2655         /* Ensure it accepts Legacy IP connections too */
2656         if (family == AF_INET6 &&
2657                         setsockopt(sk, SOL_IPV6, IPV6_V6ONLY,
2658                                         &v6only, sizeof(v6only)) < 0) {
2659                 connman_error("Failed to clear V6ONLY on %s listener socket",
2660                               proto);
2661                 close(sk);
2662                 return -EIO;
2663         }
2664
2665         if (family == AF_INET) {
2666                 memset(&s.sin, 0, sizeof(s.sin));
2667                 s.sin.sin_family = AF_INET;
2668                 s.sin.sin_port = htons(53);
2669                 s.sin.sin_addr.s_addr = htonl(INADDR_ANY);
2670                 slen = sizeof(s.sin);
2671         } else {
2672                 memset(&s.sin6, 0, sizeof(s.sin6));
2673                 s.sin6.sin6_family = AF_INET6;
2674                 s.sin6.sin6_port = htons(53);
2675                 s.sin6.sin6_addr = in6addr_any;
2676                 slen = sizeof(s.sin6);
2677         }
2678
2679         if (bind(sk, &s.sa, slen) < 0) {
2680                 connman_error("Failed to bind %s listener socket", proto);
2681                 close(sk);
2682                 return -EIO;
2683         }
2684
2685         if (protocol == IPPROTO_TCP && listen(sk, 10) < 0) {
2686                 connman_error("Failed to listen on TCP socket");
2687                 close(sk);
2688                 return -EIO;
2689         }
2690
2691         channel = g_io_channel_unix_new(sk);
2692         if (channel == NULL) {
2693                 connman_error("Failed to create %s listener channel", proto);
2694                 close(sk);
2695                 return -EIO;
2696         }
2697
2698         g_io_channel_set_close_on_unref(channel, TRUE);
2699
2700         if (protocol == IPPROTO_TCP) {
2701                 ifdata->tcp_listener_channel = channel;
2702                 ifdata->tcp_listener_watch = g_io_add_watch(channel,
2703                                 G_IO_IN, tcp_listener_event, (gpointer) ifdata);
2704         } else {
2705                 ifdata->udp_listener_channel = channel;
2706                 ifdata->udp_listener_watch = g_io_add_watch(channel,
2707                                 G_IO_IN, udp_listener_event, (gpointer) ifdata);
2708         }
2709
2710         return 0;
2711 }
2712
2713 static void destroy_udp_listener(struct listener_data *ifdata)
2714 {
2715         DBG("interface %s", ifdata->ifname);
2716
2717         if (ifdata->udp_listener_watch > 0)
2718                 g_source_remove(ifdata->udp_listener_watch);
2719
2720         g_io_channel_unref(ifdata->udp_listener_channel);
2721 }
2722
2723 static void destroy_tcp_listener(struct listener_data *ifdata)
2724 {
2725         DBG("interface %s", ifdata->ifname);
2726
2727         if (ifdata->tcp_listener_watch > 0)
2728                 g_source_remove(ifdata->tcp_listener_watch);
2729
2730         g_io_channel_unref(ifdata->tcp_listener_channel);
2731 }
2732
2733 static int create_listener(struct listener_data *ifdata)
2734 {
2735         int err;
2736
2737         err = create_dns_listener(IPPROTO_UDP, ifdata);
2738         if (err < 0)
2739                 return err;
2740
2741         err = create_dns_listener(IPPROTO_TCP, ifdata);
2742         if (err < 0) {
2743                 destroy_udp_listener(ifdata);
2744                 return err;
2745         }
2746
2747         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2748                 __connman_resolvfile_append("lo", NULL, "127.0.0.1");
2749
2750         return 0;
2751 }
2752
2753 static void destroy_listener(struct listener_data *ifdata)
2754 {
2755         GSList *list;
2756
2757         if (g_strcmp0(ifdata->ifname, "lo") == 0)
2758                 __connman_resolvfile_remove("lo", NULL, "127.0.0.1");
2759
2760         for (list = request_list; list; list = list->next) {
2761                 struct request_data *req = list->data;
2762
2763                 DBG("Dropping request (id 0x%04x -> 0x%04x)",
2764                                                 req->srcid, req->dstid);
2765                 destroy_request_data(req);
2766                 list->data = NULL;
2767         }
2768
2769         g_slist_free(request_list);
2770         request_list = NULL;
2771
2772         destroy_tcp_listener(ifdata);
2773         destroy_udp_listener(ifdata);
2774 }
2775
2776 int __connman_dnsproxy_add_listener(const char *interface)
2777 {
2778         struct listener_data *ifdata;
2779         int err;
2780
2781         DBG("interface %s", interface);
2782
2783         if (g_hash_table_lookup(listener_table, interface) != NULL)
2784                 return 0;
2785
2786         ifdata = g_try_new0(struct listener_data, 1);
2787         if (ifdata == NULL)
2788                 return -ENOMEM;
2789
2790         ifdata->ifname = g_strdup(interface);
2791         ifdata->udp_listener_channel = NULL;
2792         ifdata->udp_listener_watch = 0;
2793         ifdata->tcp_listener_channel = NULL;
2794         ifdata->tcp_listener_watch = 0;
2795
2796         err = create_listener(ifdata);
2797         if (err < 0) {
2798                 connman_error("Couldn't create listener for %s err %d",
2799                                 interface, err);
2800                 g_free(ifdata->ifname);
2801                 g_free(ifdata);
2802                 return err;
2803         }
2804         g_hash_table_insert(listener_table, ifdata->ifname, ifdata);
2805         return 0;
2806 }
2807
2808 void __connman_dnsproxy_remove_listener(const char *interface)
2809 {
2810         struct listener_data *ifdata;
2811
2812         DBG("interface %s", interface);
2813
2814         ifdata = g_hash_table_lookup(listener_table, interface);
2815         if (ifdata == NULL)
2816                 return;
2817
2818         destroy_listener(ifdata);
2819
2820         g_hash_table_remove(listener_table, interface);
2821 }
2822
2823 static void remove_listener(gpointer key, gpointer value, gpointer user_data)
2824 {
2825         const char *interface = key;
2826         struct listener_data *ifdata = value;
2827
2828         DBG("interface %s", interface);
2829
2830         destroy_listener(ifdata);
2831 }
2832
2833 int __connman_dnsproxy_init(void)
2834 {
2835         int err;
2836
2837         DBG("");
2838
2839         srandom(time(NULL));
2840
2841         listener_table = g_hash_table_new_full(g_str_hash, g_str_equal,
2842                                                         g_free, g_free);
2843         err = __connman_dnsproxy_add_listener("lo");
2844         if (err < 0)
2845                 return err;
2846
2847         err = connman_notifier_register(&dnsproxy_notifier);
2848         if (err < 0)
2849                 goto destroy;
2850
2851         return 0;
2852
2853 destroy:
2854         __connman_dnsproxy_remove_listener("lo");
2855         g_hash_table_destroy(listener_table);
2856
2857         return err;
2858 }
2859
2860 void __connman_dnsproxy_cleanup(void)
2861 {
2862         DBG("");
2863
2864         connman_notifier_unregister(&dnsproxy_notifier);
2865
2866         g_hash_table_foreach(listener_table, remove_listener, NULL);
2867
2868         g_hash_table_destroy(listener_table);
2869 }