third_party/libevent/evdns.c

   1 /* $Id: evdns.c 6979 2006-08-04 18:31:13Z nickm $ */
   2
   3 /* The original version of this module was written by Adam Langley; for
   4  * a history of modifications, check out the subversion logs.
   5  *
   6  * When editing this module, try to keep it re-mergeable by Adam.  Don't
   7  * reformat the whitespace, add Tor dependencies, or so on.
   8  *
   9  * TODO:
  10  *   - Support IPv6 and PTR records.
  11  *   - Replace all externally visible magic numbers with #defined constants.
  12  *   - Write doccumentation for APIs of all external functions.
  13  */
  14
  15 /* Async DNS Library
  16  * Adam Langley <agl@imperialviolet.org>
  17  * http://www.imperialviolet.org/eventdns.html
  18  * Public Domain code
  19  *
  20  * This software is Public Domain. To view a copy of the public domain dedication,
  21  * visit http://creativecommons.org/licenses/publicdomain/ or send a letter to
  22  * Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA.
  23  *
  24  * I ask and expect, but do not require, that all derivative works contain an
  25  * attribution similar to:
  26  *      Parts developed by Adam Langley <agl@imperialviolet.org>
  27  *
  28  * You may wish to replace the word "Parts" with something else depending on
  29  * the amount of original code.
  30  *
  31  * (Derivative works does not include programs which link against, run or include
  32  * the source verbatim in their source distributions)
  33  *
  34  * Version: 0.1b
  35  */
  36
  37 #include <sys/types.h>
  38 #ifdef HAVE_CONFIG_H
  39 #include "config.h"
  40 #endif
  41
  42 #ifdef DNS_USE_FTIME_FOR_ID
  43 #include <sys/timeb.h>
  44 #endif
  45
  46 #ifndef DNS_USE_CPU_CLOCK_FOR_ID
  47 #ifndef DNS_USE_GETTIMEOFDAY_FOR_ID
  48 #ifndef DNS_USE_OPENSSL_FOR_ID
  49 #ifndef DNS_USE_FTIME_FOR_ID
  50 #error Must configure at least one id generation method.
  51 #error Please see the documentation.
  52 #endif
  53 #endif
  54 #endif
  55 #endif
  56
  57 /* #define _POSIX_C_SOURCE 200507 */
  58 #if !defined(_GNU_SOURCE)
  59 #define _GNU_SOURCE
  60 #endif
  61
  62 #ifdef DNS_USE_CPU_CLOCK_FOR_ID
  63 #ifdef DNS_USE_OPENSSL_FOR_ID
  64 #error Multiple id options selected
  65 #endif
  66 #ifdef DNS_USE_GETTIMEOFDAY_FOR_ID
  67 #error Multiple id options selected
  68 #endif
  69 #include <time.h>
  70 #endif
  71
  72 #ifdef DNS_USE_OPENSSL_FOR_ID
  73 #ifdef DNS_USE_GETTIMEOFDAY_FOR_ID
  74 #error Multiple id options selected
  75 #endif
  76 #include <openssl/rand.h>
  77 #endif
  78
  79 #ifndef _FORTIFY_SOURCE
  80 #define _FORTIFY_SOURCE 3
  81 #endif
  82
  83 #include <string.h>
  84 #include <fcntl.h>
  85 #ifdef HAVE_SYS_TIME_H
  86 #include <sys/time.h>
  87 #endif
  88 #ifdef HAVE_STDINT_H
  89 #include <stdint.h>
  90 #endif
  91 #include <stdlib.h>
  92 #include <string.h>
  93 #include <errno.h>
  94 #include <assert.h>
  95 #ifdef HAVE_UNISTD_H
  96 #include <unistd.h>
  97 #endif
  98 #include <limits.h>
  99 #include <sys/stat.h>
 100 #include <ctype.h>
 101 #include <stdio.h>
 102 #include <stdarg.h>
 103
 104 #include "evdns.h"
 105 #include "evutil.h"
 106 #include "log.h"
 107 #ifdef WIN32
 108 #include <winsock2.h>
 109 #include <windows.h>
 110 #include <iphlpapi.h>
 111 #include <io.h>
 112 #else
 113 #include <sys/socket.h>
 114 #include <netinet/in.h>
 115 #include <arpa/inet.h>
 116 #endif
 117
 118 #ifdef HAVE_NETINET_IN6_H
 119 #include <netinet/in6.h>
 120 #endif
 121
 122 #define EVDNS_LOG_DEBUG 0
 123 #define EVDNS_LOG_WARN 1
 124
 125 #ifndef HOST_NAME_MAX
 126 #define HOST_NAME_MAX 255
 127 #endif
 128
 129 #include <stdio.h>
 130
 131 #undef MIN
 132 #define MIN(a,b) ((a)<(b)?(a):(b))
 133
 134 #ifdef __USE_ISOC99B
 135 /* libevent doesn't work without this */
 136 typedef ev_uint8_t u_char;
 137 typedef unsigned int uint;
 138 #endif
 139 #include "event.h"
 140
 141 #define u64 ev_uint64_t
 142 #define u32 ev_uint32_t
 143 #define u16 ev_uint16_t
 144 #define u8  ev_uint8_t
 145
 146 #ifdef WIN32
 147 #define open _open
 148 #define read _read
 149 #define close _close
 150 #define strdup _strdup
 151 #endif
 152
 153 #define MAX_ADDRS 32  /* maximum number of addresses from a single packet */
 154 /* which we bother recording */
 155
 156 #define TYPE_A         EVDNS_TYPE_A
 157 #define TYPE_CNAME     5
 158 #define TYPE_PTR       EVDNS_TYPE_PTR
 159 #define TYPE_AAAA      EVDNS_TYPE_AAAA
 160
 161 #define CLASS_INET     EVDNS_CLASS_INET
 162
 163 struct request {
 164         u8 *request;  /* the dns packet data */
 165         unsigned int request_len;
 166         int reissue_count;
 167         int tx_count;  /* the number of times that this packet has been sent */
 168         unsigned int request_type; /* TYPE_PTR or TYPE_A */
 169         void *user_pointer;  /* the pointer given to us for this request */
 170         evdns_callback_type user_callback;
 171         struct nameserver *ns;  /* the server which we last sent it */
 172
 173         /* elements used by the searching code */
 174         int search_index;
 175         struct search_state *search_state;
 176         char *search_origname;  /* needs to be free()ed */
 177         int search_flags;
 178
 179         /* these objects are kept in a circular list */
 180         struct request *next, *prev;
 181
 182         struct event timeout_event;
 183
 184         u16 trans_id;  /* the transaction id */
 185         char request_appended;  /* true if the request pointer is data which follows this struct */
 186         char transmit_me;  /* needs to be transmitted */
 187 };
 188
 189 #ifndef HAVE_STRUCT_IN6_ADDR
 190 struct in6_addr {
 191         u8 s6_addr[16];
 192 };
 193 #endif
 194
 195 struct reply {
 196         unsigned int type;
 197         unsigned int have_answer;
 198         union {
 199                 struct {
 200                         u32 addrcount;
 201                         u32 addresses[MAX_ADDRS];
 202                 } a;
 203                 struct {
 204                         u32 addrcount;
 205                         struct in6_addr addresses[MAX_ADDRS];
 206                 } aaaa;
 207                 struct {
 208                         char name[HOST_NAME_MAX];
 209                 } ptr;
 210         } data;
 211 };
 212
 213 struct nameserver {
 214         int socket;  /* a connected UDP socket */
 215         u32 address;
 216         u16 port;
 217         int failed_times;  /* number of times which we have given this server a chance */
 218         int timedout;  /* number of times in a row a request has timed out */
 219         struct event event;
 220         /* these objects are kept in a circular list */
 221         struct nameserver *next, *prev;
 222         struct event timeout_event;  /* used to keep the timeout for */
 223                                      /* when we next probe this server. */
 224                                      /* Valid if state == 0 */
 225         char state;  /* zero if we think that this server is down */
 226         char choked;  /* true if we have an EAGAIN from this server's socket */
 227         char write_waiting;  /* true if we are waiting for EV_WRITE events */
 228 };
 229
 230 static struct request *req_head = NULL, *req_waiting_head = NULL;
 231 static struct nameserver *server_head = NULL;
 232
 233 /* Represents a local port where we're listening for DNS requests. Right now, */
 234 /* only UDP is supported. */
 235 struct evdns_server_port {
 236         int socket; /* socket we use to read queries and write replies. */
 237         int refcnt; /* reference count. */
 238         char choked; /* Are we currently blocked from writing? */
 239         char closing; /* Are we trying to close this port, pending writes? */
 240         evdns_request_callback_fn_type user_callback; /* Fn to handle requests */
 241         void *user_data; /* Opaque pointer passed to user_callback */
 242         struct event event; /* Read/write event */
 243         /* circular list of replies that we want to write. */
 244         struct server_request *pending_replies;
 245 };
 246
 247 /* Represents part of a reply being built.      (That is, a single RR.) */
 248 struct server_reply_item {
 249         struct server_reply_item *next; /* next item in sequence. */
 250         char *name; /* name part of the RR */
 251         u16 type : 16; /* The RR type */
 252         u16 class : 16; /* The RR class (usually CLASS_INET) */
 253         u32 ttl; /* The RR TTL */
 254         char is_name; /* True iff data is a label */
 255         u16 datalen; /* Length of data; -1 if data is a label */
 256         void *data; /* The contents of the RR */
 257 };
 258
 259 /* Represents a request that we've received as a DNS server, and holds */
 260 /* the components of the reply as we're constructing it. */
 261 struct server_request {
 262         /* Pointers to the next and previous entries on the list of replies */
 263         /* that we're waiting to write.  Only set if we have tried to respond */
 264         /* and gotten EAGAIN. */
 265         struct server_request *next_pending;
 266         struct server_request *prev_pending;
 267
 268         u16 trans_id; /* Transaction id. */
 269         struct evdns_server_port *port; /* Which port received this request on? */
 270         struct sockaddr_storage addr; /* Where to send the response */
 271         socklen_t addrlen; /* length of addr */
 272
 273         int n_answer; /* how many answer RRs have been set? */
 274         int n_authority; /* how many authority RRs have been set? */
 275         int n_additional; /* how many additional RRs have been set? */
 276
 277         struct server_reply_item *answer; /* linked list of answer RRs */
 278         struct server_reply_item *authority; /* linked list of authority RRs */
 279         struct server_reply_item *additional; /* linked list of additional RRs */
 280
 281         /* Constructed response.  Only set once we're ready to send a reply. */
 282         /* Once this is set, the RR fields are cleared, and no more should be set. */
 283         char *response;
 284         size_t response_len;
 285
 286         /* Caller-visible fields: flags, questions. */
 287         struct evdns_server_request base;
 288 };
 289
 290 /* helper macro */
 291 #define OFFSET_OF(st, member) ((off_t) (((char*)&((st*)0)->member)-(char*)0))
 292
 293 /* Given a pointer to an evdns_server_request, get the corresponding */
 294 /* server_request. */
 295 #define TO_SERVER_REQUEST(base_ptr)                                                                             \
 296         ((struct server_request*)                                                                                       \
 297          (((char*)(base_ptr) - OFFSET_OF(struct server_request, base))))
 298
 299 /* The number of good nameservers that we have */
 300 static int global_good_nameservers = 0;
 301
 302 /* inflight requests are contained in the req_head list */
 303 /* and are actually going out across the network */
 304 static int global_requests_inflight = 0;
 305 /* requests which aren't inflight are in the waiting list */
 306 /* and are counted here */
 307 static int global_requests_waiting = 0;
 308
 309 static int global_max_requests_inflight = 64;
 310
 311 static struct timeval global_timeout = {5, 0};  /* 5 seconds */
 312 static int global_max_reissues = 1;  /* a reissue occurs when we get some errors from the server */
 313 static int global_max_retransmits = 3;  /* number of times we'll retransmit a request which timed out */
 314 /* number of timeouts in a row before we consider this server to be down */
 315 static int global_max_nameserver_timeout = 3;
 316
 317 /* These are the timeout values for nameservers. If we find a nameserver is down */
 318 /* we try to probe it at intervals as given below. Values are in seconds. */
 319 static const struct timeval global_nameserver_timeouts[] = {{10, 0}, {60, 0}, {300, 0}, {900, 0}, {3600, 0}};
 320 static const int global_nameserver_timeouts_length = sizeof(global_nameserver_timeouts)/sizeof(struct timeval);
 321
 322 static struct nameserver *nameserver_pick(void);
 323 static void evdns_request_insert(struct request *req, struct request **head);
 324 static void nameserver_ready_callback(int fd, short events, void *arg);
 325 static int evdns_transmit(void);
 326 static int evdns_request_transmit(struct request *req);
 327 static void nameserver_send_probe(struct nameserver *const ns);
 328 static void search_request_finished(struct request *const);
 329 static int search_try_next(struct request *const req);
 330 static int search_request_new(int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg);
 331 static void evdns_requests_pump_waiting_queue(void);
 332 static u16 transaction_id_pick(void);
 333 static struct request *request_new(int type, const char *name, int flags, evdns_callback_type callback, void *ptr);
 334 static void request_submit(struct request *const req);
 335
 336 static int server_request_free(struct server_request *req);
 337 static void server_request_free_answers(struct server_request *req);
 338 static void server_port_free(struct evdns_server_port *port);
 339 static void server_port_ready_callback(int fd, short events, void *arg);
 340
 341 static int strtoint(const char *const str);
 342
 343 #ifdef WIN32
 344 static int
 345 last_error(int sock)
 346 {
 347         int optval, optvallen=sizeof(optval);
 348         int err = WSAGetLastError();
 349         if (err == WSAEWOULDBLOCK && sock >= 0) {
 350                 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, (void*)&optval,
 351                                &optvallen))
 352                         return err;
 353                 if (optval)
 354                         return optval;
 355         }
 356         return err;
 357
 358 }
 359 static int
 360 error_is_eagain(int err)
 361 {
 362         return err == EAGAIN || err == WSAEWOULDBLOCK;
 363 }
 364 static int
 365 inet_aton(const char *c, struct in_addr *addr)
 366 {
 367         ev_uint32_t r;
 368         if (strcmp(c, "255.255.255.255") == 0) {
 369                 addr->s_addr = 0xffffffffu;
 370         } else {
 371                 r = inet_addr(c);
 372                 if (r == INADDR_NONE)
 373                         return 0;
 374                 addr->s_addr = r;
 375         }
 376         return 1;
 377 }
 378 #else
 379 #define last_error(sock) (errno)
 380 #define error_is_eagain(err) ((err) == EAGAIN)
 381 #endif
 382 #define CLOSE_SOCKET(s) EVUTIL_CLOSESOCKET(s)
 383
 384 #define ISSPACE(c) isspace((int)(unsigned char)(c))
 385 #define ISDIGIT(c) isdigit((int)(unsigned char)(c))
 386
 387 static const char *
 388 debug_ntoa(u32 address)
 389 {
 390         static char buf[32];
 391         u32 a = ntohl(address);
 392         evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d",
 393                       (int)(u8)((a>>24)&0xff),
 394                       (int)(u8)((a>>16)&0xff),
 395                       (int)(u8)((a>>8 )&0xff),
 396                       (int)(u8)((a    )&0xff));
 397         return buf;
 398 }
 399
 400 static evdns_debug_log_fn_type evdns_log_fn = NULL;
 401
 402 void
 403 evdns_set_log_fn(evdns_debug_log_fn_type fn)
 404 {
 405   evdns_log_fn = fn;
 406 }
 407
 408 #ifdef __GNUC__
 409 #define EVDNS_LOG_CHECK  __attribute__ ((format(printf, 2, 3)))
 410 #else
 411 #define EVDNS_LOG_CHECK
 412 #endif
 413
 414 static void _evdns_log(int warn, const char *fmt, ...) EVDNS_LOG_CHECK;
 415 static void
 416 _evdns_log(int warn, const char *fmt, ...)
 417 {
 418   va_list args;
 419   static char buf[512];
 420   if (!evdns_log_fn)
 421     return;
 422   va_start(args,fmt);
 423   evutil_vsnprintf(buf, sizeof(buf), fmt, args);
 424   buf[sizeof(buf)-1] = '\0';
 425   evdns_log_fn(warn, buf);
 426   va_end(args);
 427 }
 428
 429 #define log _evdns_log
 430
 431 /* This walks the list of inflight requests to find the */
 432 /* one with a matching transaction id. Returns NULL on */
 433 /* failure */
 434 static struct request *
 435 request_find_from_trans_id(u16 trans_id) {
 436         struct request *req = req_head, *const started_at = req_head;
 437
 438         if (req) {
 439                 do {
 440                         if (req->trans_id == trans_id) return req;
 441                         req = req->next;
 442                 } while (req != started_at);
 443         }
 444
 445         return NULL;
 446 }
 447
 448 /* a libevent callback function which is called when a nameserver */
 449 /* has gone down and we want to test if it has came back to life yet */
 450 static void
 451 nameserver_prod_callback(int fd, short events, void *arg) {
 452         struct nameserver *const ns = (struct nameserver *) arg;
 453         (void)fd;
 454         (void)events;
 455
 456         nameserver_send_probe(ns);
 457 }
 458
 459 /* a libevent callback which is called when a nameserver probe (to see if */
 460 /* it has come back to life) times out. We increment the count of failed_times */
 461 /* and wait longer to send the next probe packet. */
 462 static void
 463 nameserver_probe_failed(struct nameserver *const ns) {
 464         const struct timeval * timeout;
 465         (void) evtimer_del(&ns->timeout_event);
 466         if (ns->state == 1) {
 467                 /* This can happen if the nameserver acts in a way which makes us mark */
 468                 /* it as bad and then starts sending good replies. */
 469                 return;
 470         }
 471
 472         timeout =
 473           &global_nameserver_timeouts[MIN(ns->failed_times,
 474                                           global_nameserver_timeouts_length - 1)];
 475         ns->failed_times++;
 476
 477         if (evtimer_add(&ns->timeout_event, (struct timeval *) timeout) < 0) {
 478           log(EVDNS_LOG_WARN,
 479               "Error from libevent when adding timer event for %s",
 480               debug_ntoa(ns->address));
 481           /* ???? Do more? */
 482         }
 483 }
 484
 485 /* called when a nameserver has been deemed to have failed. For example, too */
 486 /* many packets have timed out etc */
 487 static void
 488 nameserver_failed(struct nameserver *const ns, const char *msg) {
 489         struct request *req, *started_at;
 490         /* if this nameserver has already been marked as failed */
 491         /* then don't do anything */
 492         if (!ns->state) return;
 493
 494         log(EVDNS_LOG_WARN, "Nameserver %s has failed: %s",
 495             debug_ntoa(ns->address), msg);
 496         global_good_nameservers--;
 497         assert(global_good_nameservers >= 0);
 498         if (global_good_nameservers == 0) {
 499                 log(EVDNS_LOG_WARN, "All nameservers have failed");
 500         }
 501
 502         ns->state = 0;
 503         ns->failed_times = 1;
 504
 505         if (evtimer_add(&ns->timeout_event, (struct timeval *) &global_nameserver_timeouts[0]) < 0) {
 506                 log(EVDNS_LOG_WARN,
 507                     "Error from libevent when adding timer event for %s",
 508                     debug_ntoa(ns->address));
 509                 /* ???? Do more? */
 510         }
 511
 512         /* walk the list of inflight requests to see if any can be reassigned to */
 513         /* a different server. Requests in the waiting queue don't have a */
 514         /* nameserver assigned yet */
 515
 516         /* if we don't have *any* good nameservers then there's no point */
 517         /* trying to reassign requests to one */
 518         if (!global_good_nameservers) return;
 519
 520         req = req_head;
 521         started_at = req_head;
 522         if (req) {
 523                 do {
 524                         if (req->tx_count == 0 && req->ns == ns) {
 525                                 /* still waiting to go out, can be moved */
 526                                 /* to another server */
 527                                 req->ns = nameserver_pick();
 528                         }
 529                         req = req->next;
 530                 } while (req != started_at);
 531         }
 532 }
 533
 534 static void
 535 nameserver_up(struct nameserver *const ns) {
 536         if (ns->state) return;
 537         log(EVDNS_LOG_WARN, "Nameserver %s is back up",
 538             debug_ntoa(ns->address));
 539         evtimer_del(&ns->timeout_event);
 540         ns->state = 1;
 541         ns->failed_times = 0;
 542         ns->timedout = 0;
 543         global_good_nameservers++;
 544 }
 545
 546 static void
 547 request_trans_id_set(struct request *const req, const u16 trans_id) {
 548         req->trans_id = trans_id;
 549         *((u16 *) req->request) = htons(trans_id);
 550 }
 551
 552 /* Called to remove a request from a list and dealloc it. */
 553 /* head is a pointer to the head of the list it should be */
 554 /* removed from or NULL if the request isn't in a list. */
 555 static void
 556 request_finished(struct request *const req, struct request **head) {
 557         if (head) {
 558                 if (req->next == req) {
 559                         /* only item in the list */
 560                         *head = NULL;
 561                 } else {
 562                         req->next->prev = req->prev;
 563                         req->prev->next = req->next;
 564                         if (*head == req) *head = req->next;
 565                 }
 566         }
 567
 568         log(EVDNS_LOG_DEBUG, "Removing timeout for request %lx",
 569             (unsigned long) req);
 570         evtimer_del(&req->timeout_event);
 571
 572         search_request_finished(req);
 573         global_requests_inflight--;
 574
 575         if (!req->request_appended) {
 576                 /* need to free the request data on it's own */
 577                 free(req->request);
 578         } else {
 579                 /* the request data is appended onto the header */
 580                 /* so everything gets free()ed when we: */
 581         }
 582
 583         free(req);
 584
 585         evdns_requests_pump_waiting_queue();
 586 }
 587
 588 /* This is called when a server returns a funny error code. */
 589 /* We try the request again with another server. */
 590 /* */
 591 /* return: */
 592 /*   0 ok */
 593 /*   1 failed/reissue is pointless */
 594 static int
 595 request_reissue(struct request *req) {
 596         const struct nameserver *const last_ns = req->ns;
 597         /* the last nameserver should have been marked as failing */
 598         /* by the caller of this function, therefore pick will try */
 599         /* not to return it */
 600         req->ns = nameserver_pick();
 601         if (req->ns == last_ns) {
 602                 /* ... but pick did return it */
 603                 /* not a lot of point in trying again with the */
 604                 /* same server */
 605                 return 1;
 606         }
 607
 608         req->reissue_count++;
 609         req->tx_count = 0;
 610         req->transmit_me = 1;
 611
 612         return 0;
 613 }
 614
 615 /* this function looks for space on the inflight queue and promotes */
 616 /* requests from the waiting queue if it can. */
 617 static void
 618 evdns_requests_pump_waiting_queue(void) {
 619         while (global_requests_inflight < global_max_requests_inflight &&
 620             global_requests_waiting) {
 621                 struct request *req;
 622                 /* move a request from the waiting queue to the inflight queue */
 623                 assert(req_waiting_head);
 624                 if (req_waiting_head->next == req_waiting_head) {
 625                         /* only one item in the queue */
 626                         req = req_waiting_head;
 627                         req_waiting_head = NULL;
 628                 } else {
 629                         req = req_waiting_head;
 630                         req->next->prev = req->prev;
 631                         req->prev->next = req->next;
 632                         req_waiting_head = req->next;
 633                 }
 634
 635                 global_requests_waiting--;
 636                 global_requests_inflight++;
 637
 638                 req->ns = nameserver_pick();
 639                 request_trans_id_set(req, transaction_id_pick());
 640
 641                 evdns_request_insert(req, &req_head);
 642                 evdns_request_transmit(req);
 643                 evdns_transmit();
 644         }
 645 }
 646
 647 static void
 648 reply_callback(struct request *const req, u32 ttl, u32 err, struct reply *reply) {
 649         switch (req->request_type) {
 650         case TYPE_A:
 651                 if (reply)
 652                         req->user_callback(DNS_ERR_NONE, DNS_IPv4_A,
 653                                                            reply->data.a.addrcount, ttl,
 654                                                  reply->data.a.addresses,
 655                                                            req->user_pointer);
 656                 else
 657                         req->user_callback(err, 0, 0, 0, NULL, req->user_pointer);
 658                 return;
 659         case TYPE_PTR:
 660                 if (reply) {
 661                         char *name = reply->data.ptr.name;
 662                         req->user_callback(DNS_ERR_NONE, DNS_PTR, 1, ttl,
 663                                                            &name, req->user_pointer);
 664                 } else {
 665                         req->user_callback(err, 0, 0, 0, NULL,
 666                                                            req->user_pointer);
 667                 }
 668                 return;
 669         case TYPE_AAAA:
 670                 if (reply)
 671                         req->user_callback(DNS_ERR_NONE, DNS_IPv6_AAAA,
 672                                                            reply->data.aaaa.addrcount, ttl,
 673                                                            reply->data.aaaa.addresses,
 674                                                            req->user_pointer);
 675                 else
 676                         req->user_callback(err, 0, 0, 0, NULL, req->user_pointer);
 677                 return;
 678         }
 679         assert(0);
 680 }
 681
 682 /* this processes a parsed reply packet */
 683 static void
 684 reply_handle(struct request *const req, u16 flags, u32 ttl, struct reply *reply) {
 685         int error;
 686         static const int error_codes[] = {
 687                 DNS_ERR_FORMAT, DNS_ERR_SERVERFAILED, DNS_ERR_NOTEXIST,
 688                 DNS_ERR_NOTIMPL, DNS_ERR_REFUSED
 689         };
 690
 691         if (flags & 0x020f || !reply || !reply->have_answer) {
 692                 /* there was an error */
 693                 if (flags & 0x0200) {
 694                         error = DNS_ERR_TRUNCATED;
 695                 } else {
 696                         u16 error_code = (flags & 0x000f) - 1;
 697                         if (error_code > 4) {
 698                                 error = DNS_ERR_UNKNOWN;
 699                         } else {
 700                                 error = error_codes[error_code];
 701                         }
 702                 }
 703
 704                 switch(error) {
 705                 case DNS_ERR_NOTIMPL:
 706                 case DNS_ERR_REFUSED:
 707                         /* we regard these errors as marking a bad nameserver */
 708                         if (req->reissue_count < global_max_reissues) {
 709                                 char msg[64];
 710                                 evutil_snprintf(msg, sizeof(msg),
 711                                     "Bad response %d (%s)",
 712                                          error, evdns_err_to_string(error));
 713                                 nameserver_failed(req->ns, msg);
 714                                 if (!request_reissue(req)) return;
 715                         }
 716                         break;
 717                 case DNS_ERR_SERVERFAILED:
 718                         /* rcode 2 (servfailed) sometimes means "we
 719                          * are broken" and sometimes (with some binds)
 720                          * means "that request was very confusing."
 721                          * Treat this as a timeout, not a failure.
 722                          */
 723                         log(EVDNS_LOG_DEBUG, "Got a SERVERFAILED from nameserver %s; "
 724                                 "will allow the request to time out.",
 725                                 debug_ntoa(req->ns->address));
 726                         break;
 727                 default:
 728                         /* we got a good reply from the nameserver */
 729                         nameserver_up(req->ns);
 730                 }
 731
 732                 if (req->search_state && req->request_type != TYPE_PTR) {
 733                         /* if we have a list of domains to search in,
 734                          * try the next one */
 735                         if (!search_try_next(req)) {
 736                                 /* a new request was issued so this
 737                                  * request is finished and */
 738                                 /* the user callback will be made when
 739                                  * that request (or a */
 740                                 /* child of it) finishes. */
 741                                 request_finished(req, &req_head);
 742                                 return;
 743                         }
 744                 }
 745
 746                 /* all else failed. Pass the failure up */
 747                 reply_callback(req, 0, error, NULL);
 748                 request_finished(req, &req_head);
 749         } else {
 750                 /* all ok, tell the user */
 751                 reply_callback(req, ttl, 0, reply);
 752                 nameserver_up(req->ns);
 753                 request_finished(req, &req_head);
 754         }
 755 }
 756
 757 static int
 758 name_parse(u8 *packet, int length, int *idx, char *name_out, int name_out_len) {
 759         int name_end = -1;
 760         int j = *idx;
 761         int ptr_count = 0;
 762 #define GET32(x) do { if (j + 4 > length) goto err; memcpy(&_t32, packet + j, 4); j += 4; x = ntohl(_t32); } while(0)
 763 #define GET16(x) do { if (j + 2 > length) goto err; memcpy(&_t, packet + j, 2); j += 2; x = ntohs(_t); } while(0)
 764 #define GET8(x) do { if (j >= length) goto err; x = packet[j++]; } while(0)
 765
 766         char *cp = name_out;
 767         const char *const end = name_out + name_out_len;
 768
 769         /* Normally, names are a series of length prefixed strings terminated */
 770         /* with a length of 0 (the lengths are u8's < 63). */
 771         /* However, the length can start with a pair of 1 bits and that */
 772         /* means that the next 14 bits are a pointer within the current */
 773         /* packet. */
 774
 775         for(;;) {
 776                 u8 label_len;
 777                 if (j >= length) return -1;
 778                 GET8(label_len);
 779                 if (!label_len) break;
 780                 if (label_len & 0xc0) {
 781                         u8 ptr_low;
 782                         GET8(ptr_low);
 783                         if (name_end < 0) name_end = j;
 784                         j = (((int)label_len & 0x3f) << 8) + ptr_low;
 785                         /* Make sure that the target offset is in-bounds. */
 786                         if (j < 0 || j >= length) return -1;
 787                         /* If we've jumped more times than there are characters in the
 788                          * message, we must have a loop. */
 789                         if (++ptr_count > length) return -1;
 790                         continue;
 791                 }
 792                 if (label_len > 63) return -1;
 793                 if (cp != name_out) {
 794                         if (cp + 1 >= end) return -1;
 795                         *cp++ = '.';
 796                 }
 797                 if (cp + label_len >= end) return -1;
 798                 memcpy(cp, packet + j, label_len);
 799                 cp += label_len;
 800                 j += label_len;
 801         }
 802         if (cp >= end) return -1;
 803         *cp = '\0';
 804         if (name_end < 0)
 805                 *idx = j;
 806         else
 807                 *idx = name_end;
 808         return 0;
 809  err:
 810         return -1;
 811 }
 812
 813 /* parses a raw request from a nameserver */
 814 static int
 815 reply_parse(u8 *packet, int length) {
 816         int j = 0, k = 0;  /* index into packet */
 817         u16 _t;  /* used by the macros */
 818         u32 _t32;  /* used by the macros */
 819         char tmp_name[256], cmp_name[256]; /* used by the macros */
 820
 821         u16 trans_id, questions, answers, authority, additional, datalength;
 822         u16 flags = 0;
 823         u32 ttl, ttl_r = 0xffffffff;
 824         struct reply reply;
 825         struct request *req = NULL;
 826         unsigned int i;
 827
 828         GET16(trans_id);
 829         GET16(flags);
 830         GET16(questions);
 831         GET16(answers);
 832         GET16(authority);
 833         GET16(additional);
 834         (void) authority; /* suppress "unused variable" warnings. */
 835         (void) additional; /* suppress "unused variable" warnings. */
 836
 837         req = request_find_from_trans_id(trans_id);
 838         if (!req) return -1;
 839
 840         memset(&reply, 0, sizeof(reply));
 841
 842         /* If it's not an answer, it doesn't correspond to any request. */
 843         if (!(flags & 0x8000)) return -1;  /* must be an answer */
 844         if (flags & 0x020f) {
 845                 /* there was an error */
 846                 goto err;
 847         }
 848         /* if (!answers) return; */  /* must have an answer of some form */
 849
 850         /* This macro skips a name in the DNS reply. */
 851 #define SKIP_NAME \
 852         do { tmp_name[0] = '\0';                                \
 853                 if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0)\
 854                         goto err;                               \
 855         } while(0)
 856 #define TEST_NAME \
 857         do { tmp_name[0] = '\0';                                \
 858                 cmp_name[0] = '\0';                             \
 859                 k = j;                                          \
 860                 if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0)\
 861                         goto err;                                       \
 862                 if (name_parse(req->request, req->request_len, &k, cmp_name, sizeof(cmp_name))<0)       \
 863                         goto err;                               \
 864                 if (memcmp(tmp_name, cmp_name, strlen (tmp_name)) != 0) \
 865                         return (-1); /* we ignore mismatching names */  \
 866         } while(0)
 867
 868         reply.type = req->request_type;
 869
 870         /* skip over each question in the reply */
 871         for (i = 0; i < questions; ++i) {
 872                 /* the question looks like
 873                  *   <label:name><u16:type><u16:class>
 874                  */
 875                 TEST_NAME;
 876                 j += 4;
 877                 if (j > length) goto err;
 878         }
 879
 880         /* now we have the answer section which looks like
 881          * <label:name><u16:type><u16:class><u32:ttl><u16:len><data...>
 882          */
 883
 884         for (i = 0; i < answers; ++i) {
 885                 u16 type, class;
 886
 887                 SKIP_NAME;
 888                 GET16(type);
 889                 GET16(class);
 890                 GET32(ttl);
 891                 GET16(datalength);
 892
 893                 if (type == TYPE_A && class == CLASS_INET) {
 894                         int addrcount, addrtocopy;
 895                         if (req->request_type != TYPE_A) {
 896                                 j += datalength; continue;
 897                         }
 898                         if ((datalength & 3) != 0) /* not an even number of As. */
 899                             goto err;
 900                         addrcount = datalength >> 2;
 901                         addrtocopy = MIN(MAX_ADDRS - reply.data.a.addrcount, (unsigned)addrcount);
 902
 903                         ttl_r = MIN(ttl_r, ttl);
 904                         /* we only bother with the first four addresses. */
 905                         if (j + 4*addrtocopy > length) goto err;
 906                         memcpy(&reply.data.a.addresses[reply.data.a.addrcount],
 907                                    packet + j, 4*addrtocopy);
 908                         j += 4*addrtocopy;
 909                         reply.data.a.addrcount += addrtocopy;
 910                         reply.have_answer = 1;
 911                         if (reply.data.a.addrcount == MAX_ADDRS) break;
 912                 } else if (type == TYPE_PTR && class == CLASS_INET) {
 913                         if (req->request_type != TYPE_PTR) {
 914                                 j += datalength; continue;
 915                         }
 916                         if (name_parse(packet, length, &j, reply.data.ptr.name,
 917                                                    sizeof(reply.data.ptr.name))<0)
 918                                 goto err;
 919                         ttl_r = MIN(ttl_r, ttl);
 920                         reply.have_answer = 1;
 921                         break;
 922                 } else if (type == TYPE_AAAA && class == CLASS_INET) {
 923                         int addrcount, addrtocopy;
 924                         if (req->request_type != TYPE_AAAA) {
 925                                 j += datalength; continue;
 926                         }
 927                         if ((datalength & 15) != 0) /* not an even number of AAAAs. */
 928                                 goto err;
 929                         addrcount = datalength >> 4;  /* each address is 16 bytes long */
 930                         addrtocopy = MIN(MAX_ADDRS - reply.data.aaaa.addrcount, (unsigned)addrcount);
 931                         ttl_r = MIN(ttl_r, ttl);
 932
 933                         /* we only bother with the first four addresses. */
 934                         if (j + 16*addrtocopy > length) goto err;
 935                         memcpy(&reply.data.aaaa.addresses[reply.data.aaaa.addrcount],
 936                                    packet + j, 16*addrtocopy);
 937                         reply.data.aaaa.addrcount += addrtocopy;
 938                         j += 16*addrtocopy;
 939                         reply.have_answer = 1;
 940                         if (reply.data.aaaa.addrcount == MAX_ADDRS) break;
 941                 } else {
 942                         /* skip over any other type of resource */
 943                         j += datalength;
 944                 }
 945         }
 946
 947         reply_handle(req, flags, ttl_r, &reply);
 948         return 0;
 949  err:
 950         if (req)
 951                 reply_handle(req, flags, 0, NULL);
 952         return -1;
 953 }
 954
 955 /* Parse a raw request (packet,length) sent to a nameserver port (port) from */
 956 /* a DNS client (addr,addrlen), and if it's well-formed, call the corresponding */
 957 /* callback. */
 958 static int
 959 request_parse(u8 *packet, int length, struct evdns_server_port *port, struct sockaddr *addr, socklen_t addrlen)
 960 {
 961         int j = 0;      /* index into packet */
 962         u16 _t;  /* used by the macros */
 963         char tmp_name[256]; /* used by the macros */
 964
 965         int i;
 966         u16 trans_id, flags, questions, answers, authority, additional;
 967         struct server_request *server_req = NULL;
 968
 969         /* Get the header fields */
 970         GET16(trans_id);
 971         GET16(flags);
 972         GET16(questions);
 973         GET16(answers);
 974         GET16(authority);
 975         GET16(additional);
 976
 977         if (flags & 0x8000) return -1; /* Must not be an answer. */
 978         flags &= 0x0110; /* Only RD and CD get preserved. */
 979
 980         server_req = malloc(sizeof(struct server_request));
 981         if (server_req == NULL) return -1;
 982         memset(server_req, 0, sizeof(struct server_request));
 983
 984         server_req->trans_id = trans_id;
 985         memcpy(&server_req->addr, addr, addrlen);
 986         server_req->addrlen = addrlen;
 987
 988         server_req->base.flags = flags;
 989         server_req->base.nquestions = 0;
 990         server_req->base.questions = malloc(sizeof(struct evdns_server_question *) * questions);
 991         if (server_req->base.questions == NULL)
 992                 goto err;
 993
 994         for (i = 0; i < questions; ++i) {
 995                 u16 type, class;
 996                 struct evdns_server_question *q;
 997                 int namelen;
 998                 if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0)
 999                         goto err;
1000                 GET16(type);
1001                 GET16(class);
1002                 namelen = strlen(tmp_name);
1003                 q = malloc(sizeof(struct evdns_server_question) + namelen);
1004                 if (!q)
1005                         goto err;
1006                 q->type = type;
1007                 q->dns_question_class = class;
1008                 memcpy(q->name, tmp_name, namelen+1);
1009                 server_req->base.questions[server_req->base.nquestions++] = q;
1010         }
1011
1012         /* Ignore answers, authority, and additional. */
1013
1014         server_req->port = port;
1015         port->refcnt++;
1016
1017         /* Only standard queries are supported. */
1018         if (flags & 0x7800) {
1019                 evdns_server_request_respond(&(server_req->base), DNS_ERR_NOTIMPL);
1020                 return -1;
1021         }
1022
1023         port->user_callback(&(server_req->base), port->user_data);
1024
1025         return 0;
1026 err:
1027         if (server_req) {
1028                 if (server_req->base.questions) {
1029                         for (i = 0; i < server_req->base.nquestions; ++i)
1030                                 free(server_req->base.questions[i]);
1031                         free(server_req->base.questions);
1032                 }
1033                 free(server_req);
1034         }
1035         return -1;
1036
1037 #undef SKIP_NAME
1038 #undef GET32
1039 #undef GET16
1040 #undef GET8
1041 }
1042
1043 static u16
1044 default_transaction_id_fn(void)
1045 {
1046         u16 trans_id;
1047 #ifdef DNS_USE_CPU_CLOCK_FOR_ID
1048         struct timespec ts;
1049         static int clkid = -1;
1050         if (clkid == -1) {
1051                 clkid = CLOCK_REALTIME;
1052 #ifdef CLOCK_MONOTONIC
1053                 if (clock_gettime(CLOCK_MONOTONIC, &ts) != -1)
1054                         clkid = CLOCK_MONOTONIC;
1055 #endif
1056         }
1057         if (clock_gettime(clkid, &ts) == -1)
1058                 event_err(1, "clock_gettime");
1059         trans_id = ts.tv_nsec & 0xffff;
1060 #endif
1061
1062 #ifdef DNS_USE_FTIME_FOR_ID
1063         struct _timeb tb;
1064         _ftime(&tb);
1065         trans_id = tb.millitm & 0xffff;
1066 #endif
1067
1068 #ifdef DNS_USE_GETTIMEOFDAY_FOR_ID
1069         struct timeval tv;
1070         evutil_gettimeofday(&tv, NULL);
1071         trans_id = tv.tv_usec & 0xffff;
1072 #endif
1073
1074 #ifdef DNS_USE_OPENSSL_FOR_ID
1075         if (RAND_pseudo_bytes((u8 *) &trans_id, 2) == -1) {
1076                 /* in the case that the RAND call fails we back */
1077                 /* down to using gettimeofday. */
1078                 /*
1079                   struct timeval tv;
1080                   evutil_gettimeofday(&tv, NULL);
1081                   trans_id = tv.tv_usec & 0xffff;
1082                 */
1083                 abort();
1084         }
1085 #endif
1086         return trans_id;
1087 }
1088
1089 static ev_uint16_t (*trans_id_function)(void) = default_transaction_id_fn;
1090
1091 void
1092 evdns_set_transaction_id_fn(ev_uint16_t (*fn)(void))
1093 {
1094         if (fn)
1095                 trans_id_function = fn;
1096         else
1097                 trans_id_function = default_transaction_id_fn;
1098 }
1099
1100 /* Try to choose a strong transaction id which isn't already in flight */
1101 static u16
1102 transaction_id_pick(void) {
1103         for (;;) {
1104                 const struct request *req = req_head, *started_at;
1105                 u16 trans_id = trans_id_function();
1106
1107                 if (trans_id == 0xffff) continue;
1108                 /* now check to see if that id is already inflight */
1109                 req = started_at = req_head;
1110                 if (req) {
1111                         do {
1112                                 if (req->trans_id == trans_id) break;
1113                                 req = req->next;
1114                         } while (req != started_at);
1115                 }
1116                 /* we didn't find it, so this is a good id */
1117                 if (req == started_at) return trans_id;
1118         }
1119 }
1120
1121 /* choose a namesever to use. This function will try to ignore */
1122 /* nameservers which we think are down and load balance across the rest */
1123 /* by updating the server_head global each time. */
1124 static struct nameserver *
1125 nameserver_pick(void) {
1126         struct nameserver *started_at = server_head, *picked;
1127         if (!server_head) return NULL;
1128
1129         /* if we don't have any good nameservers then there's no */
1130         /* point in trying to find one. */
1131         if (!global_good_nameservers) {
1132                 server_head = server_head->next;
1133                 return server_head;
1134         }
1135
1136         /* remember that nameservers are in a circular list */
1137         for (;;) {
1138                 if (server_head->state) {
1139                         /* we think this server is currently good */
1140                         picked = server_head;
1141                         server_head = server_head->next;
1142                         return picked;
1143                 }
1144
1145                 server_head = server_head->next;
1146                 if (server_head == started_at) {
1147                         /* all the nameservers seem to be down */
1148                         /* so we just return this one and hope for the */
1149                         /* best */
1150                         assert(global_good_nameservers == 0);
1151                         picked = server_head;
1152                         server_head = server_head->next;
1153                         return picked;
1154                 }
1155         }
1156 }
1157
1158 static int
1159 address_is_correct(struct nameserver *ns, struct sockaddr *sa, socklen_t slen)
1160 {
1161         struct sockaddr_in *sin = (struct sockaddr_in*) sa;
1162         if (sa->sa_family != AF_INET || slen != sizeof(struct sockaddr_in))
1163                 return 0;
1164         if (sin->sin_addr.s_addr != ns->address)
1165                 return 0;
1166         return 1;
1167 }
1168
1169 /* this is called when a namesever socket is ready for reading */
1170 static void
1171 nameserver_read(struct nameserver *ns) {
1172         u8 packet[1500];
1173         struct sockaddr_storage ss;
1174         socklen_t addrlen = sizeof(ss);
1175
1176         for (;;) {
1177                 const int r = recvfrom(ns->socket, packet, sizeof(packet), 0,
1178                     (struct sockaddr*)&ss, &addrlen);
1179                 if (r < 0) {
1180                         int err = last_error(ns->socket);
1181                         if (error_is_eagain(err)) return;
1182                         nameserver_failed(ns, strerror(err));
1183                         return;
1184                 }
1185                 if (!address_is_correct(ns, (struct sockaddr*)&ss, addrlen)) {
1186                         log(EVDNS_LOG_WARN, "Address mismatch on received "
1187                             "DNS packet.");
1188                         return;
1189                 }
1190                 ns->timedout = 0;
1191                 reply_parse(packet, r);
1192         }
1193 }
1194
1195 /* Read a packet from a DNS client on a server port s, parse it, and */
1196 /* act accordingly. */
1197 static void
1198 server_port_read(struct evdns_server_port *s) {
1199         u8 packet[1500];
1200         struct sockaddr_storage addr;
1201         socklen_t addrlen;
1202         int r;
1203
1204         for (;;) {
1205                 addrlen = sizeof(struct sockaddr_storage);
1206                 r = recvfrom(s->socket, packet, sizeof(packet), 0,
1207                                          (struct sockaddr*) &addr, &addrlen);
1208                 if (r < 0) {
1209                         int err = last_error(s->socket);
1210                         if (error_is_eagain(err)) return;
1211                         log(EVDNS_LOG_WARN, "Error %s (%d) while reading request.",
1212                                 strerror(err), err);
1213                         return;
1214                 }
1215                 request_parse(packet, r, s, (struct sockaddr*) &addr, addrlen);
1216         }
1217 }
1218
1219 /* Try to write all pending replies on a given DNS server port. */
1220 static void
1221 server_port_flush(struct evdns_server_port *port)
1222 {
1223         while (port->pending_replies) {
1224                 struct server_request *req = port->pending_replies;
1225                 int r = sendto(port->socket, req->response, req->response_len, 0,
1226                            (struct sockaddr*) &req->addr, req->addrlen);
1227                 if (r < 0) {
1228                         int err = last_error(port->socket);
1229                         if (error_is_eagain(err))
1230                                 return;
1231                         log(EVDNS_LOG_WARN, "Error %s (%d) while writing response to port; dropping", strerror(err), err);
1232                 }
1233                 if (server_request_free(req)) {
1234                         /* we released the last reference to req->port. */
1235                         return;
1236                 }
1237         }
1238
1239         /* We have no more pending requests; stop listening for 'writeable' events. */
1240         (void) event_del(&port->event);
1241         event_set(&port->event, port->socket, EV_READ | EV_PERSIST,
1242                           server_port_ready_callback, port);
1243         if (event_add(&port->event, NULL) < 0) {
1244                 log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server.");
1245                 /* ???? Do more? */
1246         }
1247 }
1248
1249 /* set if we are waiting for the ability to write to this server. */
1250 /* if waiting is true then we ask libevent for EV_WRITE events, otherwise */
1251 /* we stop these events. */
1252 static void
1253 nameserver_write_waiting(struct nameserver *ns, char waiting) {
1254         if (ns->write_waiting == waiting) return;
1255
1256         ns->write_waiting = waiting;
1257         (void) event_del(&ns->event);
1258         event_set(&ns->event, ns->socket, EV_READ | (waiting ? EV_WRITE : 0) | EV_PERSIST,
1259                         nameserver_ready_callback, ns);
1260         if (event_add(&ns->event, NULL) < 0) {
1261           log(EVDNS_LOG_WARN, "Error from libevent when adding event for %s",
1262               debug_ntoa(ns->address));
1263           /* ???? Do more? */
1264         }
1265 }
1266
1267 /* a callback function. Called by libevent when the kernel says that */
1268 /* a nameserver socket is ready for writing or reading */
1269 static void
1270 nameserver_ready_callback(int fd, short events, void *arg) {
1271         struct nameserver *ns = (struct nameserver *) arg;
1272         (void)fd;
1273
1274         if (events & EV_WRITE) {
1275                 ns->choked = 0;
1276                 if (!evdns_transmit()) {
1277                         nameserver_write_waiting(ns, 0);
1278                 }
1279         }
1280         if (events & EV_READ) {
1281                 nameserver_read(ns);
1282         }
1283 }
1284
1285 /* a callback function. Called by libevent when the kernel says that */
1286 /* a server socket is ready for writing or reading. */
1287 static void
1288 server_port_ready_callback(int fd, short events, void *arg) {
1289         struct evdns_server_port *port = (struct evdns_server_port *) arg;
1290         (void) fd;
1291
1292         if (events & EV_WRITE) {
1293                 port->choked = 0;
1294                 server_port_flush(port);
1295         }
1296         if (events & EV_READ) {
1297                 server_port_read(port);
1298         }
1299 }
1300
1301 /* This is an inefficient representation; only use it via the dnslabel_table_*
1302  * functions, so that is can be safely replaced with something smarter later. */
1303 #define MAX_LABELS 128
1304 /* Structures used to implement name compression */
1305 struct dnslabel_entry { char *v; off_t pos; };
1306 struct dnslabel_table {
1307         int n_labels; /* number of current entries */
1308         /* map from name to position in message */
1309         struct dnslabel_entry labels[MAX_LABELS];
1310 };
1311
1312 /* Initialize dnslabel_table. */
1313 static void
1314 dnslabel_table_init(struct dnslabel_table *table)
1315 {
1316         table->n_labels = 0;
1317 }
1318
1319 /* Free all storage held by table, but not the table itself. */
1320 static void
1321 dnslabel_clear(struct dnslabel_table *table)
1322 {
1323         int i;
1324         for (i = 0; i < table->n_labels; ++i)
1325                 free(table->labels[i].v);
1326         table->n_labels = 0;
1327 }
1328
1329 /* return the position of the label in the current message, or -1 if the label */
1330 /* hasn't been used yet. */
1331 static int
1332 dnslabel_table_get_pos(const struct dnslabel_table *table, const char *label)
1333 {
1334         int i;
1335         for (i = 0; i < table->n_labels; ++i) {
1336                 if (!strcmp(label, table->labels[i].v))
1337                         return table->labels[i].pos;
1338         }
1339         return -1;
1340 }
1341
1342 /* remember that we've used the label at position pos */
1343 static int
1344 dnslabel_table_add(struct dnslabel_table *table, const char *label, off_t pos)
1345 {
1346         char *v;
1347         int p;
1348         if (table->n_labels == MAX_LABELS)
1349                 return (-1);
1350         v = strdup(label);
1351         if (v == NULL)
1352                 return (-1);
1353         p = table->n_labels++;
1354         table->labels[p].v = v;
1355         table->labels[p].pos = pos;
1356
1357         return (0);
1358 }
1359
1360 /* Converts a string to a length-prefixed set of DNS labels, starting */
1361 /* at buf[j]. name and buf must not overlap. name_len should be the length */
1362 /* of name.      table is optional, and is used for compression. */
1363 /* */
1364 /* Input: abc.def */
1365 /* Output: <3>abc<3>def<0> */
1366 /* */
1367 /* Returns the first index after the encoded name, or negative on error. */
1368 /*       -1      label was > 63 bytes */
1369 /*       -2      name too long to fit in buffer. */
1370 /* */
1371 static off_t
1372 dnsname_to_labels(u8 *const buf, size_t buf_len, off_t j,
1373                                   const char *name, const int name_len,
1374                                   struct dnslabel_table *table) {
1375         const char *end = name + name_len;
1376         int ref = 0;
1377         u16 _t;
1378
1379 #define APPEND16(x) do {                                                   \
1380                 if (j + 2 > (off_t)buf_len)                                \
1381                         goto overflow;                                             \
1382                 _t = htons(x);                                                     \
1383                 memcpy(buf + j, &_t, 2);                                   \
1384                 j += 2;                                                                    \
1385         } while (0)
1386 #define APPEND32(x) do {                                                   \
1387                 if (j + 4 > (off_t)buf_len)                                \
1388                         goto overflow;                                             \
1389                 _t32 = htonl(x);                                                   \
1390                 memcpy(buf + j, &_t32, 4);                                 \
1391                 j += 4;                                                                    \
1392         } while (0)
1393
1394         if (name_len > 255) return -2;
1395
1396         for (;;) {
1397                 const char *const start = name;
1398                 if (table && (ref = dnslabel_table_get_pos(table, name)) >= 0) {
1399                         APPEND16(ref | 0xc000);
1400                         return j;
1401                 }
1402                 name = strchr(name, '.');
1403                 if (!name) {
1404                         const unsigned int label_len = end - start;
1405                         if (label_len > 63) return -1;
1406                         if ((size_t)(j+label_len+1) > buf_len) return -2;
1407                         if (table) dnslabel_table_add(table, start, j);
1408                         buf[j++] = label_len;
1409
1410                         memcpy(buf + j, start, end - start);
1411                         j += end - start;
1412                         break;
1413                 } else {
1414                         /* append length of the label. */
1415                         const unsigned int label_len = name - start;
1416                         if (label_len > 63) return -1;
1417                         if ((size_t)(j+label_len+1) > buf_len) return -2;
1418                         if (table) dnslabel_table_add(table, start, j);
1419                         buf[j++] = label_len;
1420
1421                         memcpy(buf + j, start, name - start);
1422                         j += name - start;
1423                         /* hop over the '.' */
1424                         name++;
1425                 }
1426         }
1427
1428         /* the labels must be terminated by a 0. */
1429         /* It's possible that the name ended in a . */
1430         /* in which case the zero is already there */
1431         if (!j || buf[j-1]) buf[j++] = 0;
1432         return j;
1433  overflow:
1434         return (-2);
1435 }
1436
1437 /* Finds the length of a dns request for a DNS name of the given */
1438 /* length. The actual request may be smaller than the value returned */
1439 /* here */
1440 static int
1441 evdns_request_len(const int name_len) {
1442         return 96 + /* length of the DNS standard header */
1443                 name_len + 2 +
1444                 4;  /* space for the resource type */
1445 }
1446
1447 /* build a dns request packet into buf. buf should be at least as long */
1448 /* as evdns_request_len told you it should be. */
1449 /* */
1450 /* Returns the amount of space used. Negative on error. */
1451 static int
1452 evdns_request_data_build(const char *const name, const int name_len,
1453     const u16 trans_id, const u16 type, const u16 class,
1454     u8 *const buf, size_t buf_len) {
1455         off_t j = 0;  /* current offset into buf */
1456         u16 _t;  /* used by the macros */
1457
1458         APPEND16(trans_id);
1459         APPEND16(0x0100);  /* standard query, recusion needed */
1460         APPEND16(1);  /* one question */
1461         APPEND16(0);  /* no answers */
1462         APPEND16(0);  /* no authority */
1463         APPEND16(0);  /* no additional */
1464
1465         j = dnsname_to_labels(buf, buf_len, j, name, name_len, NULL);
1466         if (j < 0) {
1467                 return (int)j;
1468         }
1469
1470         APPEND16(type);
1471         APPEND16(class);
1472
1473         return (int)j;
1474  overflow:
1475         return (-1);
1476 }
1477
1478 /* exported function */
1479 struct evdns_server_port *
1480 evdns_add_server_port(int socket, int is_tcp, evdns_request_callback_fn_type cb, void *user_data)
1481 {
1482         struct evdns_server_port *port;
1483         if (!(port = malloc(sizeof(struct evdns_server_port))))
1484                 return NULL;
1485         memset(port, 0, sizeof(struct evdns_server_port));
1486
1487         assert(!is_tcp); /* TCP sockets not yet implemented */
1488         port->socket = socket;
1489         port->refcnt = 1;
1490         port->choked = 0;
1491         port->closing = 0;
1492         port->user_callback = cb;
1493         port->user_data = user_data;
1494         port->pending_replies = NULL;
1495
1496         event_set(&port->event, port->socket, EV_READ | EV_PERSIST,
1497                           server_port_ready_callback, port);
1498         event_add(&port->event, NULL); /* check return. */
1499         return port;
1500 }
1501
1502 /* exported function */
1503 void
1504 evdns_close_server_port(struct evdns_server_port *port)
1505 {
1506         if (--port->refcnt == 0)
1507                 server_port_free(port);
1508         port->closing = 1;
1509 }
1510
1511 /* exported function */
1512 int
1513 evdns_server_request_add_reply(struct evdns_server_request *_req, int section, const char *name, int type, int class, int ttl, int datalen, int is_name, const char *data)
1514 {
1515         struct server_request *req = TO_SERVER_REQUEST(_req);
1516         struct server_reply_item **itemp, *item;
1517         int *countp;
1518
1519         if (req->response) /* have we already answered? */
1520                 return (-1);
1521
1522         switch (section) {
1523         case EVDNS_ANSWER_SECTION:
1524                 itemp = &req->answer;
1525                 countp = &req->n_answer;
1526                 break;
1527         case EVDNS_AUTHORITY_SECTION:
1528                 itemp = &req->authority;
1529                 countp = &req->n_authority;
1530                 break;
1531         case EVDNS_ADDITIONAL_SECTION:
1532                 itemp = &req->additional;
1533                 countp = &req->n_additional;
1534                 break;
1535         default:
1536                 return (-1);
1537         }
1538         while (*itemp) {
1539                 itemp = &((*itemp)->next);
1540         }
1541         item = malloc(sizeof(struct server_reply_item));
1542         if (!item)
1543                 return -1;
1544         item->next = NULL;
1545         if (!(item->name = strdup(name))) {
1546                 free(item);
1547                 return -1;
1548         }
1549         item->type = type;
1550         item->dns_question_class = class;
1551         item->ttl = ttl;
1552         item->is_name = is_name != 0;
1553         item->datalen = 0;
1554         item->data = NULL;
1555         if (data) {
1556                 if (item->is_name) {
1557                         if (!(item->data = strdup(data))) {
1558                                 free(item->name);
1559                                 free(item);
1560                                 return -1;
1561                         }
1562                         item->datalen = (u16)-1;
1563                 } else {
1564                         if (!(item->data = malloc(datalen))) {
1565                                 free(item->name);
1566                                 free(item);
1567                                 return -1;
1568                         }
1569                         item->datalen = datalen;
1570                         memcpy(item->data, data, datalen);
1571                 }
1572         }
1573
1574         *itemp = item;
1575         ++(*countp);
1576         return 0;
1577 }
1578
1579 /* exported function */
1580 int
1581 evdns_server_request_add_a_reply(struct evdns_server_request *req, const char *name, int n, void *addrs, int ttl)
1582 {
1583         return evdns_server_request_add_reply(
1584                   req, EVDNS_ANSWER_SECTION, name, TYPE_A, CLASS_INET,
1585                   ttl, n*4, 0, addrs);
1586 }
1587
1588 /* exported function */
1589 int
1590 evdns_server_request_add_aaaa_reply(struct evdns_server_request *req, const char *name, int n, void *addrs, int ttl)
1591 {
1592         return evdns_server_request_add_reply(
1593                   req, EVDNS_ANSWER_SECTION, name, TYPE_AAAA, CLASS_INET,
1594                   ttl, n*16, 0, addrs);
1595 }
1596
1597 /* exported function */
1598 int
1599 evdns_server_request_add_ptr_reply(struct evdns_server_request *req, struct in_addr *in, const char *inaddr_name, const char *hostname, int ttl)
1600 {
1601         u32 a;
1602         char buf[32];
1603         assert(in || inaddr_name);
1604         assert(!(in && inaddr_name));
1605         if (in) {
1606                 a = ntohl(in->s_addr);
1607                 evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
1608                                 (int)(u8)((a    )&0xff),
1609                                 (int)(u8)((a>>8 )&0xff),
1610                                 (int)(u8)((a>>16)&0xff),
1611                                 (int)(u8)((a>>24)&0xff));
1612                 inaddr_name = buf;
1613         }
1614         return evdns_server_request_add_reply(
1615                   req, EVDNS_ANSWER_SECTION, inaddr_name, TYPE_PTR, CLASS_INET,
1616                   ttl, -1, 1, hostname);
1617 }
1618
1619 /* exported function */
1620 int
1621 evdns_server_request_add_cname_reply(struct evdns_server_request *req, const char *name, const char *cname, int ttl)
1622 {
1623         return evdns_server_request_add_reply(
1624                   req, EVDNS_ANSWER_SECTION, name, TYPE_CNAME, CLASS_INET,
1625                   ttl, -1, 1, cname);
1626 }
1627
1628
1629 static int
1630 evdns_server_request_format_response(struct server_request *req, int err)
1631 {
1632         unsigned char buf[1500];
1633         size_t buf_len = sizeof(buf);
1634         off_t j = 0, r;
1635         u16 _t;
1636         u32 _t32;
1637         int i;
1638         u16 flags;
1639         struct dnslabel_table table;
1640
1641         if (err < 0 || err > 15) return -1;
1642
1643         /* Set response bit and error code; copy OPCODE and RD fields from
1644          * question; copy RA and AA if set by caller. */
1645         flags = req->base.flags;
1646         flags |= (0x8000 | err);
1647
1648         dnslabel_table_init(&table);
1649         APPEND16(req->trans_id);
1650         APPEND16(flags);
1651         APPEND16(req->base.nquestions);
1652         APPEND16(req->n_answer);
1653         APPEND16(req->n_authority);
1654         APPEND16(req->n_additional);
1655
1656         /* Add questions. */
1657         for (i=0; i < req->base.nquestions; ++i) {
1658                 const char *s = req->base.questions[i]->name;
1659                 j = dnsname_to_labels(buf, buf_len, j, s, strlen(s), &table);
1660                 if (j < 0) {
1661                         dnslabel_clear(&table);
1662                         return (int) j;
1663                 }
1664                 APPEND16(req->base.questions[i]->type);
1665                 APPEND16(req->base.questions[i]->dns_question_class);
1666         }
1667
1668         /* Add answer, authority, and additional sections. */
1669         for (i=0; i<3; ++i) {
1670                 struct server_reply_item *item;
1671                 if (i==0)
1672                         item = req->answer;
1673                 else if (i==1)
1674                         item = req->authority;
1675                 else
1676                         item = req->additional;
1677                 while (item) {
1678                         r = dnsname_to_labels(buf, buf_len, j, item->name, strlen(item->name), &table);
1679                         if (r < 0)
1680                                 goto overflow;
1681                         j = r;
1682
1683                         APPEND16(item->type);
1684                         APPEND16(item->dns_question_class);
1685                         APPEND32(item->ttl);
1686                         if (item->is_name) {
1687                                 off_t len_idx = j, name_start;
1688                                 j += 2;
1689                                 name_start = j;
1690                                 r = dnsname_to_labels(buf, buf_len, j, item->data, strlen(item->data), &table);
1691                                 if (r < 0)
1692                                         goto overflow;
1693                                 j = r;
1694                                 _t = htons( (short) (j-name_start) );
1695                                 memcpy(buf+len_idx, &_t, 2);
1696                         } else {
1697                                 APPEND16(item->datalen);
1698                                 if (j+item->datalen > (off_t)buf_len)
1699                                         goto overflow;
1700                                 memcpy(buf+j, item->data, item->datalen);
1701                                 j += item->datalen;
1702                         }
1703                         item = item->next;
1704                 }
1705         }
1706
1707         if (j > 512) {
1708 overflow:
1709                 j = 512;
1710                 buf[2] |= 0x02; /* set the truncated bit. */
1711         }
1712
1713         req->response_len = j;
1714
1715         if (!(req->response = malloc(req->response_len))) {
1716                 server_request_free_answers(req);
1717                 dnslabel_clear(&table);
1718                 return (-1);
1719         }
1720         memcpy(req->response, buf, req->response_len);
1721         server_request_free_answers(req);
1722         dnslabel_clear(&table);
1723         return (0);
1724 }
1725
1726 /* exported function */
1727 int
1728 evdns_server_request_respond(struct evdns_server_request *_req, int err)
1729 {
1730         struct server_request *req = TO_SERVER_REQUEST(_req);
1731         struct evdns_server_port *port = req->port;
1732         int r;
1733         if (!req->response) {
1734                 if ((r = evdns_server_request_format_response(req, err))<0)
1735                         return r;
1736         }
1737
1738         r = sendto(port->socket, req->response, req->response_len, 0,
1739                            (struct sockaddr*) &req->addr, req->addrlen);
1740         if (r<0) {
1741                 int sock_err = last_error(port->socket);
1742                 if (! error_is_eagain(sock_err))
1743                         return -1;
1744
1745                 if (port->pending_replies) {
1746                         req->prev_pending = port->pending_replies->prev_pending;
1747                         req->next_pending = port->pending_replies;
1748                         req->prev_pending->next_pending =
1749                                 req->next_pending->prev_pending = req;
1750                 } else {
1751                         req->prev_pending = req->next_pending = req;
1752                         port->pending_replies = req;
1753                         port->choked = 1;
1754
1755                         (void) event_del(&port->event);
1756                         event_set(&port->event, port->socket, (port->closing?0:EV_READ) | EV_WRITE | EV_PERSIST, server_port_ready_callback, port);
1757
1758                         if (event_add(&port->event, NULL) < 0) {
1759                                 log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server");
1760                         }
1761
1762                 }
1763
1764                 return 1;
1765         }
1766         if (server_request_free(req))
1767                 return 0;
1768
1769         if (port->pending_replies)
1770                 server_port_flush(port);
1771
1772         return 0;
1773 }
1774
1775 /* Free all storage held by RRs in req. */
1776 static void
1777 server_request_free_answers(struct server_request *req)
1778 {
1779         struct server_reply_item *victim, *next, **list;
1780         int i;
1781         for (i = 0; i < 3; ++i) {
1782                 if (i==0)
1783                         list = &req->answer;
1784                 else if (i==1)
1785                         list = &req->authority;
1786                 else
1787                         list = &req->additional;
1788
1789                 victim = *list;
1790                 while (victim) {
1791                         next = victim->next;
1792                         free(victim->name);
1793                         if (victim->data)
1794                                 free(victim->data);
1795                         free(victim);
1796                         victim = next;
1797                 }
1798                 *list = NULL;
1799         }
1800 }
1801
1802 /* Free all storage held by req, and remove links to it. */
1803 /* return true iff we just wound up freeing the server_port. */
1804 static int
1805 server_request_free(struct server_request *req)
1806 {
1807         int i, rc=1;
1808         if (req->base.questions) {
1809                 for (i = 0; i < req->base.nquestions; ++i)
1810                         free(req->base.questions[i]);
1811                 free(req->base.questions);
1812         }
1813
1814         if (req->port) {
1815                 if (req->port->pending_replies == req) {
1816                         if (req->next_pending)
1817                                 req->port->pending_replies = req->next_pending;
1818                         else
1819                                 req->port->pending_replies = NULL;
1820                 }
1821                 rc = --req->port->refcnt;
1822         }
1823
1824         if (req->response) {
1825                 free(req->response);
1826         }
1827
1828         server_request_free_answers(req);
1829
1830         if (req->next_pending && req->next_pending != req) {
1831                 req->next_pending->prev_pending = req->prev_pending;
1832                 req->prev_pending->next_pending = req->next_pending;
1833         }
1834
1835         if (rc == 0) {
1836                 server_port_free(req->port);
1837                 free(req);
1838                 return (1);
1839         }
1840         free(req);
1841         return (0);
1842 }
1843
1844 /* Free all storage held by an evdns_server_port.  Only called when  */
1845 static void
1846 server_port_free(struct evdns_server_port *port)
1847 {
1848         assert(port);
1849         assert(!port->refcnt);
1850         assert(!port->pending_replies);
1851         if (port->socket > 0) {
1852                 CLOSE_SOCKET(port->socket);
1853                 port->socket = -1;
1854         }
1855         (void) event_del(&port->event);
1856         /* XXXX actually free the port? -NM */
1857 }
1858
1859 /* exported function */
1860 int
1861 evdns_server_request_drop(struct evdns_server_request *_req)
1862 {
1863         struct server_request *req = TO_SERVER_REQUEST(_req);
1864         server_request_free(req);
1865         return 0;
1866 }
1867
1868 /* exported function */
1869 int
1870 evdns_server_request_get_requesting_addr(struct evdns_server_request *_req, struct sockaddr *sa, int addr_len)
1871 {
1872         struct server_request *req = TO_SERVER_REQUEST(_req);
1873         if (addr_len < (int)req->addrlen)
1874                 return -1;
1875         memcpy(sa, &(req->addr), req->addrlen);
1876         return req->addrlen;
1877 }
1878
1879 #undef APPEND16
1880 #undef APPEND32
1881
1882 /* this is a libevent callback function which is called when a request */
1883 /* has timed out. */
1884 static void
1885 evdns_request_timeout_callback(int fd, short events, void *arg) {
1886         struct request *const req = (struct request *) arg;
1887         (void) fd;
1888         (void) events;
1889
1890         log(EVDNS_LOG_DEBUG, "Request %lx timed out", (unsigned long) arg);
1891
1892         req->ns->timedout++;
1893         if (req->ns->timedout > global_max_nameserver_timeout) {
1894                 req->ns->timedout = 0;
1895                 nameserver_failed(req->ns, "request timed out.");
1896         }
1897
1898         (void) evtimer_del(&req->timeout_event);
1899         if (req->tx_count >= global_max_retransmits) {
1900                 /* this request has failed */
1901                 reply_callback(req, 0, DNS_ERR_TIMEOUT, NULL);
1902                 request_finished(req, &req_head);
1903         } else {
1904                 /* retransmit it */
1905                 evdns_request_transmit(req);
1906         }
1907 }
1908
1909 /* try to send a request to a given server. */
1910 /* */
1911 /* return: */
1912 /*   0 ok */
1913 /*   1 temporary failure */
1914 /*   2 other failure */
1915 static int
1916 evdns_request_transmit_to(struct request *req, struct nameserver *server) {
1917         struct sockaddr_in sin;
1918         int r;
1919         memset(&sin, 0, sizeof(sin));
1920         sin.sin_addr.s_addr = req->ns->address;
1921         sin.sin_port = req->ns->port;
1922         sin.sin_family = AF_INET;
1923
1924         r = sendto(server->socket, req->request, req->request_len, 0,
1925             (struct sockaddr*)&sin, sizeof(sin));
1926         if (r < 0) {
1927                 int err = last_error(server->socket);
1928                 if (error_is_eagain(err)) return 1;
1929                 nameserver_failed(req->ns, strerror(err));
1930                 return 2;
1931         } else if (r != (int)req->request_len) {
1932                 return 1;  /* short write */
1933         } else {
1934                 return 0;
1935         }
1936 }
1937
1938 /* try to send a request, updating the fields of the request */
1939 /* as needed */
1940 /* */
1941 /* return: */
1942 /*   0 ok */
1943 /*   1 failed */
1944 static int
1945 evdns_request_transmit(struct request *req) {
1946         int retcode = 0, r;
1947
1948         /* if we fail to send this packet then this flag marks it */
1949         /* for evdns_transmit */
1950         req->transmit_me = 1;
1951         if (req->trans_id == 0xffff) abort();
1952
1953         if (req->ns->choked) {
1954                 /* don't bother trying to write to a socket */
1955                 /* which we have had EAGAIN from */
1956                 return 1;
1957         }
1958
1959         r = evdns_request_transmit_to(req, req->ns);
1960         switch (r) {
1961         case 1:
1962                 /* temp failure */
1963                 req->ns->choked = 1;
1964                 nameserver_write_waiting(req->ns, 1);
1965                 return 1;
1966         case 2:
1967                 /* failed in some other way */
1968                 retcode = 1;
1969                 /* fall through */
1970         default:
1971                 /* all ok */
1972                 log(EVDNS_LOG_DEBUG,
1973                     "Setting timeout for request %lx", (unsigned long) req);
1974                 if (evtimer_add(&req->timeout_event, &global_timeout) < 0) {
1975                   log(EVDNS_LOG_WARN,
1976                       "Error from libevent when adding timer for request %lx",
1977                       (unsigned long) req);
1978                   /* ???? Do more? */
1979                 }
1980                 req->tx_count++;
1981                 req->transmit_me = 0;
1982                 return retcode;
1983         }
1984 }
1985
1986 static void
1987 nameserver_probe_callback(int result, char type, int count, int ttl, void *addresses, void *arg) {
1988         struct nameserver *const ns = (struct nameserver *) arg;
1989         (void) type;
1990         (void) count;
1991         (void) ttl;
1992         (void) addresses;
1993
1994         if (result == DNS_ERR_NONE || result == DNS_ERR_NOTEXIST) {
1995                 /* this is a good reply */
1996                 nameserver_up(ns);
1997         } else nameserver_probe_failed(ns);
1998 }
1999
2000 static void
2001 nameserver_send_probe(struct nameserver *const ns) {
2002         struct request *req;
2003         /* here we need to send a probe to a given nameserver */
2004         /* in the hope that it is up now. */
2005
2006         log(EVDNS_LOG_DEBUG, "Sending probe to %s", debug_ntoa(ns->address));
2007
2008         req = request_new(TYPE_A, "www.google.com", DNS_QUERY_NO_SEARCH, nameserver_probe_callback, ns);
2009         if (!req) return;
2010         /* we force this into the inflight queue no matter what */
2011         request_trans_id_set(req, transaction_id_pick());
2012         req->ns = ns;
2013         request_submit(req);
2014 }
2015
2016 /* returns: */
2017 /*   0 didn't try to transmit anything */
2018 /*   1 tried to transmit something */
2019 static int
2020 evdns_transmit(void) {
2021         char did_try_to_transmit = 0;
2022
2023         if (req_head) {
2024                 struct request *const started_at = req_head, *req = req_head;
2025                 /* first transmit all the requests which are currently waiting */
2026                 do {
2027                         if (req->transmit_me) {
2028                                 did_try_to_transmit = 1;
2029                                 evdns_request_transmit(req);
2030                         }
2031
2032                         req = req->next;
2033                 } while (req != started_at);
2034         }
2035
2036         return did_try_to_transmit;
2037 }
2038
2039 /* exported function */
2040 int
2041 evdns_count_nameservers(void)
2042 {
2043         const struct nameserver *server = server_head;
2044         int n = 0;
2045         if (!server)
2046                 return 0;
2047         do {
2048                 ++n;
2049                 server = server->next;
2050         } while (server != server_head);
2051         return n;
2052 }
2053
2054 /* exported function */
2055 int
2056 evdns_clear_nameservers_and_suspend(void)
2057 {
2058         struct nameserver *server = server_head, *started_at = server_head;
2059         struct request *req = req_head, *req_started_at = req_head;
2060
2061         if (!server)
2062                 return 0;
2063         while (1) {
2064                 struct nameserver *next = server->next;
2065                 (void) event_del(&server->event);
2066                 if (evtimer_initialized(&server->timeout_event))
2067                         (void) evtimer_del(&server->timeout_event);
2068                 if (server->socket >= 0)
2069                         CLOSE_SOCKET(server->socket);
2070                 free(server);
2071                 if (next == started_at)
2072                         break;
2073                 server = next;
2074         }
2075         server_head = NULL;
2076         global_good_nameservers = 0;
2077
2078         while (req) {
2079                 struct request *next = req->next;
2080                 req->tx_count = req->reissue_count = 0;
2081                 req->ns = NULL;
2082                 /* ???? What to do about searches? */
2083                 (void) evtimer_del(&req->timeout_event);
2084                 req->trans_id = 0;
2085                 req->transmit_me = 0;
2086
2087                 global_requests_waiting++;
2088                 evdns_request_insert(req, &req_waiting_head);
2089                 /* We want to insert these suspended elements at the front of
2090                  * the waiting queue, since they were pending before any of
2091                  * the waiting entries were added.  This is a circular list,
2092                  * so we can just shift the start back by one.*/
2093                 req_waiting_head = req_waiting_head->prev;
2094
2095                 if (next == req_started_at)
2096                         break;
2097                 req = next;
2098         }
2099         req_head = NULL;
2100         global_requests_inflight = 0;
2101
2102         return 0;
2103 }
2104
2105
2106 /* exported function */
2107 int
2108 evdns_resume(void)
2109 {
2110         evdns_requests_pump_waiting_queue();
2111         return 0;
2112 }
2113
2114 static int
2115 _evdns_nameserver_add_impl(unsigned long int address, int port) {
2116         /* first check to see if we already have this nameserver */
2117
2118         const struct nameserver *server = server_head, *const started_at = server_head;
2119         struct nameserver *ns;
2120         int err = 0;
2121         if (server) {
2122                 do {
2123                         if (server->address == address) return 3;
2124                         server = server->next;
2125                 } while (server != started_at);
2126         }
2127
2128         ns = (struct nameserver *) malloc(sizeof(struct nameserver));
2129         if (!ns) return -1;
2130
2131         memset(ns, 0, sizeof(struct nameserver));
2132
2133         evtimer_set(&ns->timeout_event, nameserver_prod_callback, ns);
2134
2135         ns->socket = socket(PF_INET, SOCK_DGRAM, 0);
2136         if (ns->socket < 0) { err = 1; goto out1; }
2137         evutil_make_socket_nonblocking(ns->socket);
2138
2139         ns->address = address;
2140         ns->port = htons(port);
2141         ns->state = 1;
2142         event_set(&ns->event, ns->socket, EV_READ | EV_PERSIST, nameserver_ready_callback, ns);
2143         if (event_add(&ns->event, NULL) < 0) {
2144           err = 2;
2145           goto out2;
2146         }
2147
2148         log(EVDNS_LOG_DEBUG, "Added nameserver %s", debug_ntoa(address));
2149
2150         /* insert this nameserver into the list of them */
2151         if (!server_head) {
2152                 ns->next = ns->prev = ns;
2153                 server_head = ns;
2154         } else {
2155                 ns->next = server_head->next;
2156                 ns->prev = server_head;
2157                 server_head->next = ns;
2158                 if (server_head->prev == server_head) {
2159                         server_head->prev = ns;
2160                 }
2161         }
2162
2163         global_good_nameservers++;
2164
2165         return 0;
2166
2167 out2:
2168         CLOSE_SOCKET(ns->socket);
2169 out1:
2170         free(ns);
2171         log(EVDNS_LOG_WARN, "Unable to add nameserver %s: error %d", debug_ntoa(address), err);
2172         return err;
2173 }
2174
2175 /* exported function */
2176 int
2177 evdns_nameserver_add(unsigned long int address) {
2178         return _evdns_nameserver_add_impl(address, 53);
2179 }
2180
2181 /* exported function */
2182 int
2183 evdns_nameserver_ip_add(const char *ip_as_string) {
2184         struct in_addr ina;
2185         int port;
2186         char buf[20];
2187         const char *cp;
2188         cp = strchr(ip_as_string, ':');
2189         if (! cp) {
2190                 cp = ip_as_string;
2191                 port = 53;
2192         } else {
2193                 port = strtoint(cp+1);
2194                 if (port < 0 || port > 65535) {
2195                         return 4;
2196                 }
2197                 if ((cp-ip_as_string) >= (int)sizeof(buf)) {
2198                         return 4;
2199                 }
2200                 memcpy(buf, ip_as_string, cp-ip_as_string);
2201                 buf[cp-ip_as_string] = '\0';
2202                 cp = buf;
2203         }
2204         if (!inet_aton(cp, &ina)) {
2205                 return 4;
2206         }
2207         return _evdns_nameserver_add_impl(ina.s_addr, port);
2208 }
2209
2210 /* insert into the tail of the queue */
2211 static void
2212 evdns_request_insert(struct request *req, struct request **head) {
2213         if (!*head) {
2214                 *head = req;
2215                 req->next = req->prev = req;
2216                 return;
2217         }
2218
2219         req->prev = (*head)->prev;
2220         req->prev->next = req;
2221         req->next = *head;
2222         (*head)->prev = req;
2223 }
2224
2225 static int
2226 string_num_dots(const char *s) {
2227         int count = 0;
2228         while ((s = strchr(s, '.'))) {
2229                 s++;
2230                 count++;
2231         }
2232         return count;
2233 }
2234
2235 static struct request *
2236 request_new(int type, const char *name, int flags,
2237     evdns_callback_type callback, void *user_ptr) {
2238         const char issuing_now =
2239             (global_requests_inflight < global_max_requests_inflight) ? 1 : 0;
2240
2241         const int name_len = strlen(name);
2242         const int request_max_len = evdns_request_len(name_len);
2243         const u16 trans_id = issuing_now ? transaction_id_pick() : 0xffff;
2244         /* the request data is alloced in a single block with the header */
2245         struct request *const req =
2246             (struct request *) malloc(sizeof(struct request) + request_max_len);
2247         int rlen;
2248         (void) flags;
2249
2250         if (!req) return NULL;
2251         memset(req, 0, sizeof(struct request));
2252
2253         evtimer_set(&req->timeout_event, evdns_request_timeout_callback, req);
2254
2255         /* request data lives just after the header */
2256         req->request = ((u8 *) req) + sizeof(struct request);
2257         /* denotes that the request data shouldn't be free()ed */
2258         req->request_appended = 1;
2259         rlen = evdns_request_data_build(name, name_len, trans_id,
2260             type, CLASS_INET, req->request, request_max_len);
2261         if (rlen < 0)
2262                 goto err1;
2263         req->request_len = rlen;
2264         req->trans_id = trans_id;
2265         req->tx_count = 0;
2266         req->request_type = type;
2267         req->user_pointer = user_ptr;
2268         req->user_callback = callback;
2269         req->ns = issuing_now ? nameserver_pick() : NULL;
2270         req->next = req->prev = NULL;
2271
2272         return req;
2273 err1:
2274         free(req);
2275         return NULL;
2276 }
2277
2278 static void
2279 request_submit(struct request *const req) {
2280         if (req->ns) {
2281                 /* if it has a nameserver assigned then this is going */
2282                 /* straight into the inflight queue */
2283                 evdns_request_insert(req, &req_head);
2284                 global_requests_inflight++;
2285                 evdns_request_transmit(req);
2286         } else {
2287                 evdns_request_insert(req, &req_waiting_head);
2288                 global_requests_waiting++;
2289         }
2290 }
2291
2292 /* exported function */
2293 int evdns_resolve_ipv4(const char *name, int flags,
2294     evdns_callback_type callback, void *ptr) {
2295         log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
2296         if (flags & DNS_QUERY_NO_SEARCH) {
2297                 struct request *const req =
2298                         request_new(TYPE_A, name, flags, callback, ptr);
2299                 if (req == NULL)
2300                         return (1);
2301                 request_submit(req);
2302                 return (0);
2303         } else {
2304                 return (search_request_new(TYPE_A, name, flags, callback, ptr));
2305         }
2306 }
2307
2308 /* exported function */
2309 int evdns_resolve_ipv6(const char *name, int flags,
2310                                            evdns_callback_type callback, void *ptr) {
2311         log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
2312         if (flags & DNS_QUERY_NO_SEARCH) {
2313                 struct request *const req =
2314                         request_new(TYPE_AAAA, name, flags, callback, ptr);
2315                 if (req == NULL)
2316                         return (1);
2317                 request_submit(req);
2318                 return (0);
2319         } else {
2320                 return (search_request_new(TYPE_AAAA, name, flags, callback, ptr));
2321         }
2322 }
2323
2324 int evdns_resolve_reverse(const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) {
2325         char buf[32];
2326         struct request *req;
2327         u32 a;
2328         assert(in);
2329         a = ntohl(in->s_addr);
2330         evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
2331                         (int)(u8)((a    )&0xff),
2332                         (int)(u8)((a>>8 )&0xff),
2333                         (int)(u8)((a>>16)&0xff),
2334                         (int)(u8)((a>>24)&0xff));
2335         log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
2336         req = request_new(TYPE_PTR, buf, flags, callback, ptr);
2337         if (!req) return 1;
2338         request_submit(req);
2339         return 0;
2340 }
2341
2342 int evdns_resolve_reverse_ipv6(const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) {
2343         /* 32 nybbles, 32 periods, "ip6.arpa", NUL. */
2344         char buf[73];
2345         char *cp;
2346         struct request *req;
2347         int i;
2348         assert(in);
2349         cp = buf;
2350         for (i=15; i >= 0; --i) {
2351                 u8 byte = in->s6_addr[i];
2352                 *cp++ = "0123456789abcdef"[byte & 0x0f];
2353                 *cp++ = '.';
2354                 *cp++ = "0123456789abcdef"[byte >> 4];
2355                 *cp++ = '.';
2356         }
2357         assert(cp + strlen("ip6.arpa") < buf+sizeof(buf));
2358         memcpy(cp, "ip6.arpa", strlen("ip6.arpa")+1);
2359         log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
2360         req = request_new(TYPE_PTR, buf, flags, callback, ptr);
2361         if (!req) return 1;
2362         request_submit(req);
2363         return 0;
2364 }
2365
2366 /*/////////////////////////////////////////////////////////////////// */
2367 /* Search support */
2368 /* */
2369 /* the libc resolver has support for searching a number of domains */
2370 /* to find a name. If nothing else then it takes the single domain */
2371 /* from the gethostname() call. */
2372 /* */
2373 /* It can also be configured via the domain and search options in a */
2374 /* resolv.conf. */
2375 /* */
2376 /* The ndots option controls how many dots it takes for the resolver */
2377 /* to decide that a name is non-local and so try a raw lookup first. */
2378
2379 struct search_domain {
2380         int len;
2381         struct search_domain *next;
2382         /* the text string is appended to this structure */
2383 };
2384
2385 struct search_state {
2386         int refcount;
2387         int ndots;
2388         int num_domains;
2389         struct search_domain *head;
2390 };
2391
2392 static struct search_state *global_search_state = NULL;
2393
2394 static void
2395 search_state_decref(struct search_state *const state) {
2396         if (!state) return;
2397         state->refcount--;
2398         if (!state->refcount) {
2399                 struct search_domain *next, *dom;
2400                 for (dom = state->head; dom; dom = next) {
2401                         next = dom->next;
2402                         free(dom);
2403                 }
2404                 free(state);
2405         }
2406 }
2407
2408 static struct search_state *
2409 search_state_new(void) {
2410         struct search_state *state = (struct search_state *) malloc(sizeof(struct search_state));
2411         if (!state) return NULL;
2412         memset(state, 0, sizeof(struct search_state));
2413         state->refcount = 1;
2414         state->ndots = 1;
2415
2416         return state;
2417 }
2418
2419 static void
2420 search_postfix_clear(void) {
2421         search_state_decref(global_search_state);
2422
2423         global_search_state = search_state_new();
2424 }
2425
2426 /* exported function */
2427 void
2428 evdns_search_clear(void) {
2429         search_postfix_clear();
2430 }
2431
2432 static void
2433 search_postfix_add(const char *domain) {
2434         int domain_len;
2435         struct search_domain *sdomain;
2436         while (domain[0] == '.') domain++;
2437         domain_len = strlen(domain);
2438
2439         if (!global_search_state) global_search_state = search_state_new();
2440         if (!global_search_state) return;
2441         global_search_state->num_domains++;
2442
2443         sdomain = (struct search_domain *) malloc(sizeof(struct search_domain) + domain_len);
2444         if (!sdomain) return;
2445         memcpy( ((u8 *) sdomain) + sizeof(struct search_domain), domain, domain_len);
2446         sdomain->next = global_search_state->head;
2447         sdomain->len = domain_len;
2448
2449         global_search_state->head = sdomain;
2450 }
2451
2452 /* reverse the order of members in the postfix list. This is needed because, */
2453 /* when parsing resolv.conf we push elements in the wrong order */
2454 static void
2455 search_reverse(void) {
2456         struct search_domain *cur, *prev = NULL, *next;
2457         cur = global_search_state->head;
2458         while (cur) {
2459                 next = cur->next;
2460                 cur->next = prev;
2461                 prev = cur;
2462                 cur = next;
2463         }
2464
2465         global_search_state->head = prev;
2466 }
2467
2468 /* exported function */
2469 void
2470 evdns_search_add(const char *domain) {
2471         search_postfix_add(domain);
2472 }
2473
2474 /* exported function */
2475 void
2476 evdns_search_ndots_set(const int ndots) {
2477         if (!global_search_state) global_search_state = search_state_new();
2478         if (!global_search_state) return;
2479         global_search_state->ndots = ndots;
2480 }
2481
2482 static void
2483 search_set_from_hostname(void) {
2484         char hostname[HOST_NAME_MAX + 1], *domainname;
2485
2486         search_postfix_clear();
2487         if (gethostname(hostname, sizeof(hostname))) return;
2488         domainname = strchr(hostname, '.');
2489         if (!domainname) return;
2490         search_postfix_add(domainname);
2491 }
2492
2493 /* warning: returns malloced string */
2494 static char *
2495 search_make_new(const struct search_state *const state, int n, const char *const base_name) {
2496         const int base_len = strlen(base_name);
2497         const char need_to_append_dot = base_name[base_len - 1] == '.' ? 0 : 1;
2498         struct search_domain *dom;
2499
2500         for (dom = state->head; dom; dom = dom->next) {
2501                 if (!n--) {
2502                         /* this is the postfix we want */
2503                         /* the actual postfix string is kept at the end of the structure */
2504                         const u8 *const postfix = ((u8 *) dom) + sizeof(struct search_domain);
2505                         const int postfix_len = dom->len;
2506                         char *const newname = (char *) malloc(base_len + need_to_append_dot + postfix_len + 1);
2507                         if (!newname) return NULL;
2508                         memcpy(newname, base_name, base_len);
2509                         if (need_to_append_dot) newname[base_len] = '.';
2510                         memcpy(newname + base_len + need_to_append_dot, postfix, postfix_len);
2511                         newname[base_len + need_to_append_dot + postfix_len] = 0;
2512                         return newname;
2513                 }
2514         }
2515
2516         /* we ran off the end of the list and still didn't find the requested string */
2517         abort();
2518         return NULL; /* unreachable; stops warnings in some compilers. */
2519 }
2520
2521 static int
2522 search_request_new(int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg) {
2523         assert(type == TYPE_A || type == TYPE_AAAA);
2524         if ( ((flags & DNS_QUERY_NO_SEARCH) == 0) &&
2525              global_search_state &&
2526                  global_search_state->num_domains) {
2527                 /* we have some domains to search */
2528                 struct request *req;
2529                 if (string_num_dots(name) >= global_search_state->ndots) {
2530                         req = request_new(type, name, flags, user_callback, user_arg);
2531                         if (!req) return 1;
2532                         req->search_index = -1;
2533                 } else {
2534                         char *const new_name = search_make_new(global_search_state, 0, name);
2535                         if (!new_name) return 1;
2536                         req = request_new(type, new_name, flags, user_callback, user_arg);
2537                         free(new_name);
2538                         if (!req) return 1;
2539                         req->search_index = 0;
2540                 }
2541                 req->search_origname = strdup(name);
2542                 req->search_state = global_search_state;
2543                 req->search_flags = flags;
2544                 global_search_state->refcount++;
2545                 request_submit(req);
2546                 return 0;
2547         } else {
2548                 struct request *const req = request_new(type, name, flags, user_callback, user_arg);
2549                 if (!req) return 1;
2550                 request_submit(req);
2551                 return 0;
2552         }
2553 }
2554
2555 /* this is called when a request has failed to find a name. We need to check */
2556 /* if it is part of a search and, if so, try the next name in the list */
2557 /* returns: */
2558 /*   0 another request has been submitted */
2559 /*   1 no more requests needed */
2560 static int
2561 search_try_next(struct request *const req) {
2562         if (req->search_state) {
2563                 /* it is part of a search */
2564                 char *new_name;
2565                 struct request *newreq;
2566                 req->search_index++;
2567                 if (req->search_index >= req->search_state->num_domains) {
2568                         /* no more postfixes to try, however we may need to try */
2569                         /* this name without a postfix */
2570                         if (string_num_dots(req->search_origname) < req->search_state->ndots) {
2571                                 /* yep, we need to try it raw */
2572                                 newreq = request_new(req->request_type, req->search_origname, req->search_flags, req->user_callback, req->user_pointer);
2573                                 log(EVDNS_LOG_DEBUG, "Search: trying raw query %s", req->search_origname);
2574                                 if (newreq) {
2575                                         request_submit(newreq);
2576                                         return 0;
2577                                 }
2578                         }
2579                         return 1;
2580                 }
2581
2582                 new_name = search_make_new(req->search_state, req->search_index, req->search_origname);
2583                 if (!new_name) return 1;
2584                 log(EVDNS_LOG_DEBUG, "Search: now trying %s (%d)", new_name, req->search_index);
2585                 newreq = request_new(req->request_type, new_name, req->search_flags, req->user_callback, req->user_pointer);
2586                 free(new_name);
2587                 if (!newreq) return 1;
2588                 newreq->search_origname = req->search_origname;
2589                 req->search_origname = NULL;
2590                 newreq->search_state = req->search_state;
2591                 newreq->search_flags = req->search_flags;
2592                 newreq->search_index = req->search_index;
2593                 newreq->search_state->refcount++;
2594                 request_submit(newreq);
2595                 return 0;
2596         }
2597         return 1;
2598 }
2599
2600 static void
2601 search_request_finished(struct request *const req) {
2602         if (req->search_state) {
2603                 search_state_decref(req->search_state);
2604                 req->search_state = NULL;
2605         }
2606         if (req->search_origname) {
2607                 free(req->search_origname);
2608                 req->search_origname = NULL;
2609         }
2610 }
2611
2612 /*/////////////////////////////////////////////////////////////////// */
2613 /* Parsing resolv.conf files */
2614
2615 static void
2616 evdns_resolv_set_defaults(int flags) {
2617         /* if the file isn't found then we assume a local resolver */
2618         if (flags & DNS_OPTION_SEARCH) search_set_from_hostname();
2619         if (flags & DNS_OPTION_NAMESERVERS) evdns_nameserver_ip_add("127.0.0.1");
2620 }
2621
2622 #ifndef HAVE_STRTOK_R
2623 static char *
2624 strtok_r(char *s, const char *delim, char **state) {
2625         return strtok(s, delim);
2626 }
2627 #endif
2628
2629 /* helper version of atoi which returns -1 on error */
2630 static int
2631 strtoint(const char *const str) {
2632         char *endptr;
2633         const int r = strtol(str, &endptr, 10);
2634         if (*endptr) return -1;
2635         return r;
2636 }
2637
2638 /* helper version of atoi that returns -1 on error and clips to bounds. */
2639 static int
2640 strtoint_clipped(const char *const str, int min, int max)
2641 {
2642         int r = strtoint(str);
2643         if (r == -1)
2644                 return r;
2645         else if (r<min)
2646                 return min;
2647         else if (r>max)
2648                 return max;
2649         else
2650                 return r;
2651 }
2652
2653 /* exported function */
2654 int
2655 evdns_set_option(const char *option, const char *val, int flags)
2656 {
2657         if (!strncmp(option, "ndots:", 6)) {
2658                 const int ndots = strtoint(val);
2659                 if (ndots == -1) return -1;
2660                 if (!(flags & DNS_OPTION_SEARCH)) return 0;
2661                 log(EVDNS_LOG_DEBUG, "Setting ndots to %d", ndots);
2662                 if (!global_search_state) global_search_state = search_state_new();
2663                 if (!global_search_state) return -1;
2664                 global_search_state->ndots = ndots;
2665         } else if (!strncmp(option, "timeout:", 8)) {
2666                 const int timeout = strtoint(val);
2667                 if (timeout == -1) return -1;
2668                 if (!(flags & DNS_OPTION_MISC)) return 0;
2669                 log(EVDNS_LOG_DEBUG, "Setting timeout to %d", timeout);
2670                 global_timeout.tv_sec = timeout;
2671         } else if (!strncmp(option, "max-timeouts:", 12)) {
2672                 const int maxtimeout = strtoint_clipped(val, 1, 255);
2673                 if (maxtimeout == -1) return -1;
2674                 if (!(flags & DNS_OPTION_MISC)) return 0;
2675                 log(EVDNS_LOG_DEBUG, "Setting maximum allowed timeouts to %d",
2676                         maxtimeout);
2677                 global_max_nameserver_timeout = maxtimeout;
2678         } else if (!strncmp(option, "max-inflight:", 13)) {
2679                 const int maxinflight = strtoint_clipped(val, 1, 65000);
2680                 if (maxinflight == -1) return -1;
2681                 if (!(flags & DNS_OPTION_MISC)) return 0;
2682                 log(EVDNS_LOG_DEBUG, "Setting maximum inflight requests to %d",
2683                         maxinflight);
2684                 global_max_requests_inflight = maxinflight;
2685         } else if (!strncmp(option, "attempts:", 9)) {
2686                 int retries = strtoint(val);
2687                 if (retries == -1) return -1;
2688                 if (retries > 255) retries = 255;
2689                 if (!(flags & DNS_OPTION_MISC)) return 0;
2690                 log(EVDNS_LOG_DEBUG, "Setting retries to %d", retries);
2691                 global_max_retransmits = retries;
2692         }
2693         return 0;
2694 }
2695
2696 static void
2697 resolv_conf_parse_line(char *const start, int flags) {
2698         char *strtok_state;
2699         static const char *const delims = " \t";
2700 #define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state)
2701
2702         char *const first_token = strtok_r(start, delims, &strtok_state);
2703         if (!first_token) return;
2704
2705         if (!strcmp(first_token, "nameserver") && (flags & DNS_OPTION_NAMESERVERS)) {
2706                 const char *const nameserver = NEXT_TOKEN;
2707                 struct in_addr ina;
2708
2709                 if (inet_aton(nameserver, &ina)) {
2710                         /* address is valid */
2711                         evdns_nameserver_add(ina.s_addr);
2712                 }
2713         } else if (!strcmp(first_token, "domain") && (flags & DNS_OPTION_SEARCH)) {
2714                 const char *const domain = NEXT_TOKEN;
2715                 if (domain) {
2716                         search_postfix_clear();
2717                         search_postfix_add(domain);
2718                 }
2719         } else if (!strcmp(first_token, "search") && (flags & DNS_OPTION_SEARCH)) {
2720                 const char *domain;
2721                 search_postfix_clear();
2722
2723                 while ((domain = NEXT_TOKEN)) {
2724                         search_postfix_add(domain);
2725                 }
2726                 search_reverse();
2727         } else if (!strcmp(first_token, "options")) {
2728                 const char *option;
2729                 while ((option = NEXT_TOKEN)) {
2730                         const char *val = strchr(option, ':');
2731                         evdns_set_option(option, val ? val+1 : "", flags);
2732                 }
2733         }
2734 #undef NEXT_TOKEN
2735 }
2736
2737 /* exported function */
2738 /* returns: */
2739 /*   0 no errors */
2740 /*   1 failed to open file */
2741 /*   2 failed to stat file */
2742 /*   3 file too large */
2743 /*   4 out of memory */
2744 /*   5 short read from file */
2745 int
2746 evdns_resolv_conf_parse(int flags, const char *const filename) {
2747         struct stat st;
2748         int fd, n, r;
2749         u8 *resolv;
2750         char *start;
2751         int err = 0;
2752
2753         log(EVDNS_LOG_DEBUG, "Parsing resolv.conf file %s", filename);
2754
2755         fd = open(filename, O_RDONLY);
2756         if (fd < 0) {
2757                 evdns_resolv_set_defaults(flags);
2758                 return 1;
2759         }
2760
2761         if (fstat(fd, &st)) { err = 2; goto out1; }
2762         if (!st.st_size) {
2763                 evdns_resolv_set_defaults(flags);
2764                 err = (flags & DNS_OPTION_NAMESERVERS) ? 6 : 0;
2765                 goto out1;
2766         }
2767         if (st.st_size > 65535) { err = 3; goto out1; }  /* no resolv.conf should be any bigger */
2768
2769         resolv = (u8 *) malloc((size_t)st.st_size + 1);
2770         if (!resolv) { err = 4; goto out1; }
2771
2772         n = 0;
2773         while ((r = read(fd, resolv+n, (size_t)st.st_size-n)) > 0) {
2774                 n += r;
2775                 if (n == st.st_size)
2776                         break;
2777                 assert(n < st.st_size);
2778         }
2779         if (r < 0) { err = 5; goto out2; }
2780         resolv[n] = 0;   /* we malloced an extra byte; this should be fine. */
2781
2782         start = (char *) resolv;
2783         for (;;) {
2784                 char *const newline = strchr(start, '\n');
2785                 if (!newline) {
2786                         resolv_conf_parse_line(start, flags);
2787                         break;
2788                 } else {
2789                         *newline = 0;
2790                         resolv_conf_parse_line(start, flags);
2791                         start = newline + 1;
2792                 }
2793         }
2794
2795         if (!server_head && (flags & DNS_OPTION_NAMESERVERS)) {
2796                 /* no nameservers were configured. */
2797                 evdns_nameserver_ip_add("127.0.0.1");
2798                 err = 6;
2799         }
2800         if (flags & DNS_OPTION_SEARCH && (!global_search_state || global_search_state->num_domains == 0)) {
2801                 search_set_from_hostname();
2802         }
2803
2804 out2:
2805         free(resolv);
2806 out1:
2807         close(fd);
2808         return err;
2809 }
2810
2811 #ifdef WIN32
2812 /* Add multiple nameservers from a space-or-comma-separated list. */
2813 static int
2814 evdns_nameserver_ip_add_line(const char *ips) {
2815         const char *addr;
2816         char *buf;
2817         int r;
2818         while (*ips) {
2819                 while (ISSPACE(*ips) || *ips == ',' || *ips == '\t')
2820                         ++ips;
2821                 addr = ips;
2822                 while (ISDIGIT(*ips) || *ips == '.' || *ips == ':')
2823                         ++ips;
2824                 buf = malloc(ips-addr+1);
2825                 if (!buf) return 4;
2826                 memcpy(buf, addr, ips-addr);
2827                 buf[ips-addr] = '\0';
2828                 r = evdns_nameserver_ip_add(buf);
2829                 free(buf);
2830                 if (r) return r;
2831         }
2832         return 0;
2833 }
2834
2835 typedef DWORD(WINAPI *GetNetworkParams_fn_t)(FIXED_INFO *, DWORD*);
2836
2837 /* Use the windows GetNetworkParams interface in iphlpapi.dll to */
2838 /* figure out what our nameservers are. */
2839 static int
2840 load_nameservers_with_getnetworkparams(void)
2841 {
2842         /* Based on MSDN examples and inspection of  c-ares code. */
2843         FIXED_INFO *fixed;
2844         HMODULE handle = 0;
2845         ULONG size = sizeof(FIXED_INFO);
2846         void *buf = NULL;
2847         int status = 0, r, added_any;
2848         IP_ADDR_STRING *ns;
2849         GetNetworkParams_fn_t fn;
2850
2851         if (!(handle = LoadLibrary("iphlpapi.dll"))) {
2852                 log(EVDNS_LOG_WARN, "Could not open iphlpapi.dll");
2853                 status = -1;
2854                 goto done;
2855         }
2856         if (!(fn = (GetNetworkParams_fn_t) GetProcAddress(handle, "GetNetworkParams"))) {
2857                 log(EVDNS_LOG_WARN, "Could not get address of function.");
2858                 status = -1;
2859                 goto done;
2860         }
2861
2862         buf = malloc(size);
2863         if (!buf) { status = 4; goto done; }
2864         fixed = buf;
2865         r = fn(fixed, &size);
2866         if (r != ERROR_SUCCESS && r != ERROR_BUFFER_OVERFLOW) {
2867                 status = -1;
2868                 goto done;
2869         }
2870         if (r != ERROR_SUCCESS) {
2871                 free(buf);
2872                 buf = malloc(size);
2873                 if (!buf) { status = 4; goto done; }
2874                 fixed = buf;
2875                 r = fn(fixed, &size);
2876                 if (r != ERROR_SUCCESS) {
2877                         log(EVDNS_LOG_DEBUG, "fn() failed.");
2878                         status = -1;
2879                         goto done;
2880                 }
2881         }
2882
2883         assert(fixed);
2884         added_any = 0;
2885         ns = &(fixed->DnsServerList);
2886         while (ns) {
2887                 r = evdns_nameserver_ip_add_line(ns->IpAddress.String);
2888                 if (r) {
2889                         log(EVDNS_LOG_DEBUG,"Could not add nameserver %s to list,error: %d",
2890                                 (ns->IpAddress.String),(int)GetLastError());
2891                         status = r;
2892                         goto done;
2893                 } else {
2894                         log(EVDNS_LOG_DEBUG,"Succesfully added %s as nameserver",ns->IpAddress.String);
2895                 }
2896
2897                 added_any++;
2898                 ns = ns->Next;
2899         }
2900
2901         if (!added_any) {
2902                 log(EVDNS_LOG_DEBUG, "No nameservers added.");
2903                 status = -1;
2904         }
2905
2906  done:
2907         if (buf)
2908                 free(buf);
2909         if (handle)
2910                 FreeLibrary(handle);
2911         return status;
2912 }
2913
2914 static int
2915 config_nameserver_from_reg_key(HKEY key, const char *subkey)
2916 {
2917         char *buf;
2918         DWORD bufsz = 0, type = 0;
2919         int status = 0;
2920
2921         if (RegQueryValueEx(key, subkey, 0, &type, NULL, &bufsz)
2922             != ERROR_MORE_DATA)
2923                 return -1;
2924         if (!(buf = malloc(bufsz)))
2925                 return -1;
2926
2927         if (RegQueryValueEx(key, subkey, 0, &type, (LPBYTE)buf, &bufsz)
2928             == ERROR_SUCCESS && bufsz > 1) {
2929                 status = evdns_nameserver_ip_add_line(buf);
2930         }
2931
2932         free(buf);
2933         return status;
2934 }
2935
2936 #define SERVICES_KEY "System\\CurrentControlSet\\Services\\"
2937 #define WIN_NS_9X_KEY  SERVICES_KEY "VxD\\MSTCP"
2938 #define WIN_NS_NT_KEY  SERVICES_KEY "Tcpip\\Parameters"
2939
2940 static int
2941 load_nameservers_from_registry(void)
2942 {
2943         int found = 0;
2944         int r;
2945 #define TRY(k, name) \
2946         if (!found && config_nameserver_from_reg_key(k,name) == 0) {    \
2947                 log(EVDNS_LOG_DEBUG,"Found nameservers in %s/%s",#k,name); \
2948                 found = 1;                                              \
2949         } else if (!found) {                                            \
2950                 log(EVDNS_LOG_DEBUG,"Didn't find nameservers in %s/%s", \
2951                     #k,#name);                                          \
2952         }
2953
2954         if (((int)GetVersion()) > 0) { /* NT */
2955                 HKEY nt_key = 0, interfaces_key = 0;
2956
2957                 if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_NT_KEY, 0,
2958                                  KEY_READ, &nt_key) != ERROR_SUCCESS) {
2959                         log(EVDNS_LOG_DEBUG,"Couldn't open nt key, %d",(int)GetLastError());
2960                         return -1;
2961                 }
2962                 r = RegOpenKeyEx(nt_key, "Interfaces", 0,
2963                              KEY_QUERY_VALUE|KEY_ENUMERATE_SUB_KEYS,
2964                              &interfaces_key);
2965                 if (r != ERROR_SUCCESS) {
2966                         log(EVDNS_LOG_DEBUG,"Couldn't open interfaces key, %d",(int)GetLastError());
2967                         return -1;
2968                 }
2969                 TRY(nt_key, "NameServer");
2970                 TRY(nt_key, "DhcpNameServer");
2971                 TRY(interfaces_key, "NameServer");
2972                 TRY(interfaces_key, "DhcpNameServer");
2973                 RegCloseKey(interfaces_key);
2974                 RegCloseKey(nt_key);
2975         } else {
2976                 HKEY win_key = 0;
2977                 if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_9X_KEY, 0,
2978                                  KEY_READ, &win_key) != ERROR_SUCCESS) {
2979                         log(EVDNS_LOG_DEBUG, "Couldn't open registry key, %d", (int)GetLastError());
2980                         return -1;
2981                 }
2982                 TRY(win_key, "NameServer");
2983                 RegCloseKey(win_key);
2984         }
2985
2986         if (found == 0) {
2987                 log(EVDNS_LOG_WARN,"Didn't find any nameservers.");
2988         }
2989
2990         return found ? 0 : -1;
2991 #undef TRY
2992 }
2993
2994 int
2995 evdns_config_windows_nameservers(void)
2996 {
2997         if (load_nameservers_with_getnetworkparams() == 0)
2998                 return 0;
2999         return load_nameservers_from_registry();
3000 }
3001 #endif
3002
3003 int
3004 evdns_init(void)
3005 {
3006         int res = 0;
3007 #ifdef WIN32
3008         res = evdns_config_windows_nameservers();
3009 #else
3010         res = evdns_resolv_conf_parse(DNS_OPTIONS_ALL, "/etc/resolv.conf");
3011 #endif
3012
3013         return (res);
3014 }
3015
3016 const char *
3017 evdns_err_to_string(int err)
3018 {
3019     switch (err) {
3020         case DNS_ERR_NONE: return "no error";
3021         case DNS_ERR_FORMAT: return "misformatted query";
3022         case DNS_ERR_SERVERFAILED: return "server failed";
3023         case DNS_ERR_NOTEXIST: return "name does not exist";
3024         case DNS_ERR_NOTIMPL: return "query not implemented";
3025         case DNS_ERR_REFUSED: return "refused";
3026
3027         case DNS_ERR_TRUNCATED: return "reply truncated or ill-formed";
3028         case DNS_ERR_UNKNOWN: return "unknown";
3029         case DNS_ERR_TIMEOUT: return "request timed out";
3030         case DNS_ERR_SHUTDOWN: return "dns subsystem shut down";
3031         default: return "[Unknown error code]";
3032     }
3033 }
3034
3035 void
3036 evdns_shutdown(int fail_requests)
3037 {
3038         struct nameserver *server, *server_next;
3039         struct search_domain *dom, *dom_next;
3040
3041         while (req_head) {
3042                 if (fail_requests)
3043                         reply_callback(req_head, 0, DNS_ERR_SHUTDOWN, NULL);
3044                 request_finished(req_head, &req_head);
3045         }
3046         while (req_waiting_head) {
3047                 if (fail_requests)
3048                         reply_callback(req_waiting_head, 0, DNS_ERR_SHUTDOWN, NULL);
3049                 request_finished(req_waiting_head, &req_waiting_head);
3050         }
3051         global_requests_inflight = global_requests_waiting = 0;
3052
3053         for (server = server_head; server; server = server_next) {
3054                 server_next = server->next;
3055                 if (server->socket >= 0)
3056                         CLOSE_SOCKET(server->socket);
3057                 (void) event_del(&server->event);
3058                 if (server->state == 0)
3059                         (void) event_del(&server->timeout_event);
3060                 free(server);
3061                 if (server_next == server_head)
3062                         break;
3063         }
3064         server_head = NULL;
3065         global_good_nameservers = 0;
3066
3067         if (global_search_state) {
3068                 for (dom = global_search_state->head; dom; dom = dom_next) {
3069                         dom_next = dom->next;
3070                         free(dom);
3071                 }
3072                 free(global_search_state);
3073                 global_search_state = NULL;
3074         }
3075         evdns_log_fn = NULL;
3076 }
3077
3078 #ifdef EVDNS_MAIN
3079 void
3080 main_callback(int result, char type, int count, int ttl,
3081                           void *addrs, void *orig) {
3082         char *n = (char*)orig;
3083         int i;
3084         for (i = 0; i < count; ++i) {
3085                 if (type == DNS_IPv4_A) {
3086                         printf("%s: %s\n", n, debug_ntoa(((u32*)addrs)[i]));
3087                 } else if (type == DNS_PTR) {
3088                         printf("%s: %s\n", n, ((char**)addrs)[i]);
3089                 }
3090         }
3091         if (!count) {
3092                 printf("%s: No answer (%d)\n", n, result);
3093         }
3094         fflush(stdout);
3095 }
3096 void
3097 evdns_server_callback(struct evdns_server_request *req, void *data)
3098 {
3099         int i, r;
3100         (void)data;
3101         /* dummy; give 192.168.11.11 as an answer for all A questions,
3102          *      give foo.bar.example.com as an answer for all PTR questions. */
3103         for (i = 0; i < req->nquestions; ++i) {
3104                 u32 ans = htonl(0xc0a80b0bUL);
3105                 if (req->questions[i]->type == EVDNS_TYPE_A &&
3106                         req->questions[i]->dns_question_class == EVDNS_CLASS_INET) {
3107                         printf(" -- replying for %s (A)\n", req->questions[i]->name);
3108                         r = evdns_server_request_add_a_reply(req, req->questions[i]->name,
3109                                                                                   1, &ans, 10);
3110                         if (r<0)
3111                                 printf("eeep, didn't work.\n");
3112                 } else if (req->questions[i]->type == EVDNS_TYPE_PTR &&
3113                                    req->questions[i]->dns_question_class == EVDNS_CLASS_INET) {
3114                         printf(" -- replying for %s (PTR)\n", req->questions[i]->name);
3115                         r = evdns_server_request_add_ptr_reply(req, NULL, req->questions[i]->name,
3116                                                                                         "foo.bar.example.com", 10);
3117                 } else {
3118                         printf(" -- skipping %s [%d %d]\n", req->questions[i]->name,
3119                                    req->questions[i]->type, req->questions[i]->dns_question_class);
3120                 }
3121         }
3122
3123         r = evdns_request_respond(req, 0);
3124         if (r<0)
3125                 printf("eeek, couldn't send reply.\n");
3126 }
3127
3128 void
3129 logfn(int is_warn, const char *msg) {
3130   (void) is_warn;
3131   fprintf(stderr, "%s\n", msg);
3132 }
3133 int
3134 main(int c, char **v) {
3135         int idx;
3136         int reverse = 0, verbose = 1, servertest = 0;
3137         if (c<2) {
3138                 fprintf(stderr, "syntax: %s [-x] [-v] hostname\n", v[0]);
3139                 fprintf(stderr, "syntax: %s [-servertest]\n", v[0]);
3140                 return 1;
3141         }
3142         idx = 1;
3143         while (idx < c && v[idx][0] == '-') {
3144                 if (!strcmp(v[idx], "-x"))
3145                         reverse = 1;
3146                 else if (!strcmp(v[idx], "-v"))
3147                         verbose = 1;
3148                 else if (!strcmp(v[idx], "-servertest"))
3149                         servertest = 1;
3150                 else
3151                         fprintf(stderr, "Unknown option %s\n", v[idx]);
3152                 ++idx;
3153         }
3154         event_init();
3155         if (verbose)
3156                 evdns_set_log_fn(logfn);
3157         evdns_resolv_conf_parse(DNS_OPTION_NAMESERVERS, "/etc/resolv.conf");
3158         if (servertest) {
3159                 int sock;
3160                 struct sockaddr_in my_addr;
3161                 sock = socket(PF_INET, SOCK_DGRAM, 0);
3162                 evutil_make_socket_nonblocking(sock);
3163                 my_addr.sin_family = AF_INET;
3164                 my_addr.sin_port = htons(10053);
3165                 my_addr.sin_addr.s_addr = INADDR_ANY;
3166                 if (bind(sock, (struct sockaddr*)&my_addr, sizeof(my_addr))<0) {
3167                         perror("bind");
3168                         exit(1);
3169                 }
3170                 evdns_add_server_port(sock, 0, evdns_server_callback, NULL);
3171         }
3172         for (; idx < c; ++idx) {
3173                 if (reverse) {
3174                         struct in_addr addr;
3175                         if (!inet_aton(v[idx], &addr)) {
3176                                 fprintf(stderr, "Skipping non-IP %s\n", v[idx]);
3177                                 continue;
3178                         }
3179                         fprintf(stderr, "resolving %s...\n",v[idx]);
3180                         evdns_resolve_reverse(&addr, 0, main_callback, v[idx]);
3181                 } else {
3182                         fprintf(stderr, "resolving (fwd) %s...\n",v[idx]);
3183                         evdns_resolve_ipv4(v[idx], 0, main_callback, v[idx]);
3184                 }
3185         }
3186         fflush(stdout);
3187         event_dispatch();
3188         return 0;
3189 }
3190 #endif