dns.c

/* Copyright (c) 2003-2004, Roger Dingledine.
 * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
 * Copyright (c) 2007-2011, The Tor Project, Inc. */
/* See LICENSE for licensing information */

/**
 * \file dns.c
 * \brief Implements a local cache for DNS results for Tor servers.
 * This is implemented as a wrapper around Adam Langley's eventdns.c code.
 * (We can't just use gethostbyname() and friends because we really need to
 * be nonblocking.)
 **/

#include "or.h"
#include "circuitlist.h"
#include "circuituse.h"
#include "config.h"
#include "connection.h"
#include "connection_edge.h"
#include "control.h"
#include "dns.h"
#include "main.h"
#include "policies.h"
#include "relay.h"
#include "router.h"
#include "ht.h"
#ifdef HAVE_EVENT2_DNS_H
#include <event2/event.h>
#include <event2/dns.h>
#else
#include <event.h>
#include "eventdns.h"
#ifndef HAVE_EVDNS_SET_DEFAULT_OUTGOING_BIND_ADDRESS
#define HAVE_EVDNS_SET_DEFAULT_OUTGOING_BIND_ADDRESS
#endif
#endif

#ifndef HAVE_EVENT2_DNS_H
struct evdns_base;
struct evdns_request;
#define evdns_base_new(x,y) tor_malloc(1)
#define evdns_base_clear_nameservers_and_suspend(base) \
  evdns_clear_nameservers_and_suspend()
#define evdns_base_search_clear(base) evdns_search_clear()
#define evdns_base_set_default_outgoing_bind_address(base, a, len)  \
  evdns_set_default_outgoing_bind_address((a),(len))
#define evdns_base_resolv_conf_parse(base, options, fname) \
  evdns_resolv_conf_parse((options), (fname))
#define evdns_base_count_nameservers(base)      \
  evdns_count_nameservers()
#define evdns_base_resume(base)                 \
  evdns_resume()
#define evdns_base_config_windows_nameservers(base)     \
  evdns_config_windows_nameservers()
#define evdns_base_set_option_(base, opt, val) \
  evdns_set_option((opt),(val),DNS_OPTIONS_ALL)
/* Note: our internal eventdns.c, plus Libevent 1.4, used a 1 return to
 * signify failure to launch a resolve. Libevent 2.0 uses a -1 return to
 * signify a failure on a resolve, though if we're on Libevent 2.0, we should
 * have event2/dns.h and never hit these macros.  Regardless, 0 is success. */
#define evdns_base_resolve_ipv4(base, addr, options, cb, ptr) \
  ((evdns_resolve_ipv4((addr), (options), (cb), (ptr))!=0)    \
   ? NULL : ((void*)1))
#define evdns_base_resolve_reverse(base, addr, options, cb, ptr)        \
  ((evdns_resolve_reverse((addr), (options), (cb), (ptr))!=0)           \
   ? NULL : ((void*)1))
#define evdns_base_resolve_reverse_ipv6(base, addr, options, cb, ptr)   \
  ((evdns_resolve_reverse_ipv6((addr), (options), (cb), (ptr))!=0)      \
   ? NULL : ((void*)1))

#elif defined(LIBEVENT_VERSION_NUMBER) && LIBEVENT_VERSION_NUMBER < 0x02000303
#define evdns_base_set_option_(base, opt, val) \
  evdns_base_set_option((base), (opt),(val),DNS_OPTIONS_ALL)

#else
#define evdns_base_set_option_ evdns_base_set_option

#endif

/** Longest hostname we're willing to resolve. */
#define MAX_ADDRESSLEN 256

/** How long will we wait for an answer from the resolver before we decide
 * that the resolver is wedged? */
#define RESOLVE_MAX_TIMEOUT 300

/** Possible outcomes from hostname lookup: permanent failure,
 * transient (retryable) failure, and success. */
#define DNS_RESOLVE_FAILED_TRANSIENT 1
#define DNS_RESOLVE_FAILED_PERMANENT 2
#define DNS_RESOLVE_SUCCEEDED 3

/** Our evdns_base; this structure handles all our name lookups. */
static struct evdns_base *the_evdns_base = NULL;

/** Have we currently configured nameservers with eventdns? */
static int nameservers_configured = 0;
/** Did our most recent attempt to configure nameservers with eventdns fail? */
static int nameserver_config_failed = 0;
/** What was the resolv_conf fname we last used when configuring the
 * nameservers? Used to check whether we need to reconfigure. */
static char *resolv_conf_fname = NULL;
/** What was the mtime on the resolv.conf file we last used when configuring
 * the nameservers?  Used to check whether we need to reconfigure. */
static time_t resolv_conf_mtime = 0;

/** Linked list of connections waiting for a DNS answer. */
typedef struct pending_connection_t {
  edge_connection_t *conn;
  struct pending_connection_t *next;
} pending_connection_t;

/** Value of 'magic' field for cached_resolve_t.  Used to try to catch bad
 * pointers and memory stomping. */
#define CACHED_RESOLVE_MAGIC 0x1234F00D

/* Possible states for a cached resolve_t */
/** We are waiting for the resolver system to tell us an answer here.
 * When we get one, or when we time out, the state of this cached_resolve_t
 * will become "DONE" and we'll possibly add a CACHED_VALID or a CACHED_FAILED
 * entry. This cached_resolve_t will be in the hash table so that we will
 * know not to launch more requests for this addr, but rather to add more
 * connections to the pending list for the addr. */
#define CACHE_STATE_PENDING 0
/** This used to be a pending cached_resolve_t, and we got an answer for it.
 * Now we're waiting for this cached_resolve_t to expire.  This should
 * have no pending connections, and should not appear in the hash table. */
#define CACHE_STATE_DONE 1
/** We are caching an answer for this address. This should have no pending
 * connections, and should appear in the hash table. */
#define CACHE_STATE_CACHED_VALID 2
/** We are caching a failure for this address. This should have no pending
 * connections, and should appear in the hash table */
#define CACHE_STATE_CACHED_FAILED 3

/** A DNS request: possibly completed, possibly pending; cached_resolve
 * structs are stored at the OR side in a hash table, and as a linked
 * list from oldest to newest.
 */
typedef struct cached_resolve_t {
  HT_ENTRY(cached_resolve_t) node;
  uint32_t magic;
  char address[MAX_ADDRESSLEN]; /**< The hostname to be resolved. */
  union {
    struct {
      struct in6_addr addr6; /**< IPv6 addr for <b>address</b>. */
      uint32_t addr;  /**< IPv4 addr for <b>address</b>. */
    } a;
    char *hostname; /**< Hostname for <b>address</b> (if a reverse lookup) */
  } result;
  uint8_t state; /**< Is this cached entry pending/done/valid/failed? */
  uint8_t is_reverse; /**< Is this a reverse (addr-to-hostname) lookup? */
  time_t expire; /**< Remove items from cache after this time. */
  uint32_t ttl; /**< What TTL did the nameserver tell us? */
  /** Connections that want to know when we get an answer for this resolve. */
  pending_connection_t *pending_connections;
  /** Position of this element in the heap*/
  int minheap_idx;
} cached_resolve_t;

static void purge_expired_resolves(time_t now);
static void dns_found_answer(const char *address, uint8_t is_reverse,
                             uint32_t addr, const char *hostname, char outcome,
                             uint32_t ttl);
static void send_resolved_cell(edge_connection_t *conn, uint8_t answer_type);
static int launch_resolve(edge_connection_t *exitconn);
static void add_wildcarded_test_address(const char *address);
static int configure_nameservers(int force);
static int answer_is_wildcarded(const char *ip);
static int dns_resolve_impl(edge_connection_t *exitconn, int is_resolve,
                            or_circuit_t *oncirc, char **resolved_to_hostname);
#ifdef DEBUG_DNS_CACHE
static void _assert_cache_ok(void);
#define assert_cache_ok() _assert_cache_ok()
#else
#define assert_cache_ok() STMT_NIL
#endif
static void assert_resolve_ok(cached_resolve_t *resolve);

/** Hash table of cached_resolve objects. */
static HT_HEAD(cache_map, cached_resolve_t) cache_root;

/** Function to compare hashed resolves on their addresses; used to
 * implement hash tables. */
static INLINE int
cached_resolves_eq(cached_resolve_t *a, cached_resolve_t *b)
{
  /* make this smarter one day? */
  assert_resolve_ok(a); // Not b; b may be just a search.
  return !strncmp(a->address, b->address, MAX_ADDRESSLEN);
}

/** Hash function for cached_resolve objects */
static INLINE unsigned int
cached_resolve_hash(cached_resolve_t *a)
{
  return ht_string_hash(a->address);
}

HT_PROTOTYPE(cache_map, cached_resolve_t, node, cached_resolve_hash,
             cached_resolves_eq)
HT_GENERATE(cache_map, cached_resolve_t, node, cached_resolve_hash,
            cached_resolves_eq, 0.6, malloc, realloc, free)

/** Initialize the DNS cache. */
static void
init_cache_map(void)
{
  HT_INIT(cache_map, &cache_root);
}

/** Helper: called by eventdns when eventdns wants to log something. */
static void
evdns_log_cb(int warn, const char *msg)
{
  const char *cp;
  static int all_down = 0;
  int severity = warn ? LOG_WARN : LOG_INFO;
  if (!strcmpstart(msg, "Resolve requested for") &&
      get_options()->SafeLogging) {
    log(LOG_INFO, LD_EXIT, "eventdns: Resolve requested.");
    return;
  } else if (!strcmpstart(msg, "Search: ")) {
    return;
  }
  if (!strcmpstart(msg, "Nameserver ") && (cp=strstr(msg, " has failed: "))) {
    char *ns = tor_strndup(msg+11, cp-(msg+11));
    const char *err = strchr(cp, ':')+2;
    tor_assert(err);
    /* Don't warn about a single failed nameserver; we'll warn with 'all
     * nameservers have failed' if we're completely out of nameservers;
     * otherwise, the situation is tolerable. */
    severity = LOG_INFO;
    control_event_server_status(LOG_NOTICE,
                                "NAMESERVER_STATUS NS=%s STATUS=DOWN ERR=%s",
                                ns, escaped(err));
    tor_free(ns);
  } else if (!strcmpstart(msg, "Nameserver ") &&
             (cp=strstr(msg, " is back up"))) {
    char *ns = tor_strndup(msg+11, cp-(msg+11));
    severity = (all_down && warn) ? LOG_NOTICE : LOG_INFO;
    all_down = 0;
    control_event_server_status(LOG_NOTICE,
                                "NAMESERVER_STATUS NS=%s STATUS=UP", ns);
    tor_free(ns);
  } else if (!strcmp(msg, "All nameservers have failed")) {
    control_event_server_status(LOG_WARN, "NAMESERVER_ALL_DOWN");
    all_down = 1;
  }
  log(severity, LD_EXIT, "eventdns: %s", msg);
}

/** Helper: passed to eventdns.c as a callback so it can generate random
 * numbers for transaction IDs and 0x20-hack coding. */
static void
_dns_randfn(char *b, size_t n)
{
  crypto_rand(b,n);
}

/** Initialize the DNS subsystem; called by the OR process. */
int
dns_init(void)
{
  init_cache_map();
  evdns_set_random_bytes_fn(_dns_randfn);
  if (server_mode(get_options())) {
    int r = configure_nameservers(1);
    return r;
  }
  return 0;
}

/** Called when DNS-related options change (or may have changed).  Returns -1
 * on failure, 0 on success. */
int
dns_reset(void)
{
  const or_options_t *options = get_options();
  if (! server_mode(options)) {

    if (!the_evdns_base) {
      if (!(the_evdns_base = evdns_base_new(tor_libevent_get_base(), 0))) {
        log_err(LD_BUG, "Couldn't create an evdns_base");
        return -1;
      }
    }

    evdns_base_clear_nameservers_and_suspend(the_evdns_base);
    evdns_base_search_clear(the_evdns_base);
    nameservers_configured = 0;
    tor_free(resolv_conf_fname);
    resolv_conf_mtime = 0;
  } else {
    if (configure_nameservers(0) < 0) {
      return -1;
    }
  }
  return 0;
}

/** Return true iff the most recent attempt to initialize the DNS subsystem
 * failed. */
int
has_dns_init_failed(void)
{
  return nameserver_config_failed;
}

/** Helper: Given a TTL from a DNS response, determine what TTL to give the
 * OP that asked us to resolve it. */
uint32_t
dns_clip_ttl(uint32_t ttl)
{
  if (ttl < MIN_DNS_TTL)
    return MIN_DNS_TTL;
  else if (ttl > MAX_DNS_TTL)
    return MAX_DNS_TTL;
  else
    return ttl;
}

/** Helper: Given a TTL from a DNS response, determine how long to hold it in
 * our cache. */
static uint32_t
dns_get_expiry_ttl(uint32_t ttl)
{
  if (ttl < MIN_DNS_TTL)
    return MIN_DNS_TTL;
  else if (ttl > MAX_DNS_ENTRY_AGE)
    return MAX_DNS_ENTRY_AGE;
  else
    return ttl;
}

/** Helper: free storage held by an entry in the DNS cache. */
static void
_free_cached_resolve(cached_resolve_t *r)
{
  if (!r)
    return;
  while (r->pending_connections) {
    pending_connection_t *victim = r->pending_connections;
    r->pending_connections = victim->next;
    tor_free(victim);
  }
  if (r->is_reverse)
    tor_free(r->result.hostname);
  r->magic = 0xFF00FF00;
  tor_free(r);
}

/** Compare two cached_resolve_t pointers by expiry time, and return
 * less-than-zero, zero, or greater-than-zero as appropriate. Used for
 * the priority queue implementation. */
static int
_compare_cached_resolves_by_expiry(const void *_a, const void *_b)
{
  const cached_resolve_t *a = _a, *b = _b;
  if (a->expire < b->expire)
    return -1;
  else if (a->expire == b->expire)
    return 0;
  else
    return 1;
}

/** Priority queue of cached_resolve_t objects to let us know when they
 * will expire. */
static smartlist_t *cached_resolve_pqueue = NULL;

/** Set an expiry time for a cached_resolve_t, and add it to the expiry
 * priority queue */
static void
set_expiry(cached_resolve_t *resolve, time_t expires)
{
  tor_assert(resolve && resolve->expire == 0);
  if (!cached_resolve_pqueue)
    cached_resolve_pqueue = smartlist_create();
  resolve->expire = expires;
  smartlist_pqueue_add(cached_resolve_pqueue,
                       _compare_cached_resolves_by_expiry,
                       STRUCT_OFFSET(cached_resolve_t, minheap_idx),
                       resolve);
}

/** Free all storage held in the DNS cache and related structures. */
void
dns_free_all(void)
{
  cached_resolve_t **ptr, **next, *item;
  assert_cache_ok();
  if (cached_resolve_pqueue) {
    SMARTLIST_FOREACH(cached_resolve_pqueue, cached_resolve_t *, res,
      {
        if (res->state == CACHE_STATE_DONE)
          _free_cached_resolve(res);
      });
  }
  for (ptr = HT_START(cache_map, &cache_root); ptr != NULL; ptr = next) {
    item = *ptr;
    next = HT_NEXT_RMV(cache_map, &cache_root, ptr);
    _free_cached_resolve(item);
  }
  HT_CLEAR(cache_map, &cache_root);
  smartlist_free(cached_resolve_pqueue);
  cached_resolve_pqueue = NULL;
  tor_free(resolv_conf_fname);
}

/** Remove every cached_resolve whose <b>expire</b> time is before or
 * equal to <b>now</b> from the cache. */
static void
purge_expired_resolves(time_t now)
{
  cached_resolve_t *resolve, *removed;
  pending_connection_t *pend;
  edge_connection_t *pendconn;

  assert_cache_ok();
  if (!cached_resolve_pqueue)
    return;

  while (smartlist_len(cached_resolve_pqueue)) {
    resolve = smartlist_get(cached_resolve_pqueue, 0);
    if (resolve->expire > now)
      break;
    smartlist_pqueue_pop(cached_resolve_pqueue,
                         _compare_cached_resolves_by_expiry,
                         STRUCT_OFFSET(cached_resolve_t, minheap_idx));

    if (resolve->state == CACHE_STATE_PENDING) {
      log_debug(LD_EXIT,
                "Expiring a dns resolve %s that's still pending. Forgot to "
                "cull it? DNS resolve didn't tell us about the timeout?",
                escaped_safe_str(resolve->address));
    } else if (resolve->state == CACHE_STATE_CACHED_VALID ||
               resolve->state == CACHE_STATE_CACHED_FAILED) {
      log_debug(LD_EXIT,
                "Forgetting old cached resolve (address %s, expires %lu)",
                escaped_safe_str(resolve->address),
                (unsigned long)resolve->expire);
      tor_assert(!resolve->pending_connections);
    } else {
      tor_assert(resolve->state == CACHE_STATE_DONE);
      tor_assert(!resolve->pending_connections);
    }

    if (resolve->pending_connections) {
      log_debug(LD_EXIT,
                "Closing pending connections on timed-out DNS resolve!");
      tor_fragile_assert();
      while (resolve->pending_connections) {
        pend = resolve->pending_connections;
        resolve->pending_connections = pend->next;
        /* Connections should only be pending if they have no socket. */
        tor_assert(pend->conn->_base.s == -1);
        pendconn = pend->conn;
        connection_edge_end(pendconn, END_STREAM_REASON_TIMEOUT);
        circuit_detach_stream(circuit_get_by_edge_conn(pendconn), pendconn);
        connection_free(TO_CONN(pendconn));
        tor_free(pend);
      }
    }

    if (resolve->state == CACHE_STATE_CACHED_VALID ||
        resolve->state == CACHE_STATE_CACHED_FAILED ||
        resolve->state == CACHE_STATE_PENDING) {
      removed = HT_REMOVE(cache_map, &cache_root, resolve);
      if (removed != resolve) {
        log_err(LD_BUG, "The expired resolve we purged didn't match any in"
                " the cache. Tried to purge %s (%p); instead got %s (%p).",
                resolve->address, (void*)resolve,
                removed ? removed->address : "NULL", (void*)removed);
      }
      tor_assert(removed == resolve);
    } else {
      /* This should be in state DONE. Make sure it's not in the cache. */
      cached_resolve_t *tmp = HT_FIND(cache_map, &cache_root, resolve);
      tor_assert(tmp != resolve);
    }
    if (resolve->is_reverse)
      tor_free(resolve->result.hostname);
    resolve->magic = 0xF0BBF0BB;
    tor_free(resolve);
  }

  assert_cache_ok();
}

/** Send a response to the RESOLVE request of a connection.
 * <b>answer_type</b> must be one of
 * RESOLVED_TYPE_(IPV4|ERROR|ERROR_TRANSIENT).
 *
 * If <b>circ</b> is provided, and we have a cached answer, send the
 * answer back along circ; otherwise, send the answer back along
 * <b>conn</b>'s attached circuit.
 */
static void
send_resolved_cell(edge_connection_t *conn, uint8_t answer_type)
{
  char buf[RELAY_PAYLOAD_SIZE];
  size_t buflen;
  uint32_t ttl;

  buf[0] = answer_type;
  ttl = dns_clip_ttl(conn->address_ttl);

  switch (answer_type)
    {
    case RESOLVED_TYPE_IPV4:
      buf[1] = 4;
      set_uint32(buf+2, tor_addr_to_ipv4n(&conn->_base.addr));
      set_uint32(buf+6, htonl(ttl));
      buflen = 10;
      break;
    /*XXXX IP6 need ipv6 implementation */
    case RESOLVED_TYPE_ERROR_TRANSIENT:
    case RESOLVED_TYPE_ERROR:
      {
        const char *errmsg = "Error resolving hostname";
        size_t msglen = strlen(errmsg);

        buf[1] = msglen;
        strlcpy(buf+2, errmsg, sizeof(buf)-2);
        set_uint32(buf+2+msglen, htonl(ttl));
        buflen = 6+msglen;
        break;
      }
    default:
      tor_assert(0);
      return;
    }
  // log_notice(LD_EXIT, "Sending a regular RESOLVED reply: ");

  connection_edge_send_command(conn, RELAY_COMMAND_RESOLVED, buf, buflen);
}

/** Send a response to the RESOLVE request of a connection for an in-addr.arpa
 * address on connection <b>conn</b> which yielded the result <b>hostname</b>.
 * The answer type will be RESOLVED_HOSTNAME.
 *
 * If <b>circ</b> is provided, and we have a cached answer, send the
 * answer back along circ; otherwise, send the answer back along
 * <b>conn</b>'s attached circuit.
 */
static void
send_resolved_hostname_cell(edge_connection_t *conn, const char *hostname)
{
  char buf[RELAY_PAYLOAD_SIZE];
  size_t buflen;
  uint32_t ttl;
  size_t namelen = strlen(hostname);
  tor_assert(hostname);

  tor_assert(namelen < 256);
  ttl = dns_clip_ttl(conn->address_ttl);

  buf[0] = RESOLVED_TYPE_HOSTNAME;
  buf[1] = (uint8_t)namelen;
  memcpy(buf+2, hostname, namelen);
  set_uint32(buf+2+namelen, htonl(ttl));
  buflen = 2+namelen+4;

  // log_notice(LD_EXIT, "Sending a reply RESOLVED reply: %s", hostname);
  connection_edge_send_command(conn, RELAY_COMMAND_RESOLVED, buf, buflen);
  // log_notice(LD_EXIT, "Sent");
}

/** See if we have a cache entry for <b>exitconn</b>-\>address. if so,
 * if resolve valid, put it into <b>exitconn</b>-\>addr and return 1.
 * If resolve failed, free exitconn and return -1.
 *
 * (For EXIT_PURPOSE_RESOLVE connections, send back a RESOLVED error cell
 * on returning -1.  For EXIT_PURPOSE_CONNECT connections, there's no
 * need to send back an END cell, since connection_exit_begin_conn will
 * do that for us.)
 *
 * If we have a cached answer, send the answer back along <b>exitconn</b>'s
 * circuit.
 *
 * Else, if seen before and pending, add conn to the pending list,
 * and return 0.
 *
 * Else, if not seen before, add conn to pending list, hand to
 * dns farm, and return 0.
 *
 * Exitconn's on_circuit field must be set, but exitconn should not
 * yet be linked onto the n_streams/resolving_streams list of that circuit.
 * On success, link the connection to n_streams if it's an exit connection.
 * On "pending", link the connection to resolving streams.  Otherwise,
 * clear its on_circuit field.
 */
int
dns_resolve(edge_connection_t *exitconn)
{
  or_circuit_t *oncirc = TO_OR_CIRCUIT(exitconn->on_circuit);
  int is_resolve, r;
  char *hostname = NULL;
  is_resolve = exitconn->_base.purpose == EXIT_PURPOSE_RESOLVE;

  r = dns_resolve_impl(exitconn, is_resolve, oncirc, &hostname);

  switch (r) {
    case 1:
      /* We got an answer without a lookup -- either the answer was
       * cached, or it was obvious (like an IP address). */
      if (is_resolve) {
        /* Send the answer back right now, and detach. */
        if (hostname)
          send_resolved_hostname_cell(exitconn, hostname);
        else
          send_resolved_cell(exitconn, RESOLVED_TYPE_IPV4);
        exitconn->on_circuit = NULL;
      } else {
        /* Add to the n_streams list; the calling function will send back a
         * connected cell. */
        exitconn->next_stream = oncirc->n_streams;
        oncirc->n_streams = exitconn;
      }
      break;
    case 0:
      /* The request is pending: add the connection into the linked list of
       * resolving_streams on this circuit. */
      exitconn->_base.state = EXIT_CONN_STATE_RESOLVING;
      exitconn->next_stream = oncirc->resolving_streams;
      oncirc->resolving_streams = exitconn;
      break;
    case -2:
    case -1:
      /* The request failed before it could start: cancel this connection,
       * and stop everybody waiting for the same connection. */
      if (is_resolve) {
        send_resolved_cell(exitconn,
             (r == -1) ? RESOLVED_TYPE_ERROR : RESOLVED_TYPE_ERROR_TRANSIENT);
      }

      exitconn->on_circuit = NULL;

      dns_cancel_pending_resolve(exitconn->_base.address);

      if (!exitconn->_base.marked_for_close) {
        connection_free(TO_CONN(exitconn));
        // XXX ... and we just leak exitconn otherwise? -RD
        // If it's marked for close, it's on closeable_connection_lst in
        // main.c.  If it's on the closeable list, it will get freed from
        // main.c. -NM
        // "<armadev> If that's true, there are other bugs around, where we
        //  don't check if it's marked, and will end up double-freeing."
        // On the other hand, I don't know of any actual bugs here, so this
        // shouldn't be holding up the rc. -RD
      }
      break;
    default:
      tor_assert(0);
  }

  tor_free(hostname);
  return r;
}

/** Helper function for dns_resolve: same functionality, but does not handle:
 *     - marking connections on error and clearing their on_circuit
 *     - linking connections to n_streams/resolving_streams,
 *     - sending resolved cells if we have an answer/error right away,
 *
 * Return -2 on a transient error. If it's a reverse resolve and it's
 * successful, sets *<b>hostname_out</b> to a newly allocated string
 * holding the cached reverse DNS value.
 */
static int
dns_resolve_impl(edge_connection_t *exitconn, int is_resolve,
                 or_circuit_t *oncirc, char **hostname_out)
{
  cached_resolve_t *resolve;
  cached_resolve_t search;
  pending_connection_t *pending_connection;
  const routerinfo_t *me;
  tor_addr_t addr;
  time_t now = time(NULL);
  uint8_t is_reverse = 0;
  int r;
  assert_connection_ok(TO_CONN(exitconn), 0);
  tor_assert(exitconn->_base.s == -1);
  assert_cache_ok();
  tor_assert(oncirc);

  /* first check if exitconn->_base.address is an IP. If so, we already
   * know the answer. */
  if (tor_addr_parse(&addr, exitconn->_base.address) >= 0) {
    if (tor_addr_family(&addr) == AF_INET) {
      tor_addr_copy(&exitconn->_base.addr, &addr);
      exitconn->address_ttl = DEFAULT_DNS_TTL;
      return 1;
    } else {
      /* XXXX IPv6 */
      return -1;
    }
  }

  /* If we're a non-exit, don't even do DNS lookups. */
  if (!(me = router_get_my_routerinfo()) ||
      policy_is_reject_star(me->exit_policy)) {
    return -1;
  }
  if (address_is_invalid_destination(exitconn->_base.address, 0)) {
    log(LOG_PROTOCOL_WARN, LD_EXIT,
        "Rejecting invalid destination address %s",
        escaped_safe_str(exitconn->_base.address));
    return -1;
  }

  /* then take this opportunity to see if there are any expired
   * resolves in the hash table. */
  purge_expired_resolves(now);

  /* lower-case exitconn->_base.address, so it's in canonical form */
  tor_strlower(exitconn->_base.address);

  /* Check whether this is a reverse lookup.  If it's malformed, or it's a
   * .in-addr.arpa address but this isn't a resolve request, kill the
   * connection.
   */
  if ((r = tor_addr_parse_reverse_lookup_name(&addr, exitconn->_base.address,
                                              AF_UNSPEC, 0)) != 0) {
    if (r == 1) {
      is_reverse = 1;
      if (tor_addr_is_internal(&addr, 0)) /* internal address? */
        return -1;
    }

    if (!is_reverse || !is_resolve) {
      if (!is_reverse)
        log_info(LD_EXIT, "Bad .in-addr.arpa address \"%s\"; sending error.",
                 escaped_safe_str(exitconn->_base.address));
      else if (!is_resolve)
        log_info(LD_EXIT,
                 "Attempt to connect to a .in-addr.arpa address \"%s\"; "
                 "sending error.",
                 escaped_safe_str(exitconn->_base.address));

      return -1;
    }
    //log_notice(LD_EXIT, "Looks like an address %s",
    //exitconn->_base.address);
  }

  /* now check the hash table to see if 'address' is already there. */
  strlcpy(search.address, exitconn->_base.address, sizeof(search.address));
  resolve = HT_FIND(cache_map, &cache_root, &search);
  if (resolve && resolve->expire > now) { /* already there */
    switch (resolve->state) {
      case CACHE_STATE_PENDING:
        /* add us to the pending list */
        pending_connection = tor_malloc_zero(
                                      sizeof(pending_connection_t));
        pending_connection->conn = exitconn;
        pending_connection->next = resolve->pending_connections;
        resolve->pending_connections = pending_connection;
        log_debug(LD_EXIT,"Connection (fd %d) waiting for pending DNS "
                  "resolve of %s", exitconn->_base.s,
                  escaped_safe_str(exitconn->_base.address));
        return 0;
      case CACHE_STATE_CACHED_VALID:
        log_debug(LD_EXIT,"Connection (fd %d) found cached answer for %s",
                  exitconn->_base.s,
                  escaped_safe_str(resolve->address));
        exitconn->address_ttl = resolve->ttl;
        if (resolve->is_reverse) {
          tor_assert(is_resolve);
          *hostname_out = tor_strdup(resolve->result.hostname);
        } else {
          tor_addr_from_ipv4h(&exitconn->_base.addr, resolve->result.a.addr);
        }
        return 1;
      case CACHE_STATE_CACHED_FAILED:
        log_debug(LD_EXIT,"Connection (fd %d) found cached error for %s",
                  exitconn->_base.s,
                  escaped_safe_str(exitconn->_base.address));
        return -1;
      case CACHE_STATE_DONE:
        log_err(LD_BUG, "Found a 'DONE' dns resolve still in the cache.");
        tor_fragile_assert();
    }
    tor_assert(0);
  }
  tor_assert(!resolve);
  /* not there, need to add it */
  resolve = tor_malloc_zero(sizeof(cached_resolve_t));
  resolve->magic = CACHED_RESOLVE_MAGIC;
  resolve->state = CACHE_STATE_PENDING;
  resolve->minheap_idx = -1;
  resolve->is_reverse = is_reverse;
  strlcpy(resolve->address, exitconn->_base.address, sizeof(resolve->address));

  /* add this connection to the pending list */
  pending_connection = tor_malloc_zero(sizeof(pending_connection_t));
  pending_connection->conn = exitconn;
  resolve->pending_connections = pending_connection;

  /* Add this resolve to the cache and priority queue. */
  HT_INSERT(cache_map, &cache_root, resolve);
  set_expiry(resolve, now + RESOLVE_MAX_TIMEOUT);

  log_debug(LD_EXIT,"Launching %s.",
            escaped_safe_str(exitconn->_base.address));
  assert_cache_ok();

  return launch_resolve(exitconn);
}

/** Log an error and abort if conn is waiting for a DNS resolve.
 */
void
assert_connection_edge_not_dns_pending(edge_connection_t *conn)
{
  pending_connection_t *pend;
  cached_resolve_t search;

#if 1
  cached_resolve_t *resolve;
  strlcpy(search.address, conn->_base.address, sizeof(search.address));
  resolve = HT_FIND(cache_map, &cache_root, &search);
  if (!resolve)
    return;
  for (pend = resolve->pending_connections; pend; pend = pend->next) {
    tor_assert(pend->conn != conn);
  }
#else
  cached_resolve_t **resolve;
  HT_FOREACH(resolve, cache_map, &cache_root) {
    for (pend = (*resolve)->pending_connections; pend; pend = pend->next) {
      tor_assert(pend->conn != conn);
    }
  }
#endif
}

/** Log an error and abort if any connection waiting for a DNS resolve is
 * corrupted. */
void
assert_all_pending_dns_resolves_ok(void)
{
  pending_connection_t *pend;
  cached_resolve_t **resolve;

  HT_FOREACH(resolve, cache_map, &cache_root) {
    for (pend = (*resolve)->pending_connections;
         pend;
         pend = pend->next) {
      assert_connection_ok(TO_CONN(pend->conn), 0);
      tor_assert(pend->conn->_base.s == -1);
      tor_assert(!connection_in_array(TO_CONN(pend->conn)));
    }
  }
}

/** Remove <b>conn</b> from the list of connections waiting for conn-\>address.
 */
void
connection_dns_remove(edge_connection_t *conn)
{
  pending_connection_t *pend, *victim;
  cached_resolve_t search;
  cached_resolve_t *resolve;

  tor_assert(conn->_base.type == CONN_TYPE_EXIT);
  tor_assert(conn->_base.state == EXIT_CONN_STATE_RESOLVING);

  strlcpy(search.address, conn->_base.address, sizeof(search.address));

  resolve = HT_FIND(cache_map, &cache_root, &search);
  if (!resolve) {
    log_notice(LD_BUG, "Address %s is not pending. Dropping.",